The GNU C Library


Table of Contents


Introduction

The C language provides no built-in facilities for performing such common operations as input/output, memory management, string manipulation, and the like. Instead, these facilities are defined in a standard library, which you compile and link with your programs.

The GNU C library, described in this document, defines all of the library functions that are specified by the ISO C standard, as well as additional features specific to POSIX and other derivatives of the Unix operating system, and extensions specific to the GNU system.

The purpose of this manual is to tell you how to use the facilities of the GNU library. We have mentioned which features belong to which standards to help you identify things that are potentially non-portable to other systems. But the emphasis in this manual is not on strict portability.

Getting Started

This manual is written with the assumption that you are at least somewhat familiar with the C programming language and basic programming concepts. Specifically, familiarity with ISO standard C (see section ISO C), rather than "traditional" pre-ISO C dialects, is assumed.

The GNU C library includes several header files, each of which provides definitions and declarations for a group of related facilities; this information is used by the C compiler when processing your program. For example, the header file `stdio.h' declares facilities for performing input and output, and the header file `string.h' declares string processing utilities. The organization of this manual generally follows the same division as the header files.

If you are reading this manual for the first time, you should read all of the introductory material and skim the remaining chapters. There are a lot of functions in the GNU C library and it's not realistic to expect that you will be able to remember exactly how to use each and every one of them. It's more important to become generally familiar with the kinds of facilities that the library provides, so that when you are writing your programs you can recognize when to make use of library functions, and where in this manual you can find more specific information about them.

Standards and Portability

This section discusses the various standards and other sources that the GNU C library is based upon. These sources include the ISO C and POSIX standards, and the System V and Berkeley Unix implementations.

The primary focus of this manual is to tell you how to make effective use of the GNU library facilities. But if you are concerned about making your programs compatible with these standards, or portable to operating systems other than GNU, this can affect how you use the library. This section gives you an overview of these standards, so that you will know what they are when they are mentioned in other parts of the manual.

See section Summary of Library Facilities, for an alphabetical list of the functions and other symbols provided by the library. This list also states which standards each function or symbol comes from.

ISO C

The GNU C library is compatible with the C standard adopted by the American National Standards Institute (ANSI): American National Standard X3.159-1989---"ANSI C" and later by the International Standardization Organization (ISO): ISO/IEC 9899:1990, "Programming languages--C". We here refer to the standard as ISO C since this is the more general standard in respect of ratification. The header files and library facilities that make up the GNU library are a superset of those specified by the ISO C standard.

If you are concerned about strict adherence to the ISO C standard, you should use the `-ansi' option when you compile your programs with the GNU C compiler. This tells the compiler to define only ISO standard features from the library header files, unless you explicitly ask for additional features. See section Feature Test Macros, for information on how to do this.

Being able to restrict the library to include only ISO C features is important because ISO C puts limitations on what names can be defined by the library implementation, and the GNU extensions don't fit these limitations. See section Reserved Names, for more information about these restrictions.

This manual does not attempt to give you complete details on the differences between ISO C and older dialects. It gives advice on how to write programs to work portably under multiple C dialects, but does not aim for completeness.

POSIX (The Portable Operating System Interface)

The GNU library is also compatible with the ISO POSIX family of standards, known more formally as the Portable Operating System Interface for Computer Environments (ISO/IEC 9945). They were also published as ANSI/IEEE Std 1003. POSIX is derived mostly from various versions of the Unix operating system.

The library facilities specified by the POSIX standards are a superset of those required by ISO C; POSIX specifies additional features for ISO C functions, as well as specifying new additional functions. In general, the additional requirements and functionality defined by the POSIX standards are aimed at providing lower-level support for a particular kind of operating system environment, rather than general programming language support which can run in many diverse operating system environments.

The GNU C library implements all of the functions specified in ISO/IEC 9945-1:1996, the POSIX System Application Program Interface, commonly referred to as POSIX.1. The primary extensions to the ISO C facilities specified by this standard include file system interface primitives (see section File System Interface), device-specific terminal control functions (see section Low-Level Terminal Interface), and process control functions (see section Processes).

Some facilities from ISO/IEC 9945-2:1993, the POSIX Shell and Utilities standard (POSIX.2) are also implemented in the GNU library. These include utilities for dealing with regular expressions and other pattern matching facilities (see section Pattern Matching).

Berkeley Unix

The GNU C library defines facilities from some versions of Unix which are not formally standardized, specifically from the 4.2 BSD, 4.3 BSD, and 4.4 BSD Unix systems (also known as Berkeley Unix) and from SunOS (a popular 4.2 BSD derivative that includes some Unix System V functionality). These systems support most of the ISO C and POSIX facilities, and 4.4 BSD and newer releases of SunOS in fact support them all.

The BSD facilities include symbolic links (see section Symbolic Links), the select function (see section Waiting for Input or Output), the BSD signal functions (see section BSD Signal Handling), and sockets (see section Sockets).

SVID (The System V Interface Description)

The System V Interface Description (SVID) is a document describing the AT&T Unix System V operating system. It is to some extent a superset of the POSIX standard (see section POSIX (The Portable Operating System Interface)).

The GNU C library defines most of the facilities required by the SVID that are not also required by the ISO C or POSIX standards, for compatibility with System V Unix and other Unix systems (such as SunOS) which include these facilities. However, many of the more obscure and less generally useful facilities required by the SVID are not included. (In fact, Unix System V itself does not provide them all.)

The supported facilities from System V include the methods for inter-process communication and shared memory, the hsearch and drand48 families of functions, fmtmsg and several of the mathematical functions.

XPG (The X/Open Portability Guide)

The X/Open Portability Guide, published by the X/Open Company, Ltd., is a more general standard than POSIX. X/Open owns the Unix copyright and the XPG specifies the requirements for systems which are intended to be a Unix system.

The GNU C library complies to the X/Open Portability Guide, Issue 4.2, with all extensions common to XSI (X/Open System Interface) compliant systems and also all X/Open UNIX extensions.

The additions on top of POSIX are mainly derived from functionality available in System V and BSD systems. Some of the really bad mistakes in System V systems were corrected, though. Since fulfilling the XPG standard with the Unix extensions is a precondition for getting the Unix brand chances are good that the functionality is available on commercial systems.

Using the Library

This section describes some of the practical issues involved in using the GNU C library.

Header Files

Libraries for use by C programs really consist of two parts: header files that define types and macros and declare variables and functions; and the actual library or archive that contains the definitions of the variables and functions.

(Recall that in C, a declaration merely provides information that a function or variable exists and gives its type. For a function declaration, information about the types of its arguments might be provided as well. The purpose of declarations is to allow the compiler to correctly process references to the declared variables and functions. A definition, on the other hand, actually allocates storage for a variable or says what a function does.)

In order to use the facilities in the GNU C library, you should be sure that your program source files include the appropriate header files. This is so that the compiler has declarations of these facilities available and can correctly process references to them. Once your program has been compiled, the linker resolves these references to the actual definitions provided in the archive file.

Header files are included into a program source file by the `#include' preprocessor directive. The C language supports two forms of this directive; the first,

#include "header"

is typically used to include a header file header that you write yourself; this would contain definitions and declarations describing the interfaces between the different parts of your particular application. By contrast,

#include <file.h>

is typically used to include a header file `file.h' that contains definitions and declarations for a standard library. This file would normally be installed in a standard place by your system administrator. You should use this second form for the C library header files.

Typically, `#include' directives are placed at the top of the C source file, before any other code. If you begin your source files with some comments explaining what the code in the file does (a good idea), put the `#include' directives immediately afterwards, following the feature test macro definition (see section Feature Test Macros).

For more information about the use of header files and `#include' directives, see section `Header Files' in The GNU C Preprocessor Manual.

The GNU C library provides several header files, each of which contains the type and macro definitions and variable and function declarations for a group of related facilities. This means that your programs may need to include several header files, depending on exactly which facilities you are using.

Some library header files include other library header files automatically. However, as a matter of programming style, you should not rely on this; it is better to explicitly include all the header files required for the library facilities you are using. The GNU C library header files have been written in such a way that it doesn't matter if a header file is accidentally included more than once; including a header file a second time has no effect. Likewise, if your program needs to include multiple header files, the order in which they are included doesn't matter.

Compatibility Note: Inclusion of standard header files in any order and any number of times works in any ISO C implementation. However, this has traditionally not been the case in many older C implementations.

Strictly speaking, you don't have to include a header file to use a function it declares; you could declare the function explicitly yourself, according to the specifications in this manual. But it is usually better to include the header file because it may define types and macros that are not otherwise available and because it may define more efficient macro replacements for some functions. It is also a sure way to have the correct declaration.

Macro Definitions of Functions

If we describe something as a function in this manual, it may have a macro definition as well. This normally has no effect on how your program runs--the macro definition does the same thing as the function would. In particular, macro equivalents for library functions evaluate arguments exactly once, in the same way that a function call would. The main reason for these macro definitions is that sometimes they can produce an inline expansion that is considerably faster than an actual function call.

Taking the address of a library function works even if it is also defined as a macro. This is because, in this context, the name of the function isn't followed by the left parenthesis that is syntactically necessary to recognize a macro call.

You might occasionally want to avoid using the macro definition of a function--perhaps to make your program easier to debug. There are two ways you can do this:

For example, suppose the header file `stdlib.h' declares a function named abs with

extern int abs (int);

and also provides a macro definition for abs. Then, in:

#include <stdlib.h>
int f (int *i) { return abs (++*i); }

the reference to abs might refer to either a macro or a function. On the other hand, in each of the following examples the reference is to a function and not a macro.

#include <stdlib.h>
int g (int *i) { return (abs) (++*i); }

#undef abs
int h (int *i) { return abs (++*i); }

Since macro definitions that double for a function behave in exactly the same way as the actual function version, there is usually no need for any of these methods. In fact, removing macro definitions usually just makes your program slower.

Reserved Names

The names of all library types, macros, variables and functions that come from the ISO C standard are reserved unconditionally; your program may not redefine these names. All other library names are reserved if your program explicitly includes the header file that defines or declares them. There are several reasons for these restrictions:

In addition to the names documented in this manual, reserved names include all external identifiers (global functions and variables) that begin with an underscore (`_') and all identifiers regardless of use that begin with either two underscores or an underscore followed by a capital letter are reserved names. This is so that the library and header files can define functions, variables, and macros for internal purposes without risk of conflict with names in user programs.

Some additional classes of identifier names are reserved for future extensions to the C language or the POSIX.1 environment. While using these names for your own purposes right now might not cause a problem, they do raise the possibility of conflict with future versions of the C or POSIX standards, so you should avoid these names.

In addition, some individual header files reserve names beyond those that they actually define. You only need to worry about these restrictions if your program includes that particular header file.

Feature Test Macros

The exact set of features available when you compile a source file is controlled by which feature test macros you define.

If you compile your programs using `gcc -ansi', you get only the ISO C library features, unless you explicitly request additional features by defining one or more of the feature macros. See section `GNU CC Command Options' in The GNU CC Manual, for more information about GCC options.

You should define these macros by using `#define' preprocessor directives at the top of your source code files. These directives must come before any #include of a system header file. It is best to make them the very first thing in the file, preceded only by comments. You could also use the `-D' option to GCC, but it's better if you make the source files indicate their own meaning in a self-contained way.

This system exists to allow the library to conform to multiple standards. Although the different standards are often described as supersets of each other, they are usually incompatible because larger standards require functions with names that smaller ones reserve to the user program. This is not mere pedantry -- it has been a problem in practice. For instance, some non-GNU programs define functions named getline that have nothing to do with this library's getline. They would not be compilable if all features were enabled indiscriminately.

This should not be used to verify that a program conforms to a limited standard. It is insufficient for this purpose, as it will not protect you from including header files outside the standard, or relying on semantics undefined within the standard.

Macro: _POSIX_SOURCE
If you define this macro, then the functionality from the POSIX.1 standard (IEEE Standard 1003.1) is available, as well as all of the ISO C facilities.

The state of _POSIX_SOURCE is irrelevant if you define the macro _POSIX_C_SOURCE to a positive integer.

Macro: _POSIX_C_SOURCE
Define this macro to a positive integer to control which POSIX functionality is made available. The greater the value of this macro, the more functionality is made available.

If you define this macro to a value greater than or equal to 1, then the functionality from the 1990 edition of the POSIX.1 standard (IEEE Standard 1003.1-1990) is made available.

If you define this macro to a value greater than or equal to 2, then the functionality from the 1992 edition of the POSIX.2 standard (IEEE Standard 1003.2-1992) is made available.

If you define this macro to a value greater than or equal to 199309L, then the functionality from the 1993 edition of the POSIX.1b standard (IEEE Standard 1003.1b-1993) is made available.

Greater values for _POSIX_C_SOURCE will enable future extensions. The POSIX standards process will define these values as necessary, and the GNU C Library should support them some time after they become standardized. The 1996 edition of POSIX.1 (ISO/IEC 9945-1: 1996) states that if you define _POSIX_C_SOURCE to a value greater than or equal to 199506L, then the functionality from the 1996 edition is made available.

Macro: _BSD_SOURCE
If you define this macro, functionality derived from 4.3 BSD Unix is included as well as the ISO C, POSIX.1, and POSIX.2 material.

Some of the features derived from 4.3 BSD Unix conflict with the corresponding features specified by the POSIX.1 standard. If this macro is defined, the 4.3 BSD definitions take precedence over the POSIX definitions.

Due to the nature of some of the conflicts between 4.3 BSD and POSIX.1, you need to use a special BSD compatibility library when linking programs compiled for BSD compatibility. This is because some functions must be defined in two different ways, one of them in the normal C library, and one of them in the compatibility library. If your program defines _BSD_SOURCE, you must give the option `-lbsd-compat' to the compiler or linker when linking the program, to tell it to find functions in this special compatibility library before looking for them in the normal C library.

Macro: _SVID_SOURCE
If you define this macro, functionality derived from SVID is included as well as the ISO C, POSIX.1, POSIX.2, and X/Open material.

Macro: _XOPEN_SOURCE
Macro: _XOPEN_SOURCE_EXTENDED
If you define this macro, functionality described in the X/Open Portability Guide is included. This is a superset of the POSIX.1 and POSIX.2 functionality and in fact _POSIX_SOURCE and _POSIX_C_SOURCE are automatically defined.

As the unification of all Unices, functionality only available in BSD and SVID is also included.

If the macro _XOPEN_SOURCE_EXTENDED is also defined, even more functionality is available. The extra functions will make all functions available which are necessary for the X/Open Unix brand.

If the macro _XOPEN_SOURCE has the value @math{500} this includes all functionality described so far plus some new definitions from the Single Unix Specification, version 2.

Macro: _LARGEFILE_SOURCE
If this macro is defined some extra functions are available which rectify a few shortcomings in all previous standards. Specifically, the functions fseeko and ftello are available. Without these functions the difference between the ISO C interface (fseek, ftell) and the low-level POSIX interface (lseek) would lead to problems.

This macro was introduced as part of the Large File Support extension (LFS).

Macro: _LARGEFILE64_SOURCE
If you define this macro an additional set of functions is made available which enables 32 bit systems to use files of sizes beyond the usual limit of 2GB. This interface is not available if the system does not support files that large. On systems where the natural file size limit is greater than 2GB (i.e., on 64 bit systems) the new functions are identical to the replaced functions.

The new functionality is made available by a new set of types and functions which replace the existing ones. The names of these new objects contain 64 to indicate the intention, e.g., off_t vs. off64_t and fseeko vs. fseeko64.

This macro was introduced as part of the Large File Support extension (LFS). It is a transition interface for the period when 64 bit offsets are not generally used (see _FILE_OFFSET_BITS).

Macro: _FILE_OFFSET_BITS
This macro determines which file system interface shall be used, one replacing the other. Whereas _LARGEFILE64_SOURCE makes the 64 bit interface available as an additional interface, _FILE_OFFSET_BITS allows the 64 bit interface to replace the old interface.

If _FILE_OFFSET_BITS is undefined, or if it is defined to the value 32, nothing changes. The 32 bit interface is used and types like off_t have a size of 32 bits on 32 bit systems.

If the macro is defined to the value 64, the large file interface replaces the old interface. I.e., the functions are not made available under different names (as they are with _LARGEFILE64_SOURCE). Instead the old function names now reference the new functions, e.g., a call to fseeko now indeed calls fseeko64.

This macro should only be selected if the system provides mechanisms for handling large files. On 64 bit systems this macro has no effect since the *64 functions are identical to the normal functions.

This macro was introduced as part of the Large File Support extension (LFS).

Macro: _ISOC99_SOURCE
Until the revised ISO C standard is widely adopted the new features are not automatically enabled. The GNU libc nevertheless has a complete implementation of the new standard and to enable the new features the macro _ISOC99_SOURCE should be defined.

Macro: _GNU_SOURCE
If you define this macro, everything is included: ISO C89, ISO C99, POSIX.1, POSIX.2, BSD, SVID, X/Open, LFS, and GNU extensions. In the cases where POSIX.1 conflicts with BSD, the POSIX definitions take precedence.

If you want to get the full effect of _GNU_SOURCE but make the BSD definitions take precedence over the POSIX definitions, use this sequence of definitions:

#define _GNU_SOURCE
#define _BSD_SOURCE
#define _SVID_SOURCE

Note that if you do this, you must link your program with the BSD compatibility library by passing the `-lbsd-compat' option to the compiler or linker. Note: If you forget to do this, you may get very strange errors at run time.

Macro: _REENTRANT
Macro: _THREAD_SAFE
If you define one of these macros, reentrant versions of several functions get declared. Some of the functions are specified in POSIX.1c but many others are only available on a few other systems or are unique to GNU libc. The problem is the delay in the standardization of the thread safe C library interface.

Unlike on some other systems, no special version of the C library must be used for linking. There is only one version but while compiling this it must have been specified to compile as thread safe.

We recommend you use _GNU_SOURCE in new programs. If you don't specify the `-ansi' option to GCC and don't define any of these macros explicitly, the effect is the same as defining _POSIX_C_SOURCE to 2 and _POSIX_SOURCE, _SVID_SOURCE, and _BSD_SOURCE to 1.

When you define a feature test macro to request a larger class of features, it is harmless to define in addition a feature test macro for a subset of those features. For example, if you define _POSIX_C_SOURCE, then defining _POSIX_SOURCE as well has no effect. Likewise, if you define _GNU_SOURCE, then defining either _POSIX_SOURCE or _POSIX_C_SOURCE or _SVID_SOURCE as well has no effect.

Note, however, that the features of _BSD_SOURCE are not a subset of any of the other feature test macros supported. This is because it defines BSD features that take precedence over the POSIX features that are requested by the other macros. For this reason, defining _BSD_SOURCE in addition to the other feature test macros does have an effect: it causes the BSD features to take priority over the conflicting POSIX features.

Roadmap to the Manual

Here is an overview of the contents of the remaining chapters of this manual.

If you already know the name of the facility you are interested in, you can look it up in section Summary of Library Facilities. This gives you a summary of its syntax and a pointer to where you can find a more detailed description. This appendix is particularly useful if you just want to verify the order and type of arguments to a function, for example. It also tells you what standard or system each function, variable, or macro is derived from.

Error Reporting

Many functions in the GNU C library detect and report error conditions, and sometimes your programs need to check for these error conditions. For example, when you open an input file, you should verify that the file was actually opened correctly, and print an error message or take other appropriate action if the call to the library function failed.

This chapter describes how the error reporting facility works. Your program should include the header file `errno.h' to use this facility.

Checking for Errors

Most library functions return a special value to indicate that they have failed. The special value is typically -1, a null pointer, or a constant such as EOF that is defined for that purpose. But this return value tells you only that an error has occurred. To find out what kind of error it was, you need to look at the error code stored in the variable errno. This variable is declared in the header file `errno.h'.

Variable: volatile int errno
The variable errno contains the system error number. You can change the value of errno.

Since errno is declared volatile, it might be changed asynchronously by a signal handler; see section Defining Signal Handlers. However, a properly written signal handler saves and restores the value of errno, so you generally do not need to worry about this possibility except when writing signal handlers.

The initial value of errno at program startup is zero. Many library functions are guaranteed to set it to certain nonzero values when they encounter certain kinds of errors. These error conditions are listed for each function. These functions do not change errno when they succeed; thus, the value of errno after a successful call is not necessarily zero, and you should not use errno to determine whether a call failed. The proper way to do that is documented for each function. If the call failed, you can examine errno.

Many library functions can set errno to a nonzero value as a result of calling other library functions which might fail. You should assume that any library function might alter errno when the function returns an error.

Portability Note: ISO C specifies errno as a "modifiable lvalue" rather than as a variable, permitting it to be implemented as a macro. For example, its expansion might involve a function call, like *_errno (). In fact, that is what it is on the GNU system itself. The GNU library, on non-GNU systems, does whatever is right for the particular system.

There are a few library functions, like sqrt and atan, that return a perfectly legitimate value in case of an error, but also set errno. For these functions, if you want to check to see whether an error occurred, the recommended method is to set errno to zero before calling the function, and then check its value afterward.

All the error codes have symbolic names; they are macros defined in `errno.h'. The names start with `E' and an upper-case letter or digit; you should consider names of this form to be reserved names. See section Reserved Names.

The error code values are all positive integers and are all distinct, with one exception: EWOULDBLOCK and EAGAIN are the same. Since the values are distinct, you can use them as labels in a switch statement; just don't use both EWOULDBLOCK and EAGAIN. Your program should not make any other assumptions about the specific values of these symbolic constants.

The value of errno doesn't necessarily have to correspond to any of these macros, since some library functions might return other error codes of their own for other situations. The only values that are guaranteed to be meaningful for a particular library function are the ones that this manual lists for that function.

On non-GNU systems, almost any system call can return EFAULT if it is given an invalid pointer as an argument. Since this could only happen as a result of a bug in your program, and since it will not happen on the GNU system, we have saved space by not mentioning EFAULT in the descriptions of individual functions.

In some Unix systems, many system calls can also return EFAULT if given as an argument a pointer into the stack, and the kernel for some obscure reason fails in its attempt to extend the stack. If this ever happens, you should probably try using statically or dynamically allocated memory instead of stack memory on that system.

Error Codes

The error code macros are defined in the header file `errno.h'. All of them expand into integer constant values. Some of these error codes can't occur on the GNU system, but they can occur using the GNU library on other systems.

Macro: int EPERM
Operation not permitted; only the owner of the file (or other resource) or processes with special privileges can perform the operation.

Macro: int ENOENT
No such file or directory. This is a "file doesn't exist" error for ordinary files that are referenced in contexts where they are expected to already exist.

Macro: int ESRCH
No process matches the specified process ID.

Macro: int EINTR
Interrupted function call; an asynchronous signal occurred and prevented completion of the call. When this happens, you should try the call again.

You can choose to have functions resume after a signal that is handled, rather than failing with EINTR; see section Primitives Interrupted by Signals.

Macro: int EIO
Input/output error; usually used for physical read or write errors.

Macro: int ENXIO
No such device or address. The system tried to use the device represented by a file you specified, and it couldn't find the device. This can mean that the device file was installed incorrectly, or that the physical device is missing or not correctly attached to the computer.

Macro: int E2BIG
Argument list too long; used when the arguments passed to a new program being executed with one of the exec functions (see section Executing a File) occupy too much memory space. This condition never arises in the GNU system.

Macro: int ENOEXEC
Invalid executable file format. This condition is detected by the exec functions; see section Executing a File.

Macro: int EBADF
Bad file descriptor; for example, I/O on a descriptor that has been closed or reading from a descriptor open only for writing (or vice versa).

Macro: int ECHILD
There are no child processes. This error happens on operations that are supposed to manipulate child processes, when there aren't any processes to manipulate.

Macro: int EDEADLK
Deadlock avoided; allocating a system resource would have resulted in a deadlock situation. The system does not guarantee that it will notice all such situations. This error means you got lucky and the system noticed; it might just hang. See section File Locks, for an example.

Macro: int ENOMEM
No memory available. The system cannot allocate more virtual memory because its capacity is full.

Macro: int EACCES
Permission denied; the file permissions do not allow the attempted operation.

Macro: int EFAULT
Bad address; an invalid pointer was detected. In the GNU system, this error never happens; you get a signal instead.

Macro: int ENOTBLK
A file that isn't a block special file was given in a situation that requires one. For example, trying to mount an ordinary file as a file system in Unix gives this error.

Macro: int EBUSY
Resource busy; a system resource that can't be shared is already in use. For example, if you try to delete a file that is the root of a currently mounted filesystem, you get this error.

Macro: int EEXIST
File exists; an existing file was specified in a context where it only makes sense to specify a new file.

Macro: int EXDEV
An attempt to make an improper link across file systems was detected. This happens not only when you use link (see section Hard Links) but also when you rename a file with rename (see section Renaming Files).

Macro: int ENODEV
The wrong type of device was given to a function that expects a particular sort of device.

Macro: int ENOTDIR
A file that isn't a directory was specified when a directory is required.

Macro: int EISDIR
File is a directory; you cannot open a directory for writing, or create or remove hard links to it.

Macro: int EINVAL
Invalid argument. This is used to indicate various kinds of problems with passing the wrong argument to a library function.

Macro: int EMFILE
The current process has too many files open and can't open any more. Duplicate descriptors do count toward this limit.

In BSD and GNU, the number of open files is controlled by a resource limit that can usually be increased. If you get this error, you might want to increase the RLIMIT_NOFILE limit or make it unlimited; see section Limiting Resource Usage.

Macro: int ENFILE
There are too many distinct file openings in the entire system. Note that any number of linked channels count as just one file opening; see section Linked Channels. This error never occurs in the GNU system.

Macro: int ENOTTY
Inappropriate I/O control operation, such as trying to set terminal modes on an ordinary file.

Macro: int ETXTBSY
An attempt to execute a file that is currently open for writing, or write to a file that is currently being executed. Often using a debugger to run a program is considered having it open for writing and will cause this error. (The name stands for "text file busy".) This is not an error in the GNU system; the text is copied as necessary.

Macro: int EFBIG
File too big; the size of a file would be larger than allowed by the system.

Macro: int ENOSPC
No space left on device; write operation on a file failed because the disk is full.

Macro: int ESPIPE
Invalid seek operation (such as on a pipe).

Macro: int EROFS
An attempt was made to modify something on a read-only file system.

Macro: int EMLINK
Too many links; the link count of a single file would become too large. rename can cause this error if the file being renamed already has as many links as it can take (see section Renaming Files).

Macro: int EPIPE
Broken pipe; there is no process reading from the other end of a pipe. Every library function that returns this error code also generates a SIGPIPE signal; this signal terminates the program if not handled or blocked. Thus, your program will never actually see EPIPE unless it has handled or blocked SIGPIPE.

Macro: int EDOM
Domain error; used by mathematical functions when an argument value does not fall into the domain over which the function is defined.

Macro: int ERANGE
Range error; used by mathematical functions when the result value is not representable because of overflow or underflow.

Macro: int EAGAIN
Resource temporarily unavailable; the call might work if you try again later. The macro EWOULDBLOCK is another name for EAGAIN; they are always the same in the GNU C library.

This error can happen in a few different situations:

Macro: int EWOULDBLOCK
In the GNU C library, this is another name for EAGAIN (above). The values are always the same, on every operating system.

C libraries in many older Unix systems have EWOULDBLOCK as a separate error code.

Macro: int EINPROGRESS
An operation that cannot complete immediately was initiated on an object that has non-blocking mode selected. Some functions that must always block (such as connect; see section Making a Connection) never return EAGAIN. Instead, they return EINPROGRESS to indicate that the operation has begun and will take some time. Attempts to manipulate the object before the call completes return EALREADY. You can use the select function to find out when the pending operation has completed; see section Waiting for Input or Output.

Macro: int EALREADY
An operation is already in progress on an object that has non-blocking mode selected.

Macro: int ENOTSOCK
A file that isn't a socket was specified when a socket is required.

Macro: int EMSGSIZE
The size of a message sent on a socket was larger than the supported maximum size.

Macro: int EPROTOTYPE
The socket type does not support the requested communications protocol.

Macro: int ENOPROTOOPT
You specified a socket option that doesn't make sense for the particular protocol being used by the socket. See section Socket Options.

Macro: int EPROTONOSUPPORT
The socket domain does not support the requested communications protocol (perhaps because the requested protocol is completely invalid). See section Creating a Socket.

Macro: int ESOCKTNOSUPPORT
The socket type is not supported.

Macro: int EOPNOTSUPP
The operation you requested is not supported. Some socket functions don't make sense for all types of sockets, and others may not be implemented for all communications protocols. In the GNU system, this error can happen for many calls when the object does not support the particular operation; it is a generic indication that the server knows nothing to do for that call.

Macro: int EPFNOSUPPORT
The socket communications protocol family you requested is not supported.

Macro: int EAFNOSUPPORT
The address family specified for a socket is not supported; it is inconsistent with the protocol being used on the socket. See section Sockets.

Macro: int EADDRINUSE
The requested socket address is already in use. See section Socket Addresses.

Macro: int EADDRNOTAVAIL
The requested socket address is not available; for example, you tried to give a socket a name that doesn't match the local host name. See section Socket Addresses.

Macro: int ENETDOWN
A socket operation failed because the network was down.

Macro: int ENETUNREACH
A socket operation failed because the subnet containing the remote host was unreachable.

Macro: int ENETRESET
A network connection was reset because the remote host crashed.

Macro: int ECONNABORTED
A network connection was aborted locally.

Macro: int ECONNRESET
A network connection was closed for reasons outside the control of the local host, such as by the remote machine rebooting or an unrecoverable protocol violation.

Macro: int ENOBUFS
The kernel's buffers for I/O operations are all in use. In GNU, this error is always synonymous with ENOMEM; you may get one or the other from network operations.

Macro: int EISCONN
You tried to connect a socket that is already connected. See section Making a Connection.

Macro: int ENOTCONN
The socket is not connected to anything. You get this error when you try to transmit data over a socket, without first specifying a destination for the data. For a connectionless socket (for datagram protocols, such as UDP), you get EDESTADDRREQ instead.

Macro: int EDESTADDRREQ
No default destination address was set for the socket. You get this error when you try to transmit data over a connectionless socket, without first specifying a destination for the data with connect.

Macro: int ESHUTDOWN
The socket has already been shut down.

Macro: int ETOOMANYREFS
???

Macro: int ETIMEDOUT
A socket operation with a specified timeout received no response during the timeout period.

Macro: int ECONNREFUSED
A remote host refused to allow the network connection (typically because it is not running the requested service).

Macro: int ELOOP
Too many levels of symbolic links were encountered in looking up a file name. This often indicates a cycle of symbolic links.

Macro: int ENAMETOOLONG
Filename too long (longer than PATH_MAX; see section Limits on File System Capacity) or host name too long (in gethostname or sethostname; see section Host Identification).

Macro: int EHOSTDOWN
The remote host for a requested network connection is down.

Macro: int EHOSTUNREACH
The remote host for a requested network connection is not reachable.

Macro: int ENOTEMPTY
Directory not empty, where an empty directory was expected. Typically, this error occurs when you are trying to delete a directory.

Macro: int EPROCLIM
This means that the per-user limit on new process would be exceeded by an attempted fork. See section Limiting Resource Usage, for details on the RLIMIT_NPROC limit.

Macro: int EUSERS
The file quota system is confused because there are too many users.

Macro: int EDQUOT
The user's disk quota was exceeded.

Macro: int ESTALE
Stale NFS file handle. This indicates an internal confusion in the NFS system which is due to file system rearrangements on the server host. Repairing this condition usually requires unmounting and remounting the NFS file system on the local host.

Macro: int EREMOTE
An attempt was made to NFS-mount a remote file system with a file name that already specifies an NFS-mounted file. (This is an error on some operating systems, but we expect it to work properly on the GNU system, making this error code impossible.)

Macro: int EBADRPC
???

Macro: int ERPCMISMATCH
???

Macro: int EPROGUNAVAIL
???

Macro: int EPROGMISMATCH
???

Macro: int EPROCUNAVAIL
???

Macro: int ENOLCK
No locks available. This is used by the file locking facilities; see section File Locks. This error is never generated by the GNU system, but it can result from an operation to an NFS server running another operating system.

Macro: int EFTYPE
Inappropriate file type or format. The file was the wrong type for the operation, or a data file had the wrong format.

On some systems chmod returns this error if you try to set the sticky bit on a non-directory file; see section Assigning File Permissions.

Macro: int EAUTH
???

Macro: int ENEEDAUTH
???

Macro: int ENOSYS
Function not implemented. This indicates that the function called is not implemented at all, either in the C library itself or in the operating system. When you get this error, you can be sure that this particular function will always fail with ENOSYS unless you install a new version of the C library or the operating system.

Macro: int ENOTSUP
Not supported. A function returns this error when certain parameter values are valid, but the functionality they request is not available. This can mean that the function does not implement a particular command or option value or flag bit at all. For functions that operate on some object given in a parameter, such as a file descriptor or a port, it might instead mean that only that specific object (file descriptor, port, etc.) is unable to support the other parameters given; different file descriptors might support different ranges of parameter values.

If the entire function is not available at all in the implementation, it returns ENOSYS instead.

Macro: int EILSEQ
While decoding a multibyte character the function came along an invalid or an incomplete sequence of bytes or the given wide character is invalid.

Macro: int EBACKGROUND
In the GNU system, servers supporting the term protocol return this error for certain operations when the caller is not in the foreground process group of the terminal. Users do not usually see this error because functions such as read and write translate it into a SIGTTIN or SIGTTOU signal. See section Job Control, for information on process groups and these signals.

Macro: int EDIED
In the GNU system, opening a file returns this error when the file is translated by a program and the translator program dies while starting up, before it has connected to the file.

Macro: int ED
The experienced user will know what is wrong.

Macro: int EGREGIOUS
You did what?

Macro: int EIEIO
Go home and have a glass of warm, dairy-fresh milk.

Macro: int EGRATUITOUS
This error code has no purpose.

Macro: int EBADMSG

Macro: int EIDRM

Macro: int EMULTIHOP

Macro: int ENODATA

Macro: int ENOLINK

Macro: int ENOMSG

Macro: int ENOSR

Macro: int ENOSTR

Macro: int EOVERFLOW

Macro: int EPROTO

Macro: int ETIME

The following error codes are defined by the Linux/i386 kernel. They are not yet documented.

Macro: int ERESTART

Macro: int ECHRNG

Macro: int EL2NSYNC

Macro: int EL3HLT

Macro: int EL3RST

Macro: int ELNRNG

Macro: int EUNATCH

Macro: int ENOCSI

Macro: int EL2HLT

Macro: int EBADE

Macro: int EBADR

Macro: int EXFULL

Macro: int ENOANO

Macro: int EBADRQC

Macro: int EBADSLT

Macro: int EDEADLOCK

Macro: int EBFONT

Macro: int ENONET

Macro: int ENOPKG

Macro: int EADV

Macro: int ESRMNT

Macro: int ECOMM

Macro: int EDOTDOT

Macro: int ENOTUNIQ

Macro: int EBADFD

Macro: int EREMCHG

Macro: int ELIBACC

Macro: int ELIBBAD

Macro: int ELIBSCN

Macro: int ELIBMAX

Macro: int ELIBEXEC

Macro: int ESTRPIPE

Macro: int EUCLEAN

Macro: int ENOTNAM

Macro: int ENAVAIL

Macro: int EISNAM

Macro: int EREMOTEIO

Macro: int ENOMEDIUM

Macro: int EMEDIUMTYPE

Error Messages

The library has functions and variables designed to make it easy for your program to report informative error messages in the customary format about the failure of a library call. The functions strerror and perror give you the standard error message for a given error code; the variable program_invocation_short_name gives you convenient access to the name of the program that encountered the error.

Function: char * strerror (int errnum)
The strerror function maps the error code (see section Checking for Errors) specified by the errnum argument to a descriptive error message string. The return value is a pointer to this string.

The value errnum normally comes from the variable errno.

You should not modify the string returned by strerror. Also, if you make subsequent calls to strerror, the string might be overwritten. (But it's guaranteed that no library function ever calls strerror behind your back.)

The function strerror is declared in `string.h'.

Function: char * strerror_r (int errnum, char *buf, size_t n)
The strerror_r function works like strerror but instead of returning the error message in a statically allocated buffer shared by all threads in the process, it returns a private copy for the thread. This might be either some permanent global data or a message string in the user supplied buffer starting at buf with the length of n bytes.

At most n characters are written (including the NUL byte) so it is up to the user to select the buffer large enough.

This function should always be used in multi-threaded programs since there is no way to guarantee the string returned by strerror really belongs to the last call of the current thread.

This function strerror_r is a GNU extension and it is declared in `string.h'.

Function: void perror (const char *message)
This function prints an error message to the stream stderr; see section Standard Streams.

If you call perror with a message that is either a null pointer or an empty string, perror just prints the error message corresponding to errno, adding a trailing newline.

If you supply a non-null message argument, then perror prefixes its output with this string. It adds a colon and a space character to separate the message from the error string corresponding to errno.

The function perror is declared in `stdio.h'.

strerror and perror produce the exact same message for any given error code; the precise text varies from system to system. On the GNU system, the messages are fairly short; there are no multi-line messages or embedded newlines. Each error message begins with a capital letter and does not include any terminating punctuation.

Compatibility Note: The strerror function is a new feature of ISO C. Many older C systems do not support this function yet.

Many programs that don't read input from the terminal are designed to exit if any system call fails. By convention, the error message from such a program should start with the program's name, sans directories. You can find that name in the variable program_invocation_short_name; the full file name is stored the variable program_invocation_name.

Variable: char * program_invocation_name
This variable's value is the name that was used to invoke the program running in the current process. It is the same as argv[0]. Note that this is not necessarily a useful file name; often it contains no directory names. See section Program Arguments.

Variable: char * program_invocation_short_name
This variable's value is the name that was used to invoke the program running in the current process, with directory names removed. (That is to say, it is the same as program_invocation_name minus everything up to the last slash, if any.)

The library initialization code sets up both of these variables before calling main.

Portability Note: These two variables are GNU extensions. If you want your program to work with non-GNU libraries, you must save the value of argv[0] in main, and then strip off the directory names yourself. We added these extensions to make it possible to write self-contained error-reporting subroutines that require no explicit cooperation from main.

Here is an example showing how to handle failure to open a file correctly. The function open_sesame tries to open the named file for reading and returns a stream if successful. The fopen library function returns a null pointer if it couldn't open the file for some reason. In that situation, open_sesame constructs an appropriate error message using the strerror function, and terminates the program. If we were going to make some other library calls before passing the error code to strerror, we'd have to save it in a local variable instead, because those other library functions might overwrite errno in the meantime.

#include <errno.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

FILE *
open_sesame (char *name)
{
  FILE *stream;

  errno = 0;
  stream = fopen (name, "r");
  if (stream == NULL)
    {
      fprintf (stderr, "%s: Couldn't open file %s; %s\n",
               program_invocation_short_name, name, strerror (errno));
      exit (EXIT_FAILURE);
    }
  else
    return stream;
}

Virtual Memory Allocation And Paging

This chapter describes how processes manage and use memory in a system that uses the GNU C library.

The GNU C Library has several functions for dynamically allocating virtual memory in various ways. They vary in generality and in efficiency. The library also provides functions for controlling paging and allocation of real memory.

Memory mapped I/O is not discussed in this chapter. See section Memory-mapped I/O.

Process Memory Concepts

One of the most basic resources a process has available to it is memory. There are a lot of different ways systems organize memory, but in a typical one, each process has one linear virtual address space, with addresses running from zero to some huge maximum. It need not be contiguous; i.e. not all of these addresses actually can be used to store data.

The virtual memory is divided into pages (4 kilobytes is typical). Backing each page of virtual memory is a page of real memory (called a frame) or some secondary storage, usually disk space. The disk space might be swap space or just some ordinary disk file. Actually, a page of all zeroes sometimes has nothing at all backing it -- there's just a flag saying it is all zeroes.

The same frame of real memory or backing store can back multiple virtual pages belonging to multiple processes. This is normally the case, for example, with virtual memory occupied by GNU C library code. The same real memory frame containing the printf function backs a virtual memory page in each of the existing processes that has a printf call in its program.

In order for a program to access any part of a virtual page, the page must at that moment be backed by ("connected to") a real frame. But because there is usually a lot more virtual memory than real memory, the pages must move back and forth between real memory and backing store regularly, coming into real memory when a process needs to access them and then retreating to backing store when not needed anymore. This movement is called paging.

When a program attempts to access a page which is not at that moment backed by real memory, this is known as a page fault. When a page fault occurs, the kernel suspends the process, places the page into a real page frame (this is called "paging in" or "faulting in"), then resumes the process so that from the process' point of view, the page was in real memory all along. In fact, to the process, all pages always seem to be in real memory. Except for one thing: the elapsed execution time of an instruction that would normally be a few nanoseconds is suddenly much, much, longer (because the kernel normally has to do I/O to complete the page-in). For programs sensitive to that, the functions described in section Locking Pages can control it.

Within each virtual address space, a process has to keep track of what is at which addresses, and that process is called memory allocation. Allocation usually brings to mind meting out scarce resources, but in the case of virtual memory, that's not a major goal, because there is generally much more of it than anyone needs. Memory allocation within a process is mainly just a matter of making sure that the same byte of memory isn't used to store two different things.

Processes allocate memory in two major ways: by exec and programmatically. Actually, forking is a third way, but it's not very interesting. See section Creating a Process.

Exec is the operation of creating a virtual address space for a process, loading its basic program into it, and executing the program. It is done by the "exec" family of functions (e.g. execl). The operation takes a program file (an executable), it allocates space to load all the data in the executable, loads it, and transfers control to it. That data is most notably the instructions of the program (the text), but also literals and constants in the program and even some variables: C variables with the static storage class (see section Memory Allocation in C Programs).

Once that program begins to execute, it uses programmatic allocation to gain additional memory. In a C program with the GNU C library, there are two kinds of programmatic allocation: automatic and dynamic. See section Memory Allocation in C Programs.

Memory-mapped I/O is another form of dynamic virtual memory allocation. Mapping memory to a file means declaring that the contents of certain range of a process' addresses shall be identical to the contents of a specified regular file. The system makes the virtual memory initially contain the contents of the file, and if you modify the memory, the system writes the same modification to the file. Note that due to the magic of virtual memory and page faults, there is no reason for the system to do I/O to read the file, or allocate real memory for its contents, until the program accesses the virtual memory. See section Memory-mapped I/O.

Just as it programmatically allocates memory, the program can programmatically deallocate (free) it. You can't free the memory that was allocated by exec. When the program exits or execs, you might say that all its memory gets freed, but since in both cases the address space ceases to exist, the point is really moot. See section Program Termination.

A process' virtual address space is divided into segments. A segment is a contiguous range of virtual addresses. Three important segments are:

Allocating Storage For Program Data

This section covers how ordinary programs manage storage for their data, including the famous malloc function and some fancier facilities special the GNU C library and GNU Compiler.

Memory Allocation in C Programs

The C language supports two kinds of memory allocation through the variables in C programs:

A third important kind of memory allocation, dynamic allocation, is not supported by C variables but is available via GNU C library functions.

Dynamic Memory Allocation

Dynamic memory allocation is a technique in which programs determine as they are running where to store some information. You need dynamic allocation when the amount of memory you need, or how long you continue to need it, depends on factors that are not known before the program runs.

For example, you may need a block to store a line read from an input file; since there is no limit to how long a line can be, you must allocate the memory dynamically and make it dynamically larger as you read more of the line.

Or, you may need a block for each record or each definition in the input data; since you can't know in advance how many there will be, you must allocate a new block for each record or definition as you read it.

When you use dynamic allocation, the allocation of a block of memory is an action that the program requests explicitly. You call a function or macro when you want to allocate space, and specify the size with an argument. If you want to free the space, you do so by calling another function or macro. You can do these things whenever you want, as often as you want.

Dynamic allocation is not supported by C variables; there is no storage class "dynamic", and there can never be a C variable whose value is stored in dynamically allocated space. The only way to get dynamically allocated memory is via a system call (which is generally via a GNU C library function call), and the only way to refer to dynamically allocated space is through a pointer. Because it is less convenient, and because the actual process of dynamic allocation requires more computation time, programmers generally use dynamic allocation only when neither static nor automatic allocation will serve.

For example, if you want to allocate dynamically some space to hold a struct foobar, you cannot declare a variable of type struct foobar whose contents are the dynamically allocated space. But you can declare a variable of pointer type struct foobar * and assign it the address of the space. Then you can use the operators `*' and `->' on this pointer variable to refer to the contents of the space:

{
  struct foobar *ptr
     = (struct foobar *) malloc (sizeof (struct foobar));
  ptr->name = x;
  ptr->next = current_foobar;
  current_foobar = ptr;
}

Unconstrained Allocation

The most general dynamic allocation facility is malloc. It allows you to allocate blocks of memory of any size at any time, make them bigger or smaller at any time, and free the blocks individually at any time (or never).

Basic Memory Allocation

To allocate a block of memory, call malloc. The prototype for this function is in `stdlib.h'.

Function: void * malloc (size_t size)
This function returns a pointer to a newly allocated block size bytes long, or a null pointer if the block could not be allocated.

The contents of the block are undefined; you must initialize it yourself (or use calloc instead; see section Allocating Cleared Space). Normally you would cast the value as a pointer to the kind of object that you want to store in the block. Here we show an example of doing so, and of initializing the space with zeros using the library function memset (see section Copying and Concatenation):

struct foo *ptr;
...
ptr = (struct foo *) malloc (sizeof (struct foo));
if (ptr == 0) abort ();
memset (ptr, 0, sizeof (struct foo));

You can store the result of malloc into any pointer variable without a cast, because ISO C automatically converts the type void * to another type of pointer when necessary. But the cast is necessary in contexts other than assignment operators or if you might want your code to run in traditional C.

Remember that when allocating space for a string, the argument to malloc must be one plus the length of the string. This is because a string is terminated with a null character that doesn't count in the "length" of the string but does need space. For example:

char *ptr;
...
ptr = (char *) malloc (length + 1);

See section Representation of Strings, for more information about this.

Examples of malloc

If no more space is available, malloc returns a null pointer. You should check the value of every call to malloc. It is useful to write a subroutine that calls malloc and reports an error if the value is a null pointer, returning only if the value is nonzero. This function is conventionally called xmalloc. Here it is:

void *
xmalloc (size_t size)
{
  register void *value = malloc (size);
  if (value == 0)
    fatal ("virtual memory exhausted");
  return value;
}

Here is a real example of using malloc (by way of xmalloc). The function savestring will copy a sequence of characters into a newly allocated null-terminated string:

char *
savestring (const char *ptr, size_t len)
{
  register char *value = (char *) xmalloc (len + 1);
  value[len] = '\0';
  return (char *) memcpy (value, ptr, len);
}

The block that malloc gives you is guaranteed to be aligned so that it can hold any type of data. In the GNU system, the address is always a multiple of eight on most systems, and a multiple of 16 on 64-bit systems. Only rarely is any higher boundary (such as a page boundary) necessary; for those cases, use memalign, posix_memalign or valloc (see section Allocating Aligned Memory Blocks).

Note that the memory located after the end of the block is likely to be in use for something else; perhaps a block already allocated by another call to malloc. If you attempt to treat the block as longer than you asked for it to be, you are liable to destroy the data that malloc uses to keep track of its blocks, or you may destroy the contents of another block. If you have already allocated a block and discover you want it to be bigger, use realloc (see section Changing the Size of a Block).

Freeing Memory Allocated with malloc

When you no longer need a block that you got with malloc, use the function free to make the block available to be allocated again. The prototype for this function is in `stdlib.h'.

Function: void free (void *ptr)
The free function deallocates the block of memory pointed at by ptr.

Function: void cfree (void *ptr)
This function does the same thing as free. It's provided for backward compatibility with SunOS; you should use free instead.

Freeing a block alters the contents of the block. Do not expect to find any data (such as a pointer to the next block in a chain of blocks) in the block after freeing it. Copy whatever you need out of the block before freeing it! Here is an example of the proper way to free all the blocks in a chain, and the strings that they point to:

struct chain
  {
    struct chain *next;
    char *name;
  }

void
free_chain (struct chain *chain)
{
  while (chain != 0)
    {
      struct chain *next = chain->next;
      free (chain->name);
      free (chain);
      chain = next;
    }
}

Occasionally, free can actually return memory to the operating system and make the process smaller. Usually, all it can do is allow a later call to malloc to reuse the space. In the meantime, the space remains in your program as part of a free-list used internally by malloc.

There is no point in freeing blocks at the end of a program, because all of the program's space is given back to the system when the process terminates.

Changing the Size of a Block

Often you do not know for certain how big a block you will ultimately need at the time you must begin to use the block. For example, the block might be a buffer that you use to hold a line being read from a file; no matter how long you make the buffer initially, you may encounter a line that is longer.

You can make the block longer by calling realloc. This function is declared in `stdlib.h'.

Function: void * realloc (void *ptr, size_t newsize)
The realloc function changes the size of the block whose address is ptr to be newsize.

Since the space after the end of the block may be in use, realloc may find it necessary to copy the block to a new address where more free space is available. The value of realloc is the new address of the block. If the block needs to be moved, realloc copies the old contents.

If you pass a null pointer for ptr, realloc behaves just like `malloc (newsize)'. This can be convenient, but beware that older implementations (before ISO C) may not support this behavior, and will probably crash when realloc is passed a null pointer.

Like malloc, realloc may return a null pointer if no memory space is available to make the block bigger. When this happens, the original block is untouched; it has not been modified or relocated.

In most cases it makes no difference what happens to the original block when realloc fails, because the application program cannot continue when it is out of memory, and the only thing to do is to give a fatal error message. Often it is convenient to write and use a subroutine, conventionally called xrealloc, that takes care of the error message as xmalloc does for malloc:

void *
xrealloc (void *ptr, size_t size)
{
  register void *value = realloc (ptr, size);
  if (value == 0)
    fatal ("Virtual memory exhausted");
  return value;
}

You can also use realloc to make a block smaller. The reason you would do this is to avoid tying up a lot of memory space when only a little is needed. In several allocation implementations, making a block smaller sometimes necessitates copying it, so it can fail if no other space is available.

If the new size you specify is the same as the old size, realloc is guaranteed to change nothing and return the same address that you gave.

Allocating Cleared Space

The function calloc allocates memory and clears it to zero. It is declared in `stdlib.h'.

Function: void * calloc (size_t count, size_t eltsize)
This function allocates a block long enough to contain a vector of count elements, each of size eltsize. Its contents are cleared to zero before calloc returns.

You could define calloc as follows:

void *
calloc (size_t count, size_t eltsize)
{
  size_t size = count * eltsize;
  void *value = malloc (size);
  if (value != 0)
    memset (value, 0, size);
  return value;
}

But in general, it is not guaranteed that calloc calls malloc internally. Therefore, if an application provides its own malloc/realloc/free outside the C library, it should always define calloc, too.

Efficiency Considerations for malloc

As opposed to other versions, the malloc in the GNU C Library does not round up block sizes to powers of two, neither for large nor for small sizes. Neighboring chunks can be coalesced on a free no matter what their size is. This makes the implementation suitable for all kinds of allocation patterns without generally incurring high memory waste through fragmentation.

Very large blocks (much larger than a page) are allocated with mmap (anonymous or via /dev/zero) by this implementation. This has the great advantage that these chunks are returned to the system immediately when they are freed. Therefore, it cannot happen that a large chunk becomes "locked" in between smaller ones and even after calling free wastes memory. The size threshold for mmap to be used can be adjusted with mallopt. The use of mmap can also be disabled completely.

Allocating Aligned Memory Blocks

The address of a block returned by malloc or realloc in the GNU system is always a multiple of eight (or sixteen on 64-bit systems). If you need a block whose address is a multiple of a higher power of two than that, use memalign, posix_memalign, or valloc. These functions are declared in `stdlib.h'.

With the GNU library, you can use free to free the blocks that memalign, posix_memalign, and valloc return. That does not work in BSD, however--BSD does not provide any way to free such blocks.

Function: void * memalign (size_t boundary, size_t size)
The memalign function allocates a block of size bytes whose address is a multiple of boundary. The boundary must be a power of two! The function memalign works by allocating a somewhat larger block, and then returning an address within the block that is on the specified boundary.

Function: int posix_memalign (void **memptr, size_t alignment, size_t size)
The posix_memalign function is similar to the memalign function in that it returns a buffer of size bytes aligned to a multiple of alignment. But it adds one requirement to the parameter alignment: the value must be a power of two multiple of sizeof (void *).

If the function succeeds in allocation memory a pointer to the allocated memory is returned in *memptr and the return value is zero. Otherwise the function returns an error value indicating the problem.

This function was introduced in POSIX 1003.1d.

Function: void * valloc (size_t size)
Using valloc is like using memalign and passing the page size as the value of the second argument. It is implemented like this:

void *
valloc (size_t size)
{
  return memalign (getpagesize (), size);
}

section How to get information about the memory subsystem? for more information about the memory subsystem.

Malloc Tunable Parameters

You can adjust some parameters for dynamic memory allocation with the mallopt function. This function is the general SVID/XPG interface, defined in `malloc.h'.

Function: int mallopt (int param, int value)
When calling mallopt, the param argument specifies the parameter to be set, and value the new value to be set. Possible choices for param, as defined in `malloc.h', are:

M_TRIM_THRESHOLD
This is the minimum size (in bytes) of the top-most, releasable chunk that will cause sbrk to be called with a negative argument in order to return memory to the system.
M_TOP_PAD
This parameter determines the amount of extra memory to obtain from the system when a call to sbrk is required. It also specifies the number of bytes to retain when shrinking the heap by calling sbrk with a negative argument. This provides the necessary hysteresis in heap size such that excessive amounts of system calls can be avoided.
M_MMAP_THRESHOLD
All chunks larger than this value are allocated outside the normal heap, using the mmap system call. This way it is guaranteed that the memory for these chunks can be returned to the system on free.
M_MMAP_MAX
The maximum number of chunks to allocate with mmap. Setting this to zero disables all use of mmap.

Heap Consistency Checking

You can ask malloc to check the consistency of dynamic memory by using the mcheck function. This function is a GNU extension, declared in `mcheck.h'.

Function: int mcheck (void (*abortfn) (enum mcheck_status status))
Calling mcheck tells malloc to perform occasional consistency checks. These will catch things such as writing past the end of a block that was allocated with malloc.

The abortfn argument is the function to call when an inconsistency is found. If you supply a null pointer, then mcheck uses a default function which prints a message and calls abort (see section Aborting a Program). The function you supply is called with one argument, which says what sort of inconsistency was detected; its type is described below.

It is too late to begin allocation checking once you have allocated anything with malloc. So mcheck does nothing in that case. The function returns -1 if you call it too late, and 0 otherwise (when it is successful).

The easiest way to arrange to call mcheck early enough is to use the option `-lmcheck' when you link your program; then you don't need to modify your program source at all. Alternatively you might use a debugger to insert a call to mcheck whenever the program is started, for example these gdb commands will automatically call mcheck whenever the program starts:

(gdb) break main
Breakpoint 1, main (argc=2, argv=0xbffff964) at whatever.c:10
(gdb) command 1
Type commands for when breakpoint 1 is hit, one per line.
End with a line saying just "end".
>call mcheck(0)
>continue
>end
(gdb) ...

This will however only work if no initialization function of any object involved calls any of the malloc functions since mcheck must be called before the first such function.

Function: enum mcheck_status mprobe (void *pointer)
The mprobe function lets you explicitly check for inconsistencies in a particular allocated block. You must have already called mcheck at the beginning of the program, to do its occasional checks; calling mprobe requests an additional consistency check to be done at the time of the call.

The argument pointer must be a pointer returned by malloc or realloc. mprobe returns a value that says what inconsistency, if any, was found. The values are described below.

Data Type: enum mcheck_status
This enumerated type describes what kind of inconsistency was detected in an allocated block, if any. Here are the possible values:

MCHECK_DISABLED
mcheck was not called before the first allocation. No consistency checking can be done.
MCHECK_OK
No inconsistency detected.
MCHECK_HEAD
The data immediately before the block was modified. This commonly happens when an array index or pointer is decremented too far.
MCHECK_TAIL
The data immediately after the block was modified. This commonly happens when an array index or pointer is incremented too far.
MCHECK_FREE
The block was already freed.

Another possibility to check for and guard against bugs in the use of malloc, realloc and free is to set the environment variable MALLOC_CHECK_. When MALLOC_CHECK_ is set, a special (less efficient) implementation is used which is designed to be tolerant against simple errors, such as double calls of free with the same argument, or overruns of a single byte (off-by-one bugs). Not all such errors can be protected against, however, and memory leaks can result. If MALLOC_CHECK_ is set to 0, any detected heap corruption is silently ignored; if set to 1, a diagnostic is printed on stderr; if set to 2, abort is called immediately. This can be useful because otherwise a crash may happen much later, and the true cause for the problem is then very hard to track down.

There is one problem with MALLOC_CHECK_: in SUID or SGID binaries it could possibly be exploited since diverging from the normal programs behaviour it now writes something to the standard error desriptor. Therefore the use of MALLOC_CHECK_ is disabled by default for SUID and SGID binaries. It can be enabled again by the system administrator by adding a file `/etc/suid-debug' (the content is not important it could be empty).

So, what's the difference between using MALLOC_CHECK_ and linking with `-lmcheck'? MALLOC_CHECK_ is orthogonal with respect to `-lmcheck'. `-lmcheck' has been added for backward compatibility. Both MALLOC_CHECK_ and `-lmcheck' should uncover the same bugs - but using MALLOC_CHECK_ you don't need to recompile your application.

Memory Allocation Hooks

The GNU C library lets you modify the behavior of malloc, realloc, and free by specifying appropriate hook functions. You can use these hooks to help you debug programs that use dynamic memory allocation, for example.

The hook variables are declared in `malloc.h'.

Variable: __malloc_hook
The value of this variable is a pointer to the function that malloc uses whenever it is called. You should define this function to look like malloc; that is, like:

void *function (size_t size, const void *caller)

The value of caller is the return address found on the stack when the malloc function was called. This value allows you to trace the memory consumption of the program.

Variable: __realloc_hook
The value of this variable is a pointer to function that realloc uses whenever it is called. You should define this function to look like realloc; that is, like:

void *function (void *ptr, size_t size, const void *caller)

The value of caller is the return address found on the stack when the realloc function was called. This value allows you to trace the memory consumption of the program.

Variable: __free_hook
The value of this variable is a pointer to function that free uses whenever it is called. You should define this function to look like free; that is, like:

void function (void *ptr, const void *caller)

The value of caller is the return address found on the stack when the free function was called. This value allows you to trace the memory consumption of the program.

Variable: __memalign_hook
The value of this variable is a pointer to function that memalign uses whenever it is called. You should define this function to look like memalign; that is, like:

void *function (size_t size, size_t alignment, const void *caller)

The value of caller is the return address found on the stack when the memalign function was called. This value allows you to trace the memory consumption of the program.

You must make sure that the function you install as a hook for one of these functions does not call that function recursively without restoring the old value of the hook first! Otherwise, your program will get stuck in an infinite recursion. Before calling the function recursively, one should make sure to restore all the hooks to their previous value. When coming back from the recursive call, all the hooks should be resaved since a hook might modify itself.

Variable: __malloc_initialize_hook
The value of this variable is a pointer to a function that is called once when the malloc implementation is initialized. This is a weak variable, so it can be overridden in the application with a definition like the following:

void (*__malloc_initialize_hook) (void) = my_init_hook;

An issue to look out for is the time at which the malloc hook functions can be safely installed. If the hook functions call the malloc-related functions recursively, it is necessary that malloc has already properly initialized itself at the time when __malloc_hook etc. is assigned to. On the other hand, if the hook functions provide a complete malloc implementation of their own, it is vital that the hooks are assigned to before the very first malloc call has completed, because otherwise a chunk obtained from the ordinary, un-hooked malloc may later be handed to __free_hook, for example.

In both cases, the problem can be solved by setting up the hooks from within a user-defined function pointed to by __malloc_initialize_hook---then the hooks will be set up safely at the right time.

Here is an example showing how to use __malloc_hook and __free_hook properly. It installs a function that prints out information every time malloc or free is called. We just assume here that realloc and memalign are not used in our program.

/* Prototypes for __malloc_hook, __free_hook */
#include <malloc.h>

/* Prototypes for our hooks.  */
static void *my_init_hook (void);
static void *my_malloc_hook (size_t, const void *);
static void my_free_hook (void*, const void *);

/* Override initializing hook from the C library. */
void (*__malloc_initialize_hook) (void) = my_init_hook;

static void
my_init_hook (void)
{
  old_malloc_hook = __malloc_hook;
  old_free_hook = __free_hook;
  __malloc_hook = my_malloc_hook;
  __free_hook = my_free_hook;
}

static void *
my_malloc_hook (size_t size, const void *caller)
{
  void *result;
  /* Restore all old hooks */
  __malloc_hook = old_malloc_hook;
  __free_hook = old_free_hook;
  /* Call recursively */
  result = malloc (size);
  /* Save underlaying hooks */
  old_malloc_hook = __malloc_hook;
  old_free_hook = __free_hook;
  /* printf might call malloc, so protect it too. */
  printf ("malloc (%u) returns %p\n", (unsigned int) size, result);
  /* Restore our own hooks */
  __malloc_hook = my_malloc_hook;
  __free_hook = my_free_hook;
  return result;
}

static void *
my_free_hook (void *ptr, const void *caller)
{
  /* Restore all old hooks */
  __malloc_hook = old_malloc_hook;
  __free_hook = old_free_hook;
  /* Call recursively */
  free (ptr);
  /* Save underlaying hooks */
  old_malloc_hook = __malloc_hook;
  old_free_hook = __free_hook;
  /* printf might call free, so protect it too. */
  printf ("freed pointer %p\n", ptr);
  /* Restore our own hooks */
  __malloc_hook = my_malloc_hook;
  __free_hook = my_free_hook;
}

main ()
{
  ...
}

The mcheck function (see section Heap Consistency Checking) works by installing such hooks.

Statistics for Memory Allocation with malloc

You can get information about dynamic memory allocation by calling the mallinfo function. This function and its associated data type are declared in `malloc.h'; they are an extension of the standard SVID/XPG version.

Data Type: struct mallinfo
This structure type is used to return information about the dynamic memory allocator. It contains the following members:

int arena
This is the total size of memory allocated with sbrk by malloc, in bytes.
int ordblks
This is the number of chunks not in use. (The memory allocator internally gets chunks of memory from the operating system, and then carves them up to satisfy individual malloc requests; see section Efficiency Considerations for malloc.)
int smblks
This field is unused.
int hblks
This is the total number of chunks allocated with mmap.
int hblkhd
This is the total size of memory allocated with mmap, in bytes.
int usmblks
This field is unused.
int fsmblks
This field is unused.
int uordblks
This is the total size of memory occupied by chunks handed out by malloc.
int fordblks
This is the total size of memory occupied by free (not in use) chunks.
int keepcost
This is the size of the top-most releasable chunk that normally borders the end of the heap (i.e. the high end of the virtual address space's data segment).

Function: struct mallinfo mallinfo (void)
This function returns information about the current dynamic memory usage in a structure of type struct mallinfo.

Summary of malloc-Related Functions

Here is a summary of the functions that work with malloc:

void *malloc (size_t size)
Allocate a block of size bytes. See section Basic Memory Allocation.
void free (void *addr)
Free a block previously allocated by malloc. See section Freeing Memory Allocated with malloc.
void *realloc (void *addr, size_t size)
Make a block previously allocated by malloc larger or smaller, possibly by copying it to a new location. See section Changing the Size of a Block.
void *calloc (size_t count, size_t eltsize)
Allocate a block of count * eltsize bytes using malloc, and set its contents to zero. See section Allocating Cleared Space.
void *valloc (size_t size)
Allocate a block of size bytes, starting on a page boundary. See section Allocating Aligned Memory Blocks.
void *memalign (size_t size, size_t boundary)
Allocate a block of size bytes, starting on an address that is a multiple of boundary. See section Allocating Aligned Memory Blocks.
int mallopt (int param, int value)
Adjust a tunable parameter. See section Malloc Tunable Parameters.
int mcheck (void (*abortfn) (void))
Tell malloc to perform occasional consistency checks on dynamically allocated memory, and to call abortfn when an inconsistency is found. See section Heap Consistency Checking.
void *(*__malloc_hook) (size_t size, const void *caller)
A pointer to a function that malloc uses whenever it is called.
void *(*__realloc_hook) (void *ptr, size_t size, const void *caller)
A pointer to a function that realloc uses whenever it is called.
void (*__free_hook) (void *ptr, const void *caller)
A pointer to a function that free uses whenever it is called.
void (*__memalign_hook) (size_t size, size_t alignment, const void *caller)
A pointer to a function that memalign uses whenever it is called.
struct mallinfo mallinfo (void)
Return information about the current dynamic memory usage. See section Statistics for Memory Allocation with malloc.

Allocation Debugging

A complicated task when programming with languages which do not use garbage collected dynamic memory allocation is to find memory leaks. Long running programs must assure that dynamically allocated objects are freed at the end of their lifetime. If this does not happen the system runs out of memory, sooner or later.

The malloc implementation in the GNU C library provides some simple means to detect such leaks and obtain some information to find the location. To do this the application must be started in a special mode which is enabled by an environment variable. There are no speed penalties for the program if the debugging mode is not enabled.

How to install the tracing functionality

Function: void mtrace (void)
When the mtrace function is called it looks for an environment variable named MALLOC_TRACE. This variable is supposed to contain a valid file name. The user must have write access. If the file already exists it is truncated. If the environment variable is not set or it does not name a valid file which can be opened for writing nothing is done. The behaviour of malloc etc. is not changed. For obvious reasons this also happens if the application is installed with the SUID or SGID bit set.

If the named file is successfully opened, mtrace installs special handlers for the functions malloc, realloc, and free (see section Memory Allocation Hooks). From then on, all uses of these functions are traced and protocolled into the file. There is now of course a speed penalty for all calls to the traced functions so tracing should not be enabled during normal use.

This function is a GNU extension and generally not available on other systems. The prototype can be found in `mcheck.h'.

Function: void muntrace (void)
The muntrace function can be called after mtrace was used to enable tracing the malloc calls. If no (succesful) call of mtrace was made muntrace does nothing.

Otherwise it deinstalls the handlers for malloc, realloc, and free and then closes the protocol file. No calls are protocolled anymore and the program runs again at full speed.

This function is a GNU extension and generally not available on other systems. The prototype can be found in `mcheck.h'.

Example program excerpts

Even though the tracing functionality does not influence the runtime behaviour of the program it is not a good idea to call mtrace in all programs. Just imagine that you debug a program using mtrace and all other programs used in the debugging session also trace their malloc calls. The output file would be the same for all programs and thus is unusable. Therefore one should call mtrace only if compiled for debugging. A program could therefore start like this:

#include <mcheck.h>

int
main (int argc, char *argv[])
{
#ifdef DEBUGGING
  mtrace ();
#endif
  ...
}

This is all what is needed if you want to trace the calls during the whole runtime of the program. Alternatively you can stop the tracing at any time with a call to muntrace. It is even possible to restart the tracing again with a new call to mtrace. But this can cause unreliable results since there may be calls of the functions which are not called. Please note that not only the application uses the traced functions, also libraries (including the C library itself) use these functions.

This last point is also why it is no good idea to call muntrace before the program terminated. The libraries are informed about the termination of the program only after the program returns from main or calls exit and so cannot free the memory they use before this time.

So the best thing one can do is to call mtrace as the very first function in the program and never call muntrace. So the program traces almost all uses of the malloc functions (except those calls which are executed by constructors of the program or used libraries).

Some more or less clever ideas

You know the situation. The program is prepared for debugging and in all debugging sessions it runs well. But once it is started without debugging the error shows up. A typical example is a memory leak that becomes visible only when we turn off the debugging. If you foresee such situations you can still win. Simply use something equivalent to the following little program:

#include <mcheck.h>
#include <signal.h>

static void
enable (int sig)
{
  mtrace ();
  signal (SIGUSR1, enable);
}

static void
disable (int sig)
{
  muntrace ();
  signal (SIGUSR2, disable);
}

int
main (int argc, char *argv[])
{
  ...

  signal (SIGUSR1, enable);
  signal (SIGUSR2, disable);

  ...
}

I.e., the user can start the memory debugger any time s/he wants if the program was started with MALLOC_TRACE set in the environment. The output will of course not show the allocations which happened before the first signal but if there is a memory leak this will show up nevertheless.

Interpreting the traces

If you take a look at the output it will look similar to this:

= Start
 [0x8048209] - 0x8064cc8
 [0x8048209] - 0x8064ce0
 [0x8048209] - 0x8064cf8
 [0x80481eb] + 0x8064c48 0x14
 [0x80481eb] + 0x8064c60 0x14
 [0x80481eb] + 0x8064c78 0x14
 [0x80481eb] + 0x8064c90 0x14
= End

What this all means is not really important since the trace file is not meant to be read by a human. Therefore no attention is given to readability. Instead there is a program which comes with the GNU C library which interprets the traces and outputs a summary in an user-friendly way. The program is called mtrace (it is in fact a Perl script) and it takes one or two arguments. In any case the name of the file with the trace output must be specified. If an optional argument precedes the name of the trace file this must be the name of the program which generated the trace.

drepper$ mtrace tst-mtrace log
No memory leaks.

In this case the program tst-mtrace was run and it produced a trace file `log'. The message printed by mtrace shows there are no problems with the code, all allocated memory was freed afterwards.

If we call mtrace on the example trace given above we would get a different outout:

drepper$ mtrace errlog
- 0x08064cc8 Free 2 was never alloc'd 0x8048209
- 0x08064ce0 Free 3 was never alloc'd 0x8048209
- 0x08064cf8 Free 4 was never alloc'd 0x8048209

Memory not freed:
-----------------
   Address     Size     Caller
0x08064c48     0x14  at 0x80481eb
0x08064c60     0x14  at 0x80481eb
0x08064c78     0x14  at 0x80481eb
0x08064c90     0x14  at 0x80481eb

We have called mtrace with only one argument and so the script has no chance to find out what is meant with the addresses given in the trace. We can do better:

drepper$ mtrace tst errlog
- 0x08064cc8 Free 2 was never alloc'd /home/drepper/tst.c:39
- 0x08064ce0 Free 3 was never alloc'd /home/drepper/tst.c:39
- 0x08064cf8 Free 4 was never alloc'd /home/drepper/tst.c:39

Memory not freed:
-----------------
   Address     Size     Caller
0x08064c48     0x14  at /home/drepper/tst.c:33
0x08064c60     0x14  at /home/drepper/tst.c:33
0x08064c78     0x14  at /home/drepper/tst.c:33
0x08064c90     0x14  at /home/drepper/tst.c:33

Suddenly the output makes much more sense and the user can see immediately where the function calls causing the trouble can be found.

Interpreting this output is not complicated. There are at most two different situations being detected. First, free was called for pointers which were never returned by one of the allocation functions. This is usually a very bad problem and what this looks like is shown in the first three lines of the output. Situations like this are quite rare and if they appear they show up very drastically: the program normally crashes.

The other situation which is much harder to detect are memory leaks. As you can see in the output the mtrace function collects all this information and so can say that the program calls an allocation function from line 33 in the source file `/home/drepper/tst-mtrace.c' four times without freeing this memory before the program terminates. Whether this is a real problem remains to be investigated.

Obstacks

An obstack is a pool of memory containing a stack of objects. You can create any number of separate obstacks, and then allocate objects in specified obstacks. Within each obstack, the last object allocated must always be the first one freed, but distinct obstacks are independent of each other.

Aside from this one constraint of order of freeing, obstacks are totally general: an obstack can contain any number of objects of any size. They are implemented with macros, so allocation is usually very fast as long as the objects are usually small. And the only space overhead per object is the padding needed to start each object on a suitable boundary.

Creating Obstacks

The utilities for manipulating obstacks are declared in the header file `obstack.h'.

Data Type: struct obstack
An obstack is represented by a data structure of type struct obstack. This structure has a small fixed size; it records the status of the obstack and how to find the space in which objects are allocated. It does not contain any of the objects themselves. You should not try to access the contents of the structure directly; use only the functions described in this chapter.

You can declare variables of type struct obstack and use them as obstacks, or you can allocate obstacks dynamically like any other kind of object. Dynamic allocation of obstacks allows your program to have a variable number of different stacks. (You can even allocate an obstack structure in another obstack, but this is rarely useful.)

All the functions that work with obstacks require you to specify which obstack to use. You do this with a pointer of type struct obstack *. In the following, we often say "an obstack" when strictly speaking the object at hand is such a pointer.

The objects in the obstack are packed into large blocks called chunks. The struct obstack structure points to a chain of the chunks currently in use.

The obstack library obtains a new chunk whenever you allocate an object that won't fit in the previous chunk. Since the obstack library manages chunks automatically, you don't need to pay much attention to them, but you do need to supply a function which the obstack library should use to get a chunk. Usually you supply a function which uses malloc directly or indirectly. You must also supply a function to free a chunk. These matters are described in the following section.

Preparing for Using Obstacks

Each source file in which you plan to use the obstack functions must include the header file `obstack.h', like this:

#include <obstack.h>

Also, if the source file uses the macro obstack_init, it must declare or define two functions or macros that will be called by the obstack library. One, obstack_chunk_alloc, is used to allocate the chunks of memory into which objects are packed. The other, obstack_chunk_free, is used to return chunks when the objects in them are freed. These macros should appear before any use of obstacks in the source file.

Usually these are defined to use malloc via the intermediary xmalloc (see section Unconstrained Allocation). This is done with the following pair of macro definitions:

#define obstack_chunk_alloc xmalloc
#define obstack_chunk_free free

Though the memory you get using obstacks really comes from malloc, using obstacks is faster because malloc is called less often, for larger blocks of memory. See section Obstack Chunks, for full details.

At run time, before the program can use a struct obstack object as an obstack, it must initialize the obstack by calling obstack_init.

Function: int obstack_init (struct obstack *obstack-ptr)
Initialize obstack obstack-ptr for allocation of objects. This function calls the obstack's obstack_chunk_alloc function. If allocation of memory fails, the function pointed to by obstack_alloc_failed_handler is called. The obstack_init function always returns 1 (Compatibility notice: Former versions of obstack returned 0 if allocation failed).

Here are two examples of how to allocate the space for an obstack and initialize it. First, an obstack that is a static variable:

static struct obstack myobstack;
...
obstack_init (&myobstack);

Second, an obstack that is itself dynamically allocated:

struct obstack *myobstack_ptr
  = (struct obstack *) xmalloc (sizeof (struct obstack));

obstack_init (myobstack_ptr);

Variable: obstack_alloc_failed_handler
The value of this variable is a pointer to a function that obstack uses when obstack_chunk_alloc fails to allocate memory. The default action is to print a message and abort. You should supply a function that either calls exit (see section Program Termination) or longjmp (see section Non-Local Exits) and doesn't return.

void my_obstack_alloc_failed (void)
...
obstack_alloc_failed_handler = &my_obstack_alloc_failed;

Allocation in an Obstack

The most direct way to allocate an object in an obstack is with obstack_alloc, which is invoked almost like malloc.

Function: void * obstack_alloc (struct obstack *obstack-ptr, int size)
This allocates an uninitialized block of size bytes in an obstack and returns its address. Here obstack-ptr specifies which obstack to allocate the block in; it is the address of the struct obstack object which represents the obstack. Each obstack function or macro requires you to specify an obstack-ptr as the first argument.

This function calls the obstack's obstack_chunk_alloc function if it needs to allocate a new chunk of memory; it calls obstack_alloc_failed_handler if allocation of memory by obstack_chunk_alloc failed.

For example, here is a function that allocates a copy of a string str in a specific obstack, which is in the variable string_obstack:

struct obstack string_obstack;

char *
copystring (char *string)
{
  size_t len = strlen (string) + 1;
  char *s = (char *) obstack_alloc (&string_obstack, len);
  memcpy (s, string, len);
  return s;
}

To allocate a block with specified contents, use the function obstack_copy, declared like this:

Function: void * obstack_copy (struct obstack *obstack-ptr, void *address, int size)
This allocates a block and initializes it by copying size bytes of data starting at address. It calls obstack_alloc_failed_handler if allocation of memory by obstack_chunk_alloc failed.

Function: void * obstack_copy0 (struct obstack *obstack-ptr, void *address, int size)
Like obstack_copy, but appends an extra byte containing a null character. This extra byte is not counted in the argument size.

The obstack_copy0 function is convenient for copying a sequence of characters into an obstack as a null-terminated string. Here is an example of its use:

char *
obstack_savestring (char *addr, int size)
{
  return obstack_copy0 (&myobstack, addr, size);
}

Contrast this with the previous example of savestring using malloc (see section Basic Memory Allocation).

Freeing Objects in an Obstack

To free an object allocated in an obstack, use the function obstack_free. Since the obstack is a stack of objects, freeing one object automatically frees all other objects allocated more recently in the same obstack.

Function: void obstack_free (struct obstack *obstack-ptr, void *object)
If object is a null pointer, everything allocated in the obstack is freed. Otherwise, object must be the address of an object allocated in the obstack. Then object is freed, along with everything allocated in obstack since object.

Note that if object is a null pointer, the result is an uninitialized obstack. To free all memory in an obstack but leave it valid for further allocation, call obstack_free with the address of the first object allocated on the obstack:

obstack_free (obstack_ptr, first_object_allocated_ptr);

Recall that the objects in an obstack are grouped into chunks. When all the objects in a chunk become free, the obstack library automatically frees the chunk (see section Preparing for Using Obstacks). Then other obstacks, or non-obstack allocation, can reuse the space of the chunk.

Obstack Functions and Macros

The interfaces for using obstacks may be defined either as functions or as macros, depending on the compiler. The obstack facility works with all C compilers, including both ISO C and traditional C, but there are precautions you must take if you plan to use compilers other than GNU C.

If you are using an old-fashioned non-ISO C compiler, all the obstack "functions" are actually defined only as macros. You can call these macros like functions, but you cannot use them in any other way (for example, you cannot take their address).

Calling the macros requires a special precaution: namely, the first operand (the obstack pointer) may not contain any side effects, because it may be computed more than once. For example, if you write this:

obstack_alloc (get_obstack (), 4);

you will find that get_obstack may be called several times. If you use *obstack_list_ptr++ as the obstack pointer argument, you will get very strange results since the incrementation may occur several times.

In ISO C, each function has both a macro definition and a function definition. The function definition is used if you take the address of the function without calling it. An ordinary call uses the macro definition by default, but you can request the function definition instead by writing the function name in parentheses, as shown here:

char *x;
void *(*funcp) ();
/* Use the macro.  */
x = (char *) obstack_alloc (obptr, size);
/* Call the function.  */
x = (char *) (obstack_alloc) (obptr, size);
/* Take the address of the function.  */
funcp = obstack_alloc;

This is the same situation that exists in ISO C for the standard library functions. See section Macro Definitions of Functions.

Warning: When you do use the macros, you must observe the precaution of avoiding side effects in the first operand, even in ISO C.

If you use the GNU C compiler, this precaution is not necessary, because various language extensions in GNU C permit defining the macros so as to compute each argument only once.

Growing Objects

Because memory in obstack chunks is used sequentially, it is possible to build up an object step by step, adding one or more bytes at a time to the end of the object. With this technique, you do not need to know how much data you will put in the object until you come to the end of it. We call this the technique of growing objects. The special functions for adding data to the growing object are described in this section.

You don't need to do anything special when you start to grow an object. Using one of the functions to add data to the object automatically starts it. However, it is necessary to say explicitly when the object is finished. This is done with the function obstack_finish.

The actual address of the object thus built up is not known until the object is finished. Until then, it always remains possible that you will add so much data that the object must be copied into a new chunk.

While the obstack is in use for a growing object, you cannot use it for ordinary allocation of another object. If you try to do so, the space already added to the growing object will become part of the other object.

Function: void obstack_blank (struct obstack *obstack-ptr, int size)
The most basic function for adding to a growing object is obstack_blank, which adds space without initializing it.

Function: void obstack_grow (struct obstack *obstack-ptr, void *data, int size)
To add a block of initialized space, use obstack_grow, which is the growing-object analogue of obstack_copy. It adds size bytes of data to the growing object, copying the contents from data.

Function: void obstack_grow0 (struct obstack *obstack-ptr, void *data, int size)
This is the growing-object analogue of obstack_copy0. It adds size bytes copied from data, followed by an additional null character.

Function: void obstack_1grow (struct obstack *obstack-ptr, char c)
To add one character at a time, use the function obstack_1grow. It adds a single byte containing c to the growing object.

Function: void obstack_ptr_grow (struct obstack *obstack-ptr, void *data)
Adding the value of a pointer one can use the function obstack_ptr_grow. It adds sizeof (void *) bytes containing the value of data.

Function: void obstack_int_grow (struct obstack *obstack-ptr, int data)
A single value of type int can be added by using the obstack_int_grow function. It adds sizeof (int) bytes to the growing object and initializes them with the value of data.

Function: void * obstack_finish (struct obstack *obstack-ptr)
When you are finished growing the object, use the function obstack_finish to close it off and return its final address.

Once you have finished the object, the obstack is available for ordinary allocation or for growing another object.

This function can return a null pointer under the same conditions as obstack_alloc (see section Allocation in an Obstack).

When you build an object by growing it, you will probably need to know afterward how long it became. You need not keep track of this as you grow the object, because you can find out the length from the obstack just before finishing the object with the function obstack_object_size, declared as follows:

Function: int obstack_object_size (struct obstack *obstack-ptr)
This function returns the current size of the growing object, in bytes. Remember to call this function before finishing the object. After it is finished, obstack_object_size will return zero.

If you have started growing an object and wish to cancel it, you should finish it and then free it, like this:

obstack_free (obstack_ptr, obstack_finish (obstack_ptr));

This has no effect if no object was growing.

You can use obstack_blank with a negative size argument to make the current object smaller. Just don't try to shrink it beyond zero length--there's no telling what will happen if you do that.

Extra Fast Growing Objects

The usual functions for growing objects incur overhead for checking whether there is room for the new growth in the current chunk. If you are frequently constructing objects in small steps of growth, this overhead can be significant.

You can reduce the overhead by using special "fast growth" functions that grow the object without checking. In order to have a robust program, you must do the checking yourself. If you do this checking in the simplest way each time you are about to add data to the object, you have not saved anything, because that is what the ordinary growth functions do. But if you can arrange to check less often, or check more efficiently, then you make the program faster.

The function obstack_room returns the amount of room available in the current chunk. It is declared as follows:

Function: int obstack_room (struct obstack *obstack-ptr)
This returns the number of bytes that can be added safely to the current growing object (or to an object about to be started) in obstack obstack using the fast growth functions.

While you know there is room, you can use these fast growth functions for adding data to a growing object:

Function: void obstack_1grow_fast (struct obstack *obstack-ptr, char c)
The function obstack_1grow_fast adds one byte containing the character c to the growing object in obstack obstack-ptr.

Function: void obstack_ptr_grow_fast (struct obstack *obstack-ptr, void *data)
The function obstack_ptr_grow_fast adds sizeof (void *) bytes containing the value of data to the growing object in obstack obstack-ptr.

Function: void obstack_int_grow_fast (struct obstack *obstack-ptr, int data)
The function obstack_int_grow_fast adds sizeof (int) bytes containing the value of data to the growing object in obstack obstack-ptr.

Function: void obstack_blank_fast (struct obstack *obstack-ptr, int size)
The function obstack_blank_fast adds size bytes to the growing object in obstack obstack-ptr without initializing them.

When you check for space using obstack_room and there is not enough room for what you want to add, the fast growth functions are not safe. In this case, simply use the corresponding ordinary growth function instead. Very soon this will copy the object to a new chunk; then there will be lots of room available again.

So, each time you use an ordinary growth function, check afterward for sufficient space using obstack_room. Once the object is copied to a new chunk, there will be plenty of space again, so the program will start using the fast growth functions again.

Here is an example:

void
add_string (struct obstack *obstack, const char *ptr, int len)
{
  while (len > 0)
    {
      int room = obstack_room (obstack);
      if (room == 0)
        {
          /* Not enough room. Add one character slowly,
             which may copy to a new chunk and make room.  */
          obstack_1grow (obstack, *ptr++);
          len--;
        }
      else
        {
          if (room > len)
            room = len;
          /* Add fast as much as we have room for. */
          len -= room;
          while (room-- > 0)
            obstack_1grow_fast (obstack, *ptr++);
        }
    }
}

Status of an Obstack

Here are functions that provide information on the current status of allocation in an obstack. You can use them to learn about an object while still growing it.

Function: void * obstack_base (struct obstack *obstack-ptr)
This function returns the tentative address of the beginning of the currently growing object in obstack-ptr. If you finish the object immediately, it will have that address. If you make it larger first, it may outgrow the current chunk--then its address will change!

If no object is growing, this value says where the next object you allocate will start (once again assuming it fits in the current chunk).

Function: void * obstack_next_free (struct obstack *obstack-ptr)
This function returns the address of the first free byte in the current chunk of obstack obstack-ptr. This is the end of the currently growing object. If no object is growing, obstack_next_free returns the same value as obstack_base.

Function: int obstack_object_size (struct obstack *obstack-ptr)
This function returns the size in bytes of the currently growing object. This is equivalent to

obstack_next_free (obstack-ptr) - obstack_base (obstack-ptr)

Alignment of Data in Obstacks

Each obstack has an alignment boundary; each object allocated in the obstack automatically starts on an address that is a multiple of the specified boundary. By default, this boundary is 4 bytes.

To access an obstack's alignment boundary, use the macro obstack_alignment_mask, whose function prototype looks like this:

Macro: int obstack_alignment_mask (struct obstack *obstack-ptr)
The value is a bit mask; a bit that is 1 indicates that the corresponding bit in the address of an object should be 0. The mask value should be one less than a power of 2; the effect is that all object addresses are multiples of that power of 2. The default value of the mask is 3, so that addresses are multiples of 4. A mask value of 0 means an object can start on any multiple of 1 (that is, no alignment is required).

The expansion of the macro obstack_alignment_mask is an lvalue, so you can alter the mask by assignment. For example, this statement:

obstack_alignment_mask (obstack_ptr) = 0;

has the effect of turning off alignment processing in the specified obstack.

Note that a change in alignment mask does not take effect until after the next time an object is allocated or finished in the obstack. If you are not growing an object, you can make the new alignment mask take effect immediately by calling obstack_finish. This will finish a zero-length object and then do proper alignment for the next object.

Obstack Chunks

Obstacks work by allocating space for themselves in large chunks, and then parceling out space in the chunks to satisfy your requests. Chunks are normally 4096 bytes long unless you specify a different chunk size. The chunk size includes 8 bytes of overhead that are not actually used for storing objects. Regardless of the specified size, longer chunks will be allocated when necessary for long objects.

The obstack library allocates chunks by calling the function obstack_chunk_alloc, which you must define. When a chunk is no longer needed because you have freed all the objects in it, the obstack library frees the chunk by calling obstack_chunk_free, which you must also define.

These two must be defined (as macros) or declared (as functions) in each source file that uses obstack_init (see section Creating Obstacks). Most often they are defined as macros like this:

#define obstack_chunk_alloc malloc
#define obstack_chunk_free free

Note that these are simple macros (no arguments). Macro definitions with arguments will not work! It is necessary that obstack_chunk_alloc or obstack_chunk_free, alone, expand into a function name if it is not itself a function name.

If you allocate chunks with malloc, the chunk size should be a power of 2. The default chunk size, 4096, was chosen because it is long enough to satisfy many typical requests on the obstack yet short enough not to waste too much memory in the portion of the last chunk not yet used.

Macro: int obstack_chunk_size (struct obstack *obstack-ptr)
This returns the chunk size of the given obstack.

Since this macro expands to an lvalue, you can specify a new chunk size by assigning it a new value. Doing so does not affect the chunks already allocated, but will change the size of chunks allocated for that particular obstack in the future. It is unlikely to be useful to make the chunk size smaller, but making it larger might improve efficiency if you are allocating many objects whose size is comparable to the chunk size. Here is how to do so cleanly:

if (obstack_chunk_size (obstack_ptr) < new-chunk-size)
  obstack_chunk_size (obstack_ptr) = new-chunk-size;

Summary of Obstack Functions

Here is a summary of all the functions associated with obstacks. Each takes the address of an obstack (struct obstack *) as its first argument.

void obstack_init (struct obstack *obstack-ptr)
Initialize use of an obstack. See section Creating Obstacks.
void *obstack_alloc (struct obstack *obstack-ptr, int size)
Allocate an object of size uninitialized bytes. See section Allocation in an Obstack.
void *obstack_copy (struct obstack *obstack-ptr, void *address, int size)
Allocate an object of size bytes, with contents copied from address. See section Allocation in an Obstack.
void *obstack_copy0 (struct obstack *obstack-ptr, void *address, int size)
Allocate an object of size+1 bytes, with size of them copied from address, followed by a null character at the end. See section Allocation in an Obstack.
void obstack_free (struct obstack *obstack-ptr, void *object)
Free object (and everything allocated in the specified obstack more recently than object). See section Freeing Objects in an Obstack.
void obstack_blank (struct obstack *obstack-ptr, int size)
Add size uninitialized bytes to a growing object. See section Growing Objects.
void obstack_grow (struct obstack *obstack-ptr, void *address, int size)
Add size bytes, copied from address, to a growing object. See section Growing Objects.
void obstack_grow0 (struct obstack *obstack-ptr, void *address, int size)
Add size bytes, copied from address, to a growing object, and then add another byte containing a null character. See section Growing Objects.
void obstack_1grow (struct obstack *obstack-ptr, char data-char)
Add one byte containing data-char to a growing object. See section Growing Objects.
void *obstack_finish (struct obstack *obstack-ptr)
Finalize the object that is growing and return its permanent address. See section Growing Objects.
int obstack_object_size (struct obstack *obstack-ptr)
Get the current size of the currently growing object. See section Growing Objects.
void obstack_blank_fast (struct obstack *obstack-ptr, int size)
Add size uninitialized bytes to a growing object without checking that there is enough room. See section Extra Fast Growing Objects.
void obstack_1grow_fast (struct obstack *obstack-ptr, char data-char)
Add one byte containing data-char to a growing object without checking that there is enough room. See section Extra Fast Growing Objects.
int obstack_room (struct obstack *obstack-ptr)
Get the amount of room now available for growing the current object. See section Extra Fast Growing Objects.
int obstack_alignment_mask (struct obstack *obstack-ptr)
The mask used for aligning the beginning of an object. This is an lvalue. See section Alignment of Data in Obstacks.
int obstack_chunk_size (struct obstack *obstack-ptr)
The size for allocating chunks. This is an lvalue. See section Obstack Chunks.
void *obstack_base (struct obstack *obstack-ptr)
Tentative starting address of the currently growing object. See section Status of an Obstack.
void *obstack_next_free (struct obstack *obstack-ptr)
Address just after the end of the currently growing object. See section Status of an Obstack.

Automatic Storage with Variable Size

The function alloca supports a kind of half-dynamic allocation in which blocks are allocated dynamically but freed automatically.

Allocating a block with alloca is an explicit action; you can allocate as many blocks as you wish, and compute the size at run time. But all the blocks are freed when you exit the function that alloca was called from, just as if they were automatic variables declared in that function. There is no way to free the space explicitly.

The prototype for alloca is in `stdlib.h'. This function is a BSD extension.

Function: void * alloca (size_t size);
The return value of alloca is the address of a block of size bytes of memory, allocated in the stack frame of the calling function.

Do not use alloca inside the arguments of a function call--you will get unpredictable results, because the stack space for the alloca would appear on the stack in the middle of the space for the function arguments. An example of what to avoid is foo (x, alloca (4), y).

alloca Example

As an example of the use of alloca, here is a function that opens a file name made from concatenating two argument strings, and returns a file descriptor or minus one signifying failure:

int
open2 (char *str1, char *str2, int flags, int mode)
{
  char *name = (char *) alloca (strlen (str1) + strlen (str2) + 1);
  stpcpy (stpcpy (name, str1), str2);
  return open (name, flags, mode);
}

Here is how you would get the same results with malloc and free:

int
open2 (char *str1, char *str2, int flags, int mode)
{
  char *name = (char *) malloc (strlen (str1) + strlen (str2) + 1);
  int desc;
  if (name == 0)
    fatal ("virtual memory exceeded");
  stpcpy (stpcpy (name, str1), str2);
  desc = open (name, flags, mode);
  free (name);
  return desc;
}

As you can see, it is simpler with alloca. But alloca has other, more important advantages, and some disadvantages.

Advantages of alloca

Here are the reasons why alloca may be preferable to malloc:

Disadvantages of alloca

These are the disadvantages of alloca in comparison with malloc:

GNU C Variable-Size Arrays

In GNU C, you can replace most uses of alloca with an array of variable size. Here is how open2 would look then:

int open2 (char *str1, char *str2, int flags, int mode)
{
  char name[strlen (str1) + strlen (str2) + 1];
  stpcpy (stpcpy (name, str1), str2);
  return open (name, flags, mode);
}

But alloca is not always equivalent to a variable-sized array, for several reasons:

Note: If you mix use of alloca and variable-sized arrays within one function, exiting a scope in which a variable-sized array was declared frees all blocks allocated with alloca during the execution of that scope.

Resizing the Data Segment

The symbols in this section are declared in `unistd.h'.

You will not normally use the functions in this section, because the functions described in section Allocating Storage For Program Data are easier to use. Those are interfaces to a GNU C Library memory allocator that uses the functions below itself. The functions below are simple interfaces to system calls.

Function: int brk (void *addr)

brk sets the high end of the calling process' data segment to addr.

The address of the end of a segment is defined to be the address of the last byte in the segment plus 1.

The function has no effect if addr is lower than the low end of the data segment. (This is considered success, by the way).

The function fails if it would cause the data segment to overlap another segment or exceed the process' data storage limit (see section Limiting Resource Usage).

The function is named for a common historical case where data storage and the stack are in the same segment. Data storage allocation grows upward from the bottom of the segment while the stack grows downward toward it from the top of the segment and the curtain between them is called the break.

The return value is zero on success. On failure, the return value is -1 and errno is set accordingly. The following errno values are specific to this function:

ENOMEM
The request would cause the data segment to overlap another segment or exceed the process' data storage limit.

Function: int sbrk (ptrdiff_t delta)
This function is the same as brk except that you specify the new end of the data segment as an offset delta from the current end and on success the return value is the address of the resulting end of the data segment instead of zero.

This means you can use `sbrk(0)' to find out what the current end of the data segment is.

Locking Pages

You can tell the system to associate a particular virtual memory page with a real page frame and keep it that way -- i.e. cause the page to be paged in if it isn't already and mark it so it will never be paged out and consequently will never cause a page fault. This is called locking a page.

The functions in this chapter lock and unlock the calling process' pages.

Why Lock Pages

Because page faults cause paged out pages to be paged in transparently, a process rarely needs to be concerned about locking pages. However, there are two reasons people sometimes are:

Be aware that when you lock a page, that's one fewer page frame that can be used to back other virtual memory (by the same or other processes), which can mean more page faults, which means the system runs more slowly. In fact, if you lock enough memory, some programs may not be able to run at all for lack of real memory.

Locked Memory Details

A memory lock is associated with a virtual page, not a real frame. The paging rule is: If a frame backs at least one locked page, don't page it out.

Memory locks do not stack. I.e. you can't lock a particular page twice so that it has to be unlocked twice before it is truly unlocked. It is either locked or it isn't.

A memory lock persists until the process that owns the memory explicitly unlocks it. (But process termination and exec cause the virtual memory to cease to exist, which you might say means it isn't locked any more).

Memory locks are not inherited by child processes. (But note that on a modern Unix system, immediately after a fork, the parent's and the child's virtual address space are backed by the same real page frames, so the child enjoys the parent's locks). See section Creating a Process.

Because of its ability to impact other processes, only the superuser can lock a page. Any process can unlock its own page.

The system sets limits on the amount of memory a process can have locked and the amount of real memory it can have dedicated to it. See section Limiting Resource Usage.

In Linux, locked pages aren't as locked as you might think. Two virtual pages that are not shared memory can nonetheless be backed by the same real frame. The kernel does this in the name of efficiency when it knows both virtual pages contain identical data, and does it even if one or both of the virtual pages are locked.

But when a process modifies one of those pages, the kernel must get it a separate frame and fill it with the page's data. This is known as a copy-on-write page fault. It takes a small amount of time and in a pathological case, getting that frame may require I/O.

To make sure this doesn't happen to your program, don't just lock the pages. Write to them as well, unless you know you won't write to them ever. And to make sure you have pre-allocated frames for your stack, enter a scope that declares a C automatic variable larger than the maximum stack size you will need, set it to something, then return from its scope.

Functions To Lock And Unlock Pages

The symbols in this section are declared in `sys/mman.h'. These functions are defined by POSIX.1b, but their availability depends on your kernel. If your kernel doesn't allow these functions, they exist but always fail. They are available with a Linux kernel.

Portability Note: POSIX.1b requires that when the mlock and munlock functions are available, the file `unistd.h' define the macro _POSIX_MEMLOCK_RANGE and the file limits.h define the macro PAGESIZE to be the size of a memory page in bytes. It requires that when the mlockall and munlockall functions are available, the `unistd.h' file define the macro _POSIX_MEMLOCK. The GNU C library conforms to this requirement.

Function: int mlock (const void *addr, size_t len)

mlock locks a range of the calling process' virtual pages.

The range of memory starts at address addr and is len bytes long. Actually, since you must lock whole pages, it is the range of pages that include any part of the specified range.

When the function returns successfully, each of those pages is backed by (connected to) a real frame (is resident) and is marked to stay that way. This means the function may cause page-ins and have to wait for them.

When the function fails, it does not affect the lock status of any pages.

The return value is zero if the function succeeds. Otherwise, it is -1 and errno is set accordingly. errno values specific to this function are:

ENOMEM
  • At least some of the specified address range does not exist in the calling process' virtual address space.
  • The locking would cause the process to exceed its locked page limit.
EPERM
The calling process is not superuser.
EINVAL
len is not positive.
ENOSYS
The kernel does not provide mlock capability.

You can lock all a process' memory with mlockall. You unlock memory with munlock or munlockall.

To avoid all page faults in a C program, you have to use mlockall, because some of the memory a program uses is hidden from the C code, e.g. the stack and automatic variables, and you wouldn't know what address to tell mlock.

Function: int munlock (const void *addr, size_t len)

mlock unlocks a range of the calling process' virtual pages.

munlock is the inverse of mlock and functions completely analogously to mlock, except that there is no EPERM failure.

Function: int mlockall (int flags)

mlockall locks all the pages in a process' virtual memory address space, and/or any that are added to it in the future. This includes the pages of the code, data and stack segment, as well as shared libraries, user space kernel data, shared memory, and memory mapped files.

flags is a string of single bit flags represented by the following macros. They tell mlockall which of its functions you want. All other bits must be zero.

MCL_CURRENT
Lock all pages which currently exist in the calling process' virtual address space.
MCL_FUTURE
Set a mode such that any pages added to the process' virtual address space in the future will be locked from birth. This mode does not affect future address spaces owned by the same process so exec, which replaces a process' address space, wipes out MCL_FUTURE. See section Executing a File.

When the function returns successfully, and you specified MCL_CURRENT, all of the process' pages are backed by (connected to) real frames (they are resident) and are marked to stay that way. This means the function may cause page-ins and have to wait for them.

When the process is in MCL_FUTURE mode because it successfully executed this function and specified MCL_CURRENT, any system call by the process that requires space be added to its virtual address space fails with errno = ENOMEM if locking the additional space would cause the process to exceed its locked page limit. In the case that the address space addition that can't be accomodated is stack expansion, the stack expansion fails and the kernel sends a SIGSEGV signal to the process.

When the function fails, it does not affect the lock status of any pages or the future locking mode.

The return value is zero if the function succeeds. Otherwise, it is -1 and errno is set accordingly. errno values specific to this function are:

ENOMEM
  • At least some of the specified address range does not exist in the calling process' virtual address space.
  • The locking would cause the process to exceed its locked page limit.
EPERM
The calling process is not superuser.
EINVAL
Undefined bits in flags are not zero.
ENOSYS
The kernel does not provide mlockall capability.

You can lock just specific pages with mlock. You unlock pages with munlockall and munlock.

Function: int munlockall (void)

munlockall unlocks every page in the calling process' virtual address space and turn off MCL_FUTURE future locking mode.

The return value is zero if the function succeeds. Otherwise, it is -1 and errno is set accordingly. The only way this function can fail is for generic reasons that all functions and system calls can fail, so there are no specific errno values.

Character Handling

Programs that work with characters and strings often need to classify a character--is it alphabetic, is it a digit, is it whitespace, and so on--and perform case conversion operations on characters. The functions in the header file `ctype.h' are provided for this purpose.

Since the choice of locale and character set can alter the classifications of particular character codes, all of these functions are affected by the current locale. (More precisely, they are affected by the locale currently selected for character classification--the LC_CTYPE category; see section Categories of Activities that Locales Affect.)

The ISO C standard specifies two different sets of functions. The one set works on char type characters, the other one on wchar_t wide characters (see section Introduction to Extended Characters).

Classification of Characters

This section explains the library functions for classifying characters. For example, isalpha is the function to test for an alphabetic character. It takes one argument, the character to test, and returns a nonzero integer if the character is alphabetic, and zero otherwise. You would use it like this:

if (isalpha (c))
  printf ("The character `%c' is alphabetic.\n", c);

Each of the functions in this section tests for membership in a particular class of characters; each has a name starting with `is'. Each of them takes one argument, which is a character to test, and returns an int which is treated as a boolean value. The character argument is passed as an int, and it may be the constant value EOF instead of a real character.

The attributes of any given character can vary between locales. See section Locales and Internationalization, for more information on locales.

These functions are declared in the header file `ctype.h'.

Function: int islower (int c)
Returns true if c is a lower-case letter. The letter need not be from the Latin alphabet, any alphabet representable is valid.

Function: int isupper (int c)
Returns true if c is an upper-case letter. The letter need not be from the Latin alphabet, any alphabet representable is valid.

Function: int isalpha (int c)
Returns true if c is an alphabetic character (a letter). If islower or isupper is true of a character, then isalpha is also true.

In some locales, there may be additional characters for which isalpha is true--letters which are neither upper case nor lower case. But in the standard "C" locale, there are no such additional characters.

Function: int isdigit (int c)
Returns true if c is a decimal digit (`0' through `9').

Function: int isalnum (int c)
Returns true if c is an alphanumeric character (a letter or number); in other words, if either isalpha or isdigit is true of a character, then isalnum is also true.

Function: int isxdigit (int c)
Returns true if c is a hexadecimal digit. Hexadecimal digits include the normal decimal digits `0' through `9' and the letters `A' through `F' and `a' through `f'.

Function: int ispunct (int c)
Returns true if c is a punctuation character. This means any printing character that is not alphanumeric or a space character.

Function: int isspace (int c)
Returns true if c is a whitespace character. In the standard "C" locale, isspace returns true for only the standard whitespace characters:

' '
space
'\f'
formfeed
'\n'
newline
'\r'
carriage return
'\t'
horizontal tab
'\v'
vertical tab

Function: int isblank (int c)
Returns true if c is a blank character; that is, a space or a tab. This function is a GNU extension.

Function: int isgraph (int c)
Returns true if c is a graphic character; that is, a character that has a glyph associated with it. The whitespace characters are not considered graphic.

Function: int isprint (int c)
Returns true if c is a printing character. Printing characters include all the graphic characters, plus the space (` ') character.

Function: int iscntrl (int c)
Returns true if c is a control character (that is, a character that is not a printing character).

Function: int isascii (int c)
Returns true if c is a 7-bit unsigned char value that fits into the US/UK ASCII character set. This function is a BSD extension and is also an SVID extension.

Case Conversion

This section explains the library functions for performing conversions such as case mappings on characters. For example, toupper converts any character to upper case if possible. If the character can't be converted, toupper returns it unchanged.

These functions take one argument of type int, which is the character to convert, and return the converted character as an int. If the conversion is not applicable to the argument given, the argument is returned unchanged.

Compatibility Note: In pre-ISO C dialects, instead of returning the argument unchanged, these functions may fail when the argument is not suitable for the conversion. Thus for portability, you may need to write islower(c) ? toupper(c) : c rather than just toupper(c).

These functions are declared in the header file `ctype.h'.

Function: int tolower (int c)
If c is an upper-case letter, tolower returns the corresponding lower-case letter. If c is not an upper-case letter, c is returned unchanged.

Function: int toupper (int c)
If c is a lower-case letter, toupper returns the corresponding upper-case letter. Otherwise c is returned unchanged.

Function: int toascii (int c)
This function converts c to a 7-bit unsigned char value that fits into the US/UK ASCII character set, by clearing the high-order bits. This function is a BSD extension and is also an SVID extension.

Function: int _tolower (int c)
This is identical to tolower, and is provided for compatibility with the SVID. See section SVID (The System V Interface Description).

Function: int _toupper (int c)
This is identical to toupper, and is provided for compatibility with the SVID.

Character class determination for wide characters

Amendment 1 to ISO C90 defines functions to classify wide characters. Although the original ISO C90 standard already defined the type wchar_t, no functions operating on them were defined.

The general design of the classification functions for wide characters is more general. It allows extensions to the set of available classifications, beyond those which are always available. The POSIX standard specifies how extensions can be made, and this is already implemented in the GNU C library implementation of the localedef program.

The character class functions are normally implemented with bitsets, with a bitset per character. For a given character, the appropriate bitset is read from a table and a test is performed as to whether a certain bit is set. Which bit is tested for is determined by the class.

For the wide character classification functions this is made visible. There is a type classification type defined, a function to retrieve this value for a given class, and a function to test whether a given character is in this class, using the classification value. On top of this the normal character classification functions as used for char objects can be defined.

Data type: wctype_t
The wctype_t can hold a value which represents a character class. The only defined way to generate such a value is by using the wctype function.

This type is defined in `wctype.h'.

Function: wctype_t wctype (const char *property)
The wctype returns a value representing a class of wide characters which is identified by the string property. Beside some standard properties each locale can define its own ones. In case no property with the given name is known for the current locale selected for the LC_CTYPE category, the function returns zero.

The properties known in every locale are:

@multitable @columnfractions .25 .25 .25 .25

  • "alnum" @tab "alpha" @tab "cntrl" @tab "digit"
  • "graph" @tab "lower" @tab "print" @tab "punct"
  • "space" @tab "upper" @tab "xdigit" This function is declared in `wctype.h'.
  • To test the membership of a character to one of the non-standard classes the ISO C standard defines a completely new function.
    Function: int iswctype (wint_t wc, wctype_t desc)
    This function returns a nonzero value if wc is in the character class specified by desc. desc must previously be returned by a successful call to wctype. This function is declared in `wctype.h'.
    To make it easier to use the commonly-used classification functions, they are defined in the C library. There is no need to use wctype if the property string is one of the known character classes. In some situations it is desirable to construct the property strings, and then it is important that wctype can also handle the standard classes.
    Function: int iswalnum (wint_t wc)
    This function returns a nonzero value if wc is an alphanumeric character (a letter or number); in other words, if either iswalpha or iswdigit is true of a character, then iswalnum is also true. This function can be implemented using
    iswctype (wc, wctype ("alnum"))
    
    It is declared in `wctype.h'.
    Function: int iswalpha (wint_t wc)
    Returns true if wc is an alphabetic character (a letter). If iswlower or iswupper is true of a character, then iswalpha is also true. In some locales, there may be additional characters for which iswalpha is true--letters which are neither upper case nor lower case. But in the standard "C" locale, there are no such additional characters. This function can be implemented using
    iswctype (wc, wctype ("alpha"))
    
    It is declared in `wctype.h'.
    Function: int iswcntrl (wint_t wc)
    Returns true if wc is a control character (that is, a character that is not a printing character). This function can be implemented using
    iswctype (wc, wctype ("cntrl"))
    
    It is declared in `wctype.h'.
    Function: int iswdigit (wint_t wc)
    Returns true if wc is a digit (e.g., `0' through `9'). Please note that this function does not only return a nonzero value for decimal digits, but for all kinds of digits. A consequence is that code like the following will not work unconditionally for wide characters:
    n = 0;
    while (iswdigit (*wc))
      {
        n *= 10;
        n += *wc++ - L'0';
      }
    
    This function can be implemented using
    iswctype (wc, wctype ("digit"))
    
    It is declared in `wctype.h'.
    Function: int iswgraph (wint_t wc)
    Returns true if wc is a graphic character; that is, a character that has a glyph associated with it. The whitespace characters are not considered graphic. This function can be implemented using
    iswctype (wc, wctype ("graph"))
    
    It is declared in `wctype.h'.
    Function: int iswlower (wint_t wc)
    Returns true if wc is a lower-case letter. The letter need not be from the Latin alphabet, any alphabet representable is valid. This function can be implemented using
    iswctype (wc, wctype ("lower"))
    
    It is declared in `wctype.h'.
    Function: int iswprint (wint_t wc)
    Returns true if wc is a printing character. Printing characters include all the graphic characters, plus the space (` ') character. This function can be implemented using
    iswctype (wc, wctype ("print"))
    
    It is declared in `wctype.h'.
    Function: int iswpunct (wint_t wc)
    Returns true if wc is a punctuation character. This means any printing character that is not alphanumeric or a space character. This function can be implemented using
    iswctype (wc, wctype ("punct"))
    
    It is declared in `wctype.h'.
    Function: int iswspace (wint_t wc)
    Returns true if wc is a whitespace character. In the standard "C" locale, iswspace returns true for only the standard whitespace characters:
    L' '
    space
    L'\f'
    formfeed
    L'\n'
    newline
    L'\r'
    carriage return
    L'\t'
    horizontal tab
    L'\v'
    vertical tab
    This function can be implemented using
    iswctype (wc, wctype ("space"))
    
    It is declared in `wctype.h'.
    Function: int iswupper (wint_t wc)
    Returns true if wc is an upper-case letter. The letter need not be from the Latin alphabet, any alphabet representable is valid. This function can be implemented using
    iswctype (wc, wctype ("upper"))
    
    It is declared in `wctype.h'.
    Function: int iswxdigit (wint_t wc)
    Returns true if wc is a hexadecimal digit. Hexadecimal digits include the normal decimal digits `0' through `9' and the letters `A' through `F' and `a' through `f'. This function can be implemented using
    iswctype (wc, wctype ("xdigit"))
    
    It is declared in `wctype.h'.
    The GNU C library also provides a function which is not defined in the ISO C standard but which is available as a version for single byte characters as well.
    Function: int iswblank (wint_t wc)
    Returns true if wc is a blank character; that is, a space or a tab. This function is a GNU extension. It is declared in `wchar.h'.

    Notes on using the wide character classes

    The first note is probably not astonishing but still occasionally a cause of problems. The iswXXX functions can be implemented using macros and in fact, the GNU C library does this. They are still available as real functions but when the `wctype.h' header is included the macros will be used. This is the same as the char type versions of these functions.

    The second note covers something new. It can be best illustrated by a (real-world) example. The first piece of code is an excerpt from the original code. It is truncated a bit but the intention should be clear.

    int
    is_in_class (int c, const char *class)
    {
      if (strcmp (class, "alnum") == 0)
        return isalnum (c);
      if (strcmp (class, "alpha") == 0)
        return isalpha (c);
      if (strcmp (class, "cntrl") == 0)
        return iscntrl (c);
      ...
      return 0;
    }
    

    Now, with the wctype and iswctype you can avoid the if cascades, but rewriting the code as follows is wrong:

    int
    is_in_class (int c, const char *class)
    {
      wctype_t desc = wctype (class);
      return desc ? iswctype ((wint_t) c, desc) : 0;
    }
    

    The problem is that it is not guaranteed that the wide character representation of a single-byte character can be found using casting. In fact, usually this fails miserably. The correct solution to this problem is to write the code as follows:

    int
    is_in_class (int c, const char *class)
    {
      wctype_t desc = wctype (class);
      return desc ? iswctype (btowc (c), desc) : 0;
    }
    

    See section Converting Single Characters, for more information on btowc. Note that this change probably does not improve the performance of the program a lot since the wctype function still has to make the string comparisons. It gets really interesting if the is_in_class function is called more than once for the same class name. In this case the variable desc could be computed once and reused for all the calls. Therefore the above form of the function is probably not the final one.

    Mapping of wide characters.

    The classification functions are also generalized by the ISO C standard. Instead of just allowing the two standard mappings, a locale can contain others. Again, the localedef program already supports generating such locale data files.

    Data Type: wctrans_t
    This data type is defined as a scalar type which can hold a value representing the locale-dependent character mapping. There is no way to construct such a value apar from using the return value of the wctrans function.

    This type is defined in `wctype.h'.

    Function: wctrans_t wctrans (const char *property)
    The wctrans function has to be used to find out whether a named mapping is defined in the current locale selected for the LC_CTYPE category. If the returned value is non-zero, you can use it afterwards in calls to towctrans. If the return value is zero no such mapping is known in the current locale.

    Beside locale-specific mappings there are two mappings which are guaranteed to be available in every locale:

    @multitable @columnfractions .5 .5

  • "tolower" @tab "toupper" These functions are declared in `wctype.h'.
  • Function: wint_t towctrans (wint_t wc, wctrans_t desc)
    towctrans maps the input character wc according to the rules of the mapping for which desc is a descriptor, and returns the value it finds. desc must be obtained by a successful call to wctrans. This function is declared in `wctype.h'.
    For the generally available mappings, the ISO C standard defines convenient shortcuts so that it is not necessary to call wctrans for them.
    Function: wint_t towlower (wint_t wc)
    If wc is an upper-case letter, towlower returns the corresponding lower-case letter. If wc is not an upper-case letter, wc is returned unchanged. towlower can be implemented using
    towctrans (wc, wctrans ("tolower"))
    
    This function is declared in `wctype.h'.
    Function: wint_t towupper (wint_t wc)
    If wc is a lower-case letter, towupper returns the corresponding upper-case letter. Otherwise wc is returned unchanged. towupper can be implemented using
    towctrans (wc, wctrans ("toupper"))
    
    This function is declared in `wctype.h'.
    The same warnings given in the last section for the use of the wide character classification functions apply here. It is not possible to simply cast a char type value to a wint_t and use it as an argument to towctrans calls.

    String and Array Utilities

    Operations on strings (or arrays of characters) are an important part of many programs. The GNU C library provides an extensive set of string utility functions, including functions for copying, concatenating, comparing, and searching strings. Many of these functions can also operate on arbitrary regions of storage; for example, the memcpy function can be used to copy the contents of any kind of array.

    It's fairly common for beginning C programmers to "reinvent the wheel" by duplicating this functionality in their own code, but it pays to become familiar with the library functions and to make use of them, since this offers benefits in maintenance, efficiency, and portability.

    For instance, you could easily compare one string to another in two lines of C code, but if you use the built-in strcmp function, you're less likely to make a mistake. And, since these library functions are typically highly optimized, your program may run faster too.

    Representation of Strings

    This section is a quick summary of string concepts for beginning C programmers. It describes how character strings are represented in C and some common pitfalls. If you are already familiar with this material, you can skip this section.

    A string is an array of char objects. But string-valued variables are usually declared to be pointers of type char *. Such variables do not include space for the text of a string; that has to be stored somewhere else--in an array variable, a string constant, or dynamically allocated memory (see section Allocating Storage For Program Data). It's up to you to store the address of the chosen memory space into the pointer variable. Alternatively you can store a null pointer in the pointer variable. The null pointer does not point anywhere, so attempting to reference the string it points to gets an error.

    "string" normally refers to multibyte character strings as opposed to wide character strings. Wide character strings are arrays of type wchar_t and as for multibyte character strings usually pointers of type wchar_t * are used.

    By convention, a null character, '\0', marks the end of a multibyte character string and the null wide character, L'\0', marks the end of a wide character string. For example, in testing to see whether the char * variable p points to a null character marking the end of a string, you can write !*p or *p == '\0'.

    A null character is quite different conceptually from a null pointer, although both are represented by the integer 0.

    String literals appear in C program source as strings of characters between double-quote characters (`"') where the initial double-quote character is immediately preceded by a capital `L' (ell) character (as in L"foo"). In ISO C, string literals can also be formed by string concatenation: "a" "b" is the same as "ab". For wide character strings one can either use L"a" L"b" or L"a" "b". Modification of string literals is not allowed by the GNU C compiler, because literals are placed in read-only storage.

    Character arrays that are declared const cannot be modified either. It's generally good style to declare non-modifiable string pointers to be of type const char *, since this often allows the C compiler to detect accidental modifications as well as providing some amount of documentation about what your program intends to do with the string.

    The amount of memory allocated for the character array may extend past the null character that normally marks the end of the string. In this document, the term allocated size is always used to refer to the total amount of memory allocated for the string, while the term length refers to the number of characters up to (but not including) the terminating null character.

    A notorious source of program bugs is trying to put more characters in a string than fit in its allocated size. When writing code that extends strings or moves characters into a pre-allocated array, you should be very careful to keep track of the length of the text and make explicit checks for overflowing the array. Many of the library functions do not do this for you! Remember also that you need to allocate an extra byte to hold the null character that marks the end of the string.

    Originally strings were sequences of bytes where each byte represents a single character. This is still true today if the strings are encoded using a single-byte character encoding. Things are different if the strings are encoded using a multibyte encoding (for more information on encodings see section Introduction to Extended Characters). There is no difference in the programming interface for these two kind of strings; the programmer has to be aware of this and interpret the byte sequences accordingly.

    But since there is no separate interface taking care of these differences the byte-based string functions are sometimes hard to use. Since the count parameters of these functions specify bytes a call to strncpy could cut a multibyte character in the middle and put an incomplete (and therefore unusable) byte sequence in the target buffer.

    To avoid these problems later versions of the ISO C standard introduce a second set of functions which are operating on wide characters (see section Introduction to Extended Characters). These functions don't have the problems the single-byte versions have since every wide character is a legal, interpretable value. This does not mean that cutting wide character strings at arbitrary points is without problems. It normally is for alphabet-based languages (except for non-normalized text) but languages based on syllables still have the problem that more than one wide character is necessary to complete a logical unit. This is a higher level problem which the C library functions are not designed to solve. But it is at least good that no invalid byte sequences can be created. Also, the higher level functions can also much easier operate on wide character than on multibyte characters so that a general advise is to use wide characters internally whenever text is more than simply copied.

    The remaining of this chapter will discuss the functions for handling wide character strings in parallel with the discussion of the multibyte character strings since there is almost always an exact equivalent available.

    String and Array Conventions

    This chapter describes both functions that work on arbitrary arrays or blocks of memory, and functions that are specific to null-terminated arrays of characters and wide characters.

    Functions that operate on arbitrary blocks of memory have names beginning with `mem' and `wmem' (such as memcpy and wmemcpy) and invariably take an argument which specifies the size (in bytes and wide characters respectively) of the block of memory to operate on. The array arguments and return values for these functions have type void * or wchar_t. As a matter of style, the elements of the arrays used with the `mem' functions are referred to as "bytes". You can pass any kind of pointer to these functions, and the sizeof operator is useful in computing the value for the size argument. Parameters to the `wmem' functions must be of type wchar_t *. These functions are not really usable with anything but arrays of this type.

    In contrast, functions that operate specifically on strings and wide character strings have names beginning with `str' and `wcs' respectively (such as strcpy and wcscpy) and look for a null character to terminate the string instead of requiring an explicit size argument to be passed. (Some of these functions accept a specified maximum length, but they also check for premature termination with a null character.) The array arguments and return values for these functions have type char * and wchar_t * respectively, and the array elements are referred to as "characters" and "wide characters".

    In many cases, there are both `mem' and `str'/`wcs' versions of a function. The one that is more appropriate to use depends on the exact situation. When your program is manipulating arbitrary arrays or blocks of storage, then you should always use the `mem' functions. On the other hand, when you are manipulating null-terminated strings it is usually more convenient to use the `str'/`wcs' functions, unless you already know the length of the string in advance. The `wmem' functions should be used for wide character arrays with known size.

    Some of the memory and string functions take single characters as arguments. Since a value of type char is automatically promoted into an value of type int when used as a parameter, the functions are declared with int as the type of the parameter in question. In case of the wide character function the situation is similarly: the parameter type for a single wide character is wint_t and not wchar_t. This would for many implementations not be necessary since the wchar_t is large enough to not be automatically promoted, but since the ISO C standard does not require such a choice of types the wint_t type is used.

    String Length

    You can get the length of a string using the strlen function. This function is declared in the header file `string.h'.

    Function: size_t strlen (const char *s)
    The strlen function returns the length of the null-terminated string s in bytes. (In other words, it returns the offset of the terminating null character within the array.)

    For example,

    strlen ("hello, world")
        => 12
    

    When applied to a character array, the strlen function returns the length of the string stored there, not its allocated size. You can get the allocated size of the character array that holds a string using the sizeof operator:

    char string[32] = "hello, world";
    sizeof (string)
        => 32
    strlen (string)
        => 12
    

    But beware, this will not work unless string is the character array itself, not a pointer to it. For example:

    char string[32] = "hello, world";
    char *ptr = string;
    sizeof (string)
        => 32
    sizeof (ptr)
        => 4  /* (on a machine with 4 byte pointers) */
    

    This is an easy mistake to make when you are working with functions that take string arguments; those arguments are always pointers, not arrays.

    It must also be noted that for multibyte encoded strings the return value does not have to correspond to the number of characters in the string. To get this value the string can be converted to wide characters and wcslen can be used or something like the following code can be used:

    /* The input is in string.
       The length is expected in n.  */
    {
      mbstate_t t;
      char *scopy = string;
      /* In initial state.  */
      memset (&t, '\0', sizeof (t));
      /* Determine number of characters.  */
      n = mbsrtowcs (NULL, &scopy, strlen (scopy), &t);
    }
    

    This is cumbersome to do so if the number of characters (as opposed to bytes) is needed often it is better to work with wide characters.

    The wide character equivalent is declared in `wchar.h'.

    Function: size_t wcslen (const wchar_t *ws)
    The wcslen function is the wide character equivalent to strlen. The return value is the number of wide characters in the wide character string pointed to by ws (this is also the offset of the terminating null wide character of ws).

    Since there are no multi wide character sequences making up one character the return value is not only the offset in the array, it is also the number of wide characters.

    This function was introduced in Amendment 1 to ISO C90.

    Function: size_t strnlen (const char *s, size_t maxlen)
    The strnlen function returns the length of the string s in bytes if this length is smaller than maxlen bytes. Otherwise it returns maxlen. Therefore this function is equivalent to (strlen (s) < n ? strlen (s) : maxlen) but it is more efficient and works even if the string s is not null-terminated.

    char string[32] = "hello, world";
    strnlen (string, 32)
        => 12
    strnlen (string, 5)
        => 5
    

    This function is a GNU extension and is declared in `string.h'.

    Function: size_t wcsnlen (const wchar_t *ws, size_t maxlen)
    wcsnlen is the wide character equivalent to strnlen. The maxlen parameter specifies the maximum number of wide characters.

    This function is a GNU extension and is declared in `wchar.h'.

    Copying and Concatenation

    You can use the functions described in this section to copy the contents of strings and arrays, or to append the contents of one string to another. The `str' and `mem' functions are declared in the header file `string.h' while the `wstr' and `wmem' functions are declared in the file `wchar.h'.

    A helpful way to remember the ordering of the arguments to the functions in this section is that it corresponds to an assignment expression, with the destination array specified to the left of the source array. All of these functions return the address of the destination array.

    Most of these functions do not work properly if the source and destination arrays overlap. For example, if the beginning of the destination array overlaps the end of the source array, the original contents of that part of the source array may get overwritten before it is copied. Even worse, in the case of the string functions, the null character marking the end of the string may be lost, and the copy function might get stuck in a loop trashing all the memory allocated to your program.

    All functions that have problems copying between overlapping arrays are explicitly identified in this manual. In addition to functions in this section, there are a few others like sprintf (see section Formatted Output Functions) and scanf (see section Formatted Input Functions).

    Function: void * memcpy (void *restrict to, const void *restrict from, size_t size)
    The memcpy function copies size bytes from the object beginning at from into the object beginning at to. The behavior of this function is undefined if the two arrays to and from overlap; use memmove instead if overlapping is possible.

    The value returned by memcpy is the value of to.

    Here is an example of how you might use memcpy to copy the contents of an array:

    struct foo *oldarray, *newarray;
    int arraysize;
    ...
    memcpy (new, old, arraysize * sizeof (struct foo));
    

    Function: wchar_t * wmemcpy (wchar_t *restrict wto, const wchar_t *restruct wfrom, size_t size)
    The wmemcpy function copies size wide characters from the object beginning at wfrom into the object beginning at wto. The behavior of this function is undefined if the two arrays wto and wfrom overlap; use wmemmove instead if overlapping is possible.

    The following is a possible implementation of wmemcpy but there are more optimizations possible.

    wchar_t *
    wmemcpy (wchar_t *restrict wto, const wchar_t *restrict wfrom,
             size_t size)
    {
      return (wchar_t *) memcpy (wto, wfrom, size * sizeof (wchar_t));
    }
    

    The value returned by wmemcpy is the value of wto.

    This function was introduced in Amendment 1 to ISO C90.

    Function: void * mempcpy (void *restrict to, const void *restrict from, size_t size)
    The mempcpy function is nearly identical to the memcpy function. It copies size bytes from the object beginning at from into the object pointed to by to. But instead of returning the value of to it returns a pointer to the byte following the last written byte in the object beginning at to. I.e., the value is ((void *) ((char *) to + size)).

    This function is useful in situations where a number of objects shall be copied to consecutive memory positions.

    void *
    combine (void *o1, size_t s1, void *o2, size_t s2)
    {
      void *result = malloc (s1 + s2);
      if (result != NULL)
        mempcpy (mempcpy (result, o1, s1), o2, s2);
      return result;
    }
    

    This function is a GNU extension.

    Function: wchar_t * wmempcpy (wchar_t *restrict wto, const wchar_t *restrict wfrom, size_t size)
    The wmempcpy function is nearly identical to the wmemcpy function. It copies size wide characters from the object beginning at wfrom into the object pointed to by wto. But instead of returning the value of wto it returns a pointer to the wide character following the last written wide character in the object beginning at wto. I.e., the value is wto + size.

    This function is useful in situations where a number of objects shall be copied to consecutive memory positions.

    The following is a possible implementation of wmemcpy but there are more optimizations possible.

    wchar_t *
    wmempcpy (wchar_t *restrict wto, const wchar_t *restrict wfrom,
              size_t size)
    {
      return (wchar_t *) mempcpy (wto, wfrom, size * sizeof (wchar_t));
    }
    

    This function is a GNU extension.

    Function: void * memmove (void *to, const void *from, size_t size)
    memmove copies the size bytes at from into the size bytes at to, even if those two blocks of space overlap. In the case of overlap, memmove is careful to copy the original values of the bytes in the block at from, including those bytes which also belong to the block at to.

    The value returned by memmove is the value of to.

    Function: wchar_t * wmemmove (wchar *wto, const wchar_t *wfrom, size_t size)
    wmemmove copies the size wide characters at wfrom into the size wide characters at wto, even if those two blocks of space overlap. In the case of overlap, memmove is careful to copy the original values of the wide characters in the block at wfrom, including those wide characters which also belong to the block at wto.

    The following is a possible implementation of wmemcpy but there are more optimizations possible.

    wchar_t *
    wmempcpy (wchar_t *restrict wto, const wchar_t *restrict wfrom,
              size_t size)
    {
      return (wchar_t *) mempcpy (wto, wfrom, size * sizeof (wchar_t));
    }
    

    The value returned by wmemmove is the value of wto.

    This function is a GNU extension.

    Function: void * memccpy (void *restrict to, const void *restrict from, int c, size_t size)
    This function copies no more than size bytes from from to to, stopping if a byte matching c is found. The return value is a pointer into to one byte past where c was copied, or a null pointer if no byte matching c appeared in the first size bytes of from.

    Function: void * memset (void *block, int c, size_t size)
    This function copies the value of c (converted to an unsigned char) into each of the first size bytes of the object beginning at block. It returns the value of block.

    Function: wchar_t * wmemset (wchar_t *block, wchar_t wc, size_t size)
    This function copies the value of wc into each of the first size wide characters of the object beginning at block. It returns the value of block.

    Function: char * strcpy (char *restrict to, const char *restrict from)
    This copies characters from the string from (up to and including the terminating null character) into the string to. Like memcpy, this function has undefined results if the strings overlap. The return value is the value of to.

    Function: wchar_t * wcscpy (wchar_t *restrict wto, const wchar_t *restrict wfrom)
    This copies wide characters from the string wfrom (up to and including the terminating null wide character) into the string wto. Like wmemcpy, this function has undefined results if the strings overlap. The return value is the value of wto.

    Function: char * strncpy (char *restrict to, const char *restrict from, size_t size)
    This function is similar to strcpy but always copies exactly size characters into to.

    If the length of from is more than size, then strncpy copies just the first size characters. Note that in this case there is no null terminator written into to.

    If the length of from is less than size, then strncpy copies all of from, followed by enough null characters to add up to size characters in all. This behavior is rarely useful, but it is specified by the ISO C standard.

    The behavior of strncpy is undefined if the strings overlap.

    Using strncpy as opposed to strcpy is a way to avoid bugs relating to writing past the end of the allocated space for to. However, it can also make your program much slower in one common case: copying a string which is probably small into a potentially large buffer. In this case, size may be large, and when it is, strncpy will waste a considerable amount of time copying null characters.

    Function: wchar_t * wcsncpy (wchar_t *restrict wto, const wchar_t *restrict wfrom, size_t size)
    This function is similar to wcscpy but always copies exactly size wide characters into wto.

    If the length of wfrom is more than size, then wcsncpy copies just the first size wide characters. Note that in this case there is no null terminator written into wto.

    If the length of wfrom is less than size, then wcsncpy copies all of wfrom, followed by enough null wide characters to add up to size wide characters in all. This behavior is rarely useful, but it is specified by the ISO C standard.

    The behavior of wcsncpy is undefined if the strings overlap.

    Using wcsncpy as opposed to wcscpy is a way to avoid bugs relating to writing past the end of the allocated space for wto. However, it can also make your program much slower in one common case: copying a string which is probably small into a potentially large buffer. In this case, size may be large, and when it is, wcsncpy will waste a considerable amount of time copying null wide characters.

    Function: char * strdup (const char *s)
    This function copies the null-terminated string s into a newly allocated string. The string is allocated using malloc; see section Unconstrained Allocation. If malloc cannot allocate space for the new string, strdup returns a null pointer. Otherwise it returns a pointer to the new string.

    Function: wchar_t * wcsdup (const wchar_t *ws)
    This function copies the null-terminated wide character string ws into a newly allocated string. The string is allocated using malloc; see section Unconstrained Allocation. If malloc cannot allocate space for the new string, wcsdup returns a null pointer. Otherwise it returns a pointer to the new wide character string.

    This function is a GNU extension.

    Function: char * strndup (const char *s, size_t size)
    This function is similar to strdup but always copies at most size characters into the newly allocated string.

    If the length of s is more than size, then strndup copies just the first size characters and adds a closing null terminator. Otherwise all characters are copied and the string is terminated.

    This function is different to strncpy in that it always terminates the destination string.

    strndup is a GNU extension.

    Function: char * stpcpy (char *restrict to, const char *restrict from)
    This function is like strcpy, except that it returns a pointer to the end of the string to (that is, the address of the terminating null character to + strlen (from)) rather than the beginning.

    For example, this program uses stpcpy to concatenate `foo' and `bar' to produce `foobar', which it then prints.

    #include <string.h>
    #include <stdio.h>
    
    int
    main (void)
    {
      char buffer[10];
      char *to = buffer;
      to = stpcpy (to, "foo");
      to = stpcpy (to, "bar");
      puts (buffer);
      return 0;
    }
    

    This function is not part of the ISO or POSIX standards, and is not customary on Unix systems, but we did not invent it either. Perhaps it comes from MS-DOG.

    Its behavior is undefined if the strings overlap. The function is declared in `string.h'.

    Function: wchar_t * wcpcpy (wchar_t *restrict wto, const wchar_t *restrict wfrom)
    This function is like wcscpy, except that it returns a pointer to the end of the string wto (that is, the address of the terminating null character wto + strlen (wfrom)) rather than the beginning.

    This function is not part of ISO or POSIX but was found useful while developing the GNU C Library itself.

    The behavior of wcpcpy is undefined if the strings overlap.

    wcpcpy is a GNU extension and is declared in `wchar.h'.

    Function: char * stpncpy (char *restrict to, const char *restrict from, size_t size)
    This function is similar to stpcpy but copies always exactly size characters into to.

    If the length of from is more then size, then stpncpy copies just the first size characters and returns a pointer to the character directly following the one which was copied last. Note that in this case there is no null terminator written into to.

    If the length of from is less than size, then stpncpy copies all of from, followed by enough null characters to add up to size characters in all. This behaviour is rarely useful, but it is implemented to be useful in contexts where this behaviour of the strncpy is used. stpncpy returns a pointer to the first written null character.

    This function is not part of ISO or POSIX but was found useful while developing the GNU C Library itself.

    Its behaviour is undefined if the strings overlap. The function is declared in `string.h'.

    Function: wchar_t * wcpncpy (wchar_t *restrict wto, const wchar_t *restrict wfrom, size_t size)
    This function is similar to wcpcpy but copies always exactly wsize characters into wto.

    If the length of wfrom is more then size, then wcpncpy copies just the first size wide characters and returns a pointer to the wide character directly following the one which was copied last. Note that in this case there is no null terminator written into wto.

    If the length of wfrom is less than size, then wcpncpy copies all of wfrom, followed by enough null characters to add up to size characters in all. This behaviour is rarely useful, but it is implemented to be useful in contexts where this behaviour of the wcsncpy is used. wcpncpy returns a pointer to the first written null character.

    This function is not part of ISO or POSIX but was found useful while developing the GNU C Library itself.

    Its behaviour is undefined if the strings overlap.

    wcpncpy is a GNU extension and is declared in `wchar.h'.

    Macro: char * strdupa (const char *s)
    This macro is similar to strdup but allocates the new string using alloca instead of malloc (see section Automatic Storage with Variable Size). This means of course the returned string has the same limitations as any block of memory allocated using alloca.

    For obvious reasons strdupa is implemented only as a macro; you cannot get the address of this function. Despite this limitation it is a useful function. The following code shows a situation where using malloc would be a lot more expensive.

    #include <paths.h>
    #include <string.h>
    #include <stdio.h>
    
    const char path[] = _PATH_STDPATH;
    
    int
    main (void)
    {
      char *wr_path = strdupa (path);
      char *cp = strtok (wr_path, ":");
    
      while (cp != NULL)
        {
          puts (cp);
          cp = strtok (NULL, ":");
        }
      return 0;
    }
    

    Please note that calling strtok using path directly is invalid. It is also not allowed to call strdupa in the argument list of strtok since strdupa uses alloca (see section Automatic Storage with Variable Size) can interfere with the parameter passing.

    This function is only available if GNU CC is used.

    Macro: char * strndupa (const char *s, size_t size)
    This function is similar to strndup but like strdupa it allocates the new string using alloca see section Automatic Storage with Variable Size. The same advantages and limitations of strdupa are valid for strndupa, too.

    This function is implemented only as a macro, just like strdupa. Just as strdupa this macro also must not be used inside the parameter list in a function call.

    strndupa is only available if GNU CC is used.

    Function: char * strcat (char *restrict to, const char *restrict from)
    The strcat function is similar to strcpy, except that the characters from from are concatenated or appended to the end of to, instead of overwriting it. That is, the first character from from overwrites the null character marking the end of to.

    An equivalent definition for strcat would be:

    char *
    strcat (char *restrict to, const char *restrict from)
    {
      strcpy (to + strlen (to), from);
      return to;
    }
    

    This function has undefined results if the strings overlap.

    Function: wchar_t * wcscat (wchar_t *restrict wto, const wchar_t *restrict wfrom)
    The wcscat function is similar to wcscpy, except that the characters from wfrom are concatenated or appended to the end of wto, instead of overwriting it. That is, the first character from wfrom overwrites the null character marking the end of wto.

    An equivalent definition for wcscat would be:

    wchar_t *
    wcscat (wchar_t *wto, const wchar_t *wfrom)
    {
      wcscpy (wto + wcslen (wto), wfrom);
      return wto;
    }
    

    This function has undefined results if the strings overlap.

    Programmers using the strcat or wcscat function (or the following strncat or wcsncar functions for that matter) can easily be recognized as lazy and reckless. In almost all situations the lengths of the participating strings are known (it better should be since how can one otherwise ensure the allocated size of the buffer is sufficient?) Or at least, one could know them if one keeps track of the results of the various function calls. But then it is very inefficient to use strcat/wcscat. A lot of time is wasted finding the end of the destination string so that the actual copying can start. This is a common example:

    /* This function concatenates arbitrarily many strings.  The last
       parameter must be NULL.  */
    char *
    concat (const char *str, ...)
    {
      va_list ap, ap2;
      size_t total = 1;
      const char *s;
      char *result;
    
      va_start (ap, str);
      /* Actually va_copy, but this is the name more gcc versions
         understand.  */
      __va_copy (ap2, ap);
    
      /* Determine how much space we need.  */
      for (s = str; s != NULL; s = va_arg (ap, const char *))
        total += strlen (s);
    
      va_end (ap);
    
      result = (char *) malloc (total);
      if (result != NULL)
        {
          result[0] = '\0';
    
          /* Copy the strings.  */
          for (s = str; s != NULL; s = va_arg (ap2, const char *))
            strcat (result, s);
        }
    
      va_end (ap2);
    
      return result;
    }
    

    This looks quite simple, especially the second loop where the strings are actually copied. But these innocent lines hide a major performance penalty. Just imagine that ten strings of 100 bytes each have to be concatenated. For the second string we search the already stored 100 bytes for the end of the string so that we can append the next string. For all strings in total the comparisons necessary to find the end of the intermediate results sums up to 5500! If we combine the copying with the search for the allocation we can write this function more efficient:

    char *
    concat (const char *str, ...)
    {
      va_list ap;
      size_t allocated = 100;
      char *result = (char *) malloc (allocated);
      char *wp;
    
      if (allocated != NULL)
        {
          char *newp;
    
          va_start (ap, atr);
    
          wp = result;
          for (s = str; s != NULL; s = va_arg (ap, const char *))
            {
              size_t len = strlen (s);
    
              /* Resize the allocated memory if necessary.  */
              if (wp + len + 1 > result + allocated)
                {
                  allocated = (allocated + len) * 2;
                  newp = (char *) realloc (result, allocated);
                  if (newp == NULL)
                    {
                      free (result);
                      return NULL;
                    }
                  wp = newp + (wp - result);
                  result = newp;
                }
    
              wp = mempcpy (wp, s, len);
            }
    
          /* Terminate the result string.  */
          *wp++ = '\0';
    
          /* Resize memory to the optimal size.  */
          newp = realloc (result, wp - result);
          if (newp != NULL)
            result = newp;
    
          va_end (ap);
        }
    
      return result;
    }
    

    With a bit more knowledge about the input strings one could fine-tune the memory allocation. The difference we are pointing to here is that we don't use strcat anymore. We always keep track of the length of the current intermediate result so we can safe us the search for the end of the string and use mempcpy. Please note that we also don't use stpcpy which might seem more natural since we handle with strings. But this is not necessary since we already know the length of the string and therefore can use the faster memory copying function. The example would work for wide characters the same way.

    Whenever a programmer feels the need to use strcat she or he should think twice and look through the program whether the code cannot be rewritten to take advantage of already calculated results. Again: it is almost always unnecessary to use strcat.

    Function: char * strncat (char *restrict to, const char *restrict from, size_t size)
    This function is like strcat except that not more than size characters from from are appended to the end of to. A single null character is also always appended to to, so the total allocated size of to must be at least size + 1 bytes longer than its initial length.

    The strncat function could be implemented like this:

    char *
    strncat (char *to, const char *from, size_t size)
    {
      to[strlen (to) + size] = '\0';
      strncpy (to + strlen (to), from, size);
      return to;
    }
    

    The behavior of strncat is undefined if the strings overlap.

    Function: wchar_t * wcsncat (wchar_t *restrict wto, const wchar_t *restrict wfrom, size_t size)
    This function is like wcscat except that not more than size characters from from are appended to the end of to. A single null character is also always appended to to, so the total allocated size of to must be at least size + 1 bytes longer than its initial length.

    The wcsncat function could be implemented like this:

    wchar_t *
    wcsncat (wchar_t *restrict wto, const wchar_t *restrict wfrom,
             size_t size)
    {
      wto[wcslen (to) + size] = L'\0';
      wcsncpy (wto + wcslen (wto), wfrom, size);
      return wto;
    }
    

    The behavior of wcsncat is undefined if the strings overlap.

    Here is an example showing the use of strncpy and strncat (the wide character version is equivalent). Notice how, in the call to strncat, the size parameter is computed to avoid overflowing the character array buffer.

    #include <string.h>
    #include <stdio.h>
    
    #define SIZE 10
    
    static char buffer[SIZE];
    
    main ()
    {
      strncpy (buffer, "hello", SIZE);
      puts (buffer);
      strncat (buffer, ", world", SIZE - strlen (buffer) - 1);
      puts (buffer);
    }
    

    The output produced by this program looks like:

    hello
    hello, wo
    

    Function: void bcopy (const void *from, void *to, size_t size)
    This is a partially obsolete alternative for memmove, derived from BSD. Note that it is not quite equivalent to memmove, because the arguments are not in the same order and there is no return value.

    Function: void bzero (void *block, size_t size)
    This is a partially obsolete alternative for memset, derived from BSD. Note that it is not as general as memset, because the only value it can store is zero.

    String/Array Comparison

    You can use the functions in this section to perform comparisons on the contents of strings and arrays. As well as checking for equality, these functions can also be used as the ordering functions for sorting operations. See section Searching and Sorting, for an example of this.

    Unlike most comparison operations in C, the string comparison functions return a nonzero value if the strings are not equivalent rather than if they are. The sign of the value indicates the relative ordering of the first characters in the strings that are not equivalent: a negative value indicates that the first string is "less" than the second, while a positive value indicates that the first string is "greater".

    The most common use of these functions is to check only for equality. This is canonically done with an expression like `! strcmp (s1, s2)'.

    All of these functions are declared in the header file `string.h'.

    Function: int memcmp (const void *a1, const void *a2, size_t size)
    The function memcmp compares the size bytes of memory beginning at a1 against the size bytes of memory beginning at a2. The value returned has the same sign as the difference between the first differing pair of bytes (interpreted as unsigned char objects, then promoted to int).

    If the contents of the two blocks are equal, memcmp returns 0.

    Function: int wmemcmp (const wchar_t *a1, const wchar_t *a2, size_t size)
    The function wmemcmp compares the size wide characters beginning at a1 against the size wide characters beginning at a2. The value returned is smaller than or larger than zero depending on whether the first differing wide character is a1 is smaller or larger than the corresponding character in a2.

    If the contents of the two blocks are equal, wmemcmp returns 0.

    On arbitrary arrays, the memcmp function is mostly useful for testing equality. It usually isn't meaningful to do byte-wise ordering comparisons on arrays of things other than bytes. For example, a byte-wise comparison on the bytes that make up floating-point numbers isn't likely to tell you anything about the relationship between the values of the floating-point numbers.

    wmemcmp is really only useful to compare arrays of type wchar_t since the function looks at sizeof (wchar_t) bytes at a time and this number of bytes is system dependent.

    You should also be careful about using memcmp to compare objects that can contain "holes", such as the padding inserted into structure objects to enforce alignment requirements, extra space at the end of unions, and extra characters at the ends of strings whose length is less than their allocated size. The contents of these "holes" are indeterminate and may cause strange behavior when performing byte-wise comparisons. For more predictable results, perform an explicit component-wise comparison.

    For example, given a structure type definition like:

    struct foo
      {
        unsigned char tag;
        union
          {
            double f;
            long i;
            char *p;
          } value;
      };
    

    you are better off writing a specialized comparison function to compare struct foo objects instead of comparing them with memcmp.

    Function: int strcmp (const char *s1, const char *s2)
    The strcmp function compares the string s1 against s2, returning a value that has the same sign as the difference between the first differing pair of characters (interpreted as unsigned char objects, then promoted to int).

    If the two strings are equal, strcmp returns 0.

    A consequence of the ordering used by strcmp is that if s1 is an initial substring of s2, then s1 is considered to be "less than" s2.

    strcmp does not take sorting conventions of the language the strings are written in into account. To get that one has to use strcoll.

    Function: int wcscmp (const wchar_t *ws1, const wchar_t *ws2)

    The wcscmp function compares the wide character string ws1 against ws2. The value returned is smaller than or larger than zero depending on whether the first differing wide character is ws1 is smaller or larger than the corresponding character in ws2.

    If the two strings are equal, wcscmp returns 0.

    A consequence of the ordering used by wcscmp is that if ws1 is an initial substring of ws2, then ws1 is considered to be "less than" ws2.

    wcscmp does not take sorting conventions of the language the strings are written in into account. To get that one has to use wcscoll.

    Function: int strcasecmp (const char *s1, const char *s2)
    This function is like strcmp, except that differences in case are ignored. How uppercase and lowercase characters are related is determined by the currently selected locale. In the standard "C" locale the characters @"A and @"a do not match but in a locale which regards these characters as parts of the alphabet they do match.

    strcasecmp is derived from BSD.

    Function: int wcscasecmp (const wchar_t *ws1, const wchar_T *ws2)
    This function is like wcscmp, except that differences in case are ignored. How uppercase and lowercase characters are related is determined by the currently selected locale. In the standard "C" locale the characters @"A and @"a do not match but in a locale which regards these characters as parts of the alphabet they do match.

    wcscasecmp is a GNU extension.

    Function: int strncmp (const char *s1, const char *s2, size_t size)
    This function is the similar to strcmp, except that no more than size wide characters are compared. In other words, if the two strings are the same in their first size wide characters, the return value is zero.

    Function: int wcsncmp (const wchar_t *ws1, const wchar_t *ws2, size_t size)
    This function is the similar to wcscmp, except that no more than size wide characters are compared. In other words, if the two strings are the same in their first size wide characters, the return value is zero.

    Function: int strncasecmp (const char *s1, const char *s2, size_t n)
    This function is like strncmp, except that differences in case are ignored. Like strcasecmp, it is locale dependent how uppercase and lowercase characters are related.

    strncasecmp is a GNU extension.

    Function: int wcsncasecmp (const wchar_t *ws1, const wchar_t *s2, size_t n)
    This function is like wcsncmp, except that differences in case are ignored. Like wcscasecmp, it is locale dependent how uppercase and lowercase characters are related.

    wcsncasecmp is a GNU extension.

    Here are some examples showing the use of strcmp and strncmp (equivalent examples can be constructed for the wide character functions). These examples assume the use of the ASCII character set. (If some other character set--say, EBCDIC--is used instead, then the glyphs are associated with different numeric codes, and the return values and ordering may differ.)

    strcmp ("hello", "hello")
        => 0    /* These two strings are the same. */
    strcmp ("hello", "Hello")
        => 32   /* Comparisons are case-sensitive. */
    strcmp ("hello", "world")
        => -15  /* The character 'h' comes before 'w'. */
    strcmp ("hello", "hello, world")
        => -44  /* Comparing a null character against a comma. */
    strncmp ("hello", "hello, world", 5)
        => 0    /* The initial 5 characters are the same. */
    strncmp ("hello, world", "hello, stupid world!!!", 5)
        => 0    /* The initial 5 characters are the same. */
    

    Function: int strverscmp (const char *s1, const char *s2)
    The strverscmp function compares the string s1 against s2, considering them as holding indices/version numbers. Return value follows the same conventions as found in the strverscmp function. In fact, if s1 and s2 contain no digits, strverscmp behaves like strcmp.

    Basically, we compare strings normally (character by character), until we find a digit in each string - then we enter a special comparison mode, where each sequence of digits is taken as a whole. If we reach the end of these two parts without noticing a difference, we return to the standard comparison mode. There are two types of numeric parts: "integral" and "fractional" (those begin with a '0'). The types of the numeric parts affect the way we sort them:

    strverscmp ("no digit", "no digit")
        => 0    /* same behaviour as strcmp. */
    strverscmp ("item#99", "item#100")
        => <0   /* same prefix, but 99 < 100. */
    strverscmp ("alpha1", "alpha001")
        => >0   /* fractional part inferior to integral one. */
    strverscmp ("part1_f012", "part1_f01")
        => >0   /* two fractional parts. */
    strverscmp ("foo.009", "foo.0")
        => <0   /* idem, but with leading zeroes only. */
    

    This function is especially useful when dealing with filename sorting, because filenames frequently hold indices/version numbers.

    strverscmp is a GNU extension.

    Function: int bcmp (const void *a1, const void *a2, size_t size)
    This is an obsolete alias for memcmp, derived from BSD.

    Collation Functions

    In some locales, the conventions for lexicographic ordering differ from the strict numeric ordering of character codes. For example, in Spanish most glyphs with diacritical marks such as accents are not considered distinct letters for the purposes of collation. On the other hand, the two-character sequence `ll' is treated as a single letter that is collated immediately after `l'.

    You can use the functions strcoll and strxfrm (declared in the headers file `string.h') and wcscoll and wcsxfrm (declared in the headers file `wchar') to compare strings using a collation ordering appropriate for the current locale. The locale used by these functions in particular can be specified by setting the locale for the LC_COLLATE category; see section Locales and Internationalization.

    In the standard C locale, the collation sequence for strcoll is the same as that for strcmp. Similarly, wcscoll and wcscmp are the same in this situation.

    Effectively, the way these functions work is by applying a mapping to transform the characters in a string to a byte sequence that represents the string's position in the collating sequence of the current locale. Comparing two such byte sequences in a simple fashion is equivalent to comparing the strings with the locale's collating sequence.

    The functions strcoll and wcscoll perform this translation implicitly, in order to do one comparison. By contrast, strxfrm and wcsxfrm perform the mapping explicitly. If you are making multiple comparisons using the same string or set of strings, it is likely to be more efficient to use strxfrm or wcsxfrm to transform all the strings just once, and subsequently compare the transformed strings with strcmp or wcscmp.

    Function: int strcoll (const char *s1, const char *s2)
    The strcoll function is similar to strcmp but uses the collating sequence of the current locale for collation (the LC_COLLATE locale).

    Function: int wcscoll (const wchar_t *ws1, const wchar_t *ws2)
    The wcscoll function is similar to wcscmp but uses the collating sequence of the current locale for collation (the LC_COLLATE locale).

    Here is an example of sorting an array of strings, using strcoll to compare them. The actual sort algorithm is not written here; it comes from qsort (see section Array Sort Function). The job of the code shown here is to say how to compare the strings while sorting them. (Later on in this section, we will show a way to do this more efficiently using strxfrm.)

    /* This is the comparison function used with qsort. */
    
    int
    compare_elements (char **p1, char **p2)
    {
      return strcoll (*p1, *p2);
    }
    
    /* This is the entry point--the function to sort
       strings using the locale's collating sequence. */
    
    void
    sort_strings (char **array, int nstrings)
    {
      /* Sort temp_array by comparing the strings. */
      qsort (array, nstrings,
             sizeof (char *), compare_elements);
    }
    

    Function: size_t strxfrm (char *restrict to, const char *restrict from, size_t size)
    The function strxfrm transforms the string from using the collation transformation determined by the locale currently selected for collation, and stores the transformed string in the array to. Up to size characters (including a terminating null character) are stored.

    The behavior is undefined if the strings to and from overlap; see section Copying and Concatenation.

    The return value is the length of the entire transformed string. This value is not affected by the value of size, but if it is greater or equal than size, it means that the transformed string did not entirely fit in the array to. In this case, only as much of the string as actually fits was stored. To get the whole transformed string, call strxfrm again with a bigger output array.

    The transformed string may be longer than the original string, and it may also be shorter.

    If size is zero, no characters are stored in to. In this case, strxfrm simply returns the number of characters that would be the length of the transformed string. This is useful for determining what size the allocated array should be. It does not matter what to is if size is zero; to may even be a null pointer.

    Function: size_t wcsxfrm (wchar_t *restrict wto, const wchar_t *wfrom, size_t size)
    The function wcsxfrm transforms wide character string wfrom using the collation transformation determined by the locale currently selected for collation, and stores the transformed string in the array wto. Up to size wide characters (including a terminating null character) are stored.

    The behavior is undefined if the strings wto and wfrom overlap; see section Copying and Concatenation.

    The return value is the length of the entire transformed wide character string. This value is not affected by the value of size, but if it is greater or equal than size, it means that the transformed wide character string did not entirely fit in the array wto. In this case, only as much of the wide character string as actually fits was stored. To get the whole transformed wide character string, call wcsxfrm again with a bigger output array.

    The transformed wide character string may be longer than the original wide character string, and it may also be shorter.

    If size is zero, no characters are stored in to. In this case, wcsxfrm simply returns the number of wide characters that would be the length of the transformed wide character string. This is useful for determining what size the allocated array should be (remember to multiply with sizeof (wchar_t)). It does not matter what wto is if size is zero; wto may even be a null pointer.

    Here is an example of how you can use strxfrm when you plan to do many comparisons. It does the same thing as the previous example, but much faster, because it has to transform each string only once, no matter how many times it is compared with other strings. Even the time needed to allocate and free storage is much less than the time we save, when there are many strings.

    struct sorter { char *input; char *transformed; };
    
    /* This is the comparison function used with qsort
       to sort an array of struct sorter. */
    
    int
    compare_elements (struct sorter *p1, struct sorter *p2)
    {
      return strcmp (p1->transformed, p2->transformed);
    }
    
    /* This is the entry point--the function to sort
       strings using the locale's collating sequence. */
    
    void
    sort_strings_fast (char **array, int nstrings)
    {
      struct sorter temp_array[nstrings];
      int i;
    
      /* Set up temp_array.  Each element contains
         one input string and its transformed string. */
      for (i = 0; i < nstrings; i++)
        {
          size_t length = strlen (array[i]) * 2;
          char *transformed;
          size_t transformed_length;
    
          temp_array[i].input = array[i];
    
          /* First try a buffer perhaps big enough.  */
          transformed = (char *) xmalloc (length);
    
          /* Transform array[i].  */
          transformed_length = strxfrm (transformed, array[i], length);
    
          /* If the buffer was not large enough, resize it
             and try again.  */
          if (transformed_length >= length)
            {
              /* Allocate the needed space. +1 for terminating
                 NUL character.  */
              transformed = (char *) xrealloc (transformed,
                                               transformed_length + 1);
    
              /* The return value is not interesting because we know
                 how long the transformed string is.  */
              (void) strxfrm (transformed, array[i],
                              transformed_length + 1);
            }
    
          temp_array[i].transformed = transformed;
        }
    
      /* Sort temp_array by comparing transformed strings. */
      qsort (temp_array, sizeof (struct sorter),
             nstrings, compare_elements);
    
      /* Put the elements back in the permanent array
         in their sorted order. */
      for (i = 0; i < nstrings; i++)
        array[i] = temp_array[i].input;
    
      /* Free the strings we allocated. */
      for (i = 0; i < nstrings; i++)
        free (temp_array[i].transformed);
    }
    

    The interesting part of this code for the wide character version would look like this:

    void
    sort_strings_fast (wchar_t **array, int nstrings)
    {
      ...
          /* Transform array[i].  */
          transformed_length = wcsxfrm (transformed, array[i], length);
    
          /* If the buffer was not large enough, resize it
             and try again.  */
          if (transformed_length >= length)
            {
              /* Allocate the needed space. +1 for terminating
                 NUL character.  */
              transformed = (wchar_t *) xrealloc (transformed,
                                                  (transformed_length + 1)
                                                  * sizeof (wchar_t));
    
              /* The return value is not interesting because we know
                 how long the transformed string is.  */
              (void) wcsxfrm (transformed, array[i],
                              transformed_length + 1);
            }
      ...
    

    Note the additional multiplication with sizeof (wchar_t) in the realloc call.

    Compatibility Note: The string collation functions are a new feature of ISO C90. Older C dialects have no equivalent feature. The wide character versions were introduced in Amendment 1 to ISO C90.

    Search Functions

    This section describes library functions which perform various kinds of searching operations on strings and arrays. These functions are declared in the header file `string.h'.

    Function: void * memchr (const void *block, int c, size_t size)
    This function finds the first occurrence of the byte c (converted to an unsigned char) in the initial size bytes of the object beginning at block. The return value is a pointer to the located byte, or a null pointer if no match was found.

    Function: wchar_t * wmemchr (const wchar_t *block, wchar_t wc, size_t size)
    This function finds the first occurrence of the wide character wc in the initial size wide characters of the object beginning at block. The return value is a pointer to the located wide character, or a null pointer if no match was found.

    Function: void * rawmemchr (const void *block, int c)
    Often the memchr function is used with the knowledge that the byte c is available in the memory block specified by the parameters. But this means that the size parameter is not really needed and that the tests performed with it at runtime (to check whether the end of the block is reached) are not needed.

    The rawmemchr function exists for just this situation which is surprisingly frequent. The interface is similar to memchr except that the size parameter is missing. The function will look beyond the end of the block pointed to by block in case the programmer made an error in assuming that the byte c is present in the block. In this case the result is unspecified. Otherwise the return value is a pointer to the located byte.

    This function is of special interest when looking for the end of a string. Since all strings are terminated by a null byte a call like

       rawmemchr (str, '\0')
    

    will never go beyond the end of the string.

    This function is a GNU extension.

    Function: void * memrchr (const void *block, int c, size_t size)
    The function memrchr is like memchr, except that it searches backwards from the end of the block defined by block and size (instead of forwards from the front).

    Function: char * strchr (const char *string, int c)
    The strchr function finds the first occurrence of the character c (converted to a char) in the null-terminated string beginning at string. The return value is a pointer to the located character, or a null pointer if no match was found.

    For example,

    strchr ("hello, world", 'l')
        => "llo, world"
    strchr ("hello, world", '?')
        => NULL
    

    The terminating null character is considered to be part of the string, so you can use this function get a pointer to the end of a string by specifying a null character as the value of the c argument. It would be better (but less portable) to use strchrnul in this case, though.

    Function: wchar_t * wcschr (const wchar_t *wstring, int wc)
    The wcschr function finds the first occurrence of the wide character wc in the null-terminated wide character string beginning at wstring. The return value is a pointer to the located wide character, or a null pointer if no match was found.

    The terminating null character is considered to be part of the wide character string, so you can use this function get a pointer to the end of a wide character string by specifying a null wude character as the value of the wc argument. It would be better (but less portable) to use wcschrnul in this case, though.

    Function: char * strchrnul (const char *string, int c)
    strchrnul is the same as strchr except that if it does not find the character, it returns a pointer to string's terminating null character rather than a null pointer.

    This function is a GNU extension.

    Function: wchar_t * wcschrnul (const wchar_t *wstring, wchar_t wc)
    wcschrnul is the same as wcschr except that if it does not find the wide character, it returns a pointer to wide character string's terminating null wide character rather than a null pointer.

    This function is a GNU extension.

    One useful, but unusual, use of the strchr function is when one wants to have a pointer pointing to the NUL byte terminating a string. This is often written in this way:

      s += strlen (s);
    

    This is almost optimal but the addition operation duplicated a bit of the work already done in the strlen function. A better solution is this:

      s = strchr (s, '\0');
    

    There is no restriction on the second parameter of strchr so it could very well also be the NUL character. Those readers thinking very hard about this might now point out that the strchr function is more expensive than the strlen function since we have two abort criteria. This is right. But in the GNU C library the implementation of strchr is optimized in a special way so that strchr actually is faster.

    Function: char * strrchr (const char *string, int c)
    The function strrchr is like strchr, except that it searches backwards from the end of the string string (instead of forwards from the front).

    For example,

    strrchr ("hello, world", 'l')
        => "ld"
    

    Function: wchar_t * wcsrchr (const wchar_t *wstring, wchar_t c)
    The function wcsrchr is like wcschr, except that it searches backwards from the end of the string wstring (instead of forwards from the front).

    Function: char * strstr (const char *haystack, const char *needle)
    This is like strchr, except that it searches haystack for a substring needle rather than just a single character. It returns a pointer into the string haystack that is the first character of the substring, or a null pointer if no match was found. If needle is an empty string, the function returns haystack.

    For example,

    strstr ("hello, world", "l")
        => "llo, world"
    strstr ("hello, world", "wo")
        => "world"
    

    Function: wchar_t * wcsstr (const wchar_t *haystack, const wchar_t *needle)
    This is like wcschr, except that it searches haystack for a substring needle rather than just a single wide character. It returns a pointer into the string haystack that is the first wide character of the substring, or a null pointer if no match was found. If needle is an empty string, the function returns haystack.

    Function: wchar_t * wcswcs (const wchar_t *haystack, const wchar_t *needle)
    wcsstr is an depricated alias for wcsstr. This is the name originally used in the X/Open Portability Guide before the Amendment 1 to ISO C90 was published.

    Function: char * strcasestr (const char *haystack, const char *needle)
    This is like strstr, except that it ignores case in searching for the substring. Like strcasecmp, it is locale dependent how uppercase and lowercase characters are related.

    For example,

    strstr ("hello, world", "L")
        => "llo, world"
    strstr ("hello, World", "wo")
        => "World"
    

    Function: void * memmem (const void *haystack, size_t haystack-len,
    const void *needle, size_t needle-len)
    This is like strstr, but needle and haystack are byte arrays rather than null-terminated strings. needle-len is the length of needle and haystack-len is the length of haystack.

    This function is a GNU extension.

    Function: size_t strspn (const char *string, const char *skipset)
    The strspn ("string span") function returns the length of the initial substring of string that consists entirely of characters that are members of the set specified by the string skipset. The order of the characters in skipset is not important.

    For example,

    strspn ("hello, world", "abcdefghijklmnopqrstuvwxyz")
        => 5
    

    Note that "character" is here used in the sense of byte. In a string using a multibyte character encoding (abstract) character consisting of more than one byte are not treated as an entity. Each byte is treated separately. The function is not locale-dependent.

    Function: size_t wcsspn (const wchar_t *wstring, const wchar_t *skipset)
    The wcsspn ("wide character string span") function returns the length of the initial substring of wstring that consists entirely of wide characters that are members of the set specified by the string skipset. The order of the wide characters in skipset is not important.

    Function: size_t strcspn (const char *string, const char *stopset)
    The strcspn ("string complement span") function returns the length of the initial substring of string that consists entirely of characters that are not members of the set specified by the string stopset. (In other words, it returns the offset of the first character in string that is a member of the set stopset.)

    For example,

    strcspn ("hello, world", " \t\n,.;!?")
        => 5
    

    Note that "character" is here used in the sense of byte. In a string using a multibyte character encoding (abstract) character consisting of more than one byte are not treated as an entity. Each byte is treated separately. The function is not locale-dependent.

    Function: size_t wcscspn (const wchar_t *wstring, const wchar_t *stopset)
    The wcscspn ("wide character string complement span") function returns the length of the initial substring of wstring that consists entirely of wide characters that are not members of the set specified by the string stopset. (In other words, it returns the offset of the first character in string that is a member of the set stopset.)

    Function: char * strpbrk (const char *string, const char *stopset)
    The strpbrk ("string pointer break") function is related to strcspn, except that it returns a pointer to the first character in string that is a member of the set stopset instead of the length of the initial substring. It returns a null pointer if no such character from stopset is found.

    For example,

    strpbrk ("hello, world", " \t\n,.;!?")
        => ", world"
    

    Note that "character" is here used in the sense of byte. In a string using a multibyte character encoding (abstract) character consisting of more than one byte are not treated as an entity. Each byte is treated separately. The function is not locale-dependent.

    Function: wchar_t * wcspbrk (const wchar_t *wstring, const wchar_t *stopset)
    The wcspbrk ("wide character string pointer break") function is related to wcscspn, except that it returns a pointer to the first wide character in wstring that is a member of the set stopset instead of the length of the initial substring. It returns a null pointer if no such character from stopset is found.

    Compatibility String Search Functions

    Function: char * index (const char *string, int c)
    index is another name for strchr; they are exactly the same. New code should always use strchr since this name is defined in ISO C while index is a BSD invention which never was available on System V derived systems.

    Function: char * rindex (const char *string, int c)
    rindex is another name for strrchr; they are exactly the same. New code should always use strrchr since this name is defined in ISO C while rindex is a BSD invention which never was available on System V derived systems.

    Finding Tokens in a String

    It's fairly common for programs to have a need to do some simple kinds of lexical analysis and parsing, such as splitting a command string up into tokens. You can do this with the strtok function, declared in the header file `string.h'.

    Function: char * strtok (char *restrict newstring, const char *restrict delimiters)
    A string can be split into tokens by making a series of calls to the function strtok.

    The string to be split up is passed as the newstring argument on the first call only. The strtok function uses this to set up some internal state information. Subsequent calls to get additional tokens from the same string are indicated by passing a null pointer as the newstring argument. Calling strtok with another non-null newstring argument reinitializes the state information. It is guaranteed that no other library function ever calls strtok behind your back (which would mess up this internal state information).

    The delimiters argument is a string that specifies a set of delimiters that may surround the token being extracted. All the initial characters that are members of this set are discarded. The first character that is not a member of this set of delimiters marks the beginning of the next token. The end of the token is found by looking for the next character that is a member of the delimiter set. This character in the original string newstring is overwritten by a null character, and the pointer to the beginning of the token in newstring is returned.

    On the next call to strtok, the searching begins at the next character beyond the one that marked the end of the previous token. Note that the set of delimiters delimiters do not have to be the same on every call in a series of calls to strtok.

    If the end of the string newstring is reached, or if the remainder of string consists only of delimiter characters, strtok returns a null pointer.

    Note that "character" is here used in the sense of byte. In a string using a multibyte character encoding (abstract) character consisting of more than one byte are not treated as an entity. Each byte is treated separately. The function is not locale-dependent.

    Note that "character" is here used in the sense of byte. In a string using a multibyte character encoding (abstract) character consisting of more than one byte are not treated as an entity. Each byte is treated separately. The function is not locale-dependent.

    Function: wchar_t * wcstok (wchar_t *newstring, const char *delimiters)
    A string can be split into tokens by making a series of calls to the function wcstok.

    The string to be split up is passed as the newstring argument on the first call only. The wcstok function uses this to set up some internal state information. Subsequent calls to get additional tokens from the same wide character string are indicated by passing a null pointer as the newstring argument. Calling wcstok with another non-null newstring argument reinitializes the state information. It is guaranteed that no other library function ever calls wcstok behind your back (which would mess up this internal state information).

    The delimiters argument is a wide character string that specifies a set of delimiters that may surround the token being extracted. All the initial wide characters that are members of this set are discarded. The first wide character that is not a member of this set of delimiters marks the beginning of the next token. The end of the token is found by looking for the next wide character that is a member of the delimiter set. This wide character in the original wide character string newstring is overwritten by a null wide character, and the pointer to the beginning of the token in newstring is returned.

    On the next call to wcstok, the searching begins at the next wide character beyond the one that marked the end of the previous token. Note that the set of delimiters delimiters do not have to be the same on every call in a series of calls to wcstok.

    If the end of the wide character string newstring is reached, or if the remainder of string consists only of delimiter wide characters, wcstok returns a null pointer.

    Note that "character" is here used in the sense of byte. In a string using a multibyte character encoding (abstract) character consisting of more than one byte are not treated as an entity. Each byte is treated separately. The function is not locale-dependent.

    Warning: Since strtok and wcstok alter the string they is parsing, you should always copy the string to a temporary buffer before parsing it with strtok/wcstok (see section Copying and Concatenation). If you allow strtok or wcstok to modify a string that came from another part of your program, you are asking for trouble; that string might be used for other purposes after strtok or wcstok has modified it, and it would not have the expected value.

    The string that you are operating on might even be a constant. Then when strtok or wcstok tries to modify it, your program will get a fatal signal for writing in read-only memory. See section Program Error Signals. Even if the operation of strtok or wcstok would not require a modification of the string (e.g., if there is exactly one token) the string can (and in the GNU libc case will) be modified.

    This is a special case of a general principle: if a part of a program does not have as its purpose the modification of a certain data structure, then it is error-prone to modify the data structure temporarily.

    The functions strtok and wcstok are not reentrant. See section Signal Handling and Nonreentrant Functions, for a discussion of where and why reentrancy is important.

    Here is a simple example showing the use of strtok.

    #include <string.h>
    #include <stddef.h>
    
    ...
    
    const char string[] = "words separated by spaces -- and, punctuation!";
    const char delimiters[] = " .,;:!-";
    char *token, *cp;
    
    ...
    
    cp = strdupa (string);                /* Make writable copy.  */
    token = strtok (cp, delimiters);      /* token => "words" */
    token = strtok (NULL, delimiters);    /* token => "separated" */
    token = strtok (NULL, delimiters);    /* token => "by" */
    token = strtok (NULL, delimiters);    /* token => "spaces" */
    token = strtok (NULL, delimiters);    /* token => "and" */
    token = strtok (NULL, delimiters);    /* token => "punctuation" */
    token = strtok (NULL, delimiters);    /* token => NULL */
    

    The GNU C library contains two more functions for tokenizing a string which overcome the limitation of non-reentrancy. They are only available for multibyte character strings.

    Function: char * strtok_r (char *newstring, const char *delimiters, char **save_ptr)
    Just like strtok, this function splits the string into several tokens which can be accessed by successive calls to strtok_r. The difference is that the information about the next token is stored in the space pointed to by the third argument, save_ptr, which is a pointer to a string pointer. Calling strtok_r with a null pointer for newstring and leaving save_ptr between the calls unchanged does the job without hindering reentrancy.

    This function is defined in POSIX.1 and can be found on many systems which support multi-threading.

    Function: char * strsep (char **string_ptr, const char *delimiter)
    This function has a similar functionality as strtok_r with the newstring argument replaced by the save_ptr argument. The initialization of the moving pointer has to be done by the user. Successive calls to strsep move the pointer along the tokens separated by delimiter, returning the address of the next token and updating string_ptr to point to the beginning of the next token.

    One difference between strsep and strtok_r is that if the input string contains more than one character from delimiter in a row strsep returns an empty string for each pair of characters from delimiter. This means that a program normally should test for strsep returning an empty string before processing it.

    This function was introduced in 4.3BSD and therefore is widely available.

    Here is how the above example looks like when strsep is used.

    #include <string.h>
    #include <stddef.h>
    
    ...
    
    const char string[] = "words separated by spaces -- and, punctuation!";
    const char delimiters[] = " .,;:!-";
    char *running;
    char *token;
    
    ...
    
    running = strdupa (string);
    token = strsep (&running, delimiters);    /* token => "words" */
    token = strsep (&running, delimiters);    /* token => "separated" */
    token = strsep (&running, delimiters);    /* token => "by" */
    token = strsep (&running, delimiters);    /* token => "spaces" */
    token = strsep (&running, delimiters);    /* token => "" */
    token = strsep (&running, delimiters);    /* token => "" */
    token = strsep (&running, delimiters);    /* token => "" */
    token = strsep (&running, delimiters);    /* token => "and" */
    token = strsep (&running, delimiters);    /* token => "" */
    token = strsep (&running, delimiters);    /* token => "punctuation" */
    token = strsep (&running, delimiters);    /* token => "" */
    token = strsep (&running, delimiters);    /* token => NULL */
    

    Function: char * basename (const char *filename)
    The GNU version of the basename function returns the last component of the path in filename. This function is the prefered usage, since it does not modify the argument, filename, and respects trailing slashes. The prototype for basename can be found in `string.h'. Note, this function is overriden by the XPG version, if `libgen.h' is included.

    Example of using GNU basename:

    #include <string.h>
    
    int
    main (int argc, char *argv[])
    {
      char *prog = basename (argv[0]);
    
      if (argc < 2)
        {
          fprintf (stderr, "Usage %s <arg>\n", prog);
          exit (1);
        }
    
      ...
    }
    

    Portability Note: This function may produce different results on different systems.

    Function: char * basename (char *path)
    This is the standard XPG defined basename. It is similar in spirit to the GNU version, but may modify the path by removing trailing '/' characters. If the path is made up entirely of '/' characters, then "/" will be returned. Also, if path is NULL or an empty string, then "." is returned. The prototype for the XPG version can be found in `libgen.h'.

    Example of using XPG basename:

    #include <libgen.h>
    
    int
    main (int argc, char *argv[])
    {
      char *prog;
      char *path = strdupa (argv[0]);
    
      prog = basename (path);
    
      if (argc < 2)
        {
          fprintf (stderr, "Usage %s <arg>\n", prog);
          exit (1);
        }
    
      ...
    
    }
    

    Function: char * dirname (char *path)
    The dirname function is the compliment to the XPG version of basename. It returns the parent directory of the file specified by path. If path is NULL, an empty string, or contains no '/' characters, then "." is returned. The prototype for this function can be found in `libgen.h'.

    strfry

    The function below addresses the perennial programming quandary: "How do I take good data in string form and painlessly turn it into garbage?" This is actually a fairly simple task for C programmers who do not use the GNU C library string functions, but for programs based on the GNU C library, the strfry function is the preferred method for destroying string data.

    The prototype for this function is in `string.h'.

    Function: char * strfry (char *string)

    strfry creates a pseudorandom anagram of a string, replacing the input with the anagram in place. For each position in the string, strfry swaps it with a position in the string selected at random (from a uniform distribution). The two positions may be the same.

    The return value of strfry is always string.

    Portability Note: This function is unique to the GNU C library.

    Trivial Encryption

    The memfrob function converts an array of data to something unrecognizable and back again. It is not encryption in its usual sense since it is easy for someone to convert the encrypted data back to clear text. The transformation is analogous to Usenet's "Rot13" encryption method for obscuring offensive jokes from sensitive eyes and such. Unlike Rot13, memfrob works on arbitrary binary data, not just text.

    For true encryption, See section DES Encryption and Password Handling.

    This function is declared in `string.h'.

    Function: void * memfrob (void *mem, size_t length)

    memfrob transforms (frobnicates) each byte of the data structure at mem, which is length bytes long, by bitwise exclusive oring it with binary 00101010. It does the transformation in place and its return value is always mem.

    Note that memfrob a second time on the same data structure returns it to its original state.

    This is a good function for hiding information from someone who doesn't want to see it or doesn't want to see it very much. To really prevent people from retrieving the information, use stronger encryption such as that described in See section DES Encryption and Password Handling.

    Portability Note: This function is unique to the GNU C library.

    Encode Binary Data

    To store or transfer binary data in environments which only support text one has to encode the binary data by mapping the input bytes to characters in the range allowed for storing or transfering. SVID systems (and nowadays XPG compliant systems) provide minimal support for this task.

    Function: char * l64a (long int n)
    This function encodes a 32-bit input value using characters from the basic character set. It returns a pointer to a 6 character buffer which contains an encoded version of n. To encode a series of bytes the user must copy the returned string to a destination buffer. It returns the empty string if n is zero, which is somewhat bizarre but mandated by the standard.
    Warning: Since a static buffer is used this function should not be used in multi-threaded programs. There is no thread-safe alternative to this function in the C library.
    Compatibility Note: The XPG standard states that the return value of l64a is undefined if n is negative. In the GNU implementation, l64a treats its argument as unsigned, so it will return a sensible encoding for any nonzero n; however, portable programs should not rely on this.

    To encode a large buffer l64a must be called in a loop, once for each 32-bit word of the buffer. For example, one could do something like this:

    char *
    encode (const void *buf, size_t len)
    {
      /* We know in advance how long the buffer has to be. */
      unsigned char *in = (unsigned char *) buf;
      char *out = malloc (6 + ((len + 3) / 4) * 6 + 1);
      char *cp = out;
    
      /* Encode the length. */
      /* Using `htonl' is necessary so that the data can be
         decoded even on machines with different byte order. */
    
      cp = mempcpy (cp, l64a (htonl (len)), 6);
    
      while (len > 3)
        {
          unsigned long int n = *in++;
          n = (n << 8) | *in++;
          n = (n << 8) | *in++;
          n = (n << 8) | *in++;
          len -= 4;
          if (n)
            cp = mempcpy (cp, l64a (htonl (n)), 6);
          else
                /* `l64a' returns the empty string for n==0, so we 
                   must generate its encoding ("......") by hand. */
            cp = stpcpy (cp, "......");
        }
      if (len > 0)
        {
          unsigned long int n = *in++;
          if (--len > 0)
            {
              n = (n << 8) | *in++;
              if (--len > 0)
                n = (n << 8) | *in;
            }
          memcpy (cp, l64a (htonl (n)), 6);
          cp += 6;
        }
      *cp = '\0';
      return out;
    }
    

    It is strange that the library does not provide the complete functionality needed but so be it.

    To decode data produced with l64a the following function should be used.

    Function: long int a64l (const char *string)
    The parameter string should contain a string which was produced by a call to l64a. The function processes at least 6 characters of this string, and decodes the characters it finds according to the table below. It stops decoding when it finds a character not in the table, rather like atoi; if you have a buffer which has been broken into lines, you must be careful to skip over the end-of-line characters.

    The decoded number is returned as a long int value.

    The l64a and a64l functions use a base 64 encoding, in which each character of an encoded string represents six bits of an input word. These symbols are used for the base 64 digits:

    @multitable {xxxxx} {xxx} {xxx} {xxx} {xxx} {xxx} {xxx} {xxx} {xxx}

  • @tab 0 @tab 1 @tab 2 @tab 3 @tab 4 @tab 5 @tab 6 @tab 7
  • 0 @tab . @tab / @tab 0 @tab 1 @tab 2 @tab 3 @tab 4 @tab 5
  • 8 @tab 6 @tab 7 @tab 8 @tab 9 @tab A @tab B @tab C @tab D
  • 16 @tab E @tab F @tab G @tab H @tab I @tab J @tab K @tab L
  • 24 @tab M @tab N @tab O @tab P @tab Q @tab R @tab S @tab T
  • 32 @tab U @tab V @tab W @tab X @tab Y @tab Z @tab a @tab b
  • 40 @tab c @tab d @tab e @tab f @tab g @tab h @tab i @tab j
  • 48 @tab k @tab l @tab m @tab n @tab o @tab p @tab q @tab r
  • 56 @tab s @tab t @tab u @tab v @tab w @tab x @tab y @tab z This encoding scheme is not standard. There are some other encoding methods which are much more widely used (UU encoding, MIME encoding). Generally, it is better to use one of these encodings.

    Argz and Envz Vectors

    argz vectors are vectors of strings in a contiguous block of memory, each element separated from its neighbors by null-characters ('\0').

    Envz vectors are an extension of argz vectors where each element is a name-value pair, separated by a '=' character (as in a Unix environment).

    Argz Functions

    Each argz vector is represented by a pointer to the first element, of type char *, and a size, of type size_t, both of which can be initialized to 0 to represent an empty argz vector. All argz functions accept either a pointer and a size argument, or pointers to them, if they will be modified.

    The argz functions use malloc/realloc to allocate/grow argz vectors, and so any argz vector creating using these functions may be freed by using free; conversely, any argz function that may grow a string expects that string to have been allocated using malloc (those argz functions that only examine their arguments or modify them in place will work on any sort of memory). See section Unconstrained Allocation.

    All argz functions that do memory allocation have a return type of error_t, and return 0 for success, and ENOMEM if an allocation error occurs.

    These functions are declared in the standard include file `argz.h'.

    Function: error_t argz_create (char *const argv[], char **argz, size_t *argz_len)
    The argz_create function converts the Unix-style argument vector argv (a vector of pointers to normal C strings, terminated by (char *)0; see section Program Arguments) into an argz vector with the same elements, which is returned in argz and argz_len.

    Function: error_t argz_create_sep (const char *string, int sep, char **argz, size_t *argz_len)
    The argz_create_sep function converts the null-terminated string string into an argz vector (returned in argz and argz_len) by splitting it into elements at every occurrence of the character sep.

    Function: size_t argz_count (const char *argz, size_t arg_len)
    Returns the number of elements in the argz vector argz and argz_len.

    Function: void argz_extract (char *argz, size_t argz_len, char **argv)
    The argz_extract function converts the argz vector argz and argz_len into a Unix-style argument vector stored in argv, by putting pointers to every element in argz into successive positions in argv, followed by a terminator of 0. Argv must be pre-allocated with enough space to hold all the elements in argz plus the terminating (char *)0 ((argz_count (argz, argz_len) + 1) * sizeof (char *) bytes should be enough). Note that the string pointers stored into argv point into argz---they are not copies--and so argz must be copied if it will be changed while argv is still active. This function is useful for passing the elements in argz to an exec function (see section Executing a File).

    Function: void argz_stringify (char *argz, size_t len, int sep)
    The argz_stringify converts argz into a normal string with the elements separated by the character sep, by replacing each '\0' inside argz (except the last one, which terminates the string) with sep. This is handy for printing argz in a readable manner.

    Function: error_t argz_add (char **argz, size_t *argz_len, const char *str)
    The argz_add function adds the string str to the end of the argz vector *argz, and updates *argz and *argz_len accordingly.

    Function: error_t argz_add_sep (char **argz, size_t *argz_len, const char *str, int delim)
    The argz_add_sep function is similar to argz_add, but str is split into separate elements in the result at occurrences of the character delim. This is useful, for instance, for adding the components of a Unix search path to an argz vector, by using a value of ':' for delim.

    Function: error_t argz_append (char **argz, size_t *argz_len, const char *buf, size_t buf_len)
    The argz_append function appends buf_len bytes starting at buf to the argz vector *argz, reallocating *argz to accommodate it, and adding buf_len to *argz_len.

    Function: error_t argz_delete (char **argz, size_t *argz_len, char *entry)
    If entry points to the beginning of one of the elements in the argz vector *argz, the argz_delete function will remove this entry and reallocate *argz, modifying *argz and *argz_len accordingly. Note that as destructive argz functions usually reallocate their argz argument, pointers into argz vectors such as entry will then become invalid.

    Function: error_t argz_insert (char **argz, size_t *argz_len, char *before, const char *entry)
    The argz_insert function inserts the string entry into the argz vector *argz at a point just before the existing element pointed to by before, reallocating *argz and updating *argz and *argz_len. If before is 0, entry is added to the end instead (as if by argz_add). Since the first element is in fact the same as *argz, passing in *argz as the value of before will result in entry being inserted at the beginning.

    Function: char * argz_next (char *argz, size_t argz_len, const char *entry)
    The argz_next function provides a convenient way of iterating over the elements in the argz vector argz. It returns a pointer to the next element in argz after the element entry, or 0 if there are no elements following entry. If entry is 0, the first element of argz is returned.

    This behavior suggests two styles of iteration:

        char *entry = 0;
        while ((entry = argz_next (argz, argz_len, entry)))
          action;
    

    (the double parentheses are necessary to make some C compilers shut up about what they consider a questionable while-test) and:

        char *entry;
        for (entry = argz;
             entry;
             entry = argz_next (argz, argz_len, entry))
          action;
    

    Note that the latter depends on argz having a value of 0 if it is empty (rather than a pointer to an empty block of memory); this invariant is maintained for argz vectors created by the functions here.

    Function: error_t argz_replace (char **argz, size_t *argz_len, const char *str, const char *with, unsigned *replace_count)
    Replace any occurrences of the string str in argz with with, reallocating argz as necessary. If replace_count is non-zero, *replace_count will be incremented by number of replacements performed.

    Envz Functions

    Envz vectors are just argz vectors with additional constraints on the form of each element; as such, argz functions can also be used on them, where it makes sense.

    Each element in an envz vector is a name-value pair, separated by a '=' character; if multiple '=' characters are present in an element, those after the first are considered part of the value, and treated like all other non-'\0' characters.

    If no '=' characters are present in an element, that element is considered the name of a "null" entry, as distinct from an entry with an empty value: envz_get will return 0 if given the name of null entry, whereas an entry with an empty value would result in a value of ""; envz_entry will still find such entries, however. Null entries can be removed with envz_strip function.

    As with argz functions, envz functions that may allocate memory (and thus fail) have a return type of error_t, and return either 0 or ENOMEM.

    These functions are declared in the standard include file `envz.h'.

    Function: char * envz_entry (const char *envz, size_t envz_len, const char *name)
    The envz_entry function finds the entry in envz with the name name, and returns a pointer to the whole entry--that is, the argz element which begins with name followed by a '=' character. If there is no entry with that name, 0 is returned.

    Function: char * envz_get (const char *envz, size_t envz_len, const char *name)
    The envz_get function finds the entry in envz with the name name (like envz_entry), and returns a pointer to the value portion of that entry (following the '='). If there is no entry with that name (or only a null entry), 0 is returned.

    Function: error_t envz_add (char **envz, size_t *envz_len, const char *name, const char *value)
    The envz_add function adds an entry to *envz (updating *envz and *envz_len) with the name name, and value value. If an entry with the same name already exists in envz, it is removed first. If value is 0, then the new entry will the special null type of entry (mentioned above).

    Function: error_t envz_merge (char **envz, size_t *envz_len, const char *envz2, size_t envz2_len, int override)
    The envz_merge function adds each entry in envz2 to envz, as if with envz_add, updating *envz and *envz_len. If override is true, then values in envz2 will supersede those with the same name in envz, otherwise not.

    Null entries are treated just like other entries in this respect, so a null entry in envz can prevent an entry of the same name in envz2 from being added to envz, if override is false.

    Function: void envz_strip (char **envz, size_t *envz_len)
    The envz_strip function removes any null entries from envz, updating *envz and *envz_len.

    Character Set Handling

    @ifnottex @macro cal{text} \text\

    Character sets used in the early days of computing had only six, seven, or eight bits for each character: there was never a case where more than eight bits (one byte) were used to represent a single character. The limitations of this approach became more apparent as more people grappled with non-Roman character sets, where not all the characters that make up a language's character set can be represented by @math{2^8} choices. This chapter shows the functionality which was added to the C library to support multiple character sets.

    Introduction to Extended Characters

    A variety of solutions to overcome the differences between character sets with a 1:1 relation between bytes and characters and character sets with ratios of 2:1 or 4:1 exist. The remainder of this section gives a few examples to help understand the design decisions made while developing the functionality of the C library.

    A distinction we have to make right away is between internal and external representation. Internal representation means the representation used by a program while keeping the text in memory. External representations are used when text is stored or transmitted through whatever communication channel. Examples of external representations include files lying in a directory that are going to be read and parsed.

    Traditionally there has been no difference between the two representations. It was equally comfortable and useful to use the same single-byte representation internally and externally. This changes with more and larger character sets.

    One of the problems to overcome with the internal representation is handling text that is externally encoded using different character sets. Assume a program which reads two texts and compares them using some metric. The comparison can be usefully done only if the texts are internally kept in a common format.

    For such a common format (@math{=} character set) eight bits are certainly no longer enough. So the smallest entity will have to grow: wide characters will now be used. Instead of one byte, two or four will be used instead. (Three are not good to address in memory and more than four bytes seem not to be necessary).

    As shown in some other part of this manual, there exists a completely new family of functions which can handle texts of this kind in memory. The most commonly used character sets for such internal wide character representations are Unicode and ISO 10646 (also known as UCS for Universal Character Set). Unicode was originally planned as a 16-bit character set, whereas ISO 10646 was designed to be a 31-bit large code space. The two standards are practically identical. They have the same character repertoire and code table, but Unicode specifies added semantics. At the moment, only characters in the first 0x10000 code positions (the so-called Basic Multilingual Plane, BMP) have been assigned, but the assignment of more specialized characters outside this 16-bit space is already in progress. A number of encodings have been defined for Unicode and ISO 10646 characters: UCS-2 is a 16-bit word that can only represent characters from the BMP, UCS-4 is a 32-bit word than can represent any Unicode and ISO 10646 character, UTF-8 is an ASCII compatible encoding where ASCII characters are represented by ASCII bytes and non-ASCII characters by sequences of 2-6 non-ASCII bytes, and finally UTF-16 is an extension of UCS-2 in which pairs of certain UCS-2 words can be used to encode non-BMP characters up to 0x10ffff.

    To represent wide characters the char type is not suitable. For this reason the ISO C standard introduces a new type which is designed to keep one character of a wide character string. To maintain the similarity there is also a type corresponding to int for those functions which take a single wide character.

    Data type: wchar_t
    This data type is used as the base type for wide character strings. I.e., arrays of objects of this type are the equivalent of char[] for multibyte character strings. The type is defined in `stddef.h'.

    The ISO C90 standard, where this type was introduced, does not say anything specific about the representation. It only requires that this type is capable of storing all elements of the basic character set. Therefore it would be legitimate to define wchar_t as char. This might make sense for embedded systems.

    But for GNU systems this type is always 32 bits wide. It is therefore capable of representing all UCS-4 values and therefore covering all of ISO 10646. Some Unix systems define wchar_t as a 16-bit type and thereby follow Unicode very strictly. This is perfectly fine with the standard but it also means that to represent all characters from Unicode and ISO 10646 one has to use UTF-16 surrogate characters which is in fact a multi-wide-character encoding. But this contradicts the purpose of the wchar_t type.

    Data type: wint_t
    wint_t is a data type used for parameters and variables which contain a single wide character. As the name already suggests it is the equivalent to int when using the normal char strings. The types wchar_t and wint_t have often the same representation if their size if 32 bits wide but if wchar_t is defined as char the type wint_t must be defined as int due to the parameter promotion.

    This type is defined in `wchar.h' and got introduced in Amendment 1 to ISO C90.

    As there are for the char data type there also exist macros specifying the minimum and maximum value representable in an object of type wchar_t.

    Macro: wint_t WCHAR_MIN
    The macro WCHAR_MIN evaluates to the minimum value representable by an object of type wint_t.

    This macro got introduced in Amendment 1 to ISO C90.

    Macro: wint_t WCHAR_MAX
    The macro WCHAR_MAX evaluates to the maximum value representable by an object of type wint_t.

    This macro got introduced in Amendment 1 to ISO C90.

    Another special wide character value is the equivalent to EOF.

    Macro: wint_t WEOF
    The macro WEOF evaluates to a constant expression of type wint_t whose value is different from any member of the extended character set.

    WEOF need not be the same value as EOF and unlike EOF it also need not be negative. I.e., sloppy code like

    {
      int c;
      ...
      while ((c = getc (fp)) < 0)
        ...
    }
    

    has to be rewritten to explicitly use WEOF when wide characters are used.

    {
      wint_t c;
      ...
      while ((c = wgetc (fp)) != WEOF)
        ...
    }
    

    This macro was introduced in Amendment 1 to ISO C90 and is defined in `wchar.h'.

    These internal representations present problems when it comes to storing and transmittal, since a single wide character consists of more than one byte they are effected by byte-ordering. I.e., machines with different endianesses would see different value accessing the same data. This also applies for communication protocols which are all byte-based and therefore the sender has to decide about splitting the wide character in bytes. A last (but not least important) point is that wide characters often require more storage space than an customized byte oriented character set.

    For all the above reasons, an external encoding which is different from the internal encoding is often used if the latter is UCS-2 or UCS-4. The external encoding is byte-based and can be chosen appropriately for the environment and for the texts to be handled. There exist a variety of different character sets which can be used for this external encoding. Information which will not be exhaustively presented here--instead, a description of the major groups will suffice. All of the ASCII-based character sets [_bkoz_: do you mean Roman character sets? If not, what do you mean here?] fulfill one requirement: they are "filesystem safe". This means that the character '/' is used in the encoding only to represent itself. Things are a bit different for character sets like EBCDIC (Extended Binary Coded Decimal Interchange Code, a character set family used by IBM) but if the operation system does not understand EBCDIC directly the parameters to system calls have to be converted first anyhow.

    The question remaining is: how to select the character set or encoding to use. The answer: you cannot decide about it yourself, it is decided by the developers of the system or the majority of the users. Since the goal is interoperability one has to use whatever the other people one works with use. If there are no constraints the selection is based on the requirements the expected circle of users will have. I.e., if a project is expected to only be used in, say, Russia it is fine to use KOI8-R or a similar character set. But if at the same time people from, say, Greece are participating one should use a character set which allows all people to collaborate.

    The most widely useful solution seems to be: go with the most general character set, namely ISO 10646. Use UTF-8 as the external encoding and problems about users not being able to use their own language adequately are a thing of the past.

    One final comment about the choice of the wide character representation is necessary at this point. We have said above that the natural choice is using Unicode or ISO 10646. This is not required, but at least encouraged, by the ISO C standard. The standard defines at least a macro __STDC_ISO_10646__ that is only defined on systems where the wchar_t type encodes ISO 10646 characters. If this symbol is not defined one should as much as possible avoid making assumption about the wide character representation. If the programmer uses only the functions provided by the C library to handle wide character strings there should not be any compatibility problems with other systems.

    Overview about Character Handling Functions

    A Unix C library contains three different sets of functions in two families to handle character set conversion. The one function family is specified in the ISO C standard and therefore is portable even beyond the Unix world.

    The most commonly known set of functions, coming from the ISO C90 standard, is unfortunately the least useful one. In fact, these functions should be avoided whenever possible, especially when developing libraries (as opposed to applications).

    The second family of functions got introduced in the early Unix standards (XPG2) and is still part of the latest and greatest Unix standard: Unix 98. It is also the most powerful and useful set of functions. But we will start with the functions defined in Amendment 1 to ISO C90.

    Restartable Multibyte Conversion Functions

    The ISO C standard defines functions to convert strings from a multibyte representation to wide character strings. There are a number of peculiarities:

    Despite these limitations the ISO C functions can very well be used in many contexts. In graphical user interfaces, for instance, it is not uncommon to have functions which require text to be displayed in a wide character string if it is not simple ASCII. The text itself might come from a file with translations and the user should decide about the current locale which determines the translation and therefore also the external encoding used. In such a situation (and many others) the functions described here are perfect. If more freedom while performing the conversion is necessary take a look at the iconv functions (see section Generic Charset Conversion).

    Selecting the conversion and its properties

    We already said above that the currently selected locale for the LC_CTYPE category decides about the conversion which is performed by the functions we are about to describe. Each locale uses its own character set (given as an argument to localedef) and this is the one assumed as the external multibyte encoding. The wide character character set always is UCS-4, at least on GNU systems.

    A characteristic of each multibyte character set is the maximum number of bytes which can be necessary to represent one character. This information is quite important when writing code which uses the conversion functions. In the examples below we will see some examples. The ISO C standard defines two macros which provide this information.

    Macro: int MB_LEN_MAX
    This macro specifies the maximum number of bytes in the multibyte sequence for a single character in any of the supported locales. It is a compile-time constant and it is defined in `limits.h'.

    Macro: int MB_CUR_MAX
    MB_CUR_MAX expands into a positive integer expression that is the maximum number of bytes in a multibyte character in the current locale. The value is never greater than MB_LEN_MAX. Unlike MB_LEN_MAX this macro need not be a compile-time constant and in fact, in the GNU C library it is not.

    MB_CUR_MAX is defined in `stdlib.h'.

    Two different macros are necessary since strictly ISO C90 compilers do not allow variable length array definitions but still it is desirable to avoid dynamic allocation. This incomplete piece of code shows the problem:

    {
      char buf[MB_LEN_MAX];
      ssize_t len = 0;
    
      while (! feof (fp))
        {
          fread (&buf[len], 1, MB_CUR_MAX - len, fp);
          /* ... process buf */
          len -= used;
        }
    }
    

    The code in the inner loop is expected to have always enough bytes in the array buf to convert one multibyte character. The array buf has to be sized statically since many compilers do not allow a variable size. The fread call makes sure that always MB_CUR_MAX bytes are available in buf. Note that it isn't a problem if MB_CUR_MAX is not a compile-time constant.

    Representing the state of the conversion

    In the introduction of this chapter it was said that certain character sets use a stateful encoding. I.e., the encoded values depend in some way on the previous bytes in the text.

    Since the conversion functions allow converting a text in more than one step we must have a way to pass this information from one call of the functions to another.

    Data type: mbstate_t
    A variable of type mbstate_t can contain all the information about the shift state needed from one call to a conversion function to another.

    This type is defined in `wchar.h'. It got introduced in Amendment 1 to ISO C90.

    To use objects of this type the programmer has to define such objects (normally as local variables on the stack) and pass a pointer to the object to the conversion functions. This way the conversion function can update the object if the current multibyte character set is stateful.

    There is no specific function or initializer to put the state object in any specific state. The rules are that the object should always represent the initial state before the first use and this is achieved by clearing the whole variable with code such as follows:

    {
      mbstate_t state;
      memset (&state, '\0', sizeof (state));
      /* from now on state can be used.  */
      ...
    }
    

    When using the conversion functions to generate output it is often necessary to test whether the current state corresponds to the initial state. This is necessary, for example, to decide whether or not to emit escape sequences to set the state to the initial state at certain sequence points. Communication protocols often require this.

    Function: int mbsinit (const mbstate_t *ps)
    This function determines whether the state object pointed to by ps is in the initial state or not. If ps is a null pointer or the object is in the initial state the return value is nonzero. Otherwise it is zero.

    This function was introduced in Amendment 1 to ISO C90 and is declared in `wchar.h'.

    Code using this function often looks similar to this:

    {
      mbstate_t state;
      memset (&state, '\0', sizeof (state));
      /* Use state.  */
      ...
      if (! mbsinit (&state))
        {
          /* Emit code to return to initial state.  */
          const wchar_t empty[] = L"";
          const wchar_t *srcp = empty;
          wcsrtombs (outbuf, &srcp, outbuflen, &state);
        }
      ...
    }
    

    The code to emit the escape sequence to get back to the initial state is interesting. The wcsrtombs function can be used to determine the necessary output code (see section Converting Multibyte and Wide Character Strings). Please note that on GNU systems it is not necessary to perform this extra action for the conversion from multibyte text to wide character text since the wide character encoding is not stateful. But there is nothing mentioned in any standard which prohibits making wchar_t using a stateful encoding.

    Converting Single Characters

    The most fundamental of the conversion functions are those dealing with single characters. Please note that this does not always mean single bytes. But since there is very often a subset of the multibyte character set which consists of single byte sequences there are functions to help with converting bytes. One very important and often applicable scenario is where ASCII is a subpart of the multibyte character set. I.e., all ASCII characters stand for itself and all other characters have at least a first byte which is beyond the range @math{0} to @math{127}.

    Function: wint_t btowc (int c)
    The btowc function ("byte to wide character") converts a valid single byte character c in the initial shift state into the wide character equivalent using the conversion rules from the currently selected locale of the LC_CTYPE category.

    If (unsigned char) c is no valid single byte multibyte character or if c is EOF the function returns WEOF.

    Please note the restriction of c being tested for validity only in the initial shift state. There is no mbstate_t object used from which the state information is taken and the function also does not use any static state.

    This function was introduced in Amendment 1 to ISO C90 and is declared in `wchar.h'.

    Despite the limitation that the single byte value always is interpreted in the initial state this function is actually useful most of the time. Most characters are either entirely single-byte character sets or they are extension to ASCII. But then it is possible to write code like this (not that this specific example is very useful):

    wchar_t *
    itow (unsigned long int val)
    {
      static wchar_t buf[30];
      wchar_t *wcp = &buf[29];
      *wcp = L'\0';
      while (val != 0)
        {
          *--wcp = btowc ('0' + val % 10);
          val /= 10;
        }
      if (wcp == &buf[29])
        *--wcp = L'0';
      return wcp;
    }
    

    Why is it necessary to use such a complicated implementation and not simply cast '0' + val % 10 to a wide character? The answer is that there is no guarantee that one can perform this kind of arithmetic on the character of the character set used for wchar_t representation. In other situations the bytes are not constant at compile time and so the compiler cannot do the work. In situations like this it is necessary btowc.

    There also is a function for the conversion in the other direction.

    Function: int wctob (wint_t c)
    The wctob function ("wide character to byte") takes as the parameter a valid wide character. If the multibyte representation for this character in the initial state is exactly one byte long the return value of this function is this character. Otherwise the return value is EOF.

    This function was introduced in Amendment 1 to ISO C90 and is declared in `wchar.h'.

    There are more general functions to convert single character from multibyte representation to wide characters and vice versa. These functions pose no limit on the length of the multibyte representation and they also do not require it to be in the initial state.

    Function: size_t mbrtowc (wchar_t *restrict pwc, const char *restrict s, size_t n, mbstate_t *restrict ps)
    The mbrtowc function ("multibyte restartable to wide character") converts the next multibyte character in the string pointed to by s into a wide character and stores it in the wide character string pointed to by pwc. The conversion is performed according to the locale currently selected for the LC_CTYPE category. If the conversion for the character set used in the locale requires a state the multibyte string is interpreted in the state represented by the object pointed to by ps. If ps is a null pointer, a static, internal state variable used only by the mbrtowc function is used.

    If the next multibyte character corresponds to the NUL wide character the return value of the function is @math{0} and the state object is afterwards in the initial state. If the next n or fewer bytes form a correct multibyte character the return value is the number of bytes starting from s which form the multibyte character. The conversion state is updated according to the bytes consumed in the conversion. In both cases the wide character (either the L'\0' or the one found in the conversion) is stored in the string pointer to by pwc iff pwc is not null.

    If the first n bytes of the multibyte string possibly form a valid multibyte character but there are more than n bytes needed to complete it the return value of the function is (size_t) -2 and no value is stored. Please note that this can happen even if n has a value greater or equal to MB_CUR_MAX since the input might contain redundant shift sequences.

    If the first n bytes of the multibyte string cannot possibly form a valid multibyte character also no value is stored, the global variable errno is set to the value EILSEQ and the function returns (size_t) -1. The conversion state is afterwards undefined.

    This function was introduced in Amendment 1 to ISO C90 and is declared in `wchar.h'.

    Using this function is straight forward. A function which copies a multibyte string into a wide character string while at the same time converting all lowercase character into uppercase could look like this (this is not the final version, just an example; it has no error checking, and leaks sometimes memory):

    wchar_t *
    mbstouwcs (const char *s)
    {
      size_t len = strlen (s);
      wchar_t *result = malloc ((len + 1) * sizeof (wchar_t));
      wchar_t *wcp = result;
      wchar_t tmp[1];
      mbstate_t state;
      size_t nbytes;
    
      memset (&state, '\0', sizeof (state));
      while ((nbytes = mbrtowc (tmp, s, len, &state)) > 0)
        {
          if (nbytes >= (size_t) -2)
            /* Invalid input string.  */
            return NULL;
          *result++ = towupper (tmp[0]);
          len -= nbytes;
          s += nbytes;
        }
      return result;
    }
    

    The use of mbrtowc should be clear. A single wide character is stored in tmp[0] and the number of consumed bytes is stored in the variable nbytes. In case the the conversion was successful the uppercase variant of the wide character is stored in the result array and the pointer to the input string and the number of available bytes is adjusted.

    The only non-obvious thing about the function might be the way memory is allocated for the result. The above code uses the fact that there can never be more wide characters in the converted results than there are bytes in the multibyte input string. This method yields to a pessimistic guess about the size of the result and if many wide character strings have to be constructed this way or the strings are long, the extra memory required allocated because the input string contains multibyte characters might be significant. It would be possible to resize the allocated memory block to the correct size before returning it. A better solution might be to allocate just the right amount of space for the result right away. Unfortunately there is no function to compute the length of the wide character string directly from the multibyte string. But there is a function which does part of the work.

    Function: size_t mbrlen (const char *restrict s, size_t n, mbstate_t *ps)
    The mbrlen function ("multibyte restartable length") computes the number of at most n bytes starting at s which form the next valid and complete multibyte character.

    If the next multibyte character corresponds to the NUL wide character the return value is @math{0}. If the next n bytes form a valid multibyte character the number of bytes belonging to this multibyte character byte sequence is returned.

    If the the first n bytes possibly form a valid multibyte character but it is incomplete the return value is (size_t) -2. Otherwise the multibyte character sequence is invalid and the return value is (size_t) -1.

    The multibyte sequence is interpreted in the state represented by the object pointed to by ps. If ps is a null pointer, a state object local to mbrlen is used.

    This function was introduced in Amendment 1 to ISO C90 and is declared in `wchar.h'.

    The tentative reader now will of course note that mbrlen can be implemented as

    mbrtowc (NULL, s, n, ps != NULL ? ps : &internal)
    

    This is true and in fact is mentioned in the official specification. Now, how can this function be used to determine the length of the wide character string created from a multibyte character string? It is not directly usable but we can define a function mbslen using it:

    size_t
    mbslen (const char *s)
    {
      mbstate_t state;
      size_t result = 0;
      size_t nbytes;
      memset (&state, '\0', sizeof (state));
      while ((nbytes = mbrlen (s, MB_LEN_MAX, &state)) > 0)
        {
          if (nbytes >= (size_t) -2)
            /* Something is wrong.  */
            return (size_t) -1;
          s += nbytes;
          ++result;
        }
      return result;
    }
    

    This function simply calls mbrlen for each multibyte character in the string and counts the number of function calls. Please note that we here use MB_LEN_MAX as the size argument in the mbrlen call. This is OK since a) this value is larger then the length of the longest multibyte character sequence and b) because we know that the string s ends with a NUL byte which cannot be part of any other multibyte character sequence but the one representing the NUL wide character. Therefore the mbrlen function will never read invalid memory.

    Now that this function is available (just to make this clear, this function is not part of the GNU C library) we can compute the number of wide character required to store the converted multibyte character string s using

    wcs_bytes = (mbslen (s) + 1) * sizeof (wchar_t);
    

    Please note that the mbslen function is quite inefficient. The implementation of mbstouwcs implemented using mbslen would have to perform the conversion of the multibyte character input string twice and this conversion might be quite expensive. So it is necessary to think about the consequences of using the easier but imprecise method before doing the work twice.

    Function: size_t wcrtomb (char *restrict s, wchar_t wc, mbstate_t *restrict ps)
    The wcrtomb function ("wide character restartable to multibyte") converts a single wide character into a multibyte string corresponding to that wide character.

    If s is a null pointer the function resets the the state stored in the objects pointer to by ps (or the internal mbstate_t object) to the initial state. This can also be achieved by a call like this:

    wcrtombs (temp_buf, L'\0', ps)
    

    since if s is a null pointer wcrtomb performs as if it writes into an internal buffer which is guaranteed to be large enough.

    If wc is the NUL wide character wcrtomb emits, if necessary, a shift sequence to get the state ps into the initial state followed by a single NUL byte is stored in the string s.

    Otherwise a byte sequence (possibly including shift sequences) is written into the string s. This of only happens if wc is a valid wide character, i.e., it has a multibyte representation in the character set selected by locale of the LC_CTYPE category. If wc is no valid wide character nothing is stored in the strings s, errno is set to EILSEQ, the conversion state in ps is undefined and the return value is (size_t) -1.

    If no error occurred the function returns the number of bytes stored in the string s. This includes all byte representing shift sequences.

    One word about the interface of the function: there is no parameter specifying the length of the array s. Instead the function assumes that there are at least MB_CUR_MAX bytes available since this is the maximum length of any byte sequence representing a single character. So the caller has to make sure that there is enough space available, otherwise buffer overruns can occur.

    This function was introduced in Amendment 1 to ISO C90 and is declared in `wchar.h'.

    Using this function is as easy as using mbrtowc. The following example appends a wide character string to a multibyte character string. Again, the code is not really useful (and correct), it is simply here to demonstrate the use and some problems.

    char *
    mbscatwcs (char *s, size_t len, const wchar_t *ws)
    {
      mbstate_t state;
      /* Find the end of the existing string.  */
      char *wp = strchr (s, '\0');
      len -= wp - s;
      memset (&state, '\0', sizeof (state));
      do
        {
          size_t nbytes;
          if (len < MB_CUR_LEN)
            {
              /* We cannot guarantee that the next
                 character fits into the buffer, so
                 return an error.  */
              errno = E2BIG;
              return NULL;
            }
          nbytes = wcrtomb (wp, *ws, &state);
          if (nbytes == (size_t) -1)
            /* Error in the conversion.  */
            return NULL;
          len -= nbytes;
          wp += nbytes;
        }
      while (*ws++ != L'\0');
      return s;
    }
    

    First the function has to find the end of the string currently in the array s. The strchr call does this very efficiently since a requirement for multibyte character representations is that the NUL byte never is used except to represent itself (and in this context, the end of the string).

    After initializing the state object the loop is entered where the first task is to make sure there is enough room in the array s. We abort if there are not at least MB_CUR_LEN bytes available. This is not always optimal but we have no other choice. We might have less than MB_CUR_LEN bytes available but the next multibyte character might also be only one byte long. At the time the wcrtomb call returns it is too late to decide whether the buffer was large enough or not. If this solution is really unsuitable there is a very slow but more accurate solution.

      ...
      if (len < MB_CUR_LEN)
        {
          mbstate_t temp_state;
          memcpy (&temp_state, &state, sizeof (state));
          if (wcrtomb (NULL, *ws, &temp_state) > len)
            {
              /* We cannot guarantee that the next
                 character fits into the buffer, so
                 return an error.  */
              errno = E2BIG;
              return NULL;
            }
        }
      ...
    

    Here we do perform the conversion which might overflow the buffer so that we are afterwards in the position to make an exact decision about the buffer size. Please note the NULL argument for the destination buffer in the new wcrtomb call; since we are not interested in the converted text at this point this is a nice way to express this. The most unusual thing about this piece of code certainly is the duplication of the conversion state object. But think about this: if a change of the state is necessary to emit the next multibyte character we want to have the same shift state change performed in the real conversion. Therefore we have to preserve the initial shift state information.

    There are certainly many more and even better solutions to this problem. This example is only meant for educational purposes.

    Converting Multibyte and Wide Character Strings

    The functions described in the previous section only convert a single character at a time. Most operations to be performed in real-world programs include strings and therefore the ISO C standard also defines conversions on entire strings. However, the defined set of functions is quite limited, thus the GNU C library contains a few extensions which can help in some important situations.

    Function: size_t mbsrtowcs (wchar_t *restrict dst, const char **restrict src, size_t len, mbstate_t *restrict ps)
    The mbsrtowcs function ("multibyte string restartable to wide character string") converts an NUL terminated multibyte character string at *src into an equivalent wide character string, including the NUL wide character at the end. The conversion is started using the state information from the object pointed to by ps or from an internal object of mbsrtowcs if ps is a null pointer. Before returning the state object to match the state after the last converted character. The state is the initial state if the terminating NUL byte is reached and converted.

    If dst is not a null pointer the result is stored in the array pointed to by dst, otherwise the conversion result is not available since it is stored in an internal buffer.

    If len wide characters are stored in the array dst before reaching the end of the input string the conversion stops and len is returned. If dst is a null pointer len is never checked.

    Another reason for a premature return from the function call is if the input string contains an invalid multibyte sequence. In this case the global variable errno is set to EILSEQ and the function returns (size_t) -1.

    In all other cases the function returns the number of wide characters converted during this call. If dst is not null mbsrtowcs stores in the pointer pointed to by src a null pointer (if the NUL byte in the input string was reached) or the address of the byte following the last converted multibyte character.

    This function was introduced in Amendment 1 to ISO C90 and is declared in `wchar.h'.

    The definition of this function has one limitation which has to be understood. The requirement that dst has to be a NUL terminated string provides problems if one wants to convert buffers with text. A buffer is normally no collection of NUL terminated strings but instead a continuous collection of lines, separated by newline characters. Now assume a function to convert one line from a buffer is needed. Since the line is not NUL terminated the source pointer cannot directly point into the unmodified text buffer. This means, either one inserts the NUL byte at the appropriate place for the time of the mbsrtowcs function call (which is not doable for a read-only buffer or in a multi-threaded application) or one copies the line in an extra buffer where it can be terminated by a NUL byte. Note that it is not in general possible to limit the number of characters to convert by setting the parameter len to any specific value. Since it is not known how many bytes each multibyte character sequence is in length one always could do only a guess.

    There is still a problem with the method of NUL-terminating a line right after the newline character which could lead to very strange results. As said in the description of the mbsrtowcs function above the conversion state is guaranteed to be in the initial shift state after processing the NUL byte at the end of the input string. But this NUL byte is not really part of the text. I.e., the conversion state after the newline in the original text could be something different than the initial shift state and therefore the first character of the next line is encoded using this state. But the state in question is never accessible to the user since the conversion stops after the NUL byte (which resets the state). Most stateful character sets in use today require that the shift state after a newline is the initial state--but this is not a strict guarantee. Therefore simply NUL terminating a piece of a running text is not always an adequate solution and therefore never should be used in generally used code.

    The generic conversion interface (see section Generic Charset Conversion) does not have this limitation (it simply works on buffers, not strings), and the GNU C library contains a set of functions which take additional parameters specifying the maximal number of bytes which are consumed from the input string. This way the problem of mbsrtowcs's example above could be solved by determining the line length and passing this length to the function.

    Function: size_t wcsrtombs (char *restrict dst, const wchar_t **restrict src, size_t len, mbstate_t *restrict ps)
    The wcsrtombs function ("wide character string restartable to multibyte string") converts the NUL terminated wide character string at *src into an equivalent multibyte character string and stores the result in the array pointed to by dst. The NUL wide character is also converted. The conversion starts in the state described in the object pointed to by ps or by a state object locally to wcsrtombs in case ps is a null pointer. If dst is a null pointer the conversion is performed as usual but the result is not available. If all characters of the input string were successfully converted and if dst is not a null pointer the pointer pointed to by src gets assigned a null pointer.

    If one of the wide characters in the input string has no valid multibyte character equivalent the conversion stops early, sets the global variable errno to EILSEQ, and returns (size_t) -1.

    Another reason for a premature stop is if dst is not a null pointer and the next converted character would require more than len bytes in total to the array dst. In this case (and if dest is not a null pointer) the pointer pointed to by src is assigned a value pointing to the wide character right after the last one successfully converted.

    Except in the case of an encoding error the return value of the function is the number of bytes in all the multibyte character sequences stored in dst. Before returning the state in the object pointed to by ps (or the internal object in case ps is a null pointer) is updated to reflect the state after the last conversion. The state is the initial shift state in case the terminating NUL wide character was converted.

    This function was introduced in Amendment 1 to ISO C90 and is declared in `wchar.h'.

    The restriction mentions above for the mbsrtowcs function applies also here. There is no possibility to directly control the number of input characters. One has to place the NUL wide character at the correct place or control the consumed input indirectly via the available output array size (the len parameter).

    Function: size_t mbsnrtowcs (wchar_t *restrict dst, const char **restrict src, size_t nmc, size_t len, mbstate_t *restrict ps)
    The mbsnrtowcs function is very similar to the mbsrtowcs function. All the parameters are the same except for nmc which is new. The return value is the same as for mbsrtowcs.

    This new parameter specifies how many bytes at most can be used from the multibyte character string. I.e., the multibyte character string *src need not be NUL terminated. But if a NUL byte is found within the nmc first bytes of the string the conversion stops here.

    This function is a GNU extensions. It is meant to work around the problems mentioned above. Now it is possible to convert buffer with multibyte character text piece for piece without having to care about inserting NUL bytes and the effect of NUL bytes on the conversion state.

    A function to convert a multibyte string into a wide character string and display it could be written like this (this is not a really useful example):

    void
    showmbs (const char *src, FILE *fp)
    {
      mbstate_t state;
      int cnt = 0;
      memset (&state, '\0', sizeof (state));
      while (1)
        {
          wchar_t linebuf[100];
          const char *endp = strchr (src, '\n');
          size_t n;
    
          /* Exit if there is no more line.  */
          if (endp == NULL)
            break;
    
          n = mbsnrtowcs (linebuf, &src, endp - src, 99, &state);
          linebuf[n] = L'\0';
          fprintf (fp, "line %d: \"%S\"\n", linebuf);
        }
    }
    

    There is no problem with the state after a call to mbsnrtowcs. Since we don't insert characters in the strings which were not in there right from the beginning and we use state only for the conversion of the given buffer there is no problem with altering the state.

    Function: size_t wcsnrtombs (char *restrict dst, const wchar_t **restrict src, size_t nwc, size_t len, mbstate_t *restrict ps)
    The wcsnrtombs function implements the conversion from wide character strings to multibyte character strings. It is similar to wcsrtombs but it takes, just like mbsnrtowcs, an extra parameter which specifies the length of the input string.

    No more than nwc wide characters from the input string *src are converted. If the input string contains a NUL wide character in the first nwc character to conversion stops at this place.

    This function is a GNU extension and just like mbsnrtowcs is helps in situations where no NUL terminated input strings are available.

    A Complete Multibyte Conversion Example

    The example programs given in the last sections are only brief and do not contain all the error checking etc. Presented here is a complete and documented example. It features the mbrtowc function but it should be easy to derive versions using the other functions.

    int
    file_mbsrtowcs (int input, int output)
    {
      /* Note the use of MB_LEN_MAX.
         MB_CUR_MAX cannot portably be used here.  */
      char buffer[BUFSIZ + MB_LEN_MAX];
      mbstate_t state;
      int filled = 0;
      int eof = 0;
    
      /* Initialize the state.  */
      memset (&state, '\0', sizeof (state));
    
      while (!eof)
        {
          ssize_t nread;
          ssize_t nwrite;
          char *inp = buffer;
          wchar_t outbuf[BUFSIZ];
          wchar_t *outp = outbuf;
    
          /* Fill up the buffer from the input file.  */
          nread = read (input, buffer + filled, BUFSIZ);
          if (nread < 0)
            {
              perror ("read");
              return 0;
            }
          /* If we reach end of file, make a note to read no more. */
          if (nread == 0)
            eof = 1;
    
          /* filled is now the number of bytes in buffer. */
          filled += nread;
    
          /* Convert those bytes to wide characters--as many as we can. */
          while (1)
            {
              size_t thislen = mbrtowc (outp, inp, filled, &state);
              /* Stop converting at invalid character;
                 this can mean we have read just the first part
                 of a valid character.  */
              if (thislen == (size_t) -1)
                break;
              /* We want to handle embedded NUL bytes
                 but the return value is 0.  Correct this.  */
              if (thislen == 0)
                thislen = 1;
              /* Advance past this character. */
              inp += thislen;
              filled -= thislen;
              ++outp;
            }
    
          /* Write the wide characters we just made.  */
          nwrite = write (output, outbuf,
                          (outp - outbuf) * sizeof (wchar_t));
          if (nwrite < 0)
            {
              perror ("write");
              return 0;
            }
    
          /* See if we have a real invalid character. */
          if ((eof && filled > 0) || filled >= MB_CUR_MAX)
            {
              error (0, 0, "invalid multibyte character");
              return 0;
            }
    
          /* If any characters must be carried forward,
             put them at the beginning of buffer. */
          if (filled > 0)
            memmove (inp, buffer, filled);
        }
    
      return 1;
    }
    

    Non-reentrant Conversion Function

    The functions described in the last chapter are defined in Amendment 1 to ISO C90. But the original ISO C90 standard also contained functions for character set conversion. The reason that they are not described in the first place is that they are almost entirely useless.

    The problem is that all the functions for conversion defined in ISO C90 use a local state. This implies that multiple conversions at the same time (not only when using threads) cannot be done, and that you cannot first convert single characters and then strings since you cannot tell the conversion functions which state to use.

    These functions are therefore usable only in a very limited set of situations. One must complete converting the entire string before starting a new one and each string/text must be converted with the same function (there is no problem with the library itself; it is guaranteed that no library function changes the state of any of these functions). For the above reasons it is highly requested that the functions from the last section are used in place of non-reentrant conversion functions.

    Non-reentrant Conversion of Single Characters

    Function: int mbtowc (wchar_t *restrict result, const char *restrict string, size_t size)
    The mbtowc ("multibyte to wide character") function when called with non-null string converts the first multibyte character beginning at string to its corresponding wide character code. It stores the result in *result.

    mbtowc never examines more than size bytes. (The idea is to supply for size the number of bytes of data you have in hand.)

    mbtowc with non-null string distinguishes three possibilities: the first size bytes at string start with valid multibyte character, they start with an invalid byte sequence or just part of a character, or string points to an empty string (a null character).

    For a valid multibyte character, mbtowc converts it to a wide character and stores that in *result, and returns the number of bytes in that character (always at least @math{1}, and never more than size).

    For an invalid byte sequence, mbtowc returns @math{-1}. For an empty string, it returns @math{0}, also storing '\0' in *result.

    If the multibyte character code uses shift characters, then mbtowc maintains and updates a shift state as it scans. If you call mbtowc with a null pointer for string, that initializes the shift state to its standard initial value. It also returns nonzero if the multibyte character code in use actually has a shift state. See section States in Non-reentrant Functions.

    Function: int wctomb (char *string, wchar_t wchar)
    The wctomb ("wide character to multibyte") function converts the wide character code wchar to its corresponding multibyte character sequence, and stores the result in bytes starting at string. At most MB_CUR_MAX characters are stored.

    wctomb with non-null string distinguishes three possibilities for wchar: a valid wide character code (one that can be translated to a multibyte character), an invalid code, and L'\0'.

    Given a valid code, wctomb converts it to a multibyte character, storing the bytes starting at string. Then it returns the number of bytes in that character (always at least @math{1}, and never more than MB_CUR_MAX).

    If wchar is an invalid wide character code, wctomb returns @math{-1}. If wchar is L'\0', it returns 0, also storing '\0' in *string.

    If the multibyte character code uses shift characters, then wctomb maintains and updates a shift state as it scans. If you call wctomb with a null pointer for string, that initializes the shift state to its standard initial value. It also returns nonzero if the multibyte character code in use actually has a shift state. See section States in Non-reentrant Functions.

    Calling this function with a wchar argument of zero when string is not null has the side-effect of reinitializing the stored shift state as well as storing the multibyte character '\0' and returning @math{0}.

    Similar to mbrlen there is also a non-reentrant function which computes the length of a multibyte character. It can be defined in terms of mbtowc.

    Function: int mblen (const char *string, size_t size)
    The mblen function with a non-null string argument returns the number of bytes that make up the multibyte character beginning at string, never examining more than size bytes. (The idea is to supply for size the number of bytes of data you have in hand.)

    The return value of mblen distinguishes three possibilities: the first size bytes at string start with valid multibyte character, they start with an invalid byte sequence or just part of a character, or string points to an empty string (a null character).

    For a valid multibyte character, mblen returns the number of bytes in that character (always at least 1, and never more than size). For an invalid byte sequence, mblen returns @math{-1}. For an empty string, it returns @math{0}.

    If the multibyte character code uses shift characters, then mblen maintains and updates a shift state as it scans. If you call mblen with a null pointer for string, that initializes the shift state to its standard initial value. It also returns a nonzero value if the multibyte character code in use actually has a shift state. See section States in Non-reentrant Functions.

    The function mblen is declared in `stdlib.h'.

    Non-reentrant Conversion of Strings

    For convenience reasons the ISO C90 standard defines also functions to convert entire strings instead of single characters. These functions suffer from the same problems as their reentrant counterparts from Amendment 1 to ISO C90; see section Converting Multibyte and Wide Character Strings.

    Function: size_t mbstowcs (wchar_t *wstring, const char *string, size_t size)
    The mbstowcs ("multibyte string to wide character string") function converts the null-terminated string of multibyte characters string to an array of wide character codes, storing not more than size wide characters into the array beginning at wstring. The terminating null character counts towards the size, so if size is less than the actual number of wide characters resulting from string, no terminating null character is stored.

    The conversion of characters from string begins in the initial shift state.

    If an invalid multibyte character sequence is found, this function returns a value of @math{-1}. Otherwise, it returns the number of wide characters stored in the array wstring. This number does not include the terminating null character, which is present if the number is less than size.

    Here is an example showing how to convert a string of multibyte characters, allocating enough space for the result.

    wchar_t *
    mbstowcs_alloc (const char *string)
    {
      size_t size = strlen (string) + 1;
      wchar_t *buf = xmalloc (size * sizeof (wchar_t));
    
      size = mbstowcs (buf, string, size);
      if (size == (size_t) -1)
        return NULL;
      buf = xrealloc (buf, (size + 1) * sizeof (wchar_t));
      return buf;
    }
    

    Function: size_t wcstombs (char *string, const wchar_t *wstring, size_t size)
    The wcstombs ("wide character string to multibyte string") function converts the null-terminated wide character array wstring into a string containing multibyte characters, storing not more than size bytes starting at string, followed by a terminating null character if there is room. The conversion of characters begins in the initial shift state.

    The terminating null character counts towards the size, so if size is less than or equal to the number of bytes needed in wstring, no terminating null character is stored.

    If a code that does not correspond to a valid multibyte character is found, this function returns a value of @math{-1}. Otherwise, the return value is the number of bytes stored in the array string. This number does not include the terminating null character, which is present if the number is less than size.

    States in Non-reentrant Functions

    In some multibyte character codes, the meaning of any particular byte sequence is not fixed; it depends on what other sequences have come earlier in the same string. Typically there are just a few sequences that can change the meaning of other sequences; these few are called shift sequences and we say that they set the shift state for other sequences that follow.

    To illustrate shift state and shift sequences, suppose we decide that the sequence 0200 (just one byte) enters Japanese mode, in which pairs of bytes in the range from 0240 to 0377 are single characters, while 0201 enters Latin-1 mode, in which single bytes in the range from 0240 to 0377 are characters, and interpreted according to the ISO Latin-1 character set. This is a multibyte code which has two alternative shift states ("Japanese mode" and "Latin-1 mode"), and two shift sequences that specify particular shift states.

    When the multibyte character code in use has shift states, then mblen, mbtowc and wctomb must maintain and update the current shift state as they scan the string. To make this work properly, you must follow these rules:

    Here is an example of using mblen following these rules:

    void
    scan_string (char *s)
    {
      int length = strlen (s);
    
      /* Initialize shift state.  */
      mblen (NULL, 0);
    
      while (1)
        {
          int thischar = mblen (s, length);
          /* Deal with end of string and invalid characters.  */
          if (thischar == 0)
            break;
          if (thischar == -1)
            {
              error ("invalid multibyte character");
              break;
            }
          /* Advance past this character.  */
          s += thischar;
          length -= thischar;
        }
    }
    

    The functions mblen, mbtowc and wctomb are not reentrant when using a multibyte code that uses a shift state. However, no other library functions call these functions, so you don't have to worry that the shift state will be changed mysteriously.

    Generic Charset Conversion

    The conversion functions mentioned so far in this chapter all had in common that they operate on character sets which are not directly specified by the functions. The multibyte encoding used is specified by the currently selected locale for the LC_CTYPE category. The wide character set is fixed by the implementation (in the case of GNU C library it always is UCS-4 encoded ISO 10646.

    This has of course several problems when it comes to general character conversion:

    The XPG2 standard defines a completely new set of functions which has none of these limitations. They are not at all coupled to the selected locales and they but no constraints on the character sets selected for source and destination. Only the set of available conversions is limiting them. The standard does not specify that any conversion at all must be available. It is a measure of the quality of the implementation.

    In the following text first the interface to iconv, the conversion function, will be described. Comparisons with other implementations will show what pitfalls lie on the way of portable applications. At last, the implementation is described as far as interesting to the advanced user who wants to extend the conversion capabilities.

    Generic Character Set Conversion Interface

    This set of functions follows the traditional cycle of using a resource: open--use--close. The interface consists of three functions, each of which implement one step.

    Before the interfaces are described it is necessary to introduce a datatype. Just like other open--use--close interface the functions introduced here work using a handles and the `iconv.h' header defines a special type for the handles used.

    Data Type: iconv_t
    This data type is an abstract type defined in `iconv.h'. The user must not assume anything about the definition of this type, it must be completely opaque.

    Objects of this type can get assigned handles for the conversions using the iconv functions. The objects themselves need not be freed but the conversions for which the handles stand for have to.

    The first step is the function to create a handle.

    Function: iconv_t iconv_open (const char *tocode, const char *fromcode)
    The iconv_open function has to be used before starting a conversion. The two parameters this function takes determine the source and destination character set for the conversion and if the implementation has the possibility to perform such a conversion the function returns a handle.

    If the wanted conversion is not available the function returns (iconv_t) -1. In this case the global variable errno can have the following values:

    EMFILE
    The process already has OPEN_MAX file descriptors open.
    ENFILE
    The system limit of open file is reached.
    ENOMEM
    Not enough memory to carry out the operation.
    EINVAL
    The conversion from fromcode to tocode is not supported.

    It is not possible to use the same descriptor in different threads to perform independent conversions. Within the data structures associated with the descriptor there is information about the conversion state. This must not be messed up by using it in different conversions.

    An iconv descriptor is like a file descriptor as for every use a new descriptor must be created. The descriptor does not stand for all of the conversions from fromset to toset.

    The GNU C library implementation of iconv_open has one significant extension to other implementations. To ease the extension of the set of available conversions the implementation allows storing the necessary files with data and code in arbitrarily many directories. How this extension has to be written will be explained below (see section The iconv Implementation in the GNU C library). Here it is only important to say that all directories mentioned in the GCONV_PATH environment variable are considered if they contain a file `gconv-modules'. These directories need not necessarily be created by the system administrator. In fact, this extension is introduced to help users writing and using their own, new conversions. Of course this does not work for security reasons in SUID binaries; in this case only the system directory is considered and this normally is `prefix/lib/gconv'. The GCONV_PATH environment variable is examined exactly once at the first call of the iconv_open function. Later modifications of the variable have no effect.

    This function got introduced early in the X/Open Portability Guide, version 2. It is supported by all commercial Unices as it is required for the Unix branding. However, the quality and completeness of the implementation varies widely. The function is declared in `iconv.h'.

    The iconv implementation can associate large data structure with the handle returned by iconv_open. Therefore it is crucial to free all the resources once all conversions are carried out and the conversion is not needed anymore.

    Function: int iconv_close (iconv_t cd)
    The iconv_close function frees all resources associated with the handle cd which must have been returned by a successful call to the iconv_open function.

    If the function call was successful the return value is @math{0}. Otherwise it is @math{-1} and errno is set appropriately. Defined error are:

    EBADF
    The conversion descriptor is invalid.

    This function was introduced together with the rest of the iconv functions in XPG2 and it is declared in `iconv.h'.

    The standard defines only one actual conversion function. This has therefore the most general interface: it allows conversion from one buffer to another. Conversion from a file to a buffer, vice versa, or even file to file can be implemented on top of it.

    Function: size_t iconv (iconv_t cd, char **inbuf, size_t *inbytesleft, char **outbuf, size_t *outbytesleft)
    The iconv function converts the text in the input buffer according to the rules associated with the descriptor cd and stores the result in the output buffer. It is possible to call the function for the same text several times in a row since for stateful character sets the necessary state information is kept in the data structures associated with the descriptor.

    The input buffer is specified by *inbuf and it contains *inbytesleft bytes. The extra indirection is necessary for communicating the used input back to the caller (see below). It is important to note that the buffer pointer is of type char and the length is measured in bytes even if the input text is encoded in wide characters.

    The output buffer is specified in a similar way. *outbuf points to the beginning of the buffer with at least *outbytesleft bytes room for the result. The buffer pointer again is of type char and the length is measured in bytes. If outbuf or *outbuf is a null pointer the conversion is performed but no output is available.

    If inbuf is a null pointer the iconv function performs the necessary action to put the state of the conversion into the initial state. This is obviously a no-op for non-stateful encodings, but if the encoding has a state such a function call might put some byte sequences in the output buffer which perform the necessary state changes. The next call with inbuf not being a null pointer then simply goes on from the initial state. It is important that the programmer never makes any assumption on whether the conversion has to deal with states or not. Even if the input and output character sets are not stateful the implementation might still have to keep states. This is due to the implementation chosen for the GNU C library as it is described below. Therefore an iconv call to reset the state should always be performed if some protocol requires this for the output text.

    The conversion stops for three reasons. The first is that all characters from the input buffer are converted. This actually can mean two things: really all bytes from the input buffer are consumed or there are some bytes at the end of the buffer which possibly can form a complete character but the input is incomplete. The second reason for a stop is when the output buffer is full. And the third reason is that the input contains invalid characters.

    In all these cases the buffer pointers after the last successful conversion, for input and output buffer, are stored in inbuf and outbuf and the available room in each buffer is stored in inbytesleft and outbytesleft.

    Since the character sets selected in the iconv_open call can be almost arbitrary there can be situations where the input buffer contains valid characters which have no identical representation in the output character set. The behavior in this situation is undefined. The current behavior of the GNU C library in this situation is to return with an error immediately. This certainly is not the most desirable solution. Therefore future versions will provide better ones but they are not yet finished.

    If all input from the input buffer is successfully converted and stored in the output buffer the function returns the number of non-reversible conversions performed. In all other cases the return value is (size_t) -1 and errno is set appropriately. In this case the value pointed to by inbytesleft is nonzero.

    EILSEQ
    The conversion stopped because of an invalid byte sequence in the input. After the call *inbuf points at the first byte of the invalid byte sequence.
    E2BIG
    The conversion stopped because it ran out of space in the output buffer.
    EINVAL
    The conversion stopped because of an incomplete byte sequence at the end of the input buffer.
    EBADF
    The cd argument is invalid.

    This function was introduced in the XPG2 standard and is declared in the `iconv.h' header.

    The definition of the iconv function is quite good overall. It provides quite flexible functionality. The only problems lie in the boundary cases which are incomplete byte sequences at the end of the input buffer and invalid input. A third problem, which is not really a design problem, is the way conversions are selected. The standard does not say anything about the legitimate names, a minimal set of available conversions. We will see how this negatively impacts other implementations, as is demonstrated below.

    A complete iconv example

    The example below features a solution for a common problem. Given that one knows the internal encoding used by the system for wchar_t strings one often is in the position to read text from a file and store it in wide character buffers. One can do this using mbsrtowcs but then we run into the problems discussed above.

    int
    file2wcs (int fd, const char *charset, wchar_t *outbuf, size_t avail)
    {
      char inbuf[BUFSIZ];
      size_t insize = 0;
      char *wrptr = (char *) outbuf;
      int result = 0;
      iconv_t cd;
    
      cd = iconv_open ("WCHAR_T", charset);
      if (cd == (iconv_t) -1)
        {
          /* Something went wrong.  */
          if (errno == EINVAL)
            error (0, 0, "conversion from '%s' to wchar_t not available",
                   charset);
          else
            perror ("iconv_open");
    
          /* Terminate the output string.  */
          *outbuf = L'\0';
    
          return -1;
        }
    
      while (avail > 0)
        {
          size_t nread;
          size_t nconv;
          char *inptr = inbuf;
    
          /* Read more input.  */
          nread = read (fd, inbuf + insize, sizeof (inbuf) - insize);
          if (nread == 0)
            {
              /* When we come here the file is completely read.
                 This still could mean there are some unused
                 characters in the inbuf.  Put them back.  */
              if (lseek (fd, -insize, SEEK_CUR) == -1)
                result = -1;
    
              /* Now write out the byte sequence to get into the
                 initial state if this is necessary.  */
              iconv (cd, NULL, NULL, &wrptr, &avail);
    
              break;
            }
          insize += nread;
    
          /* Do the conversion.  */
          nconv = iconv (cd, &inptr, &insize, &wrptr, &avail);
          if (nconv == (size_t) -1)
            {
              /* Not everything went right.  It might only be
                 an unfinished byte sequence at the end of the
                 buffer.  Or it is a real problem.  */
              if (errno == EINVAL)
                /* This is harmless.  Simply move the unused
                   bytes to the beginning of the buffer so that
                   they can be used in the next round.  */
                memmove (inbuf, inptr, insize);
              else
                {
                  /* It is a real problem.  Maybe we ran out of
                     space in the output buffer or we have invalid
                     input.  In any case back the file pointer to
                     the position of the last processed byte.  */
                  lseek (fd, -insize, SEEK_CUR);
                  result = -1;
                  break;
                }
            }
        }
    
      /* Terminate the output string.  */
      if (avail >= sizeof (wchar_t))
        *((wchar_t *) wrptr) = L'\0';
    
      if (iconv_close (cd) != 0)
        perror ("iconv_close");
    
      return (wchar_t *) wrptr - outbuf;
    }
    

    This example shows the most important aspects of using the iconv functions. It shows how successive calls to iconv can be used to convert large amounts of text. The user does not have to care about stateful encodings as the functions take care of everything.

    An interesting point is the case where iconv return an error and errno is set to EINVAL. This is not really an error in the transformation. It can happen whenever the input character set contains byte sequences of more than one byte for some character and texts are not processed in one piece. In this case there is a chance that a multibyte sequence is cut. The caller than can simply read the remainder of the takes and feed the offending bytes together with new character from the input to iconv and continue the work. The internal state kept in the descriptor is not unspecified after such an event as it is the case with the conversion functions from the ISO C standard.

    The example also shows the problem of using wide character strings with iconv. As explained in the description of the iconv function above the function always takes a pointer to a char array and the available space is measured in bytes. In the example the output buffer is a wide character buffer. Therefore we use a local variable wrptr of type char * which is used in the iconv calls.

    This looks rather innocent but can lead to problems on platforms which have tight restriction on alignment. Therefore the caller of iconv has to make sure that the pointers passed are suitable for access of characters from the appropriate character set. Since in the above case the input parameter to the function is a wchar_t pointer this is the case (unless the user violates alignment when computing the parameter). But in other situations, especially when writing generic functions where one does not know what type of character set one uses and therefore treats text as a sequence of bytes, it might become tricky.

    Some Details about other iconv Implementations

    This is not really the place to discuss the iconv implementation of other systems but it is necessary to know a bit about them to write portable programs. The above mentioned problems with the specification of the iconv functions can lead to portability issues.

    The first thing to notice is that due to the large number of character sets in use it is certainly not practical to encode the conversions directly in the C library. Therefore the conversion information must come from files outside the C library. This is usually done in one or both of the following ways:

    Some implementations in commercial Unices implement a mixture of these these possibilities, the majority only the second solution. Using loadable modules moves the code out of the library itself and keeps the door open for extensions and improvements. But this design is also limiting on some platforms since not many platforms support dynamic loading in statically linked programs. On platforms without his capability it is therefore not possible to use this interface in statically linked programs. The GNU C library has on ELF platforms no problems with dynamic loading in in these situations and therefore this point is moot. The danger is that one gets acquainted with this and forgets about the restrictions on other systems.

    A second thing to know about other iconv implementations is that the number of available conversions is often very limited. Some implementations provide in the standard release (not special international or developer releases) at most 100 to 200 conversion possibilities. This does not mean 200 different character sets are supported. E.g., conversions from one character set to a set of, say, 10 others counts as 10 conversion. Together with the other direction this makes already 20. One can imagine the thin coverage these platform provide. Some Unix vendors even provide only a handful of conversions which renders them useless for almost all uses.

    This directly leads to a third and probably the most problematic point. The way the iconv conversion functions are implemented on all known Unix system and the availability of the conversion functions from character set @math{@cal{A}} to @math{@cal{B}} and the conversion from @math{@cal{B}} to @math{@cal{C}} does not imply that the conversion from @math{@cal{A}} to @math{@cal{C}} is available.

    This might not seem unreasonable and problematic at first but it is a quite big problem as one will notice shortly after hitting it. To show the problem we assume to write a program which has to convert from @math{@cal{A}} to @math{@cal{C}}. A call like

    cd = iconv_open ("@math{@cal{C}}", "@math{@cal{A}}");
    

    does fail according to the assumption above. But what does the program do now? The conversion is really necessary and therefore simply giving up is no possibility.

    This is a nuisance. The iconv function should take care of this. But how should the program proceed from here on? If it would try to convert to character set @math{@cal{B}} first the two iconv_open calls

    cd1 = iconv_open ("@math{@cal{B}}", "@math{@cal{A}}");
    

    and

    cd2 = iconv_open ("@math{@cal{C}}", "@math{@cal{B}}");
    

    will succeed but how to find @math{@cal{B}}?

    Unfortunately, the answer is: there is no general solution. On some systems guessing might help. On those systems most character sets can convert to and from UTF-8 encoded ISO 10646 or Unicode text. Beside this only some very system-specific methods can help. Since the conversion functions come from loadable modules and these modules must be stored somewhere in the filesystem, one could try to find them and determine from the available file which conversions are available and whether there is an indirect route from @math{@cal{A}} to @math{@cal{C}}.

    This shows one of the design errors of iconv mentioned above. It should at least be possible to determine the list of available conversion programmatically so that if iconv_open says there is no such conversion, one could make sure this also is true for indirect routes.

    The iconv Implementation in the GNU C library

    After reading about the problems of iconv implementations in the last section it is certainly good to note that the implementation in the GNU C library has none of the problems mentioned above. What follows is a step-by-step analysis of the points raised above. The evaluation is based on the current state of the development (as of January 1999). The development of the iconv functions is not complete, but basic functionality has solidified.

    The GNU C library's iconv implementation uses shared loadable modules to implement the conversions. A very small number of conversions are built into the library itself but these are only rather trivial conversions.

    All the benefits of loadable modules are available in the GNU C library implementation. This is especially appealing since the interface is well documented (see below) and it therefore is easy to write new conversion modules. The drawback of using loadable objects is not a problem in the GNU C library, at least on ELF systems. Since the library is able to load shared objects even in statically linked binaries this means that static linking needs not to be forbidden in case one wants to use iconv.

    The second mentioned problem is the number of supported conversions. Currently, the GNU C library supports more than 150 character sets. The way the implementation is designed the number of supported conversions is greater than 22350 (@math{150} times @math{149}). If any conversion from or to a character set is missing it can easily be added.

    Particularly impressive as it may be, this high number is due to the fact that the GNU C library implementation of iconv does not have the third problem mentioned above. I.e., whenever there is a conversion from a character set @math{@cal{A}} to @math{@cal{B}} and from @math{@cal{B}} to @math{@cal{C}} it is always possible to convert from @math{@cal{A}} to @math{@cal{C}} directly. If the iconv_open returns an error and sets errno to EINVAL this really means there is no known way, directly or indirectly, to perform the wanted conversion.

    This is achieved by providing for each character set a conversion from and to UCS-4 encoded ISO 10646. Using ISO 10646 as an intermediate representation it is possible to triangulate, i.e., converting with an intermediate representation.

    There is no inherent requirement to provide a conversion to ISO 10646 for a new character set and it is also possible to provide other conversions where neither source nor destination character set is ISO 10646. The currently existing set of conversions is simply meant to cover all conversions which might be of interest.

    All currently available conversions use the triangulation method above, making conversion run unnecessarily slow. If, e.g., somebody often needs the conversion from ISO-2022-JP to EUC-JP, a quicker solution would involve direct conversion between the two character sets, skipping the input to ISO 10646 first. The two character sets of interest are much more similar to each other than to ISO 10646.

    In such a situation one can easy write a new conversion and provide it as a better alternative. The GNU C library iconv implementation would automatically use the module implementing the conversion if it is specified to be more efficient.

    Format of `gconv-modules' files

    All information about the available conversions comes from a file named `gconv-modules' which can be found in any of the directories along the GCONV_PATH. The `gconv-modules' files are line-oriented text files, where each of the lines has one of the following formats:

    Returning to the example above where one has written a module to directly convert from ISO-2022-JP to EUC-JP and back. All what has to be done is to put the new module, be its name ISO2022JP-EUCJP.so, in a directory and add a file `gconv-modules' with the following content in the same directory:

    module  ISO-2022-JP//   EUC-JP//        ISO2022JP-EUCJP    1
    module  EUC-JP//        ISO-2022-JP//   ISO2022JP-EUCJP    1
    

    To see why this is sufficient, it is necessary to understand how the conversion used by iconv (and described in the descriptor) is selected. The approach to this problem is quite simple.

    At the first call of the iconv_open function the program reads all available `gconv-modules' files and builds up two tables: one containing all the known aliases and another which contains the information about the conversions and which shared object implements them.

    Finding the conversion path in iconv

    The set of available conversions form a directed graph with weighted edges. The weights on the edges are the costs specified in the `gconv-modules' files. The iconv_open function uses an algorithm suitable for search for the best path in such a graph and so constructs a list of conversions which must be performed in succession to get the transformation from the source to the destination character set.

    Explaining why the above `gconv-modules' files allows the iconv implementation to resolve the specific ISO-2022-JP to EUC-JP conversion module instead of the conversion coming with the library itself is straightforward. Since the latter conversion takes two steps (from ISO-2022-JP to ISO 10646 and then from ISO 10646 to EUC-JP) the cost is @math{1+1 = 2}. But the above `gconv-modules' file specifies that the new conversion modules can perform this conversion with only the cost of @math{1}.

    A mysterious piece about the `gconv-modules' file above (and also the file coming with the GNU C library) are the names of the character sets specified in the module lines. Why do almost all the names end in //? And this is not all: the names can actually be regular expressions. At this point of time this mystery should not be revealed, unless you have the relevant spell-casting materials: ashes from an original DOS 6.2 boot disk burnt in effigy, a crucifix blessed by St. Emacs, assorted herbal roots from Central America, sand from Cebu, etc. Sorry! The part of the implementation where this is used is not yet finished. For now please simply follow the existing examples. It'll become clearer once it is. --drepper

    A last remark about the `gconv-modules' is about the names not ending with //. There often is a character set named INTERNAL mentioned. From the discussion above and the chosen name it should have become clear that this is the name for the representation used in the intermediate step of the triangulation. We have said that this is UCS-4 but actually it is not quite right. The UCS-4 specification also includes the specification of the byte ordering used. Since a UCS-4 value consists of four bytes a stored value is effected by byte ordering. The internal representation is not the same as UCS-4 in case the byte ordering of the processor (or at least the running process) is not the same as the one required for UCS-4. This is done for performance reasons as one does not want to perform unnecessary byte-swapping operations if one is not interested in actually seeing the result in UCS-4. To avoid trouble with endianess the internal representation consistently is named INTERNAL even on big-endian systems where the representations are identical.

    iconv module data structures

    So far this section described how modules are located and considered to be used. What remains to be described is the interface of the modules so that one can write new ones. This section describes the interface as it is in use in January 1999. The interface will change in future a bit but hopefully only in an upward compatible way.

    The definitions necessary to write new modules are publicly available in the non-standard header `gconv.h'. The following text will therefore describe the definitions from this header file. But first it is necessary to get an overview.

    From the perspective of the user of iconv the interface is quite simple: the iconv_open function returns a handle which can be used in calls to iconv and finally the handle is freed with a call to iconv_close. The problem is: the handle has to be able to represent the possibly long sequences of conversion steps and also the state of each conversion since the handle is all which is passed to the iconv function. Therefore the data structures are really the elements to understanding the implementation.

    We need two different kinds of data structures. The first describes the conversion and the second describes the state etc. There are really two type definitions like this in `gconv.h'.

    Data type: struct __gconv_step
    This data structure describes one conversion a module can perform. For each function in a loaded module with conversion functions there is exactly one object of this type. This object is shared by all users of the conversion. I.e., this object does not contain any information corresponding to an actual conversion. It only describes the conversion itself.

    struct __gconv_loaded_object *__shlib_handle
    const char *__modname
    int __counter
    All these elements of the structure are used internally in the C library to coordinate loading and unloading the shared. One must not expect any of the other elements be available or initialized.
    const char *__from_name
    const char *__to_name
    __from_name and __to_name contain the names of the source and destination character sets. They can be used to identify the actual conversion to be carried out since one module might implement conversions for more than one character set and/or direction.
    gconv_fct __fct
    gconv_init_fct __init_fct
    gconv_end_fct __end_fct
    These elements contain pointers to the functions in the loadable module. The interface will be explained below.
    int __min_needed_from
    int __max_needed_from
    int __min_needed_to
    int __max_needed_to;
    These values have to be filled in the init function of the module. The __min_needed_from value specifies how many bytes a character of the source character set at least needs. The __max_needed_from specifies the maximum value which also includes possible shift sequences. The __min_needed_to and __max_needed_to values serve the same purpose but this time for the destination character set. It is crucial that these values are accurate since otherwise the conversion functions will have problems or not work at all.
    int __stateful
    This element must also be initialized by the init function. It is nonzero if the source character set is stateful. Otherwise it is zero.
    void *__data
    This element can be used freely by the conversion functions in the module. It can be used to communicate extra information from one call to another. It need not be initialized if not needed at all. If this element gets assigned a pointer to dynamically allocated memory (presumably in the init function) it has to be made sure that the end function deallocates the memory. Otherwise the application will leak memory. It is important to be aware that this data structure is shared by all users of this specification conversion and therefore the __data element must not contain data specific to one specific use of the conversion function.

    Data type: struct __gconv_step_data
    This is the data structure which contains the information specific to each use of the conversion functions.

    char *__outbuf
    char *__outbufend
    These elements specify the output buffer for the conversion step. The __outbuf element points to the beginning of the buffer and __outbufend points to the byte following the last byte in the buffer. The conversion function must not assume anything about the size of the buffer but it can be safely assumed the there is room for at least one complete character in the output buffer. Once the conversion is finished and the conversion is the last step the __outbuf element must be modified to point after last last byte written into the buffer to signal how much output is available. If this conversion step is not the last one the element must not be modified. The __outbufend element must not be modified.
    int __is_last
    This element is nonzero if this conversion step is the last one. This information is necessary for the recursion. See the description of the conversion function internals below. This element must never be modified.
    int __invocation_counter
    The conversion function can use this element to see how many calls of the conversion function already happened. Some character sets require when generating output a certain prolog and by comparing this value with zero one can find out whether it is the first call and therefore the prolog should be emitted or not. This element must never be modified.
    int __internal_use
    This element is another one rarely used but needed in certain situations. It got assigned a nonzero value in case the conversion functions are used to implement mbsrtowcs et.al. I.e., the function is not used directly through the iconv interface. This sometimes makes a difference as it is expected that the iconv functions are used to translate entire texts while the mbsrtowcs functions are normally only used to convert single strings and might be used multiple times to convert entire texts. But in this situation we would have problem complying with some rules of the character set specification. Some character sets require a prolog which must appear exactly once for an entire text. If a number of mbsrtowcs calls are used to convert the text only the first call must add the prolog. But since there is no communication between the different calls of mbsrtowcs the conversion functions have no possibility to find this out. The situation is different for sequences of iconv calls since the handle allows access to the needed information. This element is mostly used together with __invocation_counter in a way like this:
    if (!data->__internal_use
         && data->__invocation_counter == 0)
      /* Emit prolog.  */
      ...
    
    This element must never be modified.
    mbstate_t *__statep
    The __statep element points to an object of type mbstate_t (see section Representing the state of the conversion). The conversion of an stateful character set must use the object pointed to by this element to store information about the conversion state. The __statep element itself must never be modified.
    mbstate_t __state
    This element never must be used directly. It is only part of this structure to have the needed space allocated.

    iconv module interfaces

    With the knowledge about the data structures we now can describe the conversion functions itself. To understand the interface a bit of knowledge about the functionality in the C library which loads the objects with the conversions is necessary.

    It is often the case that one conversion is used more than once. I.e., there are several iconv_open calls for the same set of character sets during one program run. The mbsrtowcs et.al. functions in the GNU C library also use the iconv functionality which increases the number of uses of the same functions even more.

    For this reason the modules do not get loaded exclusively for one conversion. Instead a module once loaded can be used by arbitrarily many iconv or mbsrtowcs calls at the same time. The splitting of the information between conversion function specific information and conversion data makes this possible. The last section showed the two data structures used to do this.

    This is of course also reflected in the interface and semantics of the functions the modules must provide. There are three functions which must have the following names:

    gconv_init
    The gconv_init function initializes the conversion function specific data structure. This very same object is shared by all conversion which use this conversion and therefore no state information about the conversion itself must be stored in here. If a module implements more than one conversion the gconv_init function will be called multiple times.
    gconv_end
    The gconv_end function is responsible to free all resources allocated by the gconv_init function. If there is nothing to do this function can be missing. Special care must be taken if the module implements more than one conversion and the gconv_init function does not allocate the same resources for all conversions.
    gconv
    This is the actual conversion function. It is called to convert one block of text. It gets passed the conversion step information initialized by gconv_init and the conversion data, specific to this use of the conversion functions.

    There are three data types defined for the three module interface function and these define the interface.

    Data type: int (*__gconv_init_fct) (struct __gconv_step *)
    This specifies the interface of the initialization function of the module. It is called exactly once for each conversion the module implements.

    As explained int the description of the struct __gconv_step data structure above the initialization function has to initialize parts of it.

    __min_needed_from
    __max_needed_from
    __min_needed_to
    __max_needed_to
    These elements must be initialized to the exact numbers of the minimum and maximum number of bytes used by one character in the source and destination character set respectively. If the characters all have the same size the minimum and maximum values are the same.
    __stateful
    This element must be initialized to an nonzero value if the source character set is stateful. Otherwise it must be zero.

    If the initialization function needs to communication some information to the conversion function this can happen using the __data element of the __gconv_step structure. But since this data is shared by all the conversion is must not be modified by the conversion function. How this can be used is shown in the example below.

    #define MIN_NEEDED_FROM         1
    #define MAX_NEEDED_FROM         4
    #define MIN_NEEDED_TO           4
    #define MAX_NEEDED_TO           4
    
    int
    gconv_init (struct __gconv_step *step)
    {
      /* Determine which direction.  */
      struct iso2022jp_data *new_data;
      enum direction dir = illegal_dir;
      enum variant var = illegal_var;
      int result;
    
      if (__strcasecmp (step->__from_name, "ISO-2022-JP//") == 0)
        {
          dir = from_iso2022jp;
          var = iso2022jp;
        }
      else if (__strcasecmp (step->__to_name, "ISO-2022-JP//") == 0)
        {
          dir = to_iso2022jp;
          var = iso2022jp;
        }
      else if (__strcasecmp (step->__from_name, "ISO-2022-JP-2//") == 0)
        {
          dir = from_iso2022jp;
          var = iso2022jp2;
        }
      else if (__strcasecmp (step->__to_name, "ISO-2022-JP-2//") == 0)
        {
          dir = to_iso2022jp;
          var = iso2022jp2;
        }
    
      result = __GCONV_NOCONV;
      if (dir != illegal_dir)
        {
          new_data = (struct iso2022jp_data *)
            malloc (sizeof (struct iso2022jp_data));
    
          result = __GCONV_NOMEM;
          if (new_data != NULL)
            {
              new_data->dir = dir;
              new_data->var = var;
              step->__data = new_data;
    
              if (dir == from_iso2022jp)
    	    {
                  step->__min_needed_from = MIN_NEEDED_FROM;
                  step->__max_needed_from = MAX_NEEDED_FROM;
                  step->__min_needed_to = MIN_NEEDED_TO;
                  step->__max_needed_to = MAX_NEEDED_TO;
    	    }
              else
                {
                  step->__min_needed_from = MIN_NEEDED_TO;
                  step->__max_needed_from = MAX_NEEDED_TO;
                  step->__min_needed_to = MIN_NEEDED_FROM;
                  step->__max_needed_to = MAX_NEEDED_FROM + 2;
                }
    
              /* Yes, this is a stateful encoding.  */
              step->__stateful = 1;
    
              result = __GCONV_OK;
            }
        }
    
      return result;
    }
    

    The function first checks which conversion is wanted. The module from which this function is taken implements four different conversion and which one is selected can be determined by comparing the names. The comparison should always be done without paying attention to the case.

    Then a data structure is allocated which contains the necessary information about which conversion is selected. The data structure struct iso2022jp_data is locally defined since outside the module this data is not used at all. Please note that if all four conversions this modules supports are requested there are four data blocks.

    One interesting thing is the initialization of the __min_ and __max_ elements of the step data object. A single ISO-2022-JP character can consist of one to four bytes. Therefore the MIN_NEEDED_FROM and MAX_NEEDED_FROM macros are defined this way. The output is always the INTERNAL character set (aka UCS-4) and therefore each character consists of exactly four bytes. For the conversion from INTERNAL to ISO-2022-JP we have to take into account that escape sequences might be necessary to switch the character sets. Therefore the __max_needed_to element for this direction gets assigned MAX_NEEDED_FROM + 2. This takes into account the two bytes needed for the escape sequences to single the switching. The asymmetry in the maximum values for the two directions can be explained easily: when reading ISO-2022-JP text escape sequences can be handled alone. I.e., it is not necessary to process a real character since the effect of the escape sequence can be recorded in the state information. The situation is different for the other direction. Since it is in general not known which character comes next one cannot emit escape sequences to change the state in advance. This means the escape sequences which have to be emitted together with the next character. Therefore one needs more room then only for the character itself.

    The possible return values of the initialization function are:

    __GCONV_OK
    The initialization succeeded
    __GCONV_NOCONV
    The requested conversion is not supported in the module. This can happen if the `gconv-modules' file has errors.
    __GCONV_NOMEM
    Memory required to store additional information could not be allocated.

    The functions called before the module is unloaded is significantly easier. It often has nothing at all to do in which case it can be left out completely.

    Data type: void (*__gconv_end_fct) (struct gconv_step *)
    The task of this function is it to free all resources allocated in the initialization function. Therefore only the __data element of the object pointed to by the argument is of interest. Continuing the example from the initialization function, the finalization function looks like this:

    void
    gconv_end (struct __gconv_step *data)
    {
      free (data->__data);
    }
    

    The most important function is the conversion function itself. It can get quite complicated for complex character sets. But since this is not of interest here we will only describe a possible skeleton for the conversion function.

    Data type: int (*__gconv_fct) (struct __gconv_step *, struct __gconv_step_data *, const char **, const char *, size_t *, int)
    The conversion function can be called for two basic reason: to convert text or to reset the state. From the description of the iconv function it can be seen why the flushing mode is necessary. What mode is selected is determined by the sixth argument, an integer. If it is nonzero it means that flushing is selected.

    Common to both mode is where the output buffer can be found. The information about this buffer is stored in the conversion step data. A pointer to this is passed as the second argument to this function. The description of the struct __gconv_step_data structure has more information on this.

    What has to be done for flushing depends on the source character set. If it is not stateful nothing has to be done. Otherwise the function has to emit a byte sequence to bring the state object in the initial state. Once this all happened the other conversion modules in the chain of conversions have to get the same chance. Whether another step follows can be determined from the __is_last element of the step data structure to which the first parameter points.

    The more interesting mode is when actually text has to be converted. The first step in this case is to convert as much text as possible from the input buffer and store the result in the output buffer. The start of the input buffer is determined by the third argument which is a pointer to a pointer variable referencing the beginning of the buffer. The fourth argument is a pointer to the byte right after the last byte in the buffer.

    The conversion has to be performed according to the current state if the character set is stateful. The state is stored in an object pointed to by the __statep element of the step data (second argument). Once either the input buffer is empty or the output buffer is full the conversion stops. At this point the pointer variable referenced by the third parameter must point to the byte following the last processed byte. I.e., if all of the input is consumed this pointer and the fourth parameter have the same value.

    What now happens depends on whether this step is the last one or not. If it is the last step the only thing which has to be done is to update the __outbuf element of the step data structure to point after the last written byte. This gives the caller the information on how much text is available in the output buffer. Beside this the variable pointed to by the fifth parameter, which is of type size_t, must be incremented by the number of characters (not bytes) which were converted in a non-reversible way. Then the function can return.

    In case the step is not the last one the later conversion functions have to get a chance to do their work. Therefore the appropriate conversion function has to be called. The information about the functions is stored in the conversion data structures, passed as the first parameter. This information and the step data are stored in arrays so the next element in both cases can be found by simple pointer arithmetic:

    int
    gconv (struct __gconv_step *step, struct __gconv_step_data *data,
           const char **inbuf, const char *inbufend, size_t *written,
           int do_flush)
    {
      struct __gconv_step *next_step = step + 1;
      struct __gconv_step_data *next_data = data + 1;
      ...
    

    The next_step pointer references the next step information and next_data the next data record. The call of the next function therefore will look similar to this:

      next_step->__fct (next_step, next_data, &outerr, outbuf,
                        written, 0)
    

    But this is not yet all. Once the function call returns the conversion function might have some more to do. If the return value of the function is __GCONV_EMPTY_INPUT this means there is more room in the output buffer. Unless the input buffer is empty the conversion functions start all over again and processes the rest of the input buffer. If the return value is not __GCONV_EMPTY_INPUT something went wrong and we have to recover from this.

    A requirement for the conversion function is that the input buffer pointer (the third argument) always points to the last character which was put in the converted form in the output buffer. This is trivially true after the conversion performed in the current step. But if the conversion functions deeper down the stream stop prematurely not all characters from the output buffer are consumed and therefore the input buffer pointers must be backed of to the right position.

    This is easy to do if the input and output character sets have a fixed width for all characters. In this situation we can compute how many characters are left in the output buffer and therefore can correct the input buffer pointer appropriate with a similar computation. Things are getting tricky if either character set has character represented with variable length byte sequences and it gets even more complicated if the conversion has to take care of the state. In these cases the conversion has to be performed once again, from the known state before the initial conversion. I.e., if necessary the state of the conversion has to be reset and the conversion loop has to be executed again. The difference now is that it is known how much input must be created and the conversion can stop before converting the first unused character. Once this is done the input buffer pointers must be updated again and the function can return.

    One final thing should be mentioned. If it is necessary for the conversion to know whether it is the first invocation (in case a prolog has to be emitted) the conversion function should just before returning to the caller increment the __invocation_counter element of the step data structure. See the description of the struct __gconv_step_data structure above for more information on how this can be used.

    The return value must be one of the following values:

    __GCONV_EMPTY_INPUT
    All input was consumed and there is room left in the output buffer.
    __GCONV_FULL_OUTPUT
    No more room in the output buffer. In case this is not the last step this value is propagated down from the call of the next conversion function in the chain.
    __GCONV_INCOMPLETE_INPUT
    The input buffer is not entirely empty since it contains an incomplete character sequence.

    The following example provides a framework for a conversion function. In case a new conversion has to be written the holes in this implementation have to be filled and that is it.

    int
    gconv (struct __gconv_step *step, struct __gconv_step_data *data,
           const char **inbuf, const char *inbufend, size_t *written,
           int do_flush)
    {
      struct __gconv_step *next_step = step + 1;
      struct __gconv_step_data *next_data = data + 1;
      gconv_fct fct = next_step->__fct;
      int status;
    
      /* If the function is called with no input this means we have
         to reset to the initial state.  The possibly partly
         converted input is dropped.  */
      if (do_flush)
        {
          status = __GCONV_OK;
    
          /* Possible emit a byte sequence which put the state object
             into the initial state.  */
    
          /* Call the steps down the chain if there are any but only
             if we successfully emitted the escape sequence.  */
          if (status == __GCONV_OK && ! data->__is_last)
            status = fct (next_step, next_data, NULL, NULL,
                          written, 1);
        }
      else
        {
          /* We preserve the initial values of the pointer variables.  */
          const char *inptr = *inbuf;
          char *outbuf = data->__outbuf;
          char *outend = data->__outbufend;
          char *outptr;
    
          do
            {
              /* Remember the start value for this round.  */
              inptr = *inbuf;
              /* The outbuf buffer is empty.  */
              outptr = outbuf;
    
              /* For stateful encodings the state must be safe here.  */
    
              /* Run the conversion loop.  status is set
                 appropriately afterwards.  */
    
              /* If this is the last step leave the loop, there is
                 nothing we can do.  */
              if (data->__is_last)
                {
                  /* Store information about how many bytes are
                     available.  */
                  data->__outbuf = outbuf;
    
                 /* If any non-reversible conversions were performed,
                    add the number to *written.  */
    
                 break;
               }
    
              /* Write out all output which was produced.  */
              if (outbuf > outptr)
                {
                  const char *outerr = data->__outbuf;
                  int result;
    
                  result = fct (next_step, next_data, &outerr,
                                outbuf, written, 0);
    
                  if (result != __GCONV_EMPTY_INPUT)
                    {
                      if (outerr != outbuf)
                        {
                          /* Reset the input buffer pointer.  We
                             document here the complex case.  */
                          size_t nstatus;
    
                          /* Reload the pointers.  */
                          *inbuf = inptr;
                          outbuf = outptr;
    
                          /* Possibly reset the state.  */
    
                          /* Redo the conversion, but this time
                             the end of the output buffer is at
                             outerr.  */
                        }
    
                      /* Change the status.  */
                      status = result;
                    }
                  else
                    /* All the output is consumed, we can make
                        another run if everything was ok.  */
                    if (status == __GCONV_FULL_OUTPUT)
                      status = __GCONV_OK;
               }
            }
          while (status == __GCONV_OK);
    
          /* We finished one use of this step.  */
          ++data->__invocation_counter;
        }
    
      return status;
    }
    

    This information should be sufficient to write new modules. Anybody doing so should also take a look at the available source code in the GNU C library sources. It contains many examples of working and optimized modules.

    Locales and Internationalization

    Different countries and cultures have varying conventions for how to communicate. These conventions range from very simple ones, such as the format for representing dates and times, to very complex ones, such as the language spoken.

    Internationalization of software means programming it to be able to adapt to the user's favorite conventions. In ISO C, internationalization works by means of locales. Each locale specifies a collection of conventions, one convention for each purpose. The user chooses a set of conventions by specifying a locale (via environment variables).

    All programs inherit the chosen locale as part of their environment. Provided the programs are written to obey the choice of locale, they will follow the conventions preferred by the user.

    What Effects a Locale Has

    Each locale specifies conventions for several purposes, including the following:

    Some aspects of adapting to the specified locale are handled automatically by the library subroutines. For example, all your program needs to do in order to use the collating sequence of the chosen locale is to use strcoll or strxfrm to compare strings.

    Other aspects of locales are beyond the comprehension of the library. For example, the library can't automatically translate your program's output messages into other languages. The only way you can support output in the user's favorite language is to program this more or less by hand. The C library provides functions to handle translations for multiple languages easily.

    This chapter discusses the mechanism by which you can modify the current locale. The effects of the current locale on specific library functions are discussed in more detail in the descriptions of those functions.

    Choosing a Locale

    The simplest way for the user to choose a locale is to set the environment variable LANG. This specifies a single locale to use for all purposes. For example, a user could specify a hypothetical locale named `espana-castellano' to use the standard conventions of most of Spain.

    The set of locales supported depends on the operating system you are using, and so do their names. We can't make any promises about what locales will exist, except for one standard locale called `C' or `POSIX'. Later we will describe how to construct locales.

    A user also has the option of specifying different locales for different purposes--in effect, choosing a mixture of multiple locales.

    For example, the user might specify the locale `espana-castellano' for most purposes, but specify the locale `usa-english' for currency formatting. This might make sense if the user is a Spanish-speaking American, working in Spanish, but representing monetary amounts in US dollars.

    Note that both locales `espana-castellano' and `usa-english', like all locales, would include conventions for all of the purposes to which locales apply. However, the user can choose to use each locale for a particular subset of those purposes.

    Categories of Activities that Locales Affect

    The purposes that locales serve are grouped into categories, so that a user or a program can choose the locale for each category independently. Here is a table of categories; each name is both an environment variable that a user can set, and a macro name that you can use as an argument to setlocale.

    LC_COLLATE
    This category applies to collation of strings (functions strcoll and strxfrm); see section Collation Functions.
    LC_CTYPE
    This category applies to classification and conversion of characters, and to multibyte and wide characters; see section Character Handling, and section Character Set Handling.
    LC_MONETARY
    This category applies to formatting monetary values; see section Generic Numeric Formatting Parameters.
    LC_NUMERIC
    This category applies to formatting numeric values that are not monetary; see section Generic Numeric Formatting Parameters.
    LC_TIME
    This category applies to formatting date and time values; see section Formatting Calendar Time.
    LC_MESSAGES
    This category applies to selecting the language used in the user interface for message translation (see section The Uniforum approach to Message Translation; see section X/Open Message Catalog Handling).
    LC_ALL
    This is not an environment variable; it is only a macro that you can use with setlocale to set a single locale for all purposes. Setting this environment variable overwrites all selections by the other LC_* variables or LANG.
    LANG
    If this environment variable is defined, its value specifies the locale to use for all purposes except as overridden by the variables above.

    When developing the message translation functions it was felt that the functionality provided by the variables above is not sufficient. For example, it should be possible to specify more than one locale name. Take a Swedish user who better speaks German than English, and a program whose messages are output in English by default. It should be possible to specify that the first choice of language is Swedish, the second German, and if this also fails to use English. This is possible with the variable LANGUAGE. For further description of this GNU extension see section User influence on gettext.

    How Programs Set the Locale

    A C program inherits its locale environment variables when it starts up. This happens automatically. However, these variables do not automatically control the locale used by the library functions, because ISO C says that all programs start by default in the standard `C' locale. To use the locales specified by the environment, you must call setlocale. Call it as follows:

    setlocale (LC_ALL, "");
    

    to select a locale based on the user choice of the appropriate environment variables.

    You can also use setlocale to specify a particular locale, for general use or for a specific category.

    The symbols in this section are defined in the header file `locale.h'.

    Function: char * setlocale (int category, const char *locale)
    The function setlocale sets the current locale for category category to locale. A list of all the locales the system provides can be created by running

      locale -a
    

    If category is LC_ALL, this specifies the locale for all purposes. The other possible values of category specify an single purpose (see section Categories of Activities that Locales Affect).

    You can also use this function to find out the current locale by passing a null pointer as the locale argument. In this case, setlocale returns a string that is the name of the locale currently selected for category category.

    The string returned by setlocale can be overwritten by subsequent calls, so you should make a copy of the string (see section Copying and Concatenation) if you want to save it past any further calls to setlocale. (The standard library is guaranteed never to call setlocale itself.)

    You should not modify the string returned by setlocale. It might be the same string that was passed as an argument in a previous call to setlocale. One requirement is that the category must be the same in the call the string was returned and the one when the string is passed in as locale parameter.

    When you read the current locale for category LC_ALL, the value encodes the entire combination of selected locales for all categories. In this case, the value is not just a single locale name. In fact, we don't make any promises about what it looks like. But if you specify the same "locale name" with LC_ALL in a subsequent call to setlocale, it restores the same combination of locale selections.

    To be sure you can use the returned string encoding the currently selected locale at a later time, you must make a copy of the string. It is not guaranteed that the returned pointer remains valid over time.

    When the locale argument is not a null pointer, the string returned by setlocale reflects the newly-modified locale.

    If you specify an empty string for locale, this means to read the appropriate environment variable and use its value to select the locale for category.

    If a nonempty string is given for locale, then the locale of that name is used if possible.

    If you specify an invalid locale name, setlocale returns a null pointer and leaves the current locale unchanged.

    Here is an example showing how you might use setlocale to temporarily switch to a new locale.

    #include <stddef.h>
    #include <locale.h>
    #include <stdlib.h>
    #include <string.h>
    
    void
    with_other_locale (char *new_locale,
                       void (*subroutine) (int),
                       int argument)
    {
      char *old_locale, *saved_locale;
    
      /* Get the name of the current locale.  */
      old_locale = setlocale (LC_ALL, NULL);
    
      /* Copy the name so it won't be clobbered by setlocale. */
      saved_locale = strdup (old_locale);
      if (saved_locale == NULL)
        fatal ("Out of memory");
    
      /* Now change the locale and do some stuff with it. */
      setlocale (LC_ALL, new_locale);
      (*subroutine) (argument);
    
      /* Restore the original locale. */
      setlocale (LC_ALL, saved_locale);
      free (saved_locale);
    }
    

    Portability Note: Some ISO C systems may define additional locale categories, and future versions of the library will do so. For portability, assume that any symbol beginning with `LC_' might be defined in `locale.h'.

    Standard Locales

    The only locale names you can count on finding on all operating systems are these three standard ones:

    "C"
    This is the standard C locale. The attributes and behavior it provides are specified in the ISO C standard. When your program starts up, it initially uses this locale by default.
    "POSIX"
    This is the standard POSIX locale. Currently, it is an alias for the standard C locale.
    ""
    The empty name says to select a locale based on environment variables. See section Categories of Activities that Locales Affect.

    Defining and installing named locales is normally a responsibility of the system administrator at your site (or the person who installed the GNU C library). It is also possible for the user to create private locales. All this will be discussed later when describing the tool to do so.

    If your program needs to use something other than the `C' locale, it will be more portable if you use whatever locale the user specifies with the environment, rather than trying to specify some non-standard locale explicitly by name. Remember, different machines might have different sets of locales installed.

    Accessing Locale Information

    There are several ways to access locale information. The simplest way is to let the C library itself do the work. Several of the functions in this library implicitly access the locale data, and use what information is provided by the currently selected locale. This is how the locale model is meant to work normally.

    As an example take the strftime function, which is meant to nicely format date and time information (see section Formatting Calendar Time). Part of the standard information contained in the LC_TIME category is the names of the months. Instead of requiring the programmer to take care of providing the translations the strftime function does this all by itself. %A in the format string is replaced by the appropriate weekday name of the locale currently selected by LC_TIME. This is an easy example, and wherever possible functions do things automatically in this way.

    But there are quite often situations when there is simply no function to perform the task, or it is simply not possible to do the work automatically. For these cases it is necessary to access the information in the locale directly. To do this the C library provides two functions: localeconv and nl_langinfo. The former is part of ISO C and therefore portable, but has a brain-damaged interface. The second is part of the Unix interface and is portable in as far as the system follows the Unix standards.

    localeconv: It is portable but ...

    Together with the setlocale function the ISO C people invented the localeconv function. It is a masterpiece of poor design. It is expensive to use, not extendable, and not generally usable as it provides access to only LC_MONETARY and LC_NUMERIC related information. Nevertheless, if it is applicable to a given situation it should be used since it is very portable. The function strfmon formats monetary amounts according to the selected locale using this information.

    Function: struct lconv * localeconv (void)
    The localeconv function returns a pointer to a structure whose components contain information about how numeric and monetary values should be formatted in the current locale.

    You should not modify the structure or its contents. The structure might be overwritten by subsequent calls to localeconv, or by calls to setlocale, but no other function in the library overwrites this value.

    Data Type: struct lconv
    localeconv's return value is of this data type. Its elements are described in the following subsections.

    If a member of the structure struct lconv has type char, and the value is CHAR_MAX, it means that the current locale has no value for that parameter.

    Generic Numeric Formatting Parameters

    These are the standard members of struct lconv; there may be others.

    char *decimal_point
    char *mon_decimal_point
    These are the decimal-point separators used in formatting non-monetary and monetary quantities, respectively. In the `C' locale, the value of decimal_point is ".", and the value of mon_decimal_point is "".
    char *thousands_sep
    char *mon_thousands_sep
    These are the separators used to delimit groups of digits to the left of the decimal point in formatting non-monetary and monetary quantities, respectively. In the `C' locale, both members have a value of "" (the empty string).
    char *grouping
    char *mon_grouping
    These are strings that specify how to group the digits to the left of the decimal point. grouping applies to non-monetary quantities and mon_grouping applies to monetary quantities. Use either thousands_sep or mon_thousands_sep to separate the digit groups. Each member of these strings is to be interpreted as an integer value of type char. Successive numbers (from left to right) give the sizes of successive groups (from right to left, starting at the decimal point.) The last member is either 0, in which case the previous member is used over and over again for all the remaining groups, or CHAR_MAX, in which case there is no more grouping--or, put another way, any remaining digits form one large group without separators. For example, if grouping is "\04\03\02", the correct grouping for the number 123456787654321 is `12', `34', `56', `78', `765', `4321'. This uses a group of 4 digits at the end, preceded by a group of 3 digits, preceded by groups of 2 digits (as many as needed). With a separator of `,', the number would be printed as `12,34,56,78,765,4321'. A value of "\03" indicates repeated groups of three digits, as normally used in the U.S. In the standard `C' locale, both grouping and mon_grouping have a value of "". This value specifies no grouping at all.
    char int_frac_digits
    char frac_digits
    These are small integers indicating how many fractional digits (to the right of the decimal point) should be displayed in a monetary value in international and local formats, respectively. (Most often, both members have the same value.) In the standard `C' locale, both of these members have the value CHAR_MAX, meaning "unspecified". The ISO standard doesn't say what to do when you find this value; we recommend printing no fractional digits. (This locale also specifies the empty string for mon_decimal_point, so printing any fractional digits would be confusing!)

    Printing the Currency Symbol

    These members of the struct lconv structure specify how to print the symbol to identify a monetary value--the international analog of `$' for US dollars.

    Each country has two standard currency symbols. The local currency symbol is used commonly within the country, while the international currency symbol is used internationally to refer to that country's currency when it is necessary to indicate the country unambiguously.

    For example, many countries use the dollar as their monetary unit, and when dealing with international currencies it's important to specify that one is dealing with (say) Canadian dollars instead of U.S. dollars or Australian dollars. But when the context is known to be Canada, there is no need to make this explicit--dollar amounts are implicitly assumed to be in Canadian dollars.

    char *currency_symbol
    The local currency symbol for the selected locale. In the standard `C' locale, this member has a value of "" (the empty string), meaning "unspecified". The ISO standard doesn't say what to do when you find this value; we recommend you simply print the empty string as you would print any other string pointed to by this variable.
    char *int_curr_symbol
    The international currency symbol for the selected locale. The value of int_curr_symbol should normally consist of a three-letter abbreviation determined by the international standard ISO 4217 Codes for the Representation of Currency and Funds, followed by a one-character separator (often a space). In the standard `C' locale, this member has a value of "" (the empty string), meaning "unspecified". We recommend you simply print the empty string as you would print any other string pointed to by this variable.
    char p_cs_precedes
    char n_cs_precedes
    char int_p_cs_precedes
    char int_n_cs_precedes
    These members are 1 if the currency_symbol or int_curr_symbol strings should precede the value of a monetary amount, or 0 if the strings should follow the value. The p_cs_precedes and int_p_cs_precedes members apply to positive amounts (or zero), and the n_cs_precedes and int_n_cs_precedes members apply to negative amounts. In the standard `C' locale, all of these members have a value of CHAR_MAX, meaning "unspecified". The ISO standard doesn't say what to do when you find this value. We recommend printing the currency symbol before the amount, which is right for most countries. In other words, treat all nonzero values alike in these members. The members with the int_ prefix apply to the int_curr_symbol while the other two apply to currency_symbol.
    char p_sep_by_space
    char n_sep_by_space
    char int_p_sep_by_space
    char int_n_sep_by_space
    These members are 1 if a space should appear between the currency_symbol or int_curr_symbol strings and the amount, or 0 if no space should appear. The p_sep_by_space and int_p_sep_by_space members apply to positive amounts (or zero), and the n_sep_by_space and int_n_sep_by_space members apply to negative amounts. In the standard `C' locale, all of these members have a value of CHAR_MAX, meaning "unspecified". The ISO standard doesn't say what you should do when you find this value; we suggest you treat it as 1 (print a space). In other words, treat all nonzero values alike in these members. The members with the int_ prefix apply to the int_curr_symbol while the other two apply to currency_symbol. There is one specialty with the int_curr_symbol, though. Since all legal values contain a space at the end the string one either printf this space (if the currency symbol must appear in front and must be separated) or one has to avoid printing this character at all (especially when at the end of the string).

    Printing the Sign of a Monetary Amount

    These members of the struct lconv structure specify how to print the sign (if any) of a monetary value.

    char *positive_sign
    char *negative_sign
    These are strings used to indicate positive (or zero) and negative monetary quantities, respectively. In the standard `C' locale, both of these members have a value of "" (the empty string), meaning "unspecified". The ISO standard doesn't say what to do when you find this value; we recommend printing positive_sign as you find it, even if it is empty. For a negative value, print negative_sign as you find it unless both it and positive_sign are empty, in which case print `-' instead. (Failing to indicate the sign at all seems rather unreasonable.)
    char p_sign_posn
    char n_sign_posn
    char int_p_sign_posn
    char int_n_sign_posn
    These members are small integers that indicate how to position the sign for nonnegative and negative monetary quantities, respectively. (The string used by the sign is what was specified with positive_sign or negative_sign.) The possible values are as follows:
    0
    The currency symbol and quantity should be surrounded by parentheses.
    1
    Print the sign string before the quantity and currency symbol.
    2
    Print the sign string after the quantity and currency symbol.
    3
    Print the sign string right before the currency symbol.
    4
    Print the sign string right after the currency symbol.
    CHAR_MAX
    "Unspecified". Both members have this value in the standard `C' locale.
    The ISO standard doesn't say what you should do when the value is CHAR_MAX. We recommend you print the sign after the currency symbol. The members with the int_ prefix apply to the int_curr_symbol while the other two apply to currency_symbol.

    Pinpoint Access to Locale Data

    When writing the X/Open Portability Guide the authors realized that the localeconv function is not enough to provide reasonable access to locale information. The information which was meant to be available in the locale (as later specified in the POSIX.1 standard) requires more ways to access it. Therefore the nl_langinfo function was introduced.

    Function: char * nl_langinfo (nl_item item)
    The nl_langinfo function can be used to access individual elements of the locale categories. Unlike the localeconv function, which returns all the information, nl_langinfo lets the caller select what information it requires. This is very fast and it is not a problem to call this function multiple times.

    A second advantage is that in addition to the numeric and monetary formatting information, information from the LC_TIME and LC_MESSAGES categories is available.

    The type nl_type is defined in `nl_types.h'. The argument item is a numeric value defined in the header `langinfo.h'. The X/Open standard defines the following values:

    CODESET
    nl_langinfo returns a string with the name of the coded character set used in the selected locale.
    ABDAY_1
    ABDAY_2
    ABDAY_3
    ABDAY_4
    ABDAY_5
    ABDAY_6
    ABDAY_7
    nl_langinfo returns the abbreviated weekday name. ABDAY_1 corresponds to Sunday.
    DAY_1
    DAY_2
    DAY_3
    DAY_4
    DAY_5
    DAY_6
    DAY_7
    Similar to ABDAY_1 etc., but here the return value is the unabbreviated weekday name.
    ABMON_1
    ABMON_2
    ABMON_3
    ABMON_4
    ABMON_5
    ABMON_6
    ABMON_7
    ABMON_8
    ABMON_9
    ABMON_10
    ABMON_11
    ABMON_12
    The return value is abbreviated name of the month. ABMON_1 corresponds to January.
    MON_1
    MON_2
    MON_3
    MON_4
    MON_5
    MON_6
    MON_7
    MON_8
    MON_9
    MON_10
    MON_11
    MON_12
    Similar to ABMON_1 etc., but here the month names are not abbreviated. Here the first value MON_1 also corresponds to January.
    AM_STR
    PM_STR
    The return values are strings which can be used in the representation of time as an hour from 1 to 12 plus an am/pm specifier. Note that in locales which do not use this time representation these strings might be empty, in which case the am/pm format cannot be used at all.
    D_T_FMT
    The return value can be used as a format string for strftime to represent time and date in a locale-specific way.
    D_FMT
    The return value can be used as a format string for strftime to represent a date in a locale-specific way.
    T_FMT
    The return value can be used as a format string for strftime to represent time in a locale-specific way.
    T_FMT_AMPM
    The return value can be used as a format string for strftime to represent time in the am/pm format. Note that if the am/pm format does not make any sense for the selected locale, the return value might be the same as the one for T_FMT.
    ERA
    The return value represents the era used in the current locale. Most locales do not define this value. An example of a locale which does define this value is the Japanese one. In Japan, the traditional representation of dates includes the name of the era corresponding to the then-emperor's reign. Normally it should not be necessary to use this value directly. Specifying the E modifier in their format strings causes the strftime functions to use this information. The format of the returned string is not specified, and therefore you should not assume knowledge of it on different systems.
    ERA_YEAR
    The return value gives the year in the relevant era of the locale. As for ERA it should not be necessary to use this value directly.
    ERA_D_T_FMT
    This return value can be used as a format string for strftime to represent dates and times in a locale-specific era-based way.
    ERA_D_FMT
    This return value can be used as a format string for strftime to represent a date in a locale-specific era-based way.
    ERA_T_FMT
    This return value can be used as a format string for strftime to represent time in a locale-specific era-based way.
    ALT_DIGITS
    The return value is a representation of up to @math{100} values used to represent the values @math{0} to @math{99}. As for ERA this value is not intended to be used directly, but instead indirectly through the strftime function. When the modifier O is used in a format which would otherwise use numerals to represent hours, minutes, seconds, weekdays, months, or weeks, the appropriate value for the locale is used instead.
    INT_CURR_SYMBOL
    The same as the value returned by localeconv in the int_curr_symbol element of the struct lconv.
    CURRENCY_SYMBOL
    CRNCYSTR
    The same as the value returned by localeconv in the currency_symbol element of the struct lconv. CRNCYSTR is a deprecated alias still required by Unix98.
    MON_DECIMAL_POINT
    The same as the value returned by localeconv in the mon_decimal_point element of the struct lconv.
    MON_THOUSANDS_SEP
    The same as the value returned by localeconv in the mon_thousands_sep element of the struct lconv.
    MON_GROUPING
    The same as the value returned by localeconv in the mon_grouping element of the struct lconv.
    POSITIVE_SIGN
    The same as the value returned by localeconv in the positive_sign element of the struct lconv.
    NEGATIVE_SIGN
    The same as the value returned by localeconv in the negative_sign element of the struct lconv.
    INT_FRAC_DIGITS
    The same as the value returned by localeconv in the int_frac_digits element of the struct lconv.
    FRAC_DIGITS
    The same as the value returned by localeconv in the frac_digits element of the struct lconv.
    P_CS_PRECEDES
    The same as the value returned by localeconv in the p_cs_precedes element of the struct lconv.
    P_SEP_BY_SPACE
    The same as the value returned by localeconv in the p_sep_by_space element of the struct lconv.
    N_CS_PRECEDES
    The same as the value returned by localeconv in the n_cs_precedes element of the struct lconv.
    N_SEP_BY_SPACE
    The same as the value returned by localeconv in the n_sep_by_space element of the struct lconv.
    P_SIGN_POSN
    The same as the value returned by localeconv in the p_sign_posn element of the struct lconv.
    N_SIGN_POSN
    The same as the value returned by localeconv in the n_sign_posn element of the struct lconv.
    INT_P_CS_PRECEDES
    The same as the value returned by localeconv in the int_p_cs_precedes element of the struct lconv.
    INT_P_SEP_BY_SPACE
    The same as the value returned by localeconv in the int_p_sep_by_space element of the struct lconv.
    INT_N_CS_PRECEDES
    The same as the value returned by localeconv in the int_n_cs_precedes element of the struct lconv.
    INT_N_SEP_BY_SPACE
    The same as the value returned by localeconv in the int_n_sep_by_space element of the struct lconv.
    INT_P_SIGN_POSN
    The same as the value returned by localeconv in the int_p_sign_posn element of the struct lconv.
    INT_N_SIGN_POSN
    The same as the value returned by localeconv in the int_n_sign_posn element of the struct lconv.
    DECIMAL_POINT
    RADIXCHAR
    The same as the value returned by localeconv in the decimal_point element of the struct lconv. The name RADIXCHAR is a deprecated alias still used in Unix98.
    THOUSANDS_SEP
    THOUSEP
    The same as the value returned by localeconv in the thousands_sep element of the struct lconv. The name THOUSEP is a deprecated alias still used in Unix98.
    GROUPING
    The same as the value returned by localeconv in the grouping element of the struct lconv.
    YESEXPR
    The return value is a regular expression which can be used with the regex function to recognize a positive response to a yes/no question.
    NOEXPR
    The return value is a regular expression which can be used with the regex function to recognize a negative response to a yes/no question.
    YESSTR
    The return value is a locale-specific translation of the positive response to a yes/no question. Using this value is deprecated since it is a very special case of message translation, and is better handled by the message translation functions (see section Message Translation). The use of this symbol is deprecated. Instead message translation should be used.
    NOSTR
    The return value is a locale-specific translation of the negative response to a yes/no question. What is said for YESSTR is also true here. The use of this symbol is deprecated. Instead message translation should be used.

    The file `langinfo.h' defines a lot more symbols but none of them is official. Using them is not portable, and the format of the return values might change. Therefore we recommended you not use them.

    Note that the return value for any valid argument can be used for in all situations (with the possible exception of the am/pm time formatting codes). If the user has not selected any locale for the appropriate category, nl_langinfo returns the information from the "C" locale. It is therefore possible to use this function as shown in the example below.

    If the argument item is not valid, a pointer to an empty string is returned.

    An example of nl_langinfo usage is a function which has to print a given date and time in a locale-specific way. At first one might think that, since strftime internally uses the locale information, writing something like the following is enough:

    size_t
    i18n_time_n_data (char *s, size_t len, const struct tm *tp)
    {
      return strftime (s, len, "%X %D", tp);
    }
    

    The format contains no weekday or month names and therefore is internationally usable. Wrong! The output produced is something like "hh:mm:ss MM/DD/YY". This format is only recognizable in the USA. Other countries use different formats. Therefore the function should be rewritten like this:

    size_t
    i18n_time_n_data (char *s, size_t len, const struct tm *tp)
    {
      return strftime (s, len, nl_langinfo (D_T_FMT), tp);
    }
    

    Now it uses the date and time format of the locale selected when the program runs. If the user selects the locale correctly there should never be a misunderstanding over the time and date format.

    A dedicated function to format numbers

    We have seen that the structure returned by localeconv as well as the values given to nl_langinfo allow you to retrieve the various pieces of locale-specific information to format numbers and monetary amounts. We have also seen that the underlying rules are quite complex.

    Therefore the X/Open standards introduce a function which uses such locale information, making it easier for the user to format numbers according to these rules.

    Function: ssize_t strfmon (char *s, size_t maxsize, const char *format, ...)
    The strfmon function is similar to the strftime function in that it takes a buffer, its size, a format string, and values to write into the buffer as text in a form specified by the format string. Like strftime, the function also returns the number of bytes written into the buffer.

    There are two differences: strfmon can take more than one argument, and, of course, the format specification is different. Like strftime, the format string consists of normal text, which is output as is, and format specifiers, which are indicated by a `%'. Immediately after the `%', you can optionally specify various flags and formatting information before the main formatting character, in a similar way to printf:

    • Immediately following the `%' there can be one or more of the following flags:
      `=f'
      The single byte character f is used for this field as the numeric fill character. By default this character is a space character. Filling with this character is only performed if a left precision is specified. It is not just to fill to the given field width.
      `^'
      The number is printed without grouping the digits according to the rules of the current locale. By default grouping is enabled.
      `+', `('
      At most one of these flags can be used. They select which format to represent the sign of a currency amount. By default, and if `+' is given, the locale equivalent of @math{+}/@math{-} is used. If `(' is given, negative amounts are enclosed in parentheses. The exact format is determined by the values of the LC_MONETARY category of the locale selected at program runtime.
      `!'
      The output will not contain the currency symbol.
      `-'
      The output will be formatted left-justified instead of right-justified if it does not fill the entire field width.

    The next part of a specification is an optional field width. If no width is specified @math{0} is taken. During output, the function first determines how much space is required. If it requires at least as many characters as given by the field width, it is output using as much space as necessary. Otherwise, it is extended to use the full width by filling with the space character. The presence or absence of the `-' flag determines the side at which such padding occurs. If present, the spaces are added at the right making the output left-justified, and vice versa.

    So far the format looks familiar, being similar to the printf and strftime formats. However, the next two optional fields introduce something new. The first one is a `#' character followed by a decimal digit string. The value of the digit string specifies the number of digit positions to the left of the decimal point (or equivalent). This does not include the grouping character when the `^' flag is not given. If the space needed to print the number does not fill the whole width, the field is padded at the left side with the fill character, which can be selected using the `=' flag and by default is a space. For example, if the field width is selected as 6 and the number is @math{123}, the fill character is `*' the result will be `***123'.

    The second optional field starts with a `.' (period) and consists of another decimal digit string. Its value describes the number of characters printed after the decimal point. The default is selected from the current locale (frac_digits, int_frac_digits, see see section Generic Numeric Formatting Parameters). If the exact representation needs more digits than given by the field width, the displayed value is rounded. If the number of fractional digits is selected to be zero, no decimal point is printed.

    As a GNU extension, the strfmon implementation in the GNU libc allows an optional `L' next as a format modifier. If this modifier is given, the argument is expected to be a long double instead of a double value.

    Finally, the last component is a format specifier. There are three specifiers defined:

    `i'
    Use the locale's rules for formatting an international currency value.
    `n'
    Use the locale's rules for formatting a national currency value.
    `%'
    Place a `%' in the output. There must be no flag, width specifier or modifier given, only `%%' is allowed.

    As for printf, the function reads the format string from left to right and uses the values passed to the function following the format string. The values are expected to be either of type double or long double, depending on the presence of the modifier `L'. The result is stored in the buffer pointed to by s. At most maxsize characters are stored.

    The return value of the function is the number of characters stored in s, including the terminating NULL byte. If the number of characters stored would exceed maxsize, the function returns @math{-1} and the content of the buffer s is unspecified. In this case errno is set to E2BIG.

    A few examples should make clear how the function works. It is assumed that all the following pieces of code are executed in a program which uses the USA locale (en_US). The simplest form of the format is this:

    strfmon (buf, 100, "@%n@%n@%n@", 123.45, -567.89, 12345.678);
    

    The output produced is

    "@$123.45@-$567.89@$12,345.68@"
    

    We can notice several things here. First, the widths of the output numbers are different. We have not specified a width in the format string, and so this is no wonder. Second, the third number is printed using thousands separators. The thousands separator for the en_US locale is a comma. The number is also rounded. @math{.678} is rounded to @math{.68} since the format does not specify a precision and the default value in the locale is @math{2}. Finally, note that the national currency symbol is printed since `%n' was used, not `i'. The next example shows how we can align the output.

    strfmon (buf, 100, "@%=*11n@%=*11n@%=*11n@", 123.45, -567.89, 12345.678);
    

    The output this time is:

    "@    $123.45@   -$567.89@ $12,345.68@"
    

    Two things stand out. Firstly, all fields have the same width (eleven characters) since this is the width given in the format and since no number required more characters to be printed. The second important point is that the fill character is not used. This is correct since the white space was not used to achieve a precision given by a `#' modifier, but instead to fill to the given width. The difference becomes obvious if we now add a width specification.

    strfmon (buf, 100, "@%=*11#5n@%=*11#5n@%=*11#5n@",
             123.45, -567.89, 12345.678);
    

    The output is

    "@ $***123.45@-$***567.89@ $12,456.68@"
    

    Here we can see that all the currency symbols are now aligned, and that the space between the currency sign and the number is filled with the selected fill character. Note that although the width is selected to be @math{5} and @math{123.45} has three digits left of the decimal point, the space is filled with three asterisks. This is correct since, as explained above, the width does not include the positions used to store thousands separators. One last example should explain the remaining functionality.

    strfmon (buf, 100, "@%=0(16#5.3i@%=0(16#5.3i@%=0(16#5.3i@",
             123.45, -567.89, 12345.678);
    

    This rather complex format string produces the following output:

    "@ USD 000123,450 @(USD 000567.890)@ USD 12,345.678 @"
    

    The most noticeable change is the alternative way of representing negative numbers. In financial circles this is often done using parentheses, and this is what the `(' flag selected. The fill character is now `0'. Note that this `0' character is not regarded as a numeric zero, and therefore the first and second numbers are not printed using a thousands separator. Since we used the format specifier `i' instead of `n', the international form of the currency symbol is used. This is a four letter string, in this case "USD ". The last point is that since the precision right of the decimal point is selected to be three, the first and second numbers are printed with an extra zero at the end and the third number is printed without rounding.

    Message Translation

    The program's interface with the human should be designed in a way to ease the human the task. One of the possibilities is to use messages in whatever language the user prefers.

    Printing messages in different languages can be implemented in different ways. One could add all the different languages in the source code and add among the variants every time a message has to be printed. This is certainly no good solution since extending the set of languages is difficult (the code must be changed) and the code itself can become really big with dozens of message sets.

    A better solution is to keep the message sets for each language are kept in separate files which are loaded at runtime depending on the language selection of the user.

    The GNU C Library provides two different sets of functions to support message translation. The problem is that neither of the interfaces is officially defined by the POSIX standard. The catgets family of functions is defined in the X/Open standard but this is derived from industry decisions and therefore not necessarily based on reasonable decisions.

    As mentioned above the message catalog handling provides easy extendibility by using external data files which contain the message translations. I.e., these files contain for each of the messages used in the program a translation for the appropriate language. So the tasks of the message handling functions are

    The two approaches mainly differ in the implementation of this last step. The design decisions made for this influences the whole rest.

    X/Open Message Catalog Handling

    The catgets functions are based on the simple scheme:

    Associate every message to translate in the source code with a unique identifier. To retrieve a message from a catalog file solely the identifier is used.

    This means for the author of the program that s/he will have to make sure the meaning of the identifier in the program code and in the message catalogs are always the same.

    Before a message can be translated the catalog file must be located. The user of the program must be able to guide the responsible function to find whatever catalog the user wants. This is separated from what the programmer had in mind.

    All the types, constants and functions for the catgets functions are defined/declared in the `nl_types.h' header file.

    The catgets function family

    Function: nl_catd catopen (const char *cat_name, int flag)
    The catgets function tries to locate the message data file names cat_name and loads it when found. The return value is of an opaque type and can be used in calls to the other functions to refer to this loaded catalog.

    The return value is (nl_catd) -1 in case the function failed and no catalog was loaded. The global variable errno contains a code for the error causing the failure. But even if the function call succeeded this does not mean that all messages can be translated.

    Locating the catalog file must happen in a way which lets the user of the program influence the decision. It is up to the user to decide about the language to use and sometimes it is useful to use alternate catalog files. All this can be specified by the user by setting some environment variables.

    The first problem is to find out where all the message catalogs are stored. Every program could have its own place to keep all the different files but usually the catalog files are grouped by languages and the catalogs for all programs are kept in the same place.

    To tell the catopen function where the catalog for the program can be found the user can set the environment variable NLSPATH to a value which describes her/his choice. Since this value must be usable for different languages and locales it cannot be a simple string. Instead it is a format string (similar to printf's). An example is

    /usr/share/locale/%L/%N:/usr/share/locale/%L/LC_MESSAGES/%N
    

    First one can see that more than one directory can be specified (with the usual syntax of separating them by colons). The next things to observe are the format string, %L and %N in this case. The catopen function knows about several of them and the replacement for all of them is of course different.

    %N
    This format element is substituted with the name of the catalog file. This is the value of the cat_name argument given to catgets.
    %L
    This format element is substituted with the name of the currently selected locale for translating messages. How this is determined is explained below.
    %l
    (This is the lowercase ell.) This format element is substituted with the language element of the locale name. The string describing the selected locale is expected to have the form lang[_terr[.codeset]] and this format uses the first part lang.
    %t
    This format element is substituted by the territory part terr of the name of the currently selected locale. See the explanation of the format above.
    %c
    This format element is substituted by the codeset part codeset of the name of the currently selected locale. See the explanation of the format above.
    %%
    Since % is used in a meta character there must be a way to express the % character in the result itself. Using %% does this just like it works for printf.

    Using NLSPATH allows arbitrary directories to be searched for message catalogs while still allowing different languages to be used. If the NLSPATH environment variable is not set, the default value is

    prefix/share/locale/%L/%N:prefix/share/locale/%L/LC_MESSAGES/%N
    

    where prefix is given to configure while installing the GNU C Library (this value is in many cases /usr or the empty string).

    The remaining problem is to decide which must be used. The value decides about the substitution of the format elements mentioned above. First of all the user can specify a path in the message catalog name (i.e., the name contains a slash character). In this situation the NLSPATH environment variable is not used. The catalog must exist as specified in the program, perhaps relative to the current working directory. This situation in not desirable and catalogs names never should be written this way. Beside this, this behavior is not portable to all other platforms providing the catgets interface.

    Otherwise the values of environment variables from the standard environment are examined (see section Standard Environment Variables). Which variables are examined is decided by the flag parameter of catopen. If the value is NL_CAT_LOCALE (which is defined in `nl_types.h') then the catopen function use the name of the locale currently selected for the LC_MESSAGES category.

    If flag is zero the LANG environment variable is examined. This is a left-over from the early days where the concept of the locales had not even reached the level of POSIX locales.

    The environment variable and the locale name should have a value of the form lang[_terr[.codeset]] as explained above. If no environment variable is set the "C" locale is used which prevents any translation.

    The return value of the function is in any case a valid string. Either it is a translation from a message catalog or it is the same as the string parameter. So a piece of code to decide whether a translation actually happened must look like this:

    {
      char *trans = catgets (desc, set, msg, input_string);
      if (trans == input_string)
        {
          /* Something went wrong.  */
        }
    }
    

    When an error occurred the global variable errno is set to

    EBADF
    The catalog does not exist.
    ENOMSG
    The set/message tuple does not name an existing element in the message catalog.

    While it sometimes can be useful to test for errors programs normally will avoid any test. If the translation is not available it is no big problem if the original, untranslated message is printed. Either the user understands this as well or s/he will look for the reason why the messages are not translated.

    Please note that the currently selected locale does not depend on a call to the setlocale function. It is not necessary that the locale data files for this locale exist and calling setlocale succeeds. The catopen function directly reads the values of the environment variables.

    Function: char * catgets (nl_catd catalog_desc, int set, int message, const char *string)
    The function catgets has to be used to access the massage catalog previously opened using the catopen function. The catalog_desc parameter must be a value previously returned by catopen.

    The next two parameters, set and message, reflect the internal organization of the message catalog files. This will be explained in detail below. For now it is interesting to know that a catalog can consists of several set and the messages in each thread are individually numbered using numbers. Neither the set number nor the message number must be consecutive. They can be arbitrarily chosen. But each message (unless equal to another one) must have its own unique pair of set and message number.

    Since it is not guaranteed that the message catalog for the language selected by the user exists the last parameter string helps to handle this case gracefully. If no matching string can be found string is returned. This means for the programmer that

    • the string parameters should contain reasonable text (this also helps to understand the program seems otherwise there would be no hint on the string which is expected to be returned.
    • all string arguments should be written in the same language.

    It is somewhat uncomfortable to write a program using the catgets functions if no supporting functionality is available. Since each set/message number tuple must be unique the programmer must keep lists of the messages at the same time the code is written. And the work between several people working on the same project must be coordinated. We will see some how these problems can be relaxed a bit (see section How to use the catgets interface).

    Function: int catclose (nl_catd catalog_desc)
    The catclose function can be used to free the resources associated with a message catalog which previously was opened by a call to catopen. If the resources can be successfully freed the function returns 0. Otherwise it return -1 and the global variable errno is set. Errors can occur if the catalog descriptor catalog_desc is not valid in which case errno is set to EBADF.

    Format of the message catalog files

    The only reasonable way the translate all the messages of a function and store the result in a message catalog file which can be read by the catopen function is to write all the message text to the translator and let her/him translate them all. I.e., we must have a file with entries which associate the set/message tuple with a specific translation. This file format is specified in the X/Open standard and is as follows:

    Important: The handling of identifiers instead of numbers for the set and messages is a GNU extension. Systems strictly following the X/Open specification do not have this feature. An example for a message catalog file is this:

    $ This is a leading comment.
    $quote "
    
    $set SetOne
    1 Message with ID 1.
    two "   Message with ID \"two\", which gets the value 2 assigned"
    
    $set SetTwo
    $ Since the last set got the number 1 assigned this set has number 2.
    4000 "The numbers can be arbitrary, they need not start at one."
    

    This small example shows various aspects:

    While this file format is pretty easy it is not the best possible for use in a running program. The catopen function would have to parser the file and handle syntactic errors gracefully. This is not so easy and the whole process is pretty slow. Therefore the catgets functions expect the data in another more compact and ready-to-use file format. There is a special program gencat which is explained in detail in the next section.

    Files in this other format are not human readable. To be easy to use by programs it is a binary file. But the format is byte order independent so translation files can be shared by systems of arbitrary architecture (as long as they use the GNU C Library).

    Details about the binary file format are not important to know since these files are always created by the gencat program. The sources of the GNU C Library also provide the sources for the gencat program and so the interested reader can look through these source files to learn about the file format.

    Generate Message Catalogs files

    The gencat program is specified in the X/Open standard and the GNU implementation follows this specification and so processes all correctly formed input files. Additionally some extension are implemented which help to work in a more reasonable way with the catgets functions.

    The gencat program can be invoked in two ways:

    `gencat [Option]... [Output-File [Input-File]...]`
    

    This is the interface defined in the X/Open standard. If no Input-File parameter is given input will be read from standard input. Multiple input files will be read as if they are concatenated. If Output-File is also missing, the output will be written to standard output. To provide the interface one is used to from other programs a second interface is provided.

    `gencat [Option]... -o Output-File [Input-File]...`
    

    The option `-o' is used to specify the output file and all file arguments are used as input files.

    Beside this one can use `-' or `/dev/stdin' for Input-File to denote the standard input. Corresponding one can use `-' and `/dev/stdout' for Output-File to denote standard output. Using `-' as a file name is allowed in X/Open while using the device names is a GNU extension.

    The gencat program works by concatenating all input files and then merge the resulting collection of message sets with a possibly existing output file. This is done by removing all messages with set/message number tuples matching any of the generated messages from the output file and then adding all the new messages. To regenerate a catalog file while ignoring the old contents therefore requires to remove the output file if it exists. If the output is written to standard output no merging takes place.

    The following table shows the options understood by the gencat program. The X/Open standard does not specify any option for the program so all of these are GNU extensions.

    `-V'
    `--version'
    Print the version information and exit.
    `-h'
    `--help'
    Print a usage message listing all available options, then exit successfully.
    `--new'
    Do never merge the new messages from the input files with the old content of the output files. The old content of the output file is discarded.
    `-H'
    `--header=name'
    This option is used to emit the symbolic names given to sets and messages in the input files for use in the program. Details about how to use this are given in the next section. The name parameter to this option specifies the name of the output file. It will contain a number of C preprocessor #defines to associate a name with a number. Please note that the generated file only contains the symbols from the input files. If the output is merged with the previous content of the output file the possibly existing symbols from the file(s) which generated the old output files are not in the generated header file.

    How to use the catgets interface

    The catgets functions can be used in two different ways. By following slavishly the X/Open specs and not relying on the extension and by using the GNU extensions. We will take a look at the former method first to understand the benefits of extensions.

    Not using symbolic names

    Since the X/Open format of the message catalog files does not allow symbol names we have to work with numbers all the time. When we start writing a program we have to replace all appearances of translatable strings with something like

    catgets (catdesc, set, msg, "string")
    

    catgets is retrieved from a call to catopen which is normally done once at the program start. The "string" is the string we want to translate. The problems start with the set and message numbers.

    In a bigger program several programmers usually work at the same time on the program and so coordinating the number allocation is crucial. Though no two different strings must be indexed by the same tuple of numbers it is highly desirable to reuse the numbers for equal strings with equal translations (please note that there might be strings which are equal in one language but have different translations due to difference contexts).

    The allocation process can be relaxed a bit by different set numbers for different parts of the program. So the number of developers who have to coordinate the allocation can be reduced. But still lists must be keep track of the allocation and errors can easily happen. These errors cannot be discovered by the compiler or the catgets functions. Only the user of the program might see wrong messages printed. In the worst cases the messages are so irritating that they cannot be recognized as wrong. Think about the translations for "true" and "false" being exchanged. This could result in a disaster.

    Using symbolic names

    The problems mentioned in the last section derive from the fact that:

    1. the numbers are allocated once and due to the possibly frequent use of them it is difficult to change a number later.
    2. the numbers do not allow to guess anything about the string and therefore collisions can easily happen.

    By constantly using symbolic names and by providing a method which maps the string content to a symbolic name (however this will happen) one can prevent both problems above. The cost of this is that the programmer has to write a complete message catalog file while s/he is writing the program itself.

    This is necessary since the symbolic names must be mapped to numbers before the program sources can be compiled. In the last section it was described how to generate a header containing the mapping of the names. E.g., for the example message file given in the last section we could call the gencat program as follow (assume `ex.msg' contains the sources).

    gencat -H ex.h -o ex.cat ex.msg
    

    This generates a header file with the following content:

    #define SetTwoSet 0x2   /* ex.msg:8 */
    
    #define SetOneSet 0x1   /* ex.msg:4 */
    #define SetOnetwo 0x2   /* ex.msg:6 */
    

    As can be seen the various symbols given in the source file are mangled to generate unique identifiers and these identifiers get numbers assigned. Reading the source file and knowing about the rules will allow to predict the content of the header file (it is deterministic) but this is not necessary. The gencat program can take care for everything. All the programmer has to do is to put the generated header file in the dependency list of the source files of her/his project and to add a rules to regenerate the header of any of the input files change.

    One word about the symbol mangling. Every symbol consists of two parts: the name of the message set plus the name of the message or the special string Set. So SetOnetwo means this macro can be used to access the translation with identifier two in the message set SetOne.

    The other names denote the names of the message sets. The special string Set is used in the place of the message identifier.

    If in the code the second string of the set SetOne is used the C code should look like this:

    catgets (catdesc, SetOneSet, SetOnetwo,
             "   Message with ID \"two\", which gets the value 2 assigned")
    

    Writing the function this way will allow to change the message number and even the set number without requiring any change in the C source code. (The text of the string is normally not the same; this is only for this example.)

    How does to this allow to develop

    To illustrate the usual way to work with the symbolic version numbers here is a little example. Assume we want to write the very complex and famous greeting program. We start by writing the code as usual:

    #include <stdio.h>
    int
    main (void)
    {
      printf ("Hello, world!\n");
      return 0;
    }
    

    Now we want to internationalize the message and therefore replace the message with whatever the user wants.

    #include <nl_types.h>
    #include <stdio.h>
    #include "msgnrs.h"
    int
    main (void)
    {
      nl_catd catdesc = catopen ("hello.cat", NL_CAT_LOCALE);
      printf (catgets (catdesc, SetMainSet, SetMainHello,
                       "Hello, world!\n"));
      catclose (catdesc);
      return 0;
    }
    

    We see how the catalog object is opened and the returned descriptor used in the other function calls. It is not really necessary to check for failure of any of the functions since even in these situations the functions will behave reasonable. They simply will be return a translation.

    What remains unspecified here are the constants SetMainSet and SetMainHello. These are the symbolic names describing the message. To get the actual definitions which match the information in the catalog file we have to create the message catalog source file and process it using the gencat program.

    $ Messages for the famous greeting program.
    $quote "
    
    $set Main
    Hello "Hallo, Welt!\n"
    

    Now we can start building the program (assume the message catalog source file is named `hello.msg' and the program source file `hello.c'):

    % gencat -H msgnrs.h -o hello.cat hello.msg
    % cat msgnrs.h
    #define MainSet 0x1     /* hello.msg:4 */
    #define MainHello 0x1   /* hello.msg:5 */
    % gcc -o hello hello.c -I.
    % cp hello.cat /usr/share/locale/de/LC_MESSAGES
    % echo $LC_ALL
    de
    % ./hello
    Hallo, Welt!
    %
    

    The call of the gencat program creates the missing header file `msgnrs.h' as well as the message catalog binary. The former is used in the compilation of `hello.c' while the later is placed in a directory in which the catopen function will try to locate it. Please check the LC_ALL environment variable and the default path for catopen presented in the description above.

    The Uniforum approach to Message Translation

    Sun Microsystems tried to standardize a different approach to message translation in the Uniforum group. There never was a real standard defined but still the interface was used in Sun's operation systems. Since this approach fits better in the development process of free software it is also used throughout the GNU project and the GNU `gettext' package provides support for this outside the GNU C Library.

    The code of the `libintl' from GNU `gettext' is the same as the code in the GNU C Library. So the documentation in the GNU `gettext' manual is also valid for the functionality here. The following text will describe the library functions in detail. But the numerous helper programs are not described in this manual. Instead people should read the GNU `gettext' manual (see section `GNU gettext utilities' in Native Language Support Library and Tools). We will only give a short overview.

    Though the catgets functions are available by default on more systems the gettext interface is at least as portable as the former. The GNU `gettext' package can be used wherever the functions are not available.

    The gettext family of functions

    The paradigms underlying the gettext approach to message translations is different from that of the catgets functions the basic functionally is equivalent. There are functions of the following categories:

    What has to be done to translate a message?

    The gettext functions have a very simple interface. The most basic function just takes the string which shall be translated as the argument and it returns the translation. This is fundamentally different from the catgets approach where an extra key is necessary and the original string is only used for the error case.

    If the string which has to be translated is the only argument this of course means the string itself is the key. I.e., the translation will be selected based on the original string. The message catalogs must therefore contain the original strings plus one translation for any such string. The task of the gettext function is it to compare the argument string with the available strings in the catalog and return the appropriate translation. Of course this process is optimized so that this process is not more expensive than an access using an atomic key like in catgets.

    The gettext approach has some advantages but also some disadvantages. Please see the GNU `gettext' manual for a detailed discussion of the pros and cons.

    All the definitions and declarations for gettext can be found in the `libintl.h' header file. On systems where these functions are not part of the C library they can be found in a separate library named `libintl.a' (or accordingly different for shared libraries).

    Function: char * gettext (const char *msgid)
    The gettext function searches the currently selected message catalogs for a string which is equal to msgid. If there is such a string available it is returned. Otherwise the argument string msgid is returned.

    Please note that all though the return value is char * the returned string must not be changed. This broken type results from the history of the function and does not reflect the way the function should be used.

    Please note that above we wrote "message catalogs" (plural). This is a specialty of the GNU implementation of these functions and we will say more about this when we talk about the ways message catalogs are selected (see section How to determine which catalog to be used).

    The gettext function does not modify the value of the global errno variable. This is necessary to make it possible to write something like

      printf (gettext ("Operation failed: %m\n"));
    

    Here the errno value is used in the printf function while processing the %m format element and if the gettext function would change this value (it is called before printf is called) we would get a wrong message.

    So there is no easy way to detect a missing message catalog beside comparing the argument string with the result. But it is normally the task of the user to react on missing catalogs. The program cannot guess when a message catalog is really necessary since for a user who speaks the language the program was developed in does not need any translation.

    The remaining two functions to access the message catalog add some functionality to select a message catalog which is not the default one. This is important if parts of the program are developed independently. Every part can have its own message catalog and all of them can be used at the same time. The C library itself is an example: internally it uses the gettext functions but since it must not depend on a currently selected default message catalog it must specify all ambiguous information.

    Function: char * dgettext (const char *domainname, const char *msgid)
    The dgettext functions acts just like the gettext function. It only takes an additional first argument domainname which guides the selection of the message catalogs which are searched for the translation. If the domainname parameter is the null pointer the dgettext function is exactly equivalent to gettext since the default value for the domain name is used.

    As for gettext the return value type is char * which is an anachronism. The returned string must never be modified.

    Function: char * dcgettext (const char *domainname, const char *msgid, int category)
    The dcgettext adds another argument to those which dgettext takes. This argument category specifies the last piece of information needed to localize the message catalog. I.e., the domain name and the locale category exactly specify which message catalog has to be used (relative to a given directory, see below).

    The dgettext function can be expressed in terms of dcgettext by using

    dcgettext (domain, string, LC_MESSAGES)
    

    instead of

    dgettext (domain, string)
    

    This also shows which values are expected for the third parameter. One has to use the available selectors for the categories available in `locale.h'. Normally the available values are LC_CTYPE, LC_COLLATE, LC_MESSAGES, LC_MONETARY, LC_NUMERIC, and LC_TIME. Please note that LC_ALL must not be used and even though the names might suggest this, there is no relation to the environments variables of this name.

    The dcgettext function is only implemented for compatibility with other systems which have gettext functions. There is not really any situation where it is necessary (or useful) to use a different value but LC_MESSAGES in for the category parameter. We are dealing with messages here and any other choice can only be irritating.

    As for gettext the return value type is char * which is an anachronism. The returned string must never be modified.

    When using the three functions above in a program it is a frequent case that the msgid argument is a constant string. So it is worth to optimize this case. Thinking shortly about this one will realize that as long as no new message catalog is loaded the translation of a message will not change. This optimization is actually implemented by the gettext, dgettext and dcgettext functions.

    How to determine which catalog to be used

    The functions to retrieve the translations for a given message have a remarkable simple interface. But to provide the user of the program still the opportunity to select exactly the translation s/he wants and also to provide the programmer the possibility to influence the way to locate the search for catalogs files there is a quite complicated underlying mechanism which controls all this. The code is complicated the use is easy.

    Basically we have two different tasks to perform which can also be performed by the catgets functions:

    1. Locate the set of message catalogs. There are a number of files for different languages and which all belong to the package. Usually they are all stored in the filesystem below a certain directory. There can be arbitrary many packages installed and they can follow different guidelines for the placement of their files.
    2. Relative to the location specified by the package the actual translation files must be searched, based on the wishes of the user. I.e., for each language the user selects the program should be able to locate the appropriate file.

    This is the functionality required by the specifications for gettext and this is also what the catgets functions are able to do. But there are some problems unresolved:

    We can divide the configuration actions in two parts: the one is performed by the programmer, the other by the user. We will start with the functions the programmer can use since the user configuration will be based on this.

    As the functions described in the last sections already mention separate sets of messages can be selected by a domain name. This is a simple string which should be unique for each program part with uses a separate domain. It is possible to use in one program arbitrary many domains at the same time. E.g., the GNU C Library itself uses a domain named libc while the program using the C Library could use a domain named foo. The important point is that at any time exactly one domain is active. This is controlled with the following function.

    Function: char * textdomain (const char *domainname)
    The textdomain function sets the default domain, which is used in all future gettext calls, to domainname. Please note that dgettext and dcgettext calls are not influenced if the domainname parameter of these functions is not the null pointer.

    Before the first call to textdomain the default domain is messages. This is the name specified in the specification of the gettext API. This name is as good as any other name. No program should ever really use a domain with this name since this can only lead to problems.

    The function returns the value which is from now on taken as the default domain. If the system went out of memory the returned value is NULL and the global variable errno is set to ENOMEM. Despite the return value type being char * the return string must not be changed. It is allocated internally by the textdomain function.

    If the domainname parameter is the null pointer no new default domain is set. Instead the currently selected default domain is returned.

    If the domainname parameter is the empty string the default domain is reset to its initial value, the domain with the name messages. This possibility is questionable to use since the domain messages really never should be used.

    Function: char * bindtextdomain (const char *domainname, const char *dirname)
    The bindtextdomain function can be used to specify the directory which contains the message catalogs for domain domainname for the different languages. To be correct, this is the directory where the hierarchy of directories is expected. Details are explained below.

    For the programmer it is important to note that the translations which come with the program have be placed in a directory hierarchy starting at, say, `/foo/bar'. Then the program should make a bindtextdomain call to bind the domain for the current program to this directory. So it is made sure the catalogs are found. A correctly running program does not depend on the user setting an environment variable.

    The bindtextdomain function can be used several times and if the domainname argument is different the previously bound domains will not be overwritten.

    If the program which wish to use bindtextdomain at some point of time use the chdir function to change the current working directory it is important that the dirname strings ought to be an absolute pathname. Otherwise the addressed directory might vary with the time.

    If the dirname parameter is the null pointer bindtextdomain returns the currently selected directory for the domain with the name domainname.

    The bindtextdomain function returns a pointer to a string containing the name of the selected directory name. The string is allocated internally in the function and must not be changed by the user. If the system went out of core during the execution of bindtextdomain the return value is NULL and the global variable errno is set accordingly.

    Additional functions for more complicated situations

    The functions of the gettext family described so far (and all the catgets functions as well) have one problem in the real world which have been neglected completely in all existing approaches. What is meant here is the handling of plural forms.

    Looking through Unix source code before the time anybody thought about internationalization (and, sadly, even afterwards) one can often find code similar to the following:

       printf ("%d file%s deleted", n, n == 1 ? "" : "s");
    

    After the first complaints from people internationalizing the code people either completely avoided formulations like this or used strings like "file(s)". Both look unnatural and should be avoided. First tries to solve the problem correctly looked like this:

       if (n == 1)
         printf ("%d file deleted", n);
       else
         printf ("%d files deleted", n);
    

    But this does not solve the problem. It helps languages where the plural form of a noun is not simply constructed by adding an `s' but that is all. Once again people fell into the trap of believing the rules their language is using are universal. But the handling of plural forms differs widely between the language families. There are two things we can differ between (and even inside language families);

    The consequence of this is that application writers should not try to solve the problem in their code. This would be localization since it is only usable for certain, hardcoded language environments. Instead the extended gettext interface should be used.

    These extra functions are taking instead of the one key string two strings and an numerical argument. The idea behind this is that using the numerical argument and the first string as a key, the implementation can select using rules specified by the translator the right plural form. The two string arguments then will be used to provide a return value in case no message catalog is found (similar to the normal gettext behavior). In this case the rules for Germanic language is used and it is assumed that the first string argument is the singular form, the second the plural form.

    This has the consequence that programs without language catalogs can display the correct strings only if the program itself is written using a Germanic language. This is a limitation but since the GNU C library (as well as the GNU gettext package) are written as part of the GNU package and the coding standards for the GNU project require program being written in English, this solution nevertheless fulfills its purpose.

    Function: char * ngettext (const char *msgid1, const char *msgid2, unsigned long int n)
    The ngettext function is similar to the gettext function as it finds the message catalogs in the same way. But it takes two extra arguments. The msgid1 parameter must contain the singular form of the string to be converted. It is also used as the key for the search in the catalog. The msgid2 parameter is the plural form. The parameter n is used to determine the plural form. If no message catalog is found msgid1 is returned if n == 1, otherwise msgid2.

    An example for the us of this function is:

      printf (ngettext ("%d file removed", "%d files removed", n), n);
    

    Please note that the numeric value n has to be passed to the printf function as well. It is not sufficient to pass it only to ngettext.

    Function: char * dngettext (const char *domain, const char *msgid1, const char *msgid2, unsigned long int n)
    The dngettext is similar to the dgettext function in the way the message catalog is selected. The difference is that it takes two extra parameter to provide the correct plural form. These two parameters are handled in the same way ngettext handles them.

    Function: char * dcngettext (const char *domain, const char *msgid1, const char *msgid2, unsigned long int n, int category)
    The dcngettext is similar to the dcgettext function in the way the message catalog is selected. The difference is that it takes two extra parameter to provide the correct plural form. These two parameters are handled in the same way ngettext handles them.

    The problem of plural forms

    A description of the problem can be found at the beginning of the last section. Now there is the question how to solve it. Without the input of linguists (which was not available) it was not possible to determine whether there are only a few different forms in which plural forms are formed or whether the number can increase with every new supported language.

    Therefore the solution implemented is to allow the translator to specify the rules of how to select the plural form. Since the formula varies with every language this is the only viable solution except for hardcoding the information in the code (which still would require the possibility of extensions to not prevent the use of new languages). The details are explained in the GNU gettext manual. Here only a a bit of information is provided.

    The information about the plural form selection has to be stored in the header entry (the one with the empty (msgid string). It looks like this:

    Plural-Forms: nplurals=2; plural=n == 1 ? 0 : 1;
    

    The nplurals value must be a decimal number which specifies how many different plural forms exist for this language. The string following plural is an expression which is using the C language syntax. Exceptions are that no negative number are allowed, numbers must be decimal, and the only variable allowed is n. This expression will be evaluated whenever one of the functions ngettext, dngettext, or dcngettext is called. The numeric value passed to these functions is then substituted for all uses of the variable n in the expression. The resulting value then must be greater or equal to zero and smaller than the value given as the value of nplurals.

    The following rules are known at this point. The language with families are listed. But this does not necessarily mean the information can be generalized for the whole family (as can be easily seen in the table below).(1).}

    Only one form:
    Some languages only require one single form. There is no distinction between the singular and plural form. An appropriate header entry would look like this:
    Plural-Forms: nplurals=1; plural=0;
    
    Languages with this property include:
    Finno-Ugric family
    Hungarian
    Asian family
    Japanese
    Turkic/Altaic family
    Turkish
    Two forms, singular used for one only
    This is the form used in most existing programs since it is what English is using. A header entry would look like this:
    Plural-Forms: nplurals=2; plural=n != 1;
    
    (Note: this uses the feature of C expressions that boolean expressions have to value zero or one.) Languages with this property include:
    Germanic family
    Danish, Dutch, English, German, Norwegian, Swedish
    Finno-Ugric family
    Estonian, Finnish
    Latin/Greek family
    Greek
    Semitic family
    Hebrew
    Romance family
    Italian, Spanish
    Artificial
    Esperanto
    Two forms, singular used for zero and one
    Exceptional case in the language family. The header entry would be:
    Plural-Forms: nplurals=2; plural=n>1;
    
    Languages with this property include:
    Romanic family
    French
    Three forms, special cases for one and two
    The header entry would be:
    Plural-Forms: nplurals=3; plural=n==1 ? 0 : n==2 ? 1 : 2;
    
    Languages with this property include:
    Celtic
    Gaeilge
    Three forms, special cases for numbers ending in 1 and 2, 3, 4, except those ending in 1[1-4]
    The header entry would look like this:
    Plural-Forms: nplurals=3; \
        plural=n%100/10==1 ? 2 : n%10==1 ? 0 : (n+9)%10>3 ? 2 : 1;
    
    Languages with this property include:
    Slavic family
    Czech, Russian, Slovak
    Three forms, special case for one and some numbers ending in 2, 3, or 4
    The header entry would look like this:
    Plural-Forms: nplurals=3; \
        plural=n==1 ? 0 : \
               n%10>=2 && n%10<=4 && (n%100<10 || n%100>=20) ? 1 : 2;
    
    (Continuation in the next line is possible.) Languages with this property include:
    Slavic family
    Polish
    Four forms, special case for one and all numbers ending in 2, 3, or 4
    The header entry would look like this:
    Plural-Forms: nplurals=4; \
        plural=n==1 ? 0 : n%10==2 ? 1 : n%10==3 || n%10==4 ? 2 : 3;
    
    Languages with this property include:
    Slavic family
    Slovenian

    How to specify the output character set gettext uses

    gettext not only looks up a translation in a message catalog. It also converts the translation on the fly to the desired output character set. This is useful if the user is working in a different character set than the translator who created the message catalog, because it avoids distributing variants of message catalogs which differ only in the character set.

    The output character set is, by default, the value of nl_langinfo (CODESET), which depends on the LC_CTYPE part of the current locale. But programs which store strings in a locale independent way (e.g. UTF-8) can request that gettext and related functions return the translations in that encoding, by use of the bind_textdomain_codeset function.

    Note that the msgid argument to gettext is not subject to character set conversion. Also, when gettext does not find a translation for msgid, it returns msgid unchanged -- independently of the current output character set. It is therefore recommended that all msgids be US-ASCII strings.

    Function: char * bind_textdomain_codeset (const char *domainname, const char *codeset)
    The bind_textdomain_codeset function can be used to specify the output character set for message catalogs for domain domainname. The codeset argument must be a valid codeset name which can be used for the iconv_open function, or a null pointer.

    If the codeset parameter is the null pointer, bind_textdomain_codeset returns the currently selected codeset for the domain with the name domainname. It returns NULL if no codeset has yet been selected.

    The bind_textdomain_codeset function can be used several times. If used multiple times with the same domainname argument, the later call overrides the settings made by the earlier one.

    The bind_textdomain_codeset function returns a pointer to a string containing the name of the selected codeset. The string is allocated internally in the function and must not be changed by the user. If the system went out of core during the execution of bind_textdomain_codeset, the return value is NULL and the global variable errno is set accordingly. @end deftypefun

    How to use gettext in GUI programs

    One place where the gettext functions, if used normally, have big problems is within programs with graphical user interfaces (GUIs). The problem is that many of the strings which have to be translated are very short. They have to appear in pull-down menus which restricts the length. But strings which are not containing entire sentences or at least large fragments of a sentence may appear in more than one situation in the program but might have different translations. This is especially true for the one-word strings which are frequently used in GUI programs.

    As a consequence many people say that the gettext approach is wrong and instead catgets should be used which indeed does not have this problem. But there is a very simple and powerful method to handle these kind of problems with the gettext functions.

    As as example consider the following fictional situation. A GUI program has a menu bar with the following entries:

    +------------+------------+--------------------------------------+
    | File       | Printer    |                                      |
    +------------+------------+--------------------------------------+
    | Open     | | Select   |
    | New      | | Open     |
    +----------+ | Connect  |
                 +----------+
    

    To have the strings File, Printer, Open, New, Select, and Connect translated there has to be at some point in the code a call to a function of the gettext family. But in two places the string passed into the function would be Open. The translations might not be the same and therefore we are in the dilemma described above.

    One solution to this problem is to artificially enlengthen the strings to make them unambiguous. But what would the program do if no translation is available? The enlengthened string is not what should be printed. So we should use a little bit modified version of the functions.

    To enlengthen the strings a uniform method should be used. E.g., in the example above the strings could be chosen as

    Menu|File
    Menu|Printer
    Menu|File|Open
    Menu|File|New
    Menu|Printer|Select
    Menu|Printer|Open
    Menu|Printer|Connect
    

    Now all the strings are different and if now instead of gettext the following little wrapper function is used, everything works just fine:

      char *
      sgettext (const char *msgid)
      {
        char *msgval = gettext (msgid);
        if (msgval == msgid)
          msgval = strrchr (msgid, '|') + 1;
        return msgval;
      }
    

    What this little function does is to recognize the case when no translation is available. This can be done very efficiently by a pointer comparison since the return value is the input value. If there is no translation we know that the input string is in the format we used for the Menu entries and therefore contains a | character. We simply search for the last occurrence of this character and return a pointer to the character following it. That's it!

    If one now consistently uses the enlengthened string form and replaces the gettext calls with calls to sgettext (this is normally limited to very few places in the GUI implementation) then it is possible to produce a program which can be internationalized.

    With advanced compilers (such as GNU C) one can write the sgettext functions as an inline function or as a macro like this:

    #define sgettext(msgid) \
      ({ const char *__msgid = (msgid);            \
         char *__msgstr = gettext (__msgid);       \
         if (__msgval == __msgid)                  \
           __msgval = strrchr (__msgid, '|') + 1;  \
         __msgval; })
    

    The other gettext functions (dgettext, dcgettext and the ngettext equivalents) can and should have corresponding functions as well which look almost identical, except for the parameters and the call to the underlying function.

    Now there is of course the question why such functions do not exist in the GNU C library? There are two parts of the answer to this question.

    • They are easy to write and therefore can be provided by the project they are used in. This is not an answer by itself and must be seen together with the second part which is:
    • There is no way the C library can contain a version which can work everywhere. The problem is the selection of the character to separate the prefix from the actual string in the enlenghtened string. The examples above used | which is a quite good choice because it resembles a notation frequently used in this context and it also is a character not often used in message strings. But what if the character is used in message strings. Or if the chose character is not available in the character set on the machine one compiles (e.g., | is not required to exist for ISO C; this is why the `iso646.h' file exists in ISO C programming environments).

    There is only one more comment to make left. The wrapper function above require that the translations strings are not enlengthened themselves. This is only logical. There is no need to disambiguate the strings (since they are never used as keys for a search) and one also saves quite some memory and disk space by doing this.

    User influence on gettext

    The last sections described what the programmer can do to internationalize the messages of the program. But it is finally up to the user to select the message s/he wants to see. S/He must understand them.

    The POSIX locale model uses the environment variables LC_COLLATE, LC_CTYPE, LC_MESSAGES, LC_MONETARY, NUMERIC, and LC_TIME to select the locale which is to be used. This way the user can influence lots of functions. As we mentioned above the gettext functions also take advantage of this.

    To understand how this happens it is necessary to take a look at the various components of the filename which gets computed to locate a message catalog. It is composed as follows:

    dir_name/locale/LC_category/domain_name.mo
    

    The default value for dir_name is system specific. It is computed from the value given as the prefix while configuring the C library. This value normally is `/usr' or `/'. For the former the complete dir_name is:

    /usr/share/locale
    

    We can use `/usr/share' since the `.mo' files containing the message catalogs are system independent, so all systems can use the same files. If the program executed the bindtextdomain function for the message domain that is currently handled, the dir_name component is exactly the value which was given to the function as the second parameter. I.e., bindtextdomain allows overwriting the only system dependent and fixed value to make it possible to address files anywhere in the filesystem.

    The category is the name of the locale category which was selected in the program code. For gettext and dgettext this is always LC_MESSAGES, for dcgettext this is selected by the value of the third parameter. As said above it should be avoided to ever use a category other than LC_MESSAGES.

    The locale component is computed based on the category used. Just like for the setlocale function here comes the user selection into the play. Some environment variables are examined in a fixed order and the first environment variable set determines the return value of the lookup process. In detail, for the category LC_xxx the following variables in this order are examined:

    LANGUAGE
    LC_ALL
    LC_xxx
    LANG

    This looks very familiar. With the exception of the LANGUAGE environment variable this is exactly the lookup order the setlocale function uses. But why introducing the LANGUAGE variable?

    The reason is that the syntax of the values these variables can have is different to what is expected by the setlocale function. If we would set LC_ALL to a value following the extended syntax that would mean the setlocale function will never be able to use the value of this variable as well. An additional variable removes this problem plus we can select the language independently of the locale setting which sometimes is useful.

    While for the LC_xxx variables the value should consist of exactly one specification of a locale the LANGUAGE variable's value can consist of a colon separated list of locale names. The attentive reader will realize that this is the way we manage to implement one of our additional demands above: we want to be able to specify an ordered list of language.

    Back to the constructed filename we have only one component missing. The domain_name part is the name which was either registered using the textdomain function or which was given to dgettext or dcgettext as the first parameter. Now it becomes obvious that a good choice for the domain name in the program code is a string which is closely related to the program/package name. E.g., for the GNU C Library the domain name is libc.

    A limit piece of example code should show how the programmer is supposed to work:

    {
      setlocale (LC_ALL, "");
      textdomain ("test-package");
      bindtextdomain ("test-package", "/usr/local/share/locale");
      puts (gettext ("Hello, world!"));
    }
    

    At the program start the default domain is messages, and the default locale is "C". The setlocale call sets the locale according to the user's environment variables; remember that correct functioning of gettext relies on the correct setting of the LC_MESSAGES locale (for looking up the message catalog) and of the LC_CTYPE locale (for the character set conversion). The textdomain call changes the default domain to test-package. The bindtextdomain call specifies that the message catalogs for the domain test-package can be found below the directory `/usr/local/share/locale'.

    If now the user set in her/his environment the variable LANGUAGE to de the gettext function will try to use the translations from the file

    /usr/local/share/locale/de/LC_MESSAGES/test-package.mo
    

    From the above descriptions it should be clear which component of this filename is determined by which source.

    In the above example we assumed that the LANGUAGE environment variable to de. This might be an appropriate selection but what happens if the user wants to use LC_ALL because of the wider usability and here the required value is de_DE.ISO-8859-1? We already mentioned above that a situation like this is not infrequent. E.g., a person might prefer reading a dialect and if this is not available fall back on the standard language.

    The gettext functions know about situations like this and can handle them gracefully. The functions recognize the format of the value of the environment variable. It can split the value is different pieces and by leaving out the only or the other part it can construct new values. This happens of course in a predictable way. To understand this one must know the format of the environment variable value. There are two more or less standardized forms:

    X/Open Format
    language[_territory[.codeset]][@modifier]
    CEN Format (European Community Standard)
    language[_territory][+audience][+special][,[sponsor][_revision]]

    The functions will automatically recognize which format is used. Less specific locale names will be stripped of in the order of the following list:

    1. revision
    2. sponsor
    3. special
    4. codeset
    5. normalized codeset
    6. territory
    7. audience/modifier

    From the last entry one can see that the meaning of the modifier field in the X/Open format and the audience format have the same meaning. Beside one can see that the language field for obvious reasons never will be dropped.

    The only new thing is the normalized codeset entry. This is another goodie which is introduced to help reducing the chaos which derives from the inability of the people to standardize the names of character sets. Instead of ISO-8859-1 one can often see 8859-1, 88591, iso8859-1, or iso_8859-1. The normalized codeset value is generated from the user-provided character set name by applying the following rules:

    1. Remove all characters beside numbers and letters.
    2. Fold letters to lowercase.
    3. If the same only contains digits prepend the string "iso".

    So all of the above name will be normalized to iso88591. This allows the program user much more freely choosing the locale name.

    Even this extended functionality still does not help to solve the problem that completely different names can be used to denote the same locale (e.g., de and german). To be of help in this situation the locale implementation and also the gettext functions know about aliases.

    The file `/usr/share/locale/locale.alias' (replace `/usr' with whatever prefix you used for configuring the C library) contains a mapping of alternative names to more regular names. The system manager is free to add new entries to fill her/his own needs. The selected locale from the environment is compared with the entries in the first column of this file ignoring the case. If they match the value of the second column is used instead for the further handling.

    In the description of the format of the environment variables we already mentioned the character set as a factor in the selection of the message catalog. In fact, only catalogs which contain text written using the character set of the system/program can be used (directly; there will come a solution for this some day). This means for the user that s/he will always have to take care for this. If in the collection of the message catalogs there are files for the same language but coded using different character sets the user has to be careful.

    Programs to handle message catalogs for gettext

    The GNU C Library does not contain the source code for the programs to handle message catalogs for the gettext functions. As part of the GNU project the GNU gettext package contains everything the developer needs. The functionality provided by the tools in this package by far exceeds the abilities of the gencat program described above for the catgets functions.

    There is a program msgfmt which is the equivalent program to the gencat program. It generates from the human-readable and -editable form of the message catalog a binary file which can be used by the gettext functions. But there are several more programs available.

    The xgettext program can be used to automatically extract the translatable messages from a source file. I.e., the programmer need not take care for the translations and the list of messages which have to be translated. S/He will simply wrap the translatable string in calls to gettext et.al and the rest will be done by xgettext. This program has a lot of option which help to customize the output or do help to understand the input better.

    Other programs help to manage development cycle when new messages appear in the source files or when a new translation of the messages appear. here it should only be noted that using all the tools in GNU gettext it is possible to completely automize the handling of message catalog. Beside marking the translatable string in the source code and generating the translations the developers do not have anything to do themselves.

    Searching and Sorting

    This chapter describes functions for searching and sorting arrays of arbitrary objects. You pass the appropriate comparison function to be applied as an argument, along with the size of the objects in the array and the total number of elements.

    Defining the Comparison Function

    In order to use the sorted array library functions, you have to describe how to compare the elements of the array.

    To do this, you supply a comparison function to compare two elements of the array. The library will call this function, passing as arguments pointers to two array elements to be compared. Your comparison function should return a value the way strcmp (see section String/Array Comparison) does: negative if the first argument is "less" than the second, zero if they are "equal", and positive if the first argument is "greater".

    Here is an example of a comparison function which works with an array of numbers of type double:

    int
    compare_doubles (const void *a, const void *b)
    {
      const double *da = (const double *) a;
      const double *db = (const double *) b;
    
      return (*da > *db) - (*da < *db);
    }
    

    The header file `stdlib.h' defines a name for the data type of comparison functions. This type is a GNU extension.

    int comparison_fn_t (const void *, const void *);
    

    Array Search Function

    Generally searching for a specific element in an array means that potentially all elements must be checked. The GNU C library contains functions to perform linear search. The prototypes for the following two functions can be found in `search.h'.

    Function: void * lfind (const void *key, void *base, size_t *nmemb, size_t size, comparison_fn_t compar)
    The lfind function searches in the array with *nmemb elements of size bytes pointed to by base for an element which matches the one pointed to by key. The function pointed to by compar is used decide whether two elements match.

    The return value is a pointer to the matching element in the array starting at base if it is found. If no matching element is available NULL is returned.

    The mean runtime of this function is *nmemb/2. This function should only be used elements often get added to or deleted from the array in which case it might not be useful to sort the array before searching.

    Function: void * lsearch (const void *key, void *base, size_t *nmemb, size_t size, comparison_fn_t compar)
    The lsearch function is similar to the lfind function. It searches the given array for an element and returns it if found. The difference is that if no matching element is found the lsearch function adds the object pointed to by key (with a size of size bytes) at the end of the array and it increments the value of *nmemb to reflect this addition.

    This means for the caller that if it is not sure that the array contains the element one is searching for the memory allocated for the array starting at base must have room for at least size more bytes. If one is sure the element is in the array it is better to use lfind so having more room in the array is always necessary when calling lsearch.

    To search a sorted array for an element matching the key, use the bsearch function. The prototype for this function is in the header file `stdlib.h'.

    Function: void * bsearch (const void *key, const void *array, size_t count, size_t size, comparison_fn_t compare)
    The bsearch function searches the sorted array array for an object that is equivalent to key. The array contains count elements, each of which is of size size bytes.

    The compare function is used to perform the comparison. This function is called with two pointer arguments and should return an integer less than, equal to, or greater than zero corresponding to whether its first argument is considered less than, equal to, or greater than its second argument. The elements of the array must already be sorted in ascending order according to this comparison function.

    The return value is a pointer to the matching array element, or a null pointer if no match is found. If the array contains more than one element that matches, the one that is returned is unspecified.

    This function derives its name from the fact that it is implemented using the binary search algorithm.

    Array Sort Function

    To sort an array using an arbitrary comparison function, use the qsort function. The prototype for this function is in `stdlib.h'.

    Function: void qsort (void *array, size_t count, size_t size, comparison_fn_t compare)
    The qsort function sorts the array array. The array contains count elements, each of which is of size size.

    The compare function is used to perform the comparison on the array elements. This function is called with two pointer arguments and should return an integer less than, equal to, or greater than zero corresponding to whether its first argument is considered less than, equal to, or greater than its second argument.

    Warning: If two objects compare as equal, their order after sorting is unpredictable. That is to say, the sorting is not stable. This can make a difference when the comparison considers only part of the elements. Two elements with the same sort key may differ in other respects.

    If you want the effect of a stable sort, you can get this result by writing the comparison function so that, lacking other reason distinguish between two elements, it compares them by their addresses. Note that doing this may make the sorting algorithm less efficient, so do it only if necessary.

    Here is a simple example of sorting an array of doubles in numerical order, using the comparison function defined above (see section Defining the Comparison Function):

    {
      double *array;
      int size;
      ...
      qsort (array, size, sizeof (double), compare_doubles);
    }
    

    The qsort function derives its name from the fact that it was originally implemented using the "quick sort" algorithm.

    The implementation of qsort in this library might not be an in-place sort and might thereby use an extra amount of memory to store the array.

    Searching and Sorting Example

    Here is an example showing the use of qsort and bsearch with an array of structures. The objects in the array are sorted by comparing their name fields with the strcmp function. Then, we can look up individual objects based on their names.

    #include <stdlib.h>
    #include <stdio.h>
    #include <string.h>
    
    /* Define an array of critters to sort. */
    
    struct critter
      {
        const char *name;
        const char *species;
      };
    
    struct critter muppets[] =
      {
        {"Kermit", "frog"},
        {"Piggy", "pig"},
        {"Gonzo", "whatever"},
        {"Fozzie", "bear"},
        {"Sam", "eagle"},
        {"Robin", "frog"},
        {"Animal", "animal"},
        {"Camilla", "chicken"},
        {"Sweetums", "monster"},
        {"Dr. Strangepork", "pig"},
        {"Link Hogthrob", "pig"},
        {"Zoot", "human"},
        {"Dr. Bunsen Honeydew", "human"},
        {"Beaker", "human"},
        {"Swedish Chef", "human"}
      };
    
    int count = sizeof (muppets) / sizeof (struct critter);
    
    /* This is the comparison function used for sorting and searching. */
    
    int 
    critter_cmp (const struct critter *c1, const struct critter *c2)
    {
      return strcmp (c1->name, c2->name);
    }
    
    /* Print information about a critter. */
    
    void 
    print_critter (const struct critter *c)
    {
      printf ("%s, the %s\n", c->name, c->species);
    }
    
    /* Do the lookup into the sorted array. */
    
    void 
    find_critter (const char *name)
    {
      struct critter target, *result;
      target.name = name;
      result = bsearch (&target, muppets, count, sizeof (struct critter),
                        critter_cmp);
      if (result)
        print_critter (result);
      else
        printf ("Couldn't find %s.\n", name);
    }
    
    /* Main program. */
    
    int
    main (void)
    {
      int i;
    
      for (i = 0; i < count; i++)
        print_critter (&muppets[i]);
      printf ("\n");
    
      qsort (muppets, count, sizeof (struct critter), critter_cmp);
    
      for (i = 0; i < count; i++)
        print_critter (&muppets[i]);
      printf ("\n");
    
      find_critter ("Kermit");
      find_critter ("Gonzo");
      find_critter ("Janice");
    
      return 0;
    }
    

    The output from this program looks like:

    Kermit, the frog
    Piggy, the pig
    Gonzo, the whatever
    Fozzie, the bear
    Sam, the eagle
    Robin, the frog
    Animal, the animal
    Camilla, the chicken
    Sweetums, the monster
    Dr. Strangepork, the pig
    Link Hogthrob, the pig
    Zoot, the human
    Dr. Bunsen Honeydew, the human
    Beaker, the human
    Swedish Chef, the human
    
    Animal, the animal
    Beaker, the human
    Camilla, the chicken
    Dr. Bunsen Honeydew, the human
    Dr. Strangepork, the pig
    Fozzie, the bear
    Gonzo, the whatever
    Kermit, the frog
    Link Hogthrob, the pig
    Piggy, the pig
    Robin, the frog
    Sam, the eagle
    Swedish Chef, the human
    Sweetums, the monster
    Zoot, the human
    
    Kermit, the frog
    Gonzo, the whatever
    Couldn't find Janice.
    

    The hsearch function.

    The functions mentioned so far in this chapter are searching in a sorted or unsorted array. There are other methods to organize information which later should be searched. The costs of insert, delete and search differ. One possible implementation is using hashing tables.

    Function: int hcreate (size_t nel)
    The hcreate function creates a hashing table which can contain at least nel elements. There is no possibility to grow this table so it is necessary to choose the value for nel wisely. The used methods to implement this function might make it necessary to make the number of elements in the hashing table larger than the expected maximal number of elements. Hashing tables usually work inefficient if they are filled 80% or more. The constant access time guaranteed by hashing can only be achieved if few collisions exist. See Knuth's "The Art of Computer Programming, Part 3: Searching and Sorting" for more information.

    The weakest aspect of this function is that there can be at most one hashing table used through the whole program. The table is allocated in local memory out of control of the programmer. As an extension the GNU C library provides an additional set of functions with an reentrant interface which provide a similar interface but which allow to keep arbitrarily many hashing tables.

    It is possible to use more than one hashing table in the program run if the former table is first destroyed by a call to hdestroy.

    The function returns a non-zero value if successful. If it return zero something went wrong. This could either mean there is already a hashing table in use or the program runs out of memory.

    Function: void hdestroy (void)
    The hdestroy function can be used to free all the resources allocated in a previous call of hcreate. After a call to this function it is again possible to call hcreate and allocate a new table with possibly different size.

    It is important to remember that the elements contained in the hashing table at the time hdestroy is called are not freed by this function. It is the responsibility of the program code to free those strings (if necessary at all). Freeing all the element memory is not possible without extra, separately kept information since there is no function to iterate through all available elements in the hashing table. If it is really necessary to free a table and all elements the programmer has to keep a list of all table elements and before calling hdestroy s/he has to free all element's data using this list. This is a very unpleasant mechanism and it also shows that this kind of hashing tables is mainly meant for tables which are created once and used until the end of the program run.

    Entries of the hashing table and keys for the search are defined using this type:

    Data type: struct ENTRY
    Both elements of this structure are pointers to zero-terminated strings. This is a limiting restriction of the functionality of the hsearch functions. They can only be used for data sets which use the NUL character always and solely to terminate the records. It is not possible to handle general binary data.

    char *key
    Pointer to a zero-terminated string of characters describing the key for the search or the element in the hashing table.
    char *data
    Pointer to a zero-terminated string of characters describing the data. If the functions will be called only for searching an existing entry this element might stay undefined since it is not used.

    Function: ENTRY * hsearch (ENTRY item, ACTION action)
    To search in a hashing table created using hcreate the hsearch function must be used. This function can perform simple search for an element (if action has the FIND) or it can alternatively insert the key element into the hashing table, possibly replacing a previous value (if action is ENTER).

    The key is denoted by a pointer to an object of type ENTRY. For locating the corresponding position in the hashing table only the key element of the structure is used.

    The return value depends on the action parameter value. If it is FIND the value is a pointer to the matching element in the hashing table or NULL if no matching element exists. If action is ENTER the return value is only NULL if the programs runs out of memory while adding the new element to the table. Otherwise the return value is a pointer to the element in the hashing table which contains newly added element based on the data in key.

    As mentioned before the hashing table used by the functions described so far is global and there can be at any time at most one hashing table in the program. A solution is to use the following functions which are a GNU extension. All have in common that they operate on a hashing table which is described by the content of an object of the type struct hsearch_data. This type should be treated as opaque, none of its members should be changed directly.

    Function: int hcreate_r (size_t nel, struct hsearch_data *htab)
    The hcreate_r function initializes the object pointed to by htab to contain a hashing table with at least nel elements. So this function is equivalent to the hcreate function except that the initialized data structure is controlled by the user.

    This allows having more than one hashing table at one time. The memory necessary for the struct hsearch_data object can be allocated dynamically.

    The return value is non-zero if the operation were successful. if the return value is zero something went wrong which probably means the programs runs out of memory.

    Function: void hdestroy_r (struct hsearch_data *htab)
    The hdestroy_r function frees all resources allocated by the hcreate_r function for this very same object htab. As for hdestroy it is the programs responsibility to free the strings for the elements of the table.

    Function: int hsearch_r (ENTRY item, ACTION action, ENTRY **retval, struct hsearch_data *htab)
    The hsearch_r function is equivalent to hsearch. The meaning of the first two arguments is identical. But instead of operating on a single global hashing table the function works on the table described by the object pointed to by htab (which is initialized by a call to hcreate_r).

    Another difference to hcreate is that the pointer to the found entry in the table is not the return value of the functions. It is returned by storing it in a pointer variables pointed to by the retval parameter. The return value of the function is an integer value indicating success if it is non-zero and failure if it is zero. In the latter case the global variable errno signals the reason for the failure.

    ENOMEM
    The table is filled and hsearch_r was called with an so far unknown key and action set to ENTER.
    ESRCH
    The action parameter is FIND and no corresponding element is found in the table.

    The tsearch function.

    Another common form to organize data for efficient search is to use trees. The tsearch function family provides a nice interface to functions to organize possibly large amounts of data by providing a mean access time proportional to the logarithm of the number of elements. The GNU C library implementation even guarantees that this bound is never exceeded even for input data which cause problems for simple binary tree implementations.

    The functions described in the chapter are all described in the System V and X/Open specifications and are therefore quite portable.

    In contrast to the hsearch functions the tsearch functions can be used with arbitrary data and not only zero-terminated strings.

    The tsearch functions have the advantage that no function to initialize data structures is necessary. A simple pointer of type void * initialized to NULL is a valid tree and can be extended or searched.

    Function: void * tsearch (const void *key, void **rootp, comparison_fn_t compar)
    The tsearch function searches in the tree pointed to by *rootp for an element matching key. The function pointed to by compar is used to determine whether two elements match. See section Defining the Comparison Function, for a specification of the functions which can be used for the compar parameter.

    If the tree does not contain a matching entry the key value will be added to the tree. tsearch does not make a copy of the object pointed to by key (how could it since the size is unknown). Instead it adds a reference to this object which means the object must be available as long as the tree data structure is used.

    The tree is represented by a pointer to a pointer since it is sometimes necessary to change the root node of the tree. So it must not be assumed that the variable pointed to by rootp has the same value after the call. This also shows that it is not safe to call the tsearch function more than once at the same time using the same tree. It is no problem to run it more than once at a time on different trees.

    The return value is a pointer to the matching element in the tree. If a new element was created the pointer points to the new data (which is in fact key). If an entry had to be created and the program ran out of space NULL is returned.

    Function: void * tfind (const void *key, void *const *rootp, comparison_fn_t compar)
    The tfind function is similar to the tsearch function. It locates an element matching the one pointed to by key and returns a pointer to this element. But if no matching element is available no new element is entered (note that the rootp parameter points to a constant pointer). Instead the function returns NULL.

    Another advantage of the tsearch function in contrast to the hsearch functions is that there is an easy way to remove elements.

    Function: void * tdelete (const void *key, void **rootp, comparison_fn_t compar)
    To remove a specific element matching key from the tree tdelete can be used. It locates the matching element using the same method as tfind. The corresponding element is then removed and a pointer to the parent of the deleted node is returned by the function. If there is no matching entry in the tree nothing can be deleted and the function returns NULL. If the root of the tree is deleted tdelete returns some unspecified value not equal to NULL.

    Function: void tdestroy (void *vroot, __free_fn_t freefct)
    If the complete search tree has to be removed one can use tdestroy. It frees all resources allocated by the tsearch function to generate the tree pointed to by vroot.

    For the data in each tree node the function freefct is called. The pointer to the data is passed as the argument to the function. If no such work is necessary freefct must point to a function doing nothing. It is called in any case.

    This function is a GNU extension and not covered by the System V or X/Open specifications.

    In addition to the function to create and destroy the tree data structure, there is another function which allows you to apply a function to all elements of the tree. The function must have this type:

    void __action_fn_t (const void *nodep, VISIT value, int level);
    

    The nodep is the data value of the current node (once given as the key argument to tsearch). level is a numeric value which corresponds to the depth of the current node in the tree. The root node has the depth @math{0} and its children have a depth of @math{1} and so on. The VISIT type is an enumeration type.

    Data Type: VISIT
    The VISIT value indicates the status of the current node in the tree and how the function is called. The status of a node is either `leaf' or `internal node'. For each leaf node the function is called exactly once, for each internal node it is called three times: before the first child is processed, after the first child is processed and after both children are processed. This makes it possible to handle all three methods of tree traversal (or even a combination of them).

    preorder
    The current node is an internal node and the function is called before the first child was processed.
    postorder
    The current node is an internal node and the function is called after the first child was processed.
    endorder
    The current node is an internal node and the function is called after the second child was processed.
    leaf
    The current node is a leaf.

    Function: void twalk (const void *root, __action_fn_t action)
    For each node in the tree with a node pointed to by root, the twalk function calls the function provided by the parameter action. For leaf nodes the function is called exactly once with value set to leaf. For internal nodes the function is called three times, setting the value parameter or action to the appropriate value. The level argument for the action function is computed while descending the tree with increasing the value by one for the descend to a child, starting with the value @math{0} for the root node.

    Since the functions used for the action parameter to twalk must not modify the tree data, it is safe to run twalk in more than one thread at the same time, working on the same tree. It is also safe to call tfind in parallel. Functions which modify the tree must not be used, otherwise the behaviour is undefined.

    Pattern Matching

    The GNU C Library provides pattern matching facilities for two kinds of patterns: regular expressions and file-name wildcards. The library also provides a facility for expanding variable and command references and parsing text into words in the way the shell does.

    Wildcard Matching

    This section describes how to match a wildcard pattern against a particular string. The result is a yes or no answer: does the string fit the pattern or not. The symbols described here are all declared in `fnmatch.h'.

    Function: int fnmatch (const char *pattern, const char *string, int flags)
    This function tests whether the string string matches the pattern pattern. It returns 0 if they do match; otherwise, it returns the nonzero value FNM_NOMATCH. The arguments pattern and string are both strings.

    The argument flags is a combination of flag bits that alter the details of matching. See below for a list of the defined flags.

    In the GNU C Library, fnmatch cannot experience an "error"---it always returns an answer for whether the match succeeds. However, other implementations of fnmatch might sometimes report "errors". They would do so by returning nonzero values that are not equal to FNM_NOMATCH.

    These are the available flags for the flags argument:

    FNM_FILE_NAME
    Treat the `/' character specially, for matching file names. If this flag is set, wildcard constructs in pattern cannot match `/' in string. Thus, the only way to match `/' is with an explicit `/' in pattern.
    FNM_PATHNAME
    This is an alias for FNM_FILE_NAME; it comes from POSIX.2. We don't recommend this name because we don't use the term "pathname" for file names.
    FNM_PERIOD
    Treat the `.' character specially if it appears at the beginning of string. If this flag is set, wildcard constructs in pattern cannot match `.' as the first character of string. If you set both FNM_PERIOD and FNM_FILE_NAME, then the special treatment applies to `.' following `/' as well as to `.' at the beginning of string. (The shell uses the FNM_PERIOD and FNM_FILE_NAME flags together for matching file names.)
    FNM_NOESCAPE
    Don't treat the `\' character specially in patterns. Normally, `\' quotes the following character, turning off its special meaning (if any) so that it matches only itself. When quoting is enabled, the pattern `\?' matches only the string `?', because the question mark in the pattern acts like an ordinary character. If you use FNM_NOESCAPE, then `\' is an ordinary character.
    FNM_LEADING_DIR
    Ignore a trailing sequence of characters starting with a `/' in string; that is to say, test whether string starts with a directory name that pattern matches. If this flag is set, either `foo*' or `foobar' as a pattern would match the string `foobar/frobozz'.
    FNM_CASEFOLD
    Ignore case in comparing string to pattern.
    FNM_EXTMATCH
    Recognize beside the normal patterns also the extended patterns introduced in `ksh'. The patterns are written in the form explained in the following table where pattern-list is a | separated list of patterns.
    ?(pattern-list)
    The pattern matches if zero or one occurences of any of the patterns in the pattern-list allow matching the input string.
    *(pattern-list)
    The pattern matches if zero or more occurences of any of the patterns in the pattern-list allow matching the input string.
    +(pattern-list)
    The pattern matches if one or more occurences of any of the patterns in the pattern-list allow matching the input string.
    @(pattern-list)
    The pattern matches if exactly one occurence of any of the patterns in the pattern-list allows matching the input string.
    !(pattern-list)
    The pattern matches if the input string cannot be matched with any of the patterns in the pattern-list.

    Globbing

    The archetypal use of wildcards is for matching against the files in a directory, and making a list of all the matches. This is called globbing.

    You could do this using fnmatch, by reading the directory entries one by one and testing each one with fnmatch. But that would be slow (and complex, since you would have to handle subdirectories by hand).

    The library provides a function glob to make this particular use of wildcards convenient. glob and the other symbols in this section are declared in `glob.h'.

    Calling glob

    The result of globbing is a vector of file names (strings). To return this vector, glob uses a special data type, glob_t, which is a structure. You pass glob the address of the structure, and it fills in the structure's fields to tell you about the results.

    Data Type: glob_t
    This data type holds a pointer to a word vector. More precisely, it records both the address of the word vector and its size. The GNU implementation contains some more fields which are non-standard extensions.

    gl_pathc
    The number of elements in the vector, excluding the initial null entries if the GLOB_DOOFFS flag is used (see gl_offs below).
    gl_pathv
    The address of the vector. This field has type char **.
    gl_offs
    The offset of the first real element of the vector, from its nominal address in the gl_pathv field. Unlike the other fields, this is always an input to glob, rather than an output from it. If you use a nonzero offset, then that many elements at the beginning of the vector are left empty. (The glob function fills them with null pointers.) The gl_offs field is meaningful only if you use the GLOB_DOOFFS flag. Otherwise, the offset is always zero regardless of what is in this field, and the first real element comes at the beginning of the vector.
    gl_closedir
    The address of an alternative implementation of the closedir function. It is used if the GLOB_ALTDIRFUNC bit is set in the flag parameter. The type of this field is void (*) (void *). This is a GNU extension.
    gl_readdir
    The address of an alternative implementation of the readdir function used to read the contents of a directory. It is used if the GLOB_ALTDIRFUNC bit is set in the flag parameter. The type of this field is struct dirent *(*) (void *). This is a GNU extension.
    gl_opendir
    The address of an alternative implementation of the opendir function. It is used if the GLOB_ALTDIRFUNC bit is set in the flag parameter. The type of this field is void *(*) (const char *). This is a GNU extension.
    gl_stat
    The address of an alternative implementation of the stat function to get information about an object in the filesystem. It is used if the GLOB_ALTDIRFUNC bit is set in the flag parameter. The type of this field is int (*) (const char *, struct stat *). This is a GNU extension.
    gl_lstat
    The address of an alternative implementation of the lstat function to get information about an object in the filesystems, not following symbolic links. It is used if the GLOB_ALTDIRFUNC bit is set in the flag parameter. The type of this field is int (*) (const char *, struct stat *). This is a GNU extension.

    For use in the glob64 function `glob.h' contains another definition for a very similar type. glob64_t differs from glob_t only in the types of the members gl_readdir, gl_stat, and gl_lstat.

    Data Type: glob64_t
    This data type holds a pointer to a word vector. More precisely, it records both the address of the word vector and its size. The GNU implementation contains some more fields which are non-standard extensions.

    gl_pathc
    The number of elements in the vector, excluding the initial null entries if the GLOB_DOOFFS flag is used (see gl_offs below).
    gl_pathv
    The address of the vector. This field has type char **.
    gl_offs
    The offset of the first real element of the vector, from its nominal address in the gl_pathv field. Unlike the other fields, this is always an input to glob, rather than an output from it. If you use a nonzero offset, then that many elements at the beginning of the vector are left empty. (The glob function fills them with null pointers.) The gl_offs field is meaningful only if you use the GLOB_DOOFFS flag. Otherwise, the offset is always zero regardless of what is in this field, and the first real element comes at the beginning of the vector.
    gl_closedir
    The address of an alternative implementation of the closedir function. It is used if the GLOB_ALTDIRFUNC bit is set in the flag parameter. The type of this field is void (*) (void *). This is a GNU extension.
    gl_readdir
    The address of an alternative implementation of the readdir64 function used to read the contents of a directory. It is used if the GLOB_ALTDIRFUNC bit is set in the flag parameter. The type of this field is struct dirent64 *(*) (void *). This is a GNU extension.
    gl_opendir
    The address of an alternative implementation of the opendir function. It is used if the GLOB_ALTDIRFUNC bit is set in the flag parameter. The type of this field is void *(*) (const char *). This is a GNU extension.
    gl_stat
    The address of an alternative implementation of the stat64 function to get information about an object in the filesystem. It is used if the GLOB_ALTDIRFUNC bit is set in the flag parameter. The type of this field is int (*) (const char *, struct stat64 *). This is a GNU extension.
    gl_lstat
    The address of an alternative implementation of the lstat64 function to get information about an object in the filesystems, not following symbolic links. It is used if the GLOB_ALTDIRFUNC bit is set in the flag parameter. The type of this field is int (*) (const char *, struct stat64 *). This is a GNU extension.

    Function: int glob (const char *pattern, int flags, int (*errfunc) (const char *filename, int error-code), glob_t *vector-ptr)
    The function glob does globbing using the pattern pattern in the current directory. It puts the result in a newly allocated vector, and stores the size and address of this vector into *vector-ptr. The argument flags is a combination of bit flags; see section Flags for Globbing, for details of the flags.

    The result of globbing is a sequence of file names. The function glob allocates a string for each resulting word, then allocates a vector of type char ** to store the addresses of these strings. The last element of the vector is a null pointer. This vector is called the word vector.

    To return this vector, glob stores both its address and its length (number of elements, not counting the terminating null pointer) into *vector-ptr.

    Normally, glob sorts the file names alphabetically before returning them. You can turn this off with the flag GLOB_NOSORT if you want to get the information as fast as possible. Usually it's a good idea to let glob sort them--if you process the files in alphabetical order, the users will have a feel for the rate of progress that your application is making.

    If glob succeeds, it returns 0. Otherwise, it returns one of these error codes:

    GLOB_ABORTED
    There was an error opening a directory, and you used the flag GLOB_ERR or your specified errfunc returned a nonzero value. See below for an explanation of the GLOB_ERR flag and errfunc.
    GLOB_NOMATCH
    The pattern didn't match any existing files. If you use the GLOB_NOCHECK flag, then you never get this error code, because that flag tells glob to pretend that the pattern matched at least one file.
    GLOB_NOSPACE
    It was impossible to allocate memory to hold the result.

    In the event of an error, glob stores information in *vector-ptr about all the matches it has found so far.

    It is important to notive that the glob function will not fail if it encounters directories or files which cannot be handled without the LFS interfaces. The implementation of glob is supposed to use these functions internally. This at least is the assumptions made by the Unix standard. The GNU extension of allowing the user to provide own directory handling and stat functions complicates things a bit. If these callback functions are used and a large file or directory is encountered glob can fail.

    Function: int glob64 (const char *pattern, int flags, int (*errfunc) (const char *filename, int error-code), glob64_t *vector-ptr)
    The glob64 function was added as part of the Large File Summit extensions but is not part of the original LFS proposal. The reason for this is simple: it is not necessary. The necessity for a glob64 function is added by the extensions of the GNU glob implementation which allows the user to provide own directory handling and stat functions. The readdir and stat functions do depend on the choice of _FILE_OFFSET_BITS since the definition of the types struct dirent and struct stat will change depending on the choice.

    Beside this difference the glob64 works just like glob in all aspects.

    This function is a GNU extension.

    Flags for Globbing

    This section describes the flags that you can specify in the flags argument to glob. Choose the flags you want, and combine them with the C bitwise OR operator |.

    GLOB_APPEND
    Append the words from this expansion to the vector of words produced by previous calls to glob. This way you can effectively expand several words as if they were concatenated with spaces between them. In order for appending to work, you must not modify the contents of the word vector structure between calls to glob. And, if you set GLOB_DOOFFS in the first call to glob, you must also set it when you append to the results. Note that the pointer stored in gl_pathv may no longer be valid after you call glob the second time, because glob might have relocated the vector. So always fetch gl_pathv from the glob_t structure after each glob call; never save the pointer across calls.
    GLOB_DOOFFS
    Leave blank slots at the beginning of the vector of words. The gl_offs field says how many slots to leave. The blank slots contain null pointers.
    GLOB_ERR
    Give up right away and report an error if there is any difficulty reading the directories that must be read in order to expand pattern fully. Such difficulties might include a directory in which you don't have the requisite access. Normally, glob tries its best to keep on going despite any errors, reading whatever directories it can. You can exercise even more control than this by specifying an error-handler function errfunc when you call glob. If errfunc is not a null pointer, then glob doesn't give up right away when it can't read a directory; instead, it calls errfunc with two arguments, like this:
    (*errfunc) (filename, error-code)
    
    The argument filename is the name of the directory that glob couldn't open or couldn't read, and error-code is the errno value that was reported to glob. If the error handler function returns nonzero, then glob gives up right away. Otherwise, it continues.
    GLOB_MARK
    If the pattern matches the name of a directory, append `/' to the directory's name when returning it.
    GLOB_NOCHECK
    If the pattern doesn't match any file names, return the pattern itself as if it were a file name that had been matched. (Normally, when the pattern doesn't match anything, glob returns that there were no matches.)
    GLOB_NOSORT
    Don't sort the file names; return them in no particular order. (In practice, the order will depend on the order of the entries in the directory.) The only reason not to sort is to save time.
    GLOB_NOESCAPE
    Don't treat the `\' character specially in patterns. Normally, `\' quotes the following character, turning off its special meaning (if any) so that it matches only itself. When quoting is enabled, the pattern `\?' matches only the string `?', because the question mark in the pattern acts like an ordinary character. If you use GLOB_NOESCAPE, then `\' is an ordinary character. glob does its work by calling the function fnmatch repeatedly. It handles the flag GLOB_NOESCAPE by turning on the FNM_NOESCAPE flag in calls to fnmatch.

    More Flags for Globbing

    Beside the flags described in the last section, the GNU implementation of glob allows a few more flags which are also defined in the `glob.h' file. Some of the extensions implement functionality which is available in modern shell implementations.

    GLOB_PERIOD
    The . character (period) is treated special. It cannot be matched by wildcards. See section Wildcard Matching, FNM_PERIOD.
    GLOB_MAGCHAR
    The GLOB_MAGCHAR value is not to be given to glob in the flags parameter. Instead, glob sets this bit in the gl_flags element of the glob_t structure provided as the result if the pattern used for matching contains any wildcard character.
    GLOB_ALTDIRFUNC
    Instead of the using the using the normal functions for accessing the filesystem the glob implementation uses the user-supplied functions specified in the structure pointed to by pglob parameter. For more information about the functions refer to the sections about directory handling see section Accessing Directories, and section Reading the Attributes of a File.
    GLOB_BRACE
    If this flag is given the handling of braces in the pattern is changed. It is now required that braces appear correctly grouped. I.e., for each opening brace there must be a closing one. Braces can be used recursively. So it is possible to define one brace expression in another one. It is important to note that the range of each brace expression is completely contained in the outer brace expression (if there is one). The string between the matching braces is separated into single expressions by splitting at , (comma) characters. The commas themself are discarded. Please note what we said above about recursive brace expressions. The commas used to separate the subexpressions must be at the same level. Commas in brace subexpressions are not matched. They are used during expansion of the brace expression of the deeper level. The example below shows this
    glob ("{foo/{,bar,biz},baz}", GLOB_BRACE, NULL, &result)
    
    is equivalent to the sequence
    glob ("foo/", GLOB_BRACE, NULL, &result)
    glob ("foo/bar", GLOB_BRACE|GLOB_APPEND, NULL, &result)
    glob ("foo/biz", GLOB_BRACE|GLOB_APPEND, NULL, &result)
    glob ("baz", GLOB_BRACE|GLOB_APPEND, NULL, &result)
    
    if we leave aside error handling.
    GLOB_NOMAGIC
    If the pattern contains no wildcard constructs (it is a literal file name), return it as the sole "matching" word, even if no file exists by that name.
    GLOB_TILDE
    If this flag is used the character ~ (tilde) is handled special if it appears at the beginning of the pattern. Instead of being taken verbatim it is used to represent the home directory of a known user. If ~ is the only character in pattern or it is followed by a / (slash), the home directory of the process owner is substituted. Using getlogin and getpwnam the information is read from the system databases. As an example take user bart with his home directory at `/home/bart'. For him a call like
    glob ("~/bin/*", GLOB_TILDE, NULL, &result)
    
    would return the contents of the directory `/home/bart/bin'. Instead of referring to the own home directory it is also possible to name the home directory of other users. To do so one has to append the user name after the tilde character. So the contents of user homer's `bin' directory can be retrieved by
    glob ("~homer/bin/*", GLOB_TILDE, NULL, &result)
    
    If the user name is not valid or the home directory cannot be determined for some reason the pattern is left untouched and itself used as the result. I.e., if in the last example home is not available the tilde expansion yields to "~homer/bin/*" and glob is not looking for a directory named ~homer. This functionality is equivalent to what is available in C-shells if the nonomatch flag is set.
    GLOB_TILDE_CHECK
    If this flag is used glob behaves like as if GLOB_TILDE is given. The only difference is that if the user name is not available or the home directory cannot be determined for other reasons this leads to an error. glob will return GLOB_NOMATCH instead of using the pattern itself as the name. This functionality is equivalent to what is available in C-shells if nonomatch flag is not set.
    GLOB_ONLYDIR
    If this flag is used the globbing function takes this as a hint that the caller is only interested in directories matching the pattern. If the information about the type of the file is easily available non-directories will be rejected but no extra work will be done to determine the information for each file. I.e., the caller must still be able to filter directories out. This functionality is only available with the GNU glob implementation. It is mainly used internally to increase the performance but might be useful for a user as well and therefore is documented here.

    Calling glob will in most cases allocate resources which are used to represent the result of the function call. If the same object of type glob_t is used in multiple call to glob the resources are freed or reused so that no leaks appear. But this does not include the time when all glob calls are done.

    Function: void globfree (glob_t *pglob)
    The globfree function frees all resources allocated by previous calls to glob associated with the object pointed to by pglob. This function should be called whenever the currently used glob_t typed object isn't used anymore.

    Function: void globfree64 (glob64_t *pglob)
    This function is equivalent to globfree but it frees records of type glob64_t which were allocated by glob64.

    Regular Expression Matching

    The GNU C library supports two interfaces for matching regular expressions. One is the standard POSIX.2 interface, and the other is what the GNU system has had for many years.

    Both interfaces are declared in the header file `regex.h'. If you define _POSIX_C_SOURCE, then only the POSIX.2 functions, structures, and constants are declared.

    POSIX Regular Expression Compilation

    Before you can actually match a regular expression, you must compile it. This is not true compilation--it produces a special data structure, not machine instructions. But it is like ordinary compilation in that its purpose is to enable you to "execute" the pattern fast. (See section Matching a Compiled POSIX Regular Expression, for how to use the compiled regular expression for matching.)

    There is a special data type for compiled regular expressions:

    Data Type: regex_t
    This type of object holds a compiled regular expression. It is actually a structure. It has just one field that your programs should look at:

    re_nsub
    This field holds the number of parenthetical subexpressions in the regular expression that was compiled.

    There are several other fields, but we don't describe them here, because only the functions in the library should use them.

    After you create a regex_t object, you can compile a regular expression into it by calling regcomp.

    Function: int regcomp (regex_t *compiled, const char *pattern, int cflags)
    The function regcomp "compiles" a regular expression into a data structure that you can use with regexec to match against a string. The compiled regular expression format is designed for efficient matching. regcomp stores it into *compiled.

    It's up to you to allocate an object of type regex_t and pass its address to regcomp.

    The argument cflags lets you specify various options that control the syntax and semantics of regular expressions. See section Flags for POSIX Regular Expressions.

    If you use the flag REG_NOSUB, then regcomp omits from the compiled regular expression the information necessary to record how subexpressions actually match. In this case, you might as well pass 0 for the matchptr and nmatch arguments when you call regexec.

    If you don't use REG_NOSUB, then the compiled regular expression does have the capacity to record how subexpressions match. Also, regcomp tells you how many subexpressions pattern has, by storing the number in compiled->re_nsub. You can use that value to decide how long an array to allocate to hold information about subexpression matches.

    regcomp returns 0 if it succeeds in compiling the regular expression; otherwise, it returns a nonzero error code (see the table below). You can use regerror to produce an error message string describing the reason for a nonzero value; see section POSIX Regexp Matching Cleanup.

    Here are the possible nonzero values that regcomp can return:

    REG_BADBR
    There was an invalid `\{...\}' construct in the regular expression. A valid `\{...\}' construct must contain either a single number, or two numbers in increasing order separated by a comma.
    REG_BADPAT
    There was a syntax error in the regular expression.
    REG_BADRPT
    A repetition operator such as `?' or `*' appeared in a bad position (with no preceding subexpression to act on).
    REG_ECOLLATE
    The regular expression referred to an invalid collating element (one not defined in the current locale for string collation). See section Categories of Activities that Locales Affect.
    REG_ECTYPE
    The regular expression referred to an invalid character class name.
    REG_EESCAPE
    The regular expression ended with `\'.
    REG_ESUBREG
    There was an invalid number in the `\digit' construct.
    REG_EBRACK
    There were unbalanced square brackets in the regular expression.
    REG_EPAREN
    An extended regular expression had unbalanced parentheses, or a basic regular expression had unbalanced `\(' and `\)'.
    REG_EBRACE
    The regular expression had unbalanced `\{' and `\}'.
    REG_ERANGE
    One of the endpoints in a range expression was invalid.
    REG_ESPACE
    regcomp ran out of memory.

    Flags for POSIX Regular Expressions

    These are the bit flags that you can use in the cflags operand when compiling a regular expression with regcomp.

    REG_EXTENDED
    Treat the pattern as an extended regular expression, rather than as a basic regular expression.
    REG_ICASE
    Ignore case when matching letters.
    REG_NOSUB
    Don't bother storing the contents of the matches-ptr array.
    REG_NEWLINE
    Treat a newline in string as dividing string into multiple lines, so that `$' can match before the newline and `^' can match after. Also, don't permit `.' to match a newline, and don't permit `[^...]' to match a newline. Otherwise, newline acts like any other ordinary character.

    Matching a Compiled POSIX Regular Expression

    Once you have compiled a regular expression, as described in section POSIX Regular Expression Compilation, you can match it against strings using regexec. A match anywhere inside the string counts as success, unless the regular expression contains anchor characters (`^' or `$').

    Function: int regexec (regex_t *compiled, char *string, size_t nmatch, regmatch_t matchptr [], int eflags)
    This function tries to match the compiled regular expression *compiled against string.

    regexec returns 0 if the regular expression matches; otherwise, it returns a nonzero value. See the table below for what nonzero values mean. You can use regerror to produce an error message string describing the reason for a nonzero value; see section POSIX Regexp Matching Cleanup.

    The argument eflags is a word of bit flags that enable various options.

    If you want to get information about what part of string actually matched the regular expression or its subexpressions, use the arguments matchptr and nmatch. Otherwise, pass 0 for nmatch, and NULL for matchptr. See section Match Results with Subexpressions.

    You must match the regular expression with the same set of current locales that were in effect when you compiled the regular expression.

    The function regexec accepts the following flags in the eflags argument:

    REG_NOTBOL
    Do not regard the beginning of the specified string as the beginning of a line; more generally, don't make any assumptions about what text might precede it.
    REG_NOTEOL
    Do not regard the end of the specified string as the end of a line; more generally, don't make any assumptions about what text might follow it.

    Here are the possible nonzero values that regexec can return:

    REG_NOMATCH
    The pattern didn't match the string. This isn't really an error.
    REG_ESPACE
    regexec ran out of memory.

    Match Results with Subexpressions

    When regexec matches parenthetical subexpressions of pattern, it records which parts of string they match. It returns that information by storing the offsets into an array whose elements are structures of type regmatch_t. The first element of the array (index 0) records the part of the string that matched the entire regular expression. Each other element of the array records the beginning and end of the part that matched a single parenthetical subexpression.

    Data Type: regmatch_t
    This is the data type of the matcharray array that you pass to regexec. It contains two structure fields, as follows:

    rm_so
    The offset in string of the beginning of a substring. Add this value to string to get the address of that part.
    rm_eo
    The offset in string of the end of the substring.

    Data Type: regoff_t
    regoff_t is an alias for another signed integer type. The fields of regmatch_t have type regoff_t.

    The regmatch_t elements correspond to subexpressions positionally; the first element (index 1) records where the first subexpression matched, the second element records the second subexpression, and so on. The order of the subexpressions is the order in which they begin.

    When you call regexec, you specify how long the matchptr array is, with the nmatch argument. This tells regexec how many elements to store. If the actual regular expression has more than nmatch subexpressions, then you won't get offset information about the rest of them. But this doesn't alter whether the pattern matches a particular string or not.

    If you don't want regexec to return any information about where the subexpressions matched, you can either supply 0 for nmatch, or use the flag REG_NOSUB when you compile the pattern with regcomp.

    Complications in Subexpression Matching

    Sometimes a subexpression matches a substring of no characters. This happens when `f\(o*\)' matches the string `fum'. (It really matches just the `f'.) In this case, both of the offsets identify the point in the string where the null substring was found. In this example, the offsets are both 1.

    Sometimes the entire regular expression can match without using some of its subexpressions at all--for example, when `ba\(na\)*' matches the string `ba', the parenthetical subexpression is not used. When this happens, regexec stores -1 in both fields of the element for that subexpression.

    Sometimes matching the entire regular expression can match a particular subexpression more than once--for example, when `ba\(na\)*' matches the string `bananana', the parenthetical subexpression matches three times. When this happens, regexec usually stores the offsets of the last part of the string that matched the subexpression. In the case of `bananana', these offsets are 6 and 8.

    But the last match is not always the one that is chosen. It's more accurate to say that the last opportunity to match is the one that takes precedence. What this means is that when one subexpression appears within another, then the results reported for the inner subexpression reflect whatever happened on the last match of the outer subexpression. For an example, consider `\(ba\(na\)*s \)*' matching the string `bananas bas '. The last time the inner expression actually matches is near the end of the first word. But it is considered again in the second word, and fails to match there. regexec reports nonuse of the "na" subexpression.

    Another place where this rule applies is when the regular expression

    \(ba\(na\)*s \|nefer\(ti\)* \)*
    

    matches `bananas nefertiti'. The "na" subexpression does match in the first word, but it doesn't match in the second word because the other alternative is used there. Once again, the second repetition of the outer subexpression overrides the first, and within that second repetition, the "na" subexpression is not used. So regexec reports nonuse of the "na" subexpression.

    POSIX Regexp Matching Cleanup

    When you are finished using a compiled regular expression, you can free the storage it uses by calling regfree.

    Function: void regfree (regex_t *compiled)
    Calling regfree frees all the storage that *compiled points to. This includes various internal fields of the regex_t structure that aren't documented in this manual.

    regfree does not free the object *compiled itself.

    You should always free the space in a regex_t structure with regfree before using the structure to compile another regular expression.

    When regcomp or regexec reports an error, you can use the function regerror to turn it into an error message string.

    Function: size_t regerror (int errcode, regex_t *compiled, char *buffer, size_t length)
    This function produces an error message string for the error code errcode, and stores the string in length bytes of memory starting at buffer. For the compiled argument, supply the same compiled regular expression structure that regcomp or regexec was working with when it got the error. Alternatively, you can supply NULL for compiled; you will still get a meaningful error message, but it might not be as detailed.

    If the error message can't fit in length bytes (including a terminating null character), then regerror truncates it. The string that regerror stores is always null-terminated even if it has been truncated.

    The return value of regerror is the minimum length needed to store the entire error message. If this is less than length, then the error message was not truncated, and you can use it. Otherwise, you should call regerror again with a larger buffer.

    Here is a function which uses regerror, but always dynamically allocates a buffer for the error message:

    char *get_regerror (int errcode, regex_t *compiled)
    {
      size_t length = regerror (errcode, compiled, NULL, 0);
      char *buffer = xmalloc (length);
      (void) regerror (errcode, compiled, buffer, length);
      return buffer;
    }
    

    Shell-Style Word Expansion

    Word expansion means the process of splitting a string into words and substituting for variables, commands, and wildcards just as the shell does.

    For example, when you write `ls -l foo.c', this string is split into three separate words---`ls', `-l' and `foo.c'. This is the most basic function of word expansion.

    When you write `ls *.c', this can become many words, because the word `*.c' can be replaced with any number of file names. This is called wildcard expansion, and it is also a part of word expansion.

    When you use `echo $PATH' to print your path, you are taking advantage of variable substitution, which is also part of word expansion.

    Ordinary programs can perform word expansion just like the shell by calling the library function wordexp.

    The Stages of Word Expansion

    When word expansion is applied to a sequence of words, it performs the following transformations in the order shown here:

    1. Tilde expansion: Replacement of `~foo' with the name of the home directory of `foo'.
    2. Next, three different transformations are applied in the same step, from left to right:
      • Variable substitution: Environment variables are substituted for references such as `$foo'.
      • Command substitution: Constructs such as ``cat foo`' and the equivalent `$(cat foo)' are replaced with the output from the inner command.
      • Arithmetic expansion: Constructs such as `$(($x-1))' are replaced with the result of the arithmetic computation.
    3. Field splitting: subdivision of the text into words.
    4. Wildcard expansion: The replacement of a construct such as `*.c' with a list of `.c' file names. Wildcard expansion applies to an entire word at a time, and replaces that word with 0 or more file names that are themselves words.
    5. Quote removal: The deletion of string-quotes, now that they have done their job by inhibiting the above transformations when appropriate.

    For the details of these transformations, and how to write the constructs that use them, see The BASH Manual (to appear).

    Calling wordexp

    All the functions, constants and data types for word expansion are declared in the header file `wordexp.h'.

    Word expansion produces a vector of words (strings). To return this vector, wordexp uses a special data type, wordexp_t, which is a structure. You pass wordexp the address of the structure, and it fills in the structure's fields to tell you about the results.

    Data Type: wordexp_t
    This data type holds a pointer to a word vector. More precisely, it records both the address of the word vector and its size.

    we_wordc
    The number of elements in the vector.
    we_wordv
    The address of the vector. This field has type char **.
    we_offs
    The offset of the first real element of the vector, from its nominal address in the we_wordv field. Unlike the other fields, this is always an input to wordexp, rather than an output from it. If you use a nonzero offset, then that many elements at the beginning of the vector are left empty. (The wordexp function fills them with null pointers.) The we_offs field is meaningful only if you use the WRDE_DOOFFS flag. Otherwise, the offset is always zero regardless of what is in this field, and the first real element comes at the beginning of the vector.

    Function: int wordexp (const char *words, wordexp_t *word-vector-ptr, int flags)
    Perform word expansion on the string words, putting the result in a newly allocated vector, and store the size and address of this vector into *word-vector-ptr. The argument flags is a combination of bit flags; see section Flags for Word Expansion, for details of the flags.

    You shouldn't use any of the characters `|&;<>' in the string words unless they are quoted; likewise for newline. If you use these characters unquoted, you will get the WRDE_BADCHAR error code. Don't use parentheses or braces unless they are quoted or part of a word expansion construct. If you use quotation characters `'"`', they should come in pairs that balance.

    The results of word expansion are a sequence of words. The function wordexp allocates a string for each resulting word, then allocates a vector of type char ** to store the addresses of these strings. The last element of the vector is a null pointer. This vector is called the word vector.

    To return this vector, wordexp stores both its address and its length (number of elements, not counting the terminating null pointer) into *word-vector-ptr.

    If wordexp succeeds, it returns 0. Otherwise, it returns one of these error codes:

    WRDE_BADCHAR
    The input string words contains an unquoted invalid character such as `|'.
    WRDE_BADVAL
    The input string refers to an undefined shell variable, and you used the flag WRDE_UNDEF to forbid such references.
    WRDE_CMDSUB
    The input string uses command substitution, and you used the flag WRDE_NOCMD to forbid command substitution.
    WRDE_NOSPACE
    It was impossible to allocate memory to hold the result. In this case, wordexp can store part of the results--as much as it could allocate room for.
    WRDE_SYNTAX
    There was a syntax error in the input string. For example, an unmatched quoting character is a syntax error.

    Function: void wordfree (wordexp_t *word-vector-ptr)
    Free the storage used for the word-strings and vector that *word-vector-ptr points to. This does not free the structure *word-vector-ptr itself--only the other data it points to.

    Flags for Word Expansion

    This section describes the flags that you can specify in the flags argument to wordexp. Choose the flags you want, and combine them with the C operator |.

    WRDE_APPEND
    Append the words from this expansion to the vector of words produced by previous calls to wordexp. This way you can effectively expand several words as if they were concatenated with spaces between them. In order for appending to work, you must not modify the contents of the word vector structure between calls to wordexp. And, if you set WRDE_DOOFFS in the first call to wordexp, you must also set it when you append to the results.
    WRDE_DOOFFS
    Leave blank slots at the beginning of the vector of words. The we_offs field says how many slots to leave. The blank slots contain null pointers.
    WRDE_NOCMD
    Don't do command substitution; if the input requests command substitution, report an error.
    WRDE_REUSE
    Reuse a word vector made by a previous call to wordexp. Instead of allocating a new vector of words, this call to wordexp will use the vector that already exists (making it larger if necessary). Note that the vector may move, so it is not safe to save an old pointer and use it again after calling wordexp. You must fetch we_pathv anew after each call.
    WRDE_SHOWERR
    Do show any error messages printed by commands run by command substitution. More precisely, allow these commands to inherit the standard error output stream of the current process. By default, wordexp gives these commands a standard error stream that discards all output.
    WRDE_UNDEF
    If the input refers to a shell variable that is not defined, report an error.

    wordexp Example

    Here is an example of using wordexp to expand several strings and use the results to run a shell command. It also shows the use of WRDE_APPEND to concatenate the expansions and of wordfree to free the space allocated by wordexp.

    int
    expand_and_execute (const char *program, const char *options)
    {
      wordexp_t result;
      pid_t pid
      int status, i;
    
      /* Expand the string for the program to run.  */
      switch (wordexp (program, &result, 0))
        {
        case 0:			/* Successful.  */
          break;
        case WRDE_NOSPACE:
          /* If the error was WRDE_NOSPACE,
             then perhaps part of the result was allocated.  */
          wordfree (&result);
        default:                    /* Some other error.  */
          return -1;
        }
    
      /* Expand the strings specified for the arguments.  */
      for (i = 0; args[i]; i++)
        {
          if (wordexp (options, &result, WRDE_APPEND))
            {
              wordfree (&result);
              return -1;
            }
        }
    
      pid = fork ();
      if (pid == 0)
        {
          /* This is the child process.  Execute the command. */
          execv (result.we_wordv[0], result.we_wordv);
          exit (EXIT_FAILURE);
        }
      else if (pid < 0)
        /* The fork failed.  Report failure.  */
        status = -1;
      else
        /* This is the parent process.  Wait for the child to complete.  */
        if (waitpid (pid, &status, 0) != pid)
          status = -1;
    
      wordfree (&result);
      return status;
    }
    

    Details of Tilde Expansion

    It's a standard part of shell syntax that you can use `~' at the beginning of a file name to stand for your own home directory. You can use `~user' to stand for user's home directory.

    Tilde expansion is the process of converting these abbreviations to the directory names that they stand for.

    Tilde expansion applies to the `~' plus all following characters up to whitespace or a slash. It takes place only at the beginning of a word, and only if none of the characters to be transformed is quoted in any way.

    Plain `~' uses the value of the environment variable HOME as the proper home directory name. `~' followed by a user name uses getpwname to look up that user in the user database, and uses whatever directory is recorded there. Thus, `~' followed by your own name can give different results from plain `~', if the value of HOME is not really your home directory.

    Details of Variable Substitution

    Part of ordinary shell syntax is the use of `$variable' to substitute the value of a shell variable into a command. This is called variable substitution, and it is one part of doing word expansion.

    There are two basic ways you can write a variable reference for substitution:

    ${variable}
    If you write braces around the variable name, then it is completely unambiguous where the variable name ends. You can concatenate additional letters onto the end of the variable value by writing them immediately after the close brace. For example, `${foo}s' expands into `tractors'.
    $variable
    If you do not put braces around the variable name, then the variable name consists of all the alphanumeric characters and underscores that follow the `$'. The next punctuation character ends the variable name. Thus, `$foo-bar' refers to the variable foo and expands into `tractor-bar'.

    When you use braces, you can also use various constructs to modify the value that is substituted, or test it in various ways.

    ${variable:-default}
    Substitute the value of variable, but if that is empty or undefined, use default instead.
    ${variable:=default}
    Substitute the value of variable, but if that is empty or undefined, use default instead and set the variable to default.
    ${variable:?message}
    If variable is defined and not empty, substitute its value. Otherwise, print message as an error message on the standard error stream, and consider word expansion a failure.
    ${variable:+replacement}
    Substitute replacement, but only if variable is defined and nonempty. Otherwise, substitute nothing for this construct.
    ${#variable}
    Substitute a numeral which expresses in base ten the number of characters in the value of variable. `${#foo}' stands for `7', because `tractor' is seven characters.

    These variants of variable substitution let you remove part of the variable's value before substituting it. The prefix and suffix are not mere strings; they are wildcard patterns, just like the patterns that you use to match multiple file names. But in this context, they match against parts of the variable value rather than against file names.

    ${variable%%suffix}
    Substitute the value of variable, but first discard from that variable any portion at the end that matches the pattern suffix. If there is more than one alternative for how to match against suffix, this construct uses the longest possible match. Thus, `${foo%%r*}' substitutes `t', because the largest match for `r*' at the end of `tractor' is `ractor'.
    ${variable%suffix}
    Substitute the value of variable, but first discard from that variable any portion at the end that matches the pattern suffix. If there is more than one alternative for how to match against suffix, this construct uses the shortest possible alternative. Thus, `${foo%%r*}' substitutes `tracto', because the shortest match for `r*' at the end of `tractor' is just `r'.
    ${variable##prefix}
    Substitute the value of variable, but first discard from that variable any portion at the beginning that matches the pattern prefix. If there is more than one alternative for how to match against prefix, this construct uses the longest possible match. Thus, `${foo%%r*}' substitutes `t', because the largest match for `r*' at the end of `tractor' is `ractor'.
    ${variable#prefix}
    Substitute the value of variable, but first discard from that variable any portion at the beginning that matches the pattern prefix. If there is more than one alternative for how to match against prefix, this construct uses the shortest possible alternative. Thus, `${foo%%r*}' substitutes `tracto', because the shortest match for `r*' at the end of `tractor' is just `r'.

    Input/Output Overview

    Most programs need to do either input (reading data) or output (writing data), or most frequently both, in order to do anything useful. The GNU C library provides such a large selection of input and output functions that the hardest part is often deciding which function is most appropriate!

    This chapter introduces concepts and terminology relating to input and output. Other chapters relating to the GNU I/O facilities are:

    • section Input/Output on Streams, which covers the high-level functions that operate on streams, including formatted input and output.
    • section Low-Level Input/Output, which covers the basic I/O and control functions on file descriptors.
    • section File System Interface, which covers functions for operating on directories and for manipulating file attributes such as access modes and ownership.
    • section Pipes and FIFOs, which includes information on the basic interprocess communication facilities.
    • section Sockets, which covers a more complicated interprocess communication facility with support for networking.
    • section Low-Level Terminal Interface, which covers functions for changing how input and output to terminals or other serial devices are processed.

    Input/Output Concepts

    Before you can read or write the contents of a file, you must establish a connection or communications channel to the file. This process is called opening the file. You can open a file for reading, writing, or both.

    The connection to an open file is represented either as a stream or as a file descriptor. You pass this as an argument to the functions that do the actual read or write operations, to tell them which file to operate on. Certain functions expect streams, and others are designed to operate on file descriptors.

    When you have finished reading to or writing from the file, you can terminate the connection by closing the file. Once you have closed a stream or file descriptor, you cannot do any more input or output operations on it.

    Streams and File Descriptors

    When you want to do input or output to a file, you have a choice of two basic mechanisms for representing the connection between your program and the file: file descriptors and streams. File descriptors are represented as objects of type int, while streams are represented as FILE * objects.

    File descriptors provide a primitive, low-level interface to input and output operations. Both file descriptors and streams can represent a connection to a device (such as a terminal), or a pipe or socket for communicating with another process, as well as a normal file. But, if you want to do control operations that are specific to a particular kind of device, you must use a file descriptor; there are no facilities to use streams in this way. You must also use file descriptors if your program needs to do input or output in special modes, such as nonblocking (or polled) input (see section File Status Flags).

    Streams provide a higher-level interface, layered on top of the primitive file descriptor facilities. The stream interface treats all kinds of files pretty much alike--the sole exception being the three styles of buffering that you can choose (see section Stream Buffering).

    The main advantage of using the stream interface is that the set of functions for performing actual input and output operations (as opposed to control operations) on streams is much richer and more powerful than the corresponding facilities for file descriptors. The file descriptor interface provides only simple functions for transferring blocks of characters, but the stream interface also provides powerful formatted input and output functions (printf and scanf) as well as functions for character- and line-oriented input and output.

    Since streams are implemented in terms of file descriptors, you can extract the file descriptor from a stream and perform low-level operations directly on the file descriptor. You can also initially open a connection as a file descriptor and then make a stream associated with that file descriptor.

    In general, you should stick with using streams rather than file descriptors, unless there is some specific operation you want to do that can only be done on a file descriptor. If you are a beginning programmer and aren't sure what functions to use, we suggest that you concentrate on the formatted input functions (see section Formatted Input) and formatted output functions (see section Formatted Output).

    If you are concerned about portability of your programs to systems other than GNU, you should also be aware that file descriptors are not as portable as streams. You can expect any system running ISO C to support streams, but non-GNU systems may not support file descriptors at all, or may only implement a subset of the GNU functions that operate on file descriptors. Most of the file descriptor functions in the GNU library are included in the POSIX.1 standard, however.

    File Position

    One of the attributes of an open file is its file position that keeps track of where in the file the next character is to be read or written. In the GNU system, and all POSIX.1 systems, the file position is simply an integer representing the number of bytes from the beginning of the file.

    The file position is normally set to the beginning of the file when it is opened, and each time a character is read or written, the file position is incremented. In other words, access to the file is normally sequential.

    Ordinary files permit read or write operations at any position within the file. Some other kinds of files may also permit this. Files which do permit this are sometimes referred to as random-access files. You can change the file position using the fseek function on a stream (see section File Positioning) or the lseek function on a file descriptor (see section Input and Output Primitives). If you try to change the file position on a file that doesn't support random access, you get the ESPIPE error.

    Streams and descriptors that are opened for append access are treated specially for output: output to such files is always appended sequentially to the end of the file, regardless of the file position. However, the file position is still used to control where in the file reading is done.

    If you think about it, you'll realize that several programs can read a given file at the same time. In order for each program to be able to read the file at its own pace, each program must have its own file pointer, which is not affected by anything the other programs do.

    In fact, each opening of a file creates a separate file position. Thus, if you open a file twice even in the same program, you get two streams or descriptors with independent file positions.

    By contrast, if you open a descriptor and then duplicate it to get another descriptor, these two descriptors share the same file position: changing the file position of one descriptor will affect the other.

    File Names

    In order to open a connection to a file, or to perform other operations such as deleting a file, you need some way to refer to the file. Nearly all files have names that are strings--even files which are actually devices such as tape drives or terminals. These strings are called file names. You specify the file name to say which file you want to open or operate on.

    This section describes the conventions for file names and how the operating system works with them.

    Directories

    In order to understand the syntax of file names, you need to understand how the file system is organized into a hierarchy of directories.

    A directory is a file that contains information to associate other files with names; these associations are called links or directory entries. Sometimes, people speak of "files in a directory", but in reality, a directory only contains pointers to files, not the files themselves.

    The name of a file contained in a directory entry is called a file name component. In general, a file name consists of a sequence of one or more such components, separated by the slash character (`/'). A file name which is just one component names a file with respect to its directory. A file name with multiple components names a directory, and then a file in that directory, and so on.

    Some other documents, such as the POSIX standard, use the term pathname for what we call a file name, and either filename or pathname component for what this manual calls a file name component. We don't use this terminology because a "path" is something completely different (a list of directories to search), and we think that "pathname" used for something else will confuse users. We always use "file name" and "file name component" (or sometimes just "component", where the context is obvious) in GNU documentation. Some macros use the POSIX terminology in their names, such as PATH_MAX. These macros are defined by the POSIX standard, so we cannot change their names.

    You can find more detailed information about operations on directories in section File System Interface.

    File Name Resolution

    A file name consists of file name components separated by slash (`/') characters. On the systems that the GNU C library supports, multiple successive `/' characters are equivalent to a single `/' character.

    The process of determining what file a file name refers to is called file name resolution. This is performed by examining the components that make up a file name in left-to-right order, and locating each successive component in the directory named by the previous component. Of course, each of the files that are referenced as directories must actually exist, be directories instead of regular files, and have the appropriate permissions to be accessible by the process; otherwise the file name resolution fails.

    If a file name begins with a `/', the first component in the file name is located in the root directory of the process (usually all processes on the system have the same root directory). Such a file name is called an absolute file name.

    Otherwise, the first component in the file name is located in the current working directory (see section Working Directory). This kind of file name is called a relative file name.

    The file name components `.' ("dot") and `..' ("dot-dot") have special meanings. Every directory has entries for these file name components. The file name component `.' refers to the directory itself, while the file name component `..' refers to its parent directory (the directory that contains the link for the directory in question). As a special case, `..' in the root directory refers to the root directory itself, since it has no parent; thus `/..' is the same as `/'.

    Here are some examples of file names:

    `/a'
    The file named `a', in the root directory.
    `/a/b'
    The file named `b', in the directory named `a' in the root directory.
    `a'
    The file named `a', in the current working directory.
    `/a/./b'
    This is the same as `/a/b'.
    `./a'
    The file named `a', in the current working directory.
    `../a'
    The file named `a', in the parent directory of the current working directory.

    A file name that names a directory may optionally end in a `/'. You can specify a file name of `/' to refer to the root directory, but the empty string is not a meaningful file name. If you want to refer to the current working directory, use a file name of `.' or `./'.

    Unlike some other operating systems, the GNU system doesn't have any built-in support for file types (or extensions) or file versions as part of its file name syntax. Many programs and utilities use conventions for file names--for example, files containing C source code usually have names suffixed with `.c'---but there is nothing in the file system itself that enforces this kind of convention.

    File Name Errors

    Functions that accept file name arguments usually detect these errno error conditions relating to the file name syntax or trouble finding the named file. These errors are referred to throughout this manual as the usual file name errors.

    EACCES
    The process does not have search permission for a directory component of the file name.
    ENAMETOOLONG
    This error is used when either the total length of a file name is greater than PATH_MAX, or when an individual file name component has a length greater than NAME_MAX. See section Limits on File System Capacity. In the GNU system, there is no imposed limit on overall file name length, but some file systems may place limits on the length of a component.
    ENOENT
    This error is reported when a file referenced as a directory component in the file name doesn't exist, or when a component is a symbolic link whose target file does not exist. See section Symbolic Links.
    ENOTDIR
    A file that is referenced as a directory component in the file name exists, but it isn't a directory.
    ELOOP
    Too many symbolic links were resolved while trying to look up the file name. The system has an arbitrary limit on the number of symbolic links that may be resolved in looking up a single file name, as a primitive way to detect loops. See section Symbolic Links.

    Portability of File Names

    The rules for the syntax of file names discussed in section File Names, are the rules normally used by the GNU system and by other POSIX systems. However, other operating systems may use other conventions.

    There are two reasons why it can be important for you to be aware of file name portability issues:

    • If your program makes assumptions about file name syntax, or contains embedded literal file name strings, it is more difficult to get it to run under other operating systems that use different syntax conventions.
    • Even if you are not concerned about running your program on machines that run other operating systems, it may still be possible to access files that use different naming conventions. For example, you may be able to access file systems on another computer running a different operating system over a network, or read and write disks in formats used by other operating systems.

    The ISO C standard says very little about file name syntax, only that file names are strings. In addition to varying restrictions on the length of file names and what characters can validly appear in a file name, different operating systems use different conventions and syntax for concepts such as structured directories and file types or extensions. Some concepts such as file versions might be supported in some operating systems and not by others.

    The POSIX.1 standard allows implementations to put additional restrictions on file name syntax, concerning what characters are permitted in file names and on the length of file name and file name component strings. However, in the GNU system, you do not need to worry about these restrictions; any character except the null character is permitted in a file name string, and there are no limits on the length of file name strings.

    Input/Output on Streams

    This chapter describes the functions for creating streams and performing input and output operations on them. As discussed in section Input/Output Overview, a stream is a fairly abstract, high-level concept representing a communications channel to a file, device, or process.

    Streams

    For historical reasons, the type of the C data structure that represents a stream is called FILE rather than "stream". Since most of the library functions deal with objects of type FILE *, sometimes the term file pointer is also used to mean "stream". This leads to unfortunate confusion over terminology in many books on C. This manual, however, is careful to use the terms "file" and "stream" only in the technical sense.

    The FILE type is declared in the header file `stdio.h'.

    Data Type: FILE
    This is the data type used to represent stream objects. A FILE object holds all of the internal state information about the connection to the associated file, including such things as the file position indicator and buffering information. Each stream also has error and end-of-file status indicators that can be tested with the ferror and feof functions; see section End-Of-File and Errors.

    FILE objects are allocated and managed internally by the input/output library functions. Don't try to create your own objects of type FILE; let the library do it. Your programs should deal only with pointers to these objects (that is, FILE * values) rather than the objects themselves.

    Standard Streams

    When the main function of your program is invoked, it already has three predefined streams open and available for use. These represent the "standard" input and output channels that have been established for the process.

    These streams are declared in the header file `stdio.h'.

    Variable: FILE * stdin
    The standard input stream, which is the normal source of input for the program.

    Variable: FILE * stdout
    The standard output stream, which is used for normal output from the program.

    Variable: FILE * stderr
    The standard error stream, which is used for error messages and diagnostics issued by the program.

    In the GNU system, you can specify what files or processes correspond to these streams using the pipe and redirection facilities provided by the shell. (The primitives shells use to implement these facilities are described in section File System Interface.) Most other operating systems provide similar mechanisms, but the details of how to use them can vary.

    In the GNU C library, stdin, stdout, and stderr are normal variables which you can set just like any others. For example, to redirect the standard output to a file, you could do:

    fclose (stdout);
    stdout = fopen ("standard-output-file", "w");
    

    Note however, that in other systems stdin, stdout, and stderr are macros that you cannot assign to in the normal way. But you can use freopen to get the effect of closing one and reopening it. See section Opening Streams.

    The three streams stdin, stdout, and stderr are not unoriented at program start (see section Streams in Internationalized Applications).

    Opening Streams

    Opening a file with the fopen function creates a new stream and establishes a connection between the stream and a file. This may involve creating a new file.

    Everything described in this section is declared in the header file `stdio.h'.

    Function: FILE * fopen (const char *filename, const char *opentype)
    The fopen function opens a stream for I/O to the file filename, and returns a pointer to the stream.

    The opentype argument is a string that controls how the file is opened and specifies attributes of the resulting stream. It must begin with one of the following sequences of characters:

    `r'
    Open an existing file for reading only.
    `w'
    Open the file for writing only. If the file already exists, it is truncated to zero length. Otherwise a new file is created.
    `a'
    Open a file for append access; that is, writing at the end of file only. If the file already exists, its initial contents are unchanged and output to the stream is appended to the end of the file. Otherwise, a new, empty file is created.
    `r+'
    Open an existing file for both reading and writing. The initial contents of the file are unchanged and the initial file position is at the beginning of the file.
    `w+'
    Open a file for both reading and writing. If the file already exists, it is truncated to zero length. Otherwise, a new file is created.
    `a+'
    Open or create file for both reading and appending. If the file exists, its initial contents are unchanged. Otherwise, a new file is created. The initial file position for reading is at the beginning of the file, but output is always appended to the end of the file.

    As you can see, `+' requests a stream that can do both input and output. The ISO standard says that when using such a stream, you must call fflush (see section Stream Buffering) or a file positioning function such as fseek (see section File Positioning) when switching from reading to writing or vice versa. Otherwise, internal buffers might not be emptied properly. The GNU C library does not have this limitation; you can do arbitrary reading and writing operations on a stream in whatever order.

    Additional characters may appear after these to specify flags for the call. Always put the mode (`r', `w+', etc.) first; that is the only part you are guaranteed will be understood by all systems.

    The GNU C library defines one additional character for use in opentype: the character `x' insists on creating a new file--if a file filename already exists, fopen fails rather than opening it. If you use `x' you are guaranteed that you will not clobber an existing file. This is equivalent to the O_EXCL option to the open function (see section Opening and Closing Files).

    The character `b' in opentype has a standard meaning; it requests a binary stream rather than a text stream. But this makes no difference in POSIX systems (including the GNU system). If both `+' and `b' are specified, they can appear in either order. See section Text and Binary Streams.

    If the opentype string contains the sequence ,ccs=STRING then STRING is taken as the name of a coded character set and fopen will mark the stream as wide-oriented which appropriate conversion functions in place to convert from and to the character set STRING is place. Any other stream is opened initially unoriented and the orientation is decided with the first file operation. If the first operation is a wide character operation, the stream is not only marked as wide-oriented, also the conversion functions to convert to the coded character set used for the current locale are loaded. This will not change anymore from this point on even if the locale selected for the LC_CTYPE category is changed.

    Any other characters in opentype are simply ignored. They may be meaningful in other systems.

    If the open fails, fopen returns a null pointer.

    When the sources are compiling with _FILE_OFFSET_BITS == 64 on a 32 bit machine this function is in fact fopen64 since the LFS interface replaces transparently the old interface.

    You can have multiple streams (or file descriptors) pointing to the same file open at the same time. If you do only input, this works straightforwardly, but you must be careful if any output streams are included. See section Dangers of Mixing Streams and Descriptors. This is equally true whether the streams are in one program (not usual) or in several programs (which can easily happen). It may be advantageous to use the file locking facilities to avoid simultaneous access. See section File Locks.

    Function: FILE * fopen64 (const char *filename, const char *opentype)
    This function is similar to fopen but the stream it returns a pointer for is opened using open64. Therefore this stream can be used even on files larger then @math{2^31} bytes on 32 bit machines.

    Please note that the return type is still FILE *. There is no special FILE type for the LFS interface.

    If the sources are compiled with _FILE_OFFSET_BITS == 64 on a 32 bits machine this function is available under the name fopen and so transparently replaces the old interface.

    Macro: int FOPEN_MAX
    The value of this macro is an integer constant expression that represents the minimum number of streams that the implementation guarantees can be open simultaneously. You might be able to open more than this many streams, but that is not guaranteed. The value of this constant is at least eight, which includes the three standard streams stdin, stdout, and stderr. In POSIX.1 systems this value is determined by the OPEN_MAX parameter; see section General Capacity Limits. In BSD and GNU, it is controlled by the RLIMIT_NOFILE resource limit; see section Limiting Resource Usage.

    Function: FILE * freopen (const char *filename, const char *opentype, FILE *stream)
    This function is like a combination of fclose and fopen. It first closes the stream referred to by stream, ignoring any errors that are detected in the process. (Because errors are ignored, you should not use freopen on an output stream if you have actually done any output using the stream.) Then the file named by filename is opened with mode opentype as for fopen, and associated with the same stream object stream.

    If the operation fails, a null pointer is returned; otherwise, freopen returns stream.

    freopen has traditionally been used to connect a standard stream such as stdin with a file of your own choice. This is useful in programs in which use of a standard stream for certain purposes is hard-coded. In the GNU C library, you can simply close the standard streams and open new ones with fopen. But other systems lack this ability, so using freopen is more portable.

    When the sources are compiling with _FILE_OFFSET_BITS == 64 on a 32 bit machine this function is in fact freopen64 since the LFS interface replaces transparently the old interface.

    Function: FILE * freopen64 (const char *filename, const char *opentype, FILE *stream)
    This function is similar to freopen. The only difference is that on 32 bit machine the stream returned is able to read beyond the @math{2^31} bytes limits imposed by the normal interface. It should be noted that the stream pointed to by stream need not be opened using fopen64 or freopen64 since its mode is not important for this function.

    If the sources are compiled with _FILE_OFFSET_BITS == 64 on a 32 bits machine this function is available under the name freopen and so transparently replaces the old interface.

    In some situations it is useful to know whether a given stream is available for reading or writing. This information is normally not available and would have to be remembered separately. Solaris introduced a few functions to get this information from the stream descriptor and these functions are also available in the GNU C library.

    Function: int __freadable (FILE *stream)
    The __freadable function determines whether the stream stream was opened to allow reading. In this case the return value is nonzero. For write-only streams the function returns zero.

    This function is declared in `stdio_ext.h'.

    Function: int __fwritable (FILE *stream)
    The __fwritable function determines whether the stream stream was opened to allow writing. In this case the return value is nonzero. For read-only streams the function returns zero.

    This function is declared in `stdio_ext.h'.

    For slightly different kind of problems there are two more functions. They provide even finer-grained information.

    Function: int __freading (FILE *stream)
    The __freading function determines whether the stream stream was last read from or whether it is opened read-only. In this case the return value is nonzero, otherwise it is zero. Determining whether a stream opened for reading and writing was last used for writing allows to draw conclusions about the content about the buffer, among other things.

    This function is declared in `stdio_ext.h'.

    Function: int __fwriting (FILE *stream)
    The __fwriting function determines whether the stream stream was last written to or whether it is opened write-only. In this case the return value is nonzero, otherwise it is zero.

    This function is declared in `stdio_ext.h'.

    Closing Streams

    When a stream is closed with fclose, the connection between the stream and the file is cancelled. After you have closed a stream, you cannot perform any additional operations on it.

    Function: int fclose (FILE *stream)
    This function causes stream to be closed and the connection to the corresponding file to be broken. Any buffered output is written and any buffered input is discarded. The fclose function returns a value of 0 if the file was closed successfully, and EOF if an error was detected.

    It is important to check for errors when you call fclose to close an output stream, because real, everyday errors can be detected at this time. For example, when fclose writes the remaining buffered output, it might get an error because the disk is full. Even if you know the buffer is empty, errors can still occur when closing a file if you are using NFS.

    The function fclose is declared in `stdio.h'.

    To close all streams currently available the GNU C Library provides another function.

    Function: int fcloseall (void)
    This function causes all open streams of the process to be closed and the connection to corresponding files to be broken. All buffered data is written and any buffered input is discarded. The fcloseall function returns a value of 0 if all the files were closed successfully, and EOF if an error was detected.

    This function should be used only in special situations, e.g., when an error occurred and the program must be aborted. Normally each single stream should be closed separately so that problems with individual streams can be identified. It is also problematic since the standard streams (see section Standard Streams) will also be closed.

    The function fcloseall is declared in `stdio.h'.

    If the main function to your program returns, or if you call the exit function (see section Normal Termination), all open streams are automatically closed properly. If your program terminates in any other manner, such as by calling the abort function (see section Aborting a Program) or from a fatal signal (see section Signal Handling), open streams might not be closed properly. Buffered output might not be flushed and files may be incomplete. For more information on buffering of streams, see section Stream Buffering.

    Streams and Threads

    Streams can be used in multi-threaded applications in the same way they are used in single-threaded applications. But the programmer must be aware of a the possible complications. It is important to know about these also if the program one writes never use threads since the design and implementation of many stream functions is heavily influenced by the requirements added by multi-threaded programming.

    The POSIX standard requires that by default the stream operations are atomic. I.e., issueing two stream operations for the same stream in two threads at the same time will cause the operations to be executed as if they were issued sequentially. The buffer operations performed while reading or writing are protected from other uses of the same stream. To do this each stream has an internal lock object which has to be (implicitly) acquired before any work can be done.

    But there are situations where this is not enough and there are also situations where this is not wanted. The implicit locking is not enough if the program requires more than one stream function call to happen atomically. One example would be if an output line a program wants to generate is created by several function calls. The functions by themselves would ensure only atomicity of their own operation, but not atomicity over all the function calls. For this it is necessary to perform the stream locking in the application code.

    Function: void flockfile (FILE *stream)
    The flockfile function acquires the internal locking object associated with the stream stream. This ensures that no other thread can explicitly through flockfile/ftrylockfile or implicit through a call of a stream function lock the stream. The thread will block until the lock is acquired. An explicit call to funlockfile has to be used to release the lock.

    Function: int ftrylockfile (FILE *stream)
    The ftrylockfile function tries to acquire the internal locking object associated with the stream stream just like flockfile. But unlike flockfile this function does not block if the lock is not available. ftrylockfile returns zero if the lock was successfully acquired. Otherwise the stream is locked by another thread.

    Function: void funlockfile (FILE *stream)
    The funlockfile function releases the internal locking object of the stream stream. The stream must have been locked before by a call to flockfile or a successful call of ftrylockfile. The implicit locking performed by the stream operations do not count. The funlockfile function does not return an error status and the behavior of a call for a stream which is not locked by the current thread is undefined.

    The following example shows how the functions above can be used to generate an output line atomically even in multi-threaded applications (yes, the same job could be done with one fprintf call but it is sometimes not possible):

    FILE *fp;
    {
       ...
       flockfile (fp);
       fputs ("This is test number ", fp);
       fprintf (fp, "%d\n", test);
       funlockfile (fp)
    }
    

    Without the explicit locking it would be possible for another thread to use the stream fp after the fputs call return and before fprintf was called with the result that the number does not follow the word `number'.

    From this description it might already be clear that the locking objects in streams are no simple mutexes. Since locking the same stream twice in the same thread is allowed the locking objects must be equivalent to recursive mutexes. These mutexes keep track of the owner and the number of times the lock is acquired. The same number of funlockfile calls by the same threads is necessary to unlock the stream completely. For instance:

    void
    foo (FILE *fp)
    {
      ftrylockfile (fp);
      fputs ("in foo\n", fp);
      /* This is very wrong!!!  */
      funlockfile (fp);
    }
    

    It is important here that the funlockfile function is only called if the ftrylockfile function succeeded in locking the stream. It is therefore always wrong to ignore the result of ftrylockfile. And it makes no sense since otherwise one would use flockfile. The result of code like that above is that either funlockfile tries to free a stream that hasn't been locked by the current thread or it frees the stream prematurely. The code should look like this:

    void
    foo (FILE *fp)
    {
      if (ftrylockfile (fp) == 0)
        {
          fputs ("in foo\n", fp);
          funlockfile (fp);
        }
    }
    

    Now that we covered why it is necessary to have these locking it is necessary to talk about situations when locking is unwanted and what can be done. The locking operations (explicit or implicit) don't come for free. Even if a lock is not taken the cost is not zero. The operations which have to be performed require memory operations which are save in multi-processor environments. With the many local caches involved in such systems this is quite costly. So it is best to avoid the locking completely if it is known that the code using the stream is never used in a context where more than one thread can use the stream at one time. This can be determined most of the time for application code; for library code which can be used in many contexts one should default to be conservative and use locking.

    There are two basic mechanisms to avoid locking. The first is to use the _unlocked variants of the stream operations. The POSIX standard defines quite a few of those and the GNU library adds a few more. These variants of the functions behave just like the functions with the name without the suffix except that they are not locking the stream. Using these functions is very desirable since they are potentially much faster. This is not only because the locking operation itself is avoided. More importantly, functions like putc and getc are very simple and tradionally (before the introduction of threads) were implemented as macros which are very fast if the buffer is not empty. With locking required these functions are now no macros anymore (the code generated would be too much). But these macros are still available with the same functionality under the new names putc_unlocked and getc_unlocked. This possibly huge difference of speed also suggests the use of the _unlocked functions even if locking is required. The difference is that the locking then has to be performed in the program:

    void
    foo (FILE *fp, char *buf)
    {
      flockfile (fp);
      while (*buf != '/')
        putc_unlocked (*buf++, fp);
      funlockfile (fp);
    }
    

    If in this example the putc function would be used and the explicit locking would be missing the putc function would have to acquire the lock in every call, potentially many times depending on when the loop terminates. Writing it the way illustrated above allows the putc_unlocked macro to be used which means no locking and direct manipulation of the buffer of the stream.

    A second way to avoid locking is by using a non-standard function which was introduced in Solaris and is available in the GNU C library as well.

    Function: int __fsetlocking (FILE *stream, int type)

    The __fsetlocking function can be used to select whether the stream operations will implicitly acquire the locking object of the stream stream. By default this is done but it can be disabled and reinstated using this function. There are three values defined for the type parameter.

    FSETLOCKING_INTERNAL
    The stream stream will from now on use the default internal locking. Every stream operation with exception of the _unlocked variants will implicitly lock the stream.
    FSETLOCKING_BYCALLER
    After the __fsetlocking function returns the user is responsible for locking the stream. None of the stream operations will implicitly do this anymore until the state is set back to FSETLOCKING_INTERNAL.
    FSETLOCKING_QUERY
    __fsetlocking only queries the current locking state of the stream. The return value will be FSETLOCKING_INTERNAL or FSETLOCKING_BYCALLER depending on the state.

    The return value of __fsetlocking is either FSETLOCKING_INTERNAL or FSETLOCKING_BYCALLER depending on the state of the stream before the call.

    This function and the values for the type parameter are declared in `stdio_ext.h'.

    This function is especially useful when program code has to be used which is written without knowledge about the _unlocked functions (or if the programmer was to lazy to use them).

    Streams in Internationalized Applications

    ISO C90 introduced the new type wchar_t to allow handling larger character sets. What was missing was a possibility to output strings of wchar_t directly. One had to convert them into multibyte strings using mbstowcs (there was no mbsrtowcs yet) and then use the normal stream functions. While this is doable it is very cumbersome since performing the conversions is not trivial and greatly increases program complexity and size.

    The Unix standard early on (I think in XPG4.2) introduced two additional format specifiers for the printf and scanf families of functions. Printing and reading of single wide characters was made possible using the %C specifier and wide character strings can be handled with %S. These modifiers behave just like %c and %s only that they expect the corresponding argument to have the wide character type and that the wide character and string are transformed into/from multibyte strings before being used.

    This was a beginning but it is still not good enough. Not always is it desirable to use printf and scanf. The other, smaller and faster functions cannot handle wide characters. Second, it is not possible to have a format string for printf and scanf consisting of wide characters. The result is that format strings would have to be generated if they have to contain non-basic characters.

    In the Amendment 1 to ISO C90 a whole new set of functions was added to solve the problem. Most of the stream functions got a counterpart which take a wide character or wide character string instead of a character or string respectively. The new functions operate on the same streams (like stdout). This is different from the model of the C++ runtime library where separate streams for wide and normal I/O are used.

    Being able to use the same stream for wide and normal operations comes with a restriction: a stream can be used either for wide operations or for normal operations. Once it is decided there is no way back. Only a call to freopen or freopen64 can reset the orientation. The orientation can be decided in three ways:

    • If any of the normal character functions is used (this includes the fread and fwrite functions) the stream is marked as not wide oriented.
    • If any of the wide character functions is used the stream is marked as wide oriented.
    • The fwide function can be used to set the orientation either way.

    It is important to never mix the use of wide and not wide operations on a stream. There are no diagnostics issued. The application behavior will simply be strange or the application will simply crash. The fwide function can help avoiding this.

    Function: int fwide (FILE *stream, int mode)

    The fwide function can be used to set and query the state of the orientation of the stream stream. If the mode parameter has a positive value the streams get wide oriented, for negative values narrow oriented. It is not possible to overwrite previous orientations with fwide. I.e., if the stream stream was already oriented before the call nothing is done.

    If mode is zero the current orientation state is queried and nothing is changed.

    The fwide function returns a negative value, zero, or a positive value if the stream is narrow, not at all, or wide oriented respectively.

    This function was introduced in Amendment 1 to ISO C90 and is declared in `wchar.h'.

    It is generally a good idea to orient a stream as early as possible. This can prevent surprise especially for the standard streams stdin, stdout, and stderr. If some library function in some situations uses one of these streams and this use orients the stream in a different way the rest of the application expects it one might end up with hard to reproduce errors. Remember that no errors are signal if the streams are used incorrectly. Leaving a stream unoriented after creation is normally only necessary for library functions which create streams which can be used in different contexts.

    When writing code which uses streams and which can be used in different contexts it is important to query the orientation of the stream before using it (unless the rules of the library interface demand a specific orientation). The following little, silly function illustrates this.

    void
    print_f (FILE *fp)
    {
      if (fwide (fp, 0) > 0)
        /* Positive return value means wide orientation.  */
        fputwc (L'f', fp);
      else
        fputc ('f', fp);
    }
    

    Note that in this case the function print_f decides about the orientation of the stream if it was unoriented before (will not happen if the advise above is followed).

    The encoding used for the wchar_t values is unspecified and the user must not make any assumptions about it. For I/O of wchar_t values this means that it is impossible to write these values directly to the stream. This is not what follows from the ISO C locale model either. What happens instead is that the bytes read from or written to the underlying media are first converted into the internal encoding chosen by the implementation for wchar_t. The external encoding is determined by the LC_CTYPE category of the current locale or by the `ccs' part of the mode specification given to fopen, fopen64, freopen, or freopen64. How and when the conversion happens is unspecified and it happens invisible to the user.

    Since a stream is created in the unoriented state it has at that point no conversion associated with it. The conversion which will be used is determined by the LC_CTYPE category selected at the time the stream is oriented. If the locales are changed at the runtime this might produce surprising results unless one pays attention. This is just another good reason to orient the stream explicitly as soon as possible, perhaps with a call to fwide.

    Simple Output by Characters or Lines

    This section describes functions for performing character- and line-oriented output.

    These narrow streams functions are declared in the header file `stdio.h' and the wide stream functions in `wchar.h'.

    Function: int fputc (int c, FILE *stream)
    The fputc function converts the character c to type unsigned char, and writes it to the stream stream. EOF is returned if a write error occurs; otherwise the character c is returned.

    Function: wint_t fputwc (wchar_t wc, FILE *stream)
    The fputwc function writes the wide character wc to the stream stream. WEOF is returned if a write error occurs; otherwise the character wc is returned.

    Function: int fputc_unlocked (int c, FILE *stream)
    The fputc_unlocked function is equivalent to the fputc function except that it does not implicitly lock the stream.

    Function: wint_t fputwc_unlocked (wint_t wc, FILE *stream)
    The fputwc_unlocked function is equivalent to the fputwc function except that it does not implicitly lock the stream.

    This function is a GNU extension.

    Function: int putc (int c, FILE *stream)
    This is just like fputc, except that most systems implement it as a macro, making it faster. One consequence is that it may evaluate the stream argument more than once, which is an exception to the general rule for macros. putc is usually the best function to use for writing a single character.

    Function: wint_t putwc (wchar_t wc, FILE *stream)
    This is just like fputwc, except that it can be implement as a macro, making it faster. One consequence is that it may evaluate the stream argument more than once, which is an exception to the general rule for macros. putwc is usually the best function to use for writing a single wide character.

    Function: int putc_unlocked (int c, FILE *stream)
    The putc_unlocked function is equivalent to the putc function except that it does not implicitly lock the stream.

    Function: wint_t putwc_unlocked (wchar_t wc, FILE *stream)
    The putwc_unlocked function is equivalent to the putwc function except that it does not implicitly lock the stream.

    This function is a GNU extension.

    Function: int putchar (int c)
    The putchar function is equivalent to putc with stdout as the value of the stream argument.

    Function: wint_t putwchar (wchar_t wc)
    The putwchar function is equivalent to putwc with stdout as the value of the stream argument.

    Function: int putchar_unlocked (int c)
    The putchar_unlocked function is equivalent to the putchar function except that it does not implicitly lock the stream.

    Function: wint_t putwchar_unlocked (wchar_t wc)
    The putwchar_unlocked function is equivalent to the putwchar function except that it does not implicitly lock the stream.

    This function is a GNU extension.

    Function: int fputs (const char *s, FILE *stream)
    The function fputs writes the string s to the stream stream. The terminating null character is not written. This function does not add a newline character, either. It outputs only the characters in the string.

    This function returns EOF if a write error occurs, and otherwise a non-negative value.

    For example:

    fputs ("Are ", stdout);
    fputs ("you ", stdout);
    fputs ("hungry?\n", stdout);
    

    outputs the text `Are you hungry?' followed by a newline.

    Function: int fputws (const wchar_t *ws, FILE *stream)
    The function fputws writes the wide character string ws to the stream stream. The terminating null character is not written. This function does not add a newline character, either. It outputs only the characters in the string.

    This function returns WEOF if a write error occurs, and otherwise a non-negative value.

    Function: int fputs_unlocked (const char *s, FILE *stream)
    The fputs_unlocked function is equivalent to the fputs function except that it does not implicitly lock the stream.

    This function is a GNU extension.

    Function: int fputws_unlocked (const wchar_t *ws, FILE *stream)
    The fputws_unlocked function is equivalent to the fputws function except that it does not implicitly lock the stream.

    This function is a GNU extension.

    Function: int puts (const char *s)
    The puts function writes the string s to the stream stdout followed by a newline. The terminating null character of the string is not written. (Note that fputs does not write a newline as this function does.)

    puts is the most convenient function for printing simple messages. For example:

    puts ("This is a message.");
    

    outputs the text `This is a message.' followed by a newline.

    Function: int putw (int w, FILE *stream)
    This function writes the word w (that is, an int) to stream. It is provided for compatibility with SVID, but we recommend you use fwrite instead (see section Block Input/Output).

    Character Input

    This section describes functions for performing character-oriented input. These narrow streams functions are declared in the header file `stdio.h' and the wide character functions are declared in `wchar.h'.

    These functions return an int or wint_t value (for narrow and wide stream functions respectively) that is either a character of input, or the special value EOF/WEOF (usually -1). For the narrow stream functions it is important to store the result of these functions in a variable of type int instead of char, even when you plan to use it only as a character. Storing EOF in a char variable truncates its value to the size of a character, so that it is no longer distinguishable from the valid character `(char) -1'. So always use an int for the result of getc and friends, and check for EOF after the call; once you've verified that the result is not EOF, you can be sure that it will fit in a `char' variable without loss of information.

    Function: int fgetc (FILE *stream)
    This function reads the next character as an unsigned char from the stream stream and returns its value, converted to an int. If an end-of-file condition or read error occurs, EOF is returned instead.

    Function: wint_t fgetwc (FILE *stream)
    This function reads the next wide character from the stream stream and returns its value. If an end-of-file condition or read error occurs, WEOF is returned instead.

    Function: int fgetc_unlocked (FILE *stream)
    The fgetc_unlocked function is equivalent to the fgetc function except that it does not implicitly lock the stream.

    Function: wint_t fgetwc_unlocked (FILE *stream)
    The fgetwc_unlocked function is equivalent to the fgetwc function except that it does not implicitly lock the stream.

    This function is a GNU extension.

    Function: int getc (FILE *stream)
    This is just like fgetc, except that it is permissible (and typical) for it to be implemented as a macro that evaluates the stream argument more than once. getc is often highly optimized, so it is usually the best function to use to read a single character.

    Function: wint_t getwc (FILE *stream)
    This is just like fgetwc, except that it is permissible for it to be implemented as a macro that evaluates the stream argument more than once. getwc can be highly optimized, so it is usually the best function to use to read a single wide character.

    Function: int getc_unlocked (FILE *stream)
    The getc_unlocked function is equivalent to the getc function except that it does not implicitly lock the stream.

    Function: wint_t getwc_unlocked (FILE *stream)
    The getwc_unlocked function is equivalent to the getwc function except that it does not implicitly lock the stream.

    This function is a GNU extension.

    Function: int getchar (void)
    The getchar function is equivalent to getc with stdin as the value of the stream argument.

    Function: wint_t getwchar (void)
    The getwchar function is equivalent to getwc with stdin as the value of the stream argument.

    Function: int getchar_unlocked (void)
    The getchar_unlocked function is equivalent to the getchar function except that it does not implicitly lock the stream.

    Function: wint_t getwchar_unlocked (void)
    The getwchar_unlocked function is equivalent to the getwchar function except that it does not implicitly lock the stream.

    This function is a GNU extension.

    Here is an example of a function that does input using fgetc. It would work just as well using getc instead, or using getchar () instead of fgetc (stdin). The code would also work the same for the wide character stream functions.

    int
    y_or_n_p (const char *question)
    {
      fputs (question, stdout);
      while (1)
        {
          int c, answer;
          /* Write a space to separate answer from question. */
          fputc (' ', stdout);
          /* Read the first character of the line.
             This should be the answer character, but might not be. */
          c = tolower (fgetc (stdin));
          answer = c;
          /* Discard rest of input line. */
          while (c != '\n' && c != EOF)
            c = fgetc (stdin);
          /* Obey the answer if it was valid. */
          if (answer == 'y')
            return 1;
          if (answer == 'n')
            return 0;
          /* Answer was invalid: ask for valid answer. */
          fputs ("Please answer y or n:", stdout);
        }
    }
    

    Function: int getw (FILE *stream)
    This function reads a word (that is, an int) from stream. It's provided for compatibility with SVID. We recommend you use fread instead (see section Block Input/Output). Unlike getc, any int value could be a valid result. getw returns EOF when it encounters end-of-file or an error, but there is no way to distinguish this from an input word with value -1.

    Line-Oriented Input

    Since many programs interpret input on the basis of lines, it is convenient to have functions to read a line of text from a stream.

    Standard C has functions to do this, but they aren't very safe: null characters and even (for gets) long lines can confuse them. So the GNU library provides the nonstandard getline function that makes it easy to read lines reliably.

    Another GNU extension, getdelim, generalizes getline. It reads a delimited record, defined as everything through the next occurrence of a specified delimiter character.

    All these functions are declared in `stdio.h'.

    Function: ssize_t getline (char **lineptr, size_t *n, FILE *stream)
    This function reads an entire line from stream, storing the text (including the newline and a terminating null character) in a buffer and storing the buffer address in *lineptr.

    Before calling getline, you should place in *lineptr the address of a buffer *n bytes long, allocated with malloc. If this buffer is long enough to hold the line, getline stores the line in this buffer. Otherwise, getline makes the buffer bigger using realloc, storing the new buffer address back in *lineptr and the increased size back in *n. See section Unconstrained Allocation.

    If you set *lineptr to a null pointer, and *n to zero, before the call, then getline allocates the initial buffer for you by calling malloc.

    In either case, when getline returns, *lineptr is a char * which points to the text of the line.

    When getline is successful, it returns the number of characters read (including the newline, but not including the terminating null). This value enables you to distinguish null characters that are part of the line from the null character inserted as a terminator.

    This function is a GNU extension, but it is the recommended way to read lines from a stream. The alternative standard functions are unreliable.

    If an error occurs or end of file is reached without any bytes read, getline returns -1.

    Function: ssize_t getdelim (char **lineptr, size_t *n, int delimiter, FILE *stream)
    This function is like getline except that the character which tells it to stop reading is not necessarily newline. The argument delimiter specifies the delimiter character; getdelim keeps reading until it sees that character (or end of file).

    The text is stored in lineptr, including the delimiter character and a terminating null. Like getline, getdelim makes lineptr bigger if it isn't big enough.

    getline is in fact implemented in terms of getdelim, just like this:

    ssize_t
    getline (char **lineptr, size_t *n, FILE *stream)
    {
      return getdelim (lineptr, n, '\n', stream);
    }
    

    Function: char * fgets (char *s, int count, FILE *stream)
    The fgets function reads characters from the stream stream up to and including a newline character and stores them in the string s, adding a null character to mark the end of the string. You must supply count characters worth of space in s, but the number of characters read is at most count - 1. The extra character space is used to hold the null character at the end of the string.

    If the system is already at end of file when you call fgets, then the contents of the array s are unchanged and a null pointer is returned. A null pointer is also returned if a read error occurs. Otherwise, the return value is the pointer s.

    Warning: If the input data has a null character, you can't tell. So don't use fgets unless you know the data cannot contain a null. Don't use it to read files edited by the user because, if the user inserts a null character, you should either handle it properly or print a clear error message. We recommend using getline instead of fgets.

    Function: wchar_t * fgetws (wchar_t *ws, int count, FILE *stream)
    The fgetws function reads wide characters from the stream stream up to and including a newline character and stores them in the string ws, adding a null wide character to mark the end of the string. You must supply count wide characters worth of space in ws, but the number of characters read is at most count - 1. The extra character space is used to hold the null wide character at the end of the string.

    If the system is already at end of file when you call fgetws, then the contents of the array ws are unchanged and a null pointer is returned. A null pointer is also returned if a read error occurs. Otherwise, the return value is the pointer ws.

    Warning: If the input data has a null wide character (which are null bytes in the input stream), you can't tell. So don't use fgetws unless you know the data cannot contain a null. Don't use it to read files edited by the user because, if the user inserts a null character, you should either handle it properly or print a clear error message.

    Function: char * fgets_unlocked (char *s, int count, FILE *stream)
    The fgets_unlocked function is equivalent to the fgets function except that it does not implicitly lock the stream.

    This function is a GNU extension.

    Function: wchar_t * fgetws_unlocked (wchar_t *ws, int count, FILE *stream)
    The fgetws_unlocked function is equivalent to the fgetws function except that it does not implicitly lock the stream.

    This function is a GNU extension.

    Deprecated function: char * gets (char *s)
    The function gets reads characters from the stream stdin up to the next newline character, and stores them in the string s. The newline character is discarded (note that this differs from the behavior of fgets, which copies the newline character into the string). If gets encounters a read error or end-of-file, it returns a null pointer; otherwise it returns s.

    Warning: The gets function is very dangerous because it provides no protection against overflowing the string s. The GNU library includes it for compatibility only. You should always use fgets or getline instead. To remind you of this, the linker (if using GNU ld) will issue a warning whenever you use gets.

    Unreading

    In parser programs it is often useful to examine the next character in the input stream without removing it from the stream. This is called "peeking ahead" at the input because your program gets a glimpse of the input it will read next.

    Using stream I/O, you can peek ahead at input by first reading it and then unreading it (also called pushing it back on the stream). Unreading a character makes it available to be input again from the stream, by the next call to fgetc or other input function on that stream.

    What Unreading Means

    Here is a pictorial explanation of unreading. Suppose you have a stream reading a file that contains just six characters, the letters `foobar'. Suppose you have read three characters so far. The situation looks like this:

    f  o  o  b  a  r
             ^
    

    so the next input character will be `b'.

    If instead of reading `b' you unread the letter `o', you get a situation like this:

    f  o  o  b  a  r
             |
          o--
          ^
    

    so that the next input characters will be `o' and `b'.

    If you unread `9' instead of `o', you get this situation:

    f  o  o  b  a  r
             |
          9--
          ^
    

    so that the next input characters will be `9' and `b'.

    Using ungetc To Do Unreading

    The function to unread a character is called ungetc, because it reverses the action of getc.

    Function: int ungetc (int c, FILE *stream)
    The ungetc function pushes back the character c onto the input stream stream. So the next input from stream will read c before anything else.

    If c is EOF, ungetc does nothing and just returns EOF. This lets you call ungetc with the return value of getc without needing to check for an error from getc.

    The character that you push back doesn't have to be the same as the last character that was actually read from the stream. In fact, it isn't necessary to actually read any characters from the stream before unreading them with ungetc! But that is a strange way to write a program; usually ungetc is used only to unread a character that was just read from the same stream.

    The GNU C library only supports one character of pushback--in other words, it does not work to call ungetc twice without doing input in between. Other systems might let you push back multiple characters; then reading from the stream retrieves the characters in the reverse order that they were pushed.

    Pushing back characters doesn't alter the file; only the internal buffering for the stream is affected. If a file positioning function (such as fseek, fseeko or rewind; see section File Positioning) is called, any pending pushed-back characters are discarded.

    Unreading a character on a stream that is at end of file clears the end-of-file indicator for the stream, because it makes the character of input available. After you read that character, trying to read again will encounter end of file.

    Function: wint_t ungetwc (wint_t wc, FILE *stream)
    The ungetwc function behaves just like ungetc just that it pushes back a wide character.

    Here is an example showing the use of getc and ungetc to skip over whitespace characters. When this function reaches a non-whitespace character, it unreads that character to be seen again on the next read operation on the stream.

    #include <stdio.h>
    #include <ctype.h>
    
    void
    skip_whitespace (FILE *stream)
    {
      int c;
      do
        /* No need to check for EOF because it is not
           isspace, and ungetc ignores EOF.  */
        c = getc (stream);
      while (isspace (c));
      ungetc (c, stream);
    }
    

    Block Input/Output

    This section describes how to do input and output operations on blocks of data. You can use these functions to read and write binary data, as well as to read and write text in fixed-size blocks instead of by characters or lines.

    Binary files are typically used to read and write blocks of data in the same format as is used to represent the data in a running program. In other words, arbitrary blocks of memory--not just character or string objects--can be written to a binary file, and meaningfully read in again by the same program.

    Storing data in binary form is often considerably more efficient than using the formatted I/O functions. Also, for floating-point numbers, the binary form avoids possible loss of precision in the conversion process. On the other hand, binary files can't be examined or modified easily using many standard file utilities (such as text editors), and are not portable between different implementations of the language, or different kinds of computers.

    These functions are declared in `stdio.h'.

    Function: size_t fread (void *data, size_t size, size_t count, FILE *stream)
    This function reads up to count objects of size size into the array data, from the stream stream. It returns the number of objects actually read, which might be less than count if a read error occurs or the end of the file is reached. This function returns a value of zero (and doesn't read anything) if either size or count is zero.

    If fread encounters end of file in the middle of an object, it returns the number of complete objects read, and discards the partial object. Therefore, the stream remains at the actual end of the file.

    Function: size_t fread_unlocked (void *data, size_t size, size_t count, FILE *stream)
    The fread_unlocked function is equivalent to the fread function except that it does not implicitly lock the stream.

    This function is a GNU extension.

    Function: size_t fwrite (const void *data, size_t size, size_t count, FILE *stream)
    This function writes up to count objects of size size from the array data, to the stream stream. The return value is normally count, if the call succeeds. Any other value indicates some sort of error, such as running out of space.

    Function: size_t fwrite_unlocked (const void *data, size_t size, size_t count, FILE *stream)
    The fwrite_unlocked function is equivalent to the fwrite function except that it does not implicitly lock the stream.

    This function is a GNU extension.

    Formatted Output

    The functions described in this section (printf and related functions) provide a convenient way to perform formatted output. You call printf with a format string or template string that specifies how to format the values of the remaining arguments.

    Unless your program is a filter that specifically performs line- or character-oriented processing, using printf or one of the other related functions described in this section is usually the easiest and most concise way to perform output. These functions are especially useful for printing error messages, tables of data, and the like.

    Formatted Output Basics

    The printf function can be used to print any number of arguments. The template string argument you supply in a call provides information not only about the number of additional arguments, but also about their types and what style should be used for printing them.

    Ordinary characters in the template string are simply written to the output stream as-is, while conversion specifications introduced by a `%' character in the template cause subsequent arguments to be formatted and written to the output stream. For example,

    int pct = 37;
    char filename[] = "foo.txt";
    printf ("Processing of `%s' is %d%% finished.\nPlease be patient.\n",
            filename, pct);
    

    produces output like

    Processing of `foo.txt' is 37% finished.
    Please be patient.
    

    This example shows the use of the `%d' conversion to specify that an int argument should be printed in decimal notation, the `%s' conversion to specify printing of a string argument, and the `%%' conversion to print a literal `%' character.

    There are also conversions for printing an integer argument as an unsigned value in octal, decimal, or hexadecimal radix (`%o', `%u', or `%x', respectively); or as a character value (`%c').

    Floating-point numbers can be printed in normal, fixed-point notation using the `%f' conversion or in exponential notation using the `%e' conversion. The `%g' conversion uses either `%e' or `%f' format, depending on what is more appropriate for the magnitude of the particular number.

    You can control formatting more precisely by writing modifiers between the `%' and the character that indicates which conversion to apply. These slightly alter the ordinary behavior of the conversion. For example, most conversion specifications permit you to specify a minimum field width and a flag indicating whether you want the result left- or right-justified within the field.

    The specific flags and modifiers that are permitted and their interpretation vary depending on the particular conversion. They're all described in more detail in the following sections. Don't worry if this all seems excessively complicated at first; you can almost always get reasonable free-format output without using any of the modifiers at all. The modifiers are mostly used to make the output look "prettier" in tables.

    Output Conversion Syntax

    This section provides details about the precise syntax of conversion specifications that can appear in a printf template string.

    Characters in the template string that are not part of a conversion specification are printed as-is to the output stream. Multibyte character sequences (see section Character Set Handling) are permitted in a template string.

    The conversion specifications in a printf template string have the general form:

    % [ param-no $] flags width [ . precision ] type conversion
    

    For example, in the conversion specifier `%-10.8ld', the `-' is a flag, `10' specifies the field width, the precision is `8', the letter `l' is a type modifier, and `d' specifies the conversion style. (This particular type specifier says to print a long int argument in decimal notation, with a minimum of 8 digits left-justified in a field at least 10 characters wide.)

    In more detail, output conversion specifications consist of an initial `%' character followed in sequence by:

    • An optional specification of the parameter used for this format. Normally the parameters to the printf function are assigned to the formats in the order of appearance in the format string. But in some situations (such as message translation) this is not desirable and this extension allows an explicit parameter to be specified. The param-no part of the format must be an integer in the range of 1 to the maximum number of arguments present to the function call. Some implementations limit this number to a certainly upper bound. The exact limit can be retrieved by the following constant.
      Macro: NL_ARGMAX
      The value of ARGMAX is the maximum value allowed for the specification of an positional parameter in a printf call. The actual value in effect at runtime can be retrieved by using sysconf using the _SC_NL_ARGMAX parameter see section Definition of sysconf. Some system have a quite low limit such as @math{9} for System V systems. The GNU C library has no real limit.
      If any of the formats has a specification for the parameter position all of them in the format string shall have one. Otherwise the behaviour is undefined.
    • Zero or more flag characters that modify the normal behavior of the conversion specification.
    • An optional decimal integer specifying the minimum field width. If the normal conversion produces fewer characters than this, the field is padded with spaces to the specified width. This is a minimum value; if the normal conversion produces more characters than this, the field is not truncated. Normally, the output is right-justified within the field. You can also specify a field width of `*'. This means that the next argument in the argument list (before the actual value to be printed) is used as the field width. The value must be an int. If the value is negative, this means to set the `-' flag (see below) and to use the absolute value as the field width.
    • An optional precision to specify the number of digits to be written for the numeric conversions. If the precision is specified, it consists of a period (`.') followed optionally by a decimal integer (which defaults to zero if omitted). You can also specify a precision of `*'. This means that the next argument in the argument list (before the actual value to be printed) is used as the precision. The value must be an int, and is ignored if it is negative. If you specify `*' for both the field width and precision, the field width argument precedes the precision argument. Other C library versions may not recognize this syntax.
    • An optional type modifier character, which is used to specify the data type of the corresponding argument if it differs from the default type. (For example, the integer conversions assume a type of int, but you can specify `h', `l', or `L' for other integer types.)
    • A character that specifies the conversion to be applied.

    The exact options that are permitted and how they are interpreted vary between the different conversion specifiers. See the descriptions of the individual conversions for information about the particular options that they use.

    With the `-Wformat' option, the GNU C compiler checks calls to printf and related functions. It examines the format string and verifies that the correct number and types of arguments are supplied. There is also a GNU C syntax to tell the compiler that a function you write uses a printf-style format string. See section `Declaring Attributes of Functions' in Using GNU CC, for more information.

    Table of Output Conversions

    Here is a table summarizing what all the different conversions do:

    `%d', `%i'
    Print an integer as a signed decimal number. See section Integer Conversions, for details. `%d' and `%i' are synonymous for output, but are different when used with scanf for input (see section Table of Input Conversions).
    `%o'
    Print an integer as an unsigned octal number. See section Integer Conversions, for details.
    `%u'
    Print an integer as an unsigned decimal number. See section Integer Conversions, for details.
    `%x', `%X'
    Print an integer as an unsigned hexadecimal number. `%x' uses lower-case letters and `%X' uses upper-case. See section Integer Conversions, for details.
    `%f'
    Print a floating-point number in normal (fixed-point) notation. See section Floating-Point Conversions, for details.
    `%e', `%E'
    Print a floating-point number in exponential notation. `%e' uses lower-case letters and `%E' uses upper-case. See section Floating-Point Conversions, for details.
    `%g', `%G'
    Print a floating-point number in either normal or exponential notation, whichever is more appropriate for its magnitude. `%g' uses lower-case letters and `%G' uses upper-case. See section Floating-Point Conversions, for details.
    `%a', `%A'
    Print a floating-point number in a hexadecimal fractional notation which the exponent to base 2 represented in decimal digits. `%a' uses lower-case letters and `%A' uses upper-case. See section Floating-Point Conversions, for details.
    `%c'
    Print a single character. See section Other Output Conversions.
    `%C'
    This is an alias for `%lc' which is supported for compatibility with the Unix standard.
    `%s'
    Print a string. See section Other Output Conversions.
    `%S'
    This is an alias for `%ls' which is supported for compatibility with the Unix standard.
    `%p'
    Print the value of a pointer. See section Other Output Conversions.
    `%n'
    Get the number of characters printed so far. See section Other Output Conversions. Note that this conversion specification never produces any output.
    `%m'
    Print the string corresponding to the value of errno. (This is a GNU extension.) See section Other Output Conversions.
    `%%'
    Print a literal `%' character. See section Other Output Conversions.

    If the syntax of a conversion specification is invalid, unpredictable things will happen, so don't do this. If there aren't enough function arguments provided to supply values for all the conversion specifications in the template string, or if the arguments are not of the correct types, the results are unpredictable. If you supply more arguments than conversion specifications, the extra argument values are simply ignored; this is sometimes useful.

    Integer Conversions

    This section describes the options for the `%d', `%i', `%o', `%u', `%x', and `%X' conversion specifications. These conversions print integers in various formats.

    The `%d' and `%i' conversion specifications both print an int argument as a signed decimal number; while `%o', `%u', and `%x' print the argument as an unsigned octal, decimal, or hexadecimal number (respectively). The `%X' conversion specification is just like `%x' except that it uses the characters `ABCDEF' as digits instead of `abcdef'.

    The following flags are meaningful:

    `-'
    Left-justify the result in the field (instead of the normal right-justification).
    `+'
    For the signed `%d' and `%i' conversions, print a plus sign if the value is positive.
    ` '
    For the signed `%d' and `%i' conversions, if the result doesn't start with a plus or minus sign, prefix it with a space character instead. Since the `+' flag ensures that the result includes a sign, this flag is ignored if you supply both of them.
    `#'
    For the `%o' conversion, this forces the leading digit to be `0', as if by increasing the precision. For `%x' or `%X', this prefixes a leading `0x' or `0X' (respectively) to the result. This doesn't do anything useful for the `%d', `%i', or `%u' conversions. Using this flag produces output which can be parsed by the strtoul function (see section Parsing of Integers) and scanf with the `%i' conversion (see section Numeric Input Conversions).
    `''
    Separate the digits into groups as specified by the locale specified for the LC_NUMERIC category; see section Generic Numeric Formatting Parameters. This flag is a GNU extension.
    `0'
    Pad the field with zeros instead of spaces. The zeros are placed after any indication of sign or base. This flag is ignored if the `-' flag is also specified, or if a precision is specified.

    If a precision is supplied, it specifies the minimum number of digits to appear; leading zeros are produced if necessary. If you don't specify a precision, the number is printed with as many digits as it needs. If you convert a value of zero with an explicit precision of zero, then no characters at all are produced.

    Without a type modifier, the corresponding argument is treated as an int (for the signed conversions `%i' and `%d') or unsigned int (for the unsigned conversions `%o', `%u', `%x', and `%X'). Recall that since printf and friends are variadic, any char and short arguments are automatically converted to int by the default argument promotions. For arguments of other integer types, you can use these modifiers:

    `hh'
    Specifies that the argument is a signed char or unsigned char, as appropriate. A char argument is converted to an int or unsigned int by the default argument promotions anyway, but the `h' modifier says to convert it back to a char again. This modifier was introduced in ISO C99.
    `h'
    Specifies that the argument is a short int or unsigned short int, as appropriate. A short argument is converted to an int or unsigned int by the default argument promotions anyway, but the `h' modifier says to convert it back to a short again.
    `j'
    Specifies that the argument is a intmax_t or uintmax_t, as appropriate. This modifier was introduced in ISO C99.
    `l'
    Specifies that the argument is a long int or unsigned long int, as appropriate. Two `l' characters is like the `L' modifier, below. If used with `%c' or `%s' the corresponding parameter is considered as a wide character or wide character string respectively. This use of `l' was introduced in Amendment 1 to ISO C90.
    `L'
    `ll'
    `q'
    Specifies that the argument is a long long int. (This type is an extension supported by the GNU C compiler. On systems that don't support extra-long integers, this is the same as long int.) The `q' modifier is another name for the same thing, which comes from 4.4 BSD; a long long int is sometimes called a "quad" int.
    `t'
    Specifies that the argument is a ptrdiff_t. This modifier was introduced in ISO C99.
    `z'
    `Z'
    Specifies that the argument is a size_t. `z' was introduced in ISO C99. `Z' is a GNU extension predating this addition and should not be used in new code.

    Here is an example. Using the template string:

    "|%5d|%-5d|%+5d|%+-5d|% 5d|%05d|%5.0d|%5.2d|%d|\n"
    

    to print numbers using the different options for the `%d' conversion gives results like:

    |    0|0    |   +0|+0   |    0|00000|     |   00|0|
    |    1|1    |   +1|+1   |    1|00001|    1|   01|1|
    |   -1|-1   |   -1|-1   |   -1|-0001|   -1|  -01|-1|
    |100000|100000|+100000| 100000|100000|100000|100000|100000|
    

    In particular, notice what happens in the last case where the number is too large to fit in the minimum field width specified.

    Here are some more examples showing how unsigned integers print under various format options, using the template string:

    "|%5u|%5o|%5x|%5X|%#5o|%#5x|%#5X|%#10.8x|\n"
    
    |    0|    0|    0|    0|    0|  0x0|  0X0|0x00000000|
    |    1|    1|    1|    1|   01|  0x1|  0X1|0x00000001|
    |100000|303240|186a0|186A0|0303240|0x186a0|0X186A0|0x000186a0|
    

    Floating-Point Conversions

    This section discusses the conversion specifications for floating-point numbers: the `%f', `%e', `%E', `%g', and `%G' conversions.

    The `%f' conversion prints its argument in fixed-point notation, producing output of the form [-]ddd.ddd, where the number of digits following the decimal point is controlled by the precision you specify.

    The `%e' conversion prints its argument in exponential notation, producing output of the form [-]d.ddde[+|-]dd. Again, the number of digits following the decimal point is controlled by the precision. The exponent always contains at least two digits. The `%E' conversion is similar but the exponent is marked with the letter `E' instead of `e'.

    The `%g' and `%G' conversions print the argument in the style of `%e' or `%E' (respectively) if the exponent would be less than -4 or greater than or equal to the precision; otherwise they use the `%f' style. Trailing zeros are removed from the fractional portion of the result and a decimal-point character appears only if it is followed by a digit.

    The `%a' and `%A' conversions are meant for representing floating-point numbers exactly in textual form so that they can be exchanged as texts between different programs and/or machines. The numbers are represented is the form [-]0xh.hhhp[+|-]dd. At the left of the decimal-point character exactly one digit is print. This character is only 0 if the number is denormalized. Otherwise the value is unspecified; it is implementation dependent how many bits are used. The number of hexadecimal digits on the right side of the decimal-point character is equal to the precision. If the precision is zero it is determined to be large enough to provide an exact representation of the number (or it is large enough to distinguish two adjacent values if the FLT_RADIX is not a power of 2, see section Floating Point Parameters). For the `%a' conversion lower-case characters are used to represent the hexadecimal number and the prefix and exponent sign are printed as 0x and p respectively. Otherwise upper-case characters are used and 0X and P are used for the representation of prefix and exponent string. The exponent to the base of two is printed as a decimal number using at least one digit but at most as many digits as necessary to represent the value exactly.

    If the value to be printed represents infinity or a NaN, the output is [-]inf or nan respectively if the conversion specifier is `%a', `%e', `%f', or `%g' and it is [-]INF or NAN respectively if the conversion is `%A', `%E', or `%G'.

    The following flags can be used to modify the behavior:

    `-'
    Left-justify the result in the field. Normally the result is right-justified.
    `+'
    Always include a plus or minus sign in the result.
    ` '
    If the result doesn't start with a plus or minus sign, prefix it with a space instead. Since the `+' flag ensures that the result includes a sign, this flag is ignored if you supply both of them.
    `#'
    Specifies that the result should always include a decimal point, even if no digits follow it. For the `%g' and `%G' conversions, this also forces trailing zeros after the decimal point to be left in place where they would otherwise be removed.
    `''
    Separate the digits of the integer part of the result into groups as specified by the locale specified for the LC_NUMERIC category; see section Generic Numeric Formatting Parameters. This flag is a GNU extension.
    `0'
    Pad the field with zeros instead of spaces; the zeros are placed after any sign. This flag is ignored if the `-' flag is also specified.

    The precision specifies how many digits follow the decimal-point character for the `%f', `%e', and `%E' conversions. For these conversions, the default precision is 6. If the precision is explicitly 0, this suppresses the decimal point character entirely. For the `%g' and `%G' conversions, the precision specifies how many significant digits to print. Significant digits are the first digit before the decimal point, and all the digits after it. If the precision is 0 or not specified for `%g' or `%G', it is treated like a value of 1. If the value being printed cannot be expressed accurately in the specified number of digits, the value is rounded to the nearest number that fits.

    Without a type modifier, the floating-point conversions use an argument of type double. (By the default argument promotions, any float arguments are automatically converted to double.) The following type modifier is supported:

    `L'
    An uppercase `L' specifies that the argument is a long double.

    Here are some examples showing how numbers print using the various floating-point conversions. All of the numbers were printed using this template string:

    "|%13.4a|%13.4f|%13.4e|%13.4g|\n"
    

    Here is the output:

    |  0x0.0000p+0|       0.0000|   0.0000e+00|            0|
    |  0x1.0000p-1|       0.5000|   5.0000e-01|          0.5|
    |  0x1.0000p+0|       1.0000|   1.0000e+00|            1|
    | -0x1.0000p+0|      -1.0000|  -1.0000e+00|           -1|
    |  0x1.9000p+6|     100.0000|   1.0000e+02|          100|
    |  0x1.f400p+9|    1000.0000|   1.0000e+03|         1000|
    | 0x1.3880p+13|   10000.0000|   1.0000e+04|        1e+04|
    | 0x1.81c8p+13|   12345.0000|   1.2345e+04|    1.234e+04|
    | 0x1.86a0p+16|  100000.0000|   1.0000e+05|        1e+05|
    | 0x1.e240p+16|  123456.0000|   1.2346e+05|    1.235e+05|
    

    Notice how the `%g' conversion drops trailing zeros.

    Other Output Conversions

    This section describes miscellaneous conversions for printf.

    The `%c' conversion prints a single character. In case there is no `l' modifier the int argument is first converted to an unsigned char. Then, if used in a wide stream function, the character is converted into the corresponding wide character. The `-' flag can be used to specify left-justification in the field, but no other flags are defined, and no precision or type modifier can be given. For example:

    printf ("%c%c%c%c%c", 'h', 'e', 'l', 'l', 'o');
    

    prints `hello'.

    If there is a `l' modifier present the argument is expected to be of type wint_t. If used in a multibyte function the wide character is converted into a multibyte character before being added to the output. In this case more than one output byte can be produced.

    The `%s' conversion prints a string. If no `l' modifier is present the corresponding argument must be of type char * (or const char *). If used in a wide stream function the string is first converted in a wide character string. A precision can be specified to indicate the maximum number of characters to write; otherwise characters in the string up to but not including the terminating null character are written to the output stream. The `-' flag can be used to specify left-justification in the field, but no other flags or type modifiers are defined for this conversion. For example:

    printf ("%3s%-6s", "no", "where");
    

    prints ` nowhere '.

    If there is a `l' modifier present the argument is expected to be of type wchar_t (or const wchar_t *).

    If you accidentally pass a null pointer as the argument for a `%s' conversion, the GNU library prints it as `(null)'. We think this is more useful than crashing. But it's not good practice to pass a null argument intentionally.

    The `%m' conversion prints the string corresponding to the error code in errno. See section Error Messages. Thus:

    fprintf (stderr, "can't open `%s': %m\n", filename);
    

    is equivalent to:

    fprintf (stderr, "can't open `%s': %s\n", filename, strerror (errno));
    

    The `%m' conversion is a GNU C library extension.

    The `%p' conversion prints a pointer value. The corresponding argument must be of type void *. In practice, you can use any type of pointer.

    In the GNU system, non-null pointers are printed as unsigned integers, as if a `%#x' conversion were used. Null pointers print as `(nil)'. (Pointers might print differently in other systems.)

    For example:

    printf ("%p", "testing");
    

    prints `0x' followed by a hexadecimal number--the address of the string constant "testing". It does not print the word `testing'.

    You can supply the `-' flag with the `%p' conversion to specify left-justification, but no other flags, precision, or type modifiers are defined.

    The `%n' conversion is unlike any of the other output conversions. It uses an argument which must be a pointer to an int, but instead of printing anything it stores the number of characters printed so far by this call at that location. The `h' and `l' type modifiers are permitted to specify that the argument is of type short int * or long int * instead of int *, but no flags, field width, or precision are permitted.

    For example,

    int nchar;
    printf ("%d %s%n\n", 3, "bears", &nchar);
    

    prints:

    3 bears
    

    and sets nchar to 7, because `3 bears' is seven characters.

    The `%%' conversion prints a literal `%' character. This conversion doesn't use an argument, and no flags, field width, precision, or type modifiers are permitted.

    Formatted Output Functions

    This section describes how to call printf and related functions. Prototypes for these functions are in the header file `stdio.h'. Because these functions take a variable number of arguments, you must declare prototypes for them before using them. Of course, the easiest way to make sure you have all the right prototypes is to just include `stdio.h'.

    Function: int printf (const char *template, ...)
    The printf function prints the optional arguments under the control of the template string template to the stream stdout. It returns the number of characters printed, or a negative value if there was an output error.

    Function: int wprintf (const wchar_t *template, ...)
    The wprintf function prints the optional arguments under the control of the wide template string template to the stream stdout. It returns the number of wide characters printed, or a negative value if there was an output error.

    Function: int fprintf (FILE *stream, const char *template, ...)
    This function is just like printf, except that the output is written to the stream stream instead of stdout.

    Function: int fwprintf (FILE *stream, const wchar_t *template, ...)
    This function is just like wprintf, except that the output is written to the stream stream instead of stdout.

    Function: int sprintf (char *s, const char *template, ...)
    This is like printf, except that the output is stored in the character array s instead of written to a stream. A null character is written to mark the end of the string.

    The sprintf function returns the number of characters stored in the array s, not including the terminating null character.

    The behavior of this function is undefined if copying takes place between objects that overlap--for example, if s is also given as an argument to be printed under control of the `%s' conversion. See section Copying and Concatenation.

    Warning: The sprintf function can be dangerous because it can potentially output more characters than can fit in the allocation size of the string s. Remember that the field width given in a conversion specification is only a minimum value.

    To avoid this problem, you can use snprintf or asprintf, described below.

    Function: int swprintf (wchar_t *s, size_t size, const wchar_t *template, ...)
    This is like wprintf, except that the output is stored in the wide character array ws instead of written to a stream. A null wide character is written to mark the end of the string. The size argument specifies the maximum number of characters to produce. The trailing null character is counted towards this limit, so you should allocate at least size wide characters for the string ws.

    The return value is the number of characters which would be generated for the given input, excluding the trailing null. If this value is greater or equal to size, not all characters from the result have been stored in ws. You should try again with a bigger output string.

    Note that the corresponding narrow stream function takes fewer parameters. swprintf in fact corresponds to the snprintf function. Since the sprintf function can be dangerous and should be avoided the ISO C committee refused to make the same mistake again and decided to not define an function exactly corresponding to sprintf.

    Function: int snprintf (char *s, size_t size, const char *template, ...)
    The snprintf function is similar to sprintf, except that the size argument specifies the maximum number of characters to produce. The trailing null character is counted towards this limit, so you should allocate at least size characters for the string s.

    The return value is the number of characters which would be generated for the given input, excluding the trailing null. If this value is greater or equal to size, not all characters from the result have been stored in s. You should try again with a bigger output string. Here is an example of doing this:

    /* Construct a message describing the value of a variable
       whose name is name and whose value is value. */
    char *
    make_message (char *name, char *value)
    {
      /* Guess we need no more than 100 chars of space. */
      int size = 100;
      char *buffer = (char *) xmalloc (size);
      int nchars;
      if (buffer == NULL)
        return NULL;
    
     /* Try to print in the allocated space. */
      nchars = snprintf (buffer, size, "value of %s is %s",
                         name, value);
      if (nchars >= size)
        {
          /* Reallocate buffer now that we know
             how much space is needed. */
          buffer = (char *) xrealloc (buffer, nchars + 1);
    
          if (buffer != NULL)
            /* Try again. */
            snprintf (buffer, size, "value of %s is %s",
                      name, value);
        }
      /* The last call worked, return the string. */
      return buffer;
    }
    

    In practice, it is often easier just to use asprintf, below.

    Attention: In versions of the GNU C library prior to 2.1 the return value is the number of characters stored, not including the terminating null; unless there was not enough space in s to store the result in which case -1 is returned. This was changed in order to comply with the ISO C99 standard.

    Dynamically Allocating Formatted Output

    The functions in this section do formatted output and place the results in dynamically allocated memory.

    Function: int asprintf (char **ptr, const char *template, ...)
    This function is similar to sprintf, except that it dynamically allocates a string (as with malloc; see section Unconstrained Allocation) to hold the output, instead of putting the output in a buffer you allocate in advance. The ptr argument should be the address of a char * object, and asprintf stores a pointer to the newly allocated string at that location.

    The return value is the number of characters allocated for the buffer, or less than zero if an error occured. Usually this means that the buffer could not be allocated.

    Here is how to use asprintf to get the same result as the snprintf example, but more easily:

    /* Construct a message describing the value of a variable
       whose name is name and whose value is value. */
    char *
    make_message (char *name, char *value)
    {
      char *result;
      if (asprintf (&result, "value of %s is %s", name, value) < 0)
        return NULL;
      return result;
    }
    

    Function: int obstack_printf (struct obstack *obstack, const char *template, ...)
    This function is similar to asprintf, except that it uses the obstack obstack to allocate the space. See section Obstacks.

    The characters are written onto the end of the current object. To get at them, you must finish the object with obstack_finish (see section Growing Objects).

    Variable Arguments Output Functions

    The functions vprintf and friends are provided so that you can define your own variadic printf-like functions that make use of the same internals as the built-in formatted output functions.

    The most natural way to define such functions would be to use a language construct to say, "Call printf and pass this template plus all of my arguments after the first five." But there is no way to do this in C, and it would be hard to provide a way, since at the C language level there is no way to tell how many arguments your function received.

    Since that method is impossible, we provide alternative functions, the vprintf series, which lets you pass a va_list to describe "all of my arguments after the first five."

    When it is sufficient to define a macro rather than a real function, the GNU C compiler provides a way to do this much more easily with macros. For example:

    #define myprintf(a, b, c, d, e, rest...) \
                printf (mytemplate , ## rest...)
    

    See section `Macros with Variable Numbers of Arguments' in Using GNU CC, for details. But this is limited to macros, and does not apply to real functions at all.

    Before calling vprintf or the other functions listed in this section, you must call va_start (see section Variadic Functions) to initialize a pointer to the variable arguments. Then you can call va_arg to fetch the arguments that you want to handle yourself. This advances the pointer past those arguments.

    Once your va_list pointer is pointing at the argument of your choice, you are ready to call vprintf. That argument and all subsequent arguments that were passed to your function are used by vprintf along with the template that you specified separately.

    In some other systems, the va_list pointer may become invalid after the call to vprintf, so you must not use va_arg after you call vprintf. Instead, you should call va_end to retire the pointer from service. However, you can safely call va_start on another pointer variable and begin fetching the arguments again through that pointer. Calling vprintf does not destroy the argument list of your function, merely the particular pointer that you passed to it.

    GNU C does not have such restrictions. You can safely continue to fetch arguments from a va_list pointer after passing it to vprintf, and va_end is a no-op. (Note, however, that subsequent va_arg calls will fetch the same arguments which vprintf previously used.)

    Prototypes for these functions are declared in `stdio.h'.

    Function: int vprintf (const char *template, va_list ap)
    This function is similar to printf except that, instead of taking a variable number of arguments directly, it takes an argument list pointer ap.

    Function: int vwprintf (const wchar_t *template, va_list ap)
    This function is similar to wprintf except that, instead of taking a variable number of arguments directly, it takes an argument list pointer ap.

    Function: int vfprintf (FILE *stream, const char *template, va_list ap)
    This is the equivalent of fprintf with the variable argument list specified directly as for vprintf.

    Function: int vfwprintf (FILE *stream, const wchar_t *template, va_list ap)
    This is the equivalent of fwprintf with the variable argument list specified directly as for vwprintf.

    Function: int vsprintf (char *s, const char *template, va_list ap)
    This is the equivalent of sprintf with the variable argument list specified directly as for vprintf.

    Function: int vswprintf (wchar_t *s, size_t size, const wchar_t *template, va_list ap)
    This is the equivalent of swprintf with the variable argument list specified directly as for vwprintf.

    Function: int vsnprintf (char *s, size_t size, const char *template, va_list ap)
    This is the equivalent of snprintf with the variable argument list specified directly as for vprintf.

    Function: int vasprintf (char **ptr, const char *template, va_list ap)
    The vasprintf function is the equivalent of asprintf with the variable argument list specified directly as for vprintf.

    Function: int obstack_vprintf (struct obstack *obstack, const char *template, va_list ap)
    The obstack_vprintf function is the equivalent of obstack_printf with the variable argument list specified directly as for vprintf.

    Here's an example showing how you might use vfprintf. This is a function that prints error messages to the stream stderr, along with a prefix indicating the name of the program (see section Error Messages, for a description of program_invocation_short_name).

    #include <stdio.h>
    #include <stdarg.h>
    
    void
    eprintf (const char *template, ...)
    {
      va_list ap;
      extern char *program_invocation_short_name;
    
      fprintf (stderr, "%s: ", program_invocation_short_name);
      va_start (ap, template);
      vfprintf (stderr, template, ap);
      va_end (ap);
    }
    

    You could call eprintf like this:

    eprintf ("file `%s' does not exist\n", filename);
    

    In GNU C, there is a special construct you can use to let the compiler know that a function uses a printf-style format string. Then it can check the number and types of arguments in each call to the function, and warn you when they do not match the format string. For example, take this declaration of eprintf:

    void eprintf (const char *template, ...)
            __attribute__ ((format (printf, 1, 2)));
    

    This tells the compiler that eprintf uses a format string like printf (as opposed to scanf; see section Formatted Input); the format string appears as the first argument; and the arguments to satisfy the format begin with the second. See section `Declaring Attributes of Functions' in Using GNU CC, for more information.

    Parsing a Template String

    You can use the function parse_printf_format to obtain information about the number and types of arguments that are expected by a given template string. This function permits interpreters that provide interfaces to printf to avoid passing along invalid arguments from the user's program, which could cause a crash.

    All the symbols described in this section are declared in the header file `printf.h'.

    Function: size_t parse_printf_format (const char *template, size_t n, int *argtypes)
    This function returns information about the number and types of arguments expected by the printf template string template. The information is stored in the array argtypes; each element of this array describes one argument. This information is encoded using the various `PA_' macros, listed below.

    The argument n specifies the number of elements in the array argtypes. This is the maximum number of elements that parse_printf_format will try to write.

    parse_printf_format returns the total number of arguments required by template. If this number is greater than n, then the information returned describes only the first n arguments. If you want information about additional arguments, allocate a bigger array and call parse_printf_format again.

    The argument types are encoded as a combination of a basic type and modifier flag bits.

    Macro: int PA_FLAG_MASK
    This macro is a bitmask for the type modifier flag bits. You can write the expression (argtypes[i] & PA_FLAG_MASK) to extract just the flag bits for an argument, or (argtypes[i] & ~PA_FLAG_MASK) to extract just the basic type code.

    Here are symbolic constants that represent the basic types; they stand for integer values.

    PA_INT
    This specifies that the base type is int.
    PA_CHAR
    This specifies that the base type is int, cast to char.
    PA_STRING
    This specifies that the base type is char *, a null-terminated string.
    PA_POINTER
    This specifies that the base type is void *, an arbitrary pointer.
    PA_FLOAT
    This specifies that the base type is float.
    PA_DOUBLE
    This specifies that the base type is double.
    PA_LAST
    You can define additional base types for your own programs as offsets from PA_LAST. For example, if you have data types `foo' and `bar' with their own specialized printf conversions, you could define encodings for these types as:
    #define PA_FOO  PA_LAST
    #define PA_BAR  (PA_LAST + 1)
    

    Here are the flag bits that modify a basic type. They are combined with the code for the basic type using inclusive-or.

    PA_FLAG_PTR
    If this bit is set, it indicates that the encoded type is a pointer to the base type, rather than an immediate value. For example, `PA_INT|PA_FLAG_PTR' represents the type `int *'.
    PA_FLAG_SHORT
    If this bit is set, it indicates that the base type is modified with short. (This corresponds to the `h' type modifier.)
    PA_FLAG_LONG
    If this bit is set, it indicates that the base type is modified with long. (This corresponds to the `l' type modifier.)
    PA_FLAG_LONG_LONG
    If this bit is set, it indicates that the base type is modified with long long. (This corresponds to the `L' type modifier.)
    PA_FLAG_LONG_DOUBLE
    This is a synonym for PA_FLAG_LONG_LONG, used by convention with a base type of PA_DOUBLE to indicate a type of long double.

    Example of Parsing a Template String

    Here is an example of decoding argument types for a format string. We assume this is part of an interpreter which contains arguments of type NUMBER, CHAR, STRING and STRUCTURE (and perhaps others which are not valid here).

    /* Test whether the nargs specified objects
       in the vector args are valid
       for the format string format:
       if so, return 1.
       If not, return 0 after printing an error message.  */
    
    int
    validate_args (char *format, int nargs, OBJECT *args)
    {
      int *argtypes;
      int nwanted;
    
      /* Get the information about the arguments.
         Each conversion specification must be at least two characters
         long, so there cannot be more specifications than half the
         length of the string.  */
    
      argtypes = (int *) alloca (strlen (format) / 2 * sizeof (int));
      nwanted = parse_printf_format (string, nelts, argtypes);
    
      /* Check the number of arguments.  */
      if (nwanted > nargs)
        {
          error ("too few arguments (at least %d required)", nwanted);
          return 0;
        }
    
      /* Check the C type wanted for each argument
         and see if the object given is suitable.  */
      for (i = 0; i < nwanted; i++)
        {
          int wanted;
    
          if (argtypes[i] & PA_FLAG_PTR)
            wanted = STRUCTURE;
          else
            switch (argtypes[i] & ~PA_FLAG_MASK)
              {
              case PA_INT:
              case PA_FLOAT:
              case PA_DOUBLE:
                wanted = NUMBER;
                break;
              case PA_CHAR:
                wanted = CHAR;
                break;
              case PA_STRING:
                wanted = STRING;
                break;
              case PA_POINTER:
                wanted = STRUCTURE;
                break;
              }
          if (TYPE (args[i]) != wanted)
            {
              error ("type mismatch for arg number %d", i);
              return 0;
            }
        }
      return 1;
    }
    

    Customizing printf

    The GNU C library lets you define your own custom conversion specifiers for printf template strings, to teach printf clever ways to print the important data structures of your program.

    The way you do this is by registering the conversion with the function register_printf_function; see section Registering New Conversions. One of the arguments you pass to this function is a pointer to a handler function that produces the actual output; see section Defining the Output Handler, for information on how to write this function.

    You can also install a function that just returns information about the number and type of arguments expected by the conversion specifier. See section Parsing a Template String, for information about this.

    The facilities of this section are declared in the header file `printf.h'.

    Portability Note: The ability to extend the syntax of printf template strings is a GNU extension. ISO standard C has nothing similar.

    Registering New Conversions

    The function to register a new output conversion is register_printf_function, declared in `printf.h'.

    Function: int register_printf_function (int spec, printf_function handler-function, printf_arginfo_function arginfo-function)
    This function defines the conversion specifier character spec. Thus, if spec is 'Y', it defines the conversion `%Y'. You can redefine the built-in conversions like `%s', but flag characters like `#' and type modifiers like `l' can never be used as conversions; calling register_printf_function for those characters has no effect. It is advisable not to use lowercase letters, since the ISO C standard warns that additional lowercase letters may be standardized in future editions of the standard.

    The handler-function is the function called by printf and friends when this conversion appears in a template string. See section Defining the Output Handler, for information about how to define a function to pass as this argument. If you specify a null pointer, any existing handler function for spec is removed.

    The arginfo-function is the function called by parse_printf_format when this conversion appears in a template string. See section Parsing a Template String, for information about this.

    Attention: In the GNU C library versions before 2.0 the arginfo-function function did not need to be installed unless the user used the parse_printf_format function. This has changed. Now a call to any of the printf functions will call this function when this format specifier appears in the format string.

    The return value is 0 on success, and -1 on failure (which occurs if spec is out of range).

    You can redefine the standard output conversions, but this is probably not a good idea because of the potential for confusion. Library routines written by other people could break if you do this.

    Conversion Specifier Options

    If you define a meaning for `%A', what if the template contains `%+23A' or `%-#A'? To implement a sensible meaning for these, the handler when called needs to be able to get the options specified in the template.

    Both the handler-function and arginfo-function accept an argument that points to a struct printf_info, which contains information about the options appearing in an instance of the conversion specifier. This data type is declared in the header file `printf.h'.

    Type: struct printf_info
    This structure is used to pass information about the options appearing in an instance of a conversion specifier in a printf template string to the handler and arginfo functions for that specifier. It contains the following members:

    int prec
    This is the precision specified. The value is -1 if no precision was specified. If the precision was given as `*', the printf_info structure passed to the handler function contains the actual value retrieved from the argument list. But the structure passed to the arginfo function contains a value of INT_MIN, since the actual value is not known.
    int width
    This is the minimum field width specified. The value is 0 if no width was specified. If the field width was given as `*', the printf_info structure passed to the handler function contains the actual value retrieved from the argument list. But the structure passed to the arginfo function contains a value of INT_MIN, since the actual value is not known.
    wchar_t spec
    This is the conversion specifier character specified. It's stored in the structure so that you can register the same handler function for multiple characters, but still have a way to tell them apart when the handler function is called.
    unsigned int is_long_double
    This is a boolean that is true if the `L', `ll', or `q' type modifier was specified. For integer conversions, this indicates long long int, as opposed to long double for floating point conversions.
    unsigned int is_char
    This is a boolean that is true if the `hh' type modifier was specified.
    unsigned int is_short
    This is a boolean that is true if the `h' type modifier was specified.
    unsigned int is_long
    This is a boolean that is true if the `l' type modifier was specified.
    unsigned int alt
    This is a boolean that is true if the `#' flag was specified.
    unsigned int space
    This is a boolean that is true if the ` ' flag was specified.
    unsigned int left
    This is a boolean that is true if the `-' flag was specified.
    unsigned int showsign
    This is a boolean that is true if the `+' flag was specified.
    unsigned int group
    This is a boolean that is true if the `'' flag was specified.
    unsigned int extra
    This flag has a special meaning depending on the context. It could be used freely by the user-defined handlers but when called from the printf function this variable always contains the value 0.
    unsigned int wide
    This flag is set if the stream is wide oriented.
    wchar_t pad
    This is the character to use for padding the output to the minimum field width. The value is '0' if the `0' flag was specified, and ' ' otherwise.

    Defining the Output Handler

    Now let's look at how to define the handler and arginfo functions which are passed as arguments to register_printf_function.

    Compatibility Note: The interface changed in GNU libc version 2.0. Previously the third argument was of type va_list *.

    You should define your handler functions with a prototype like:

    int function (FILE *stream, const struct printf_info *info,
                        const void *const *args)
    

    The stream argument passed to the handler function is the stream to which it should write output.

    The info argument is a pointer to a structure that contains information about the various options that were included with the conversion in the template string. You should not modify this structure inside your handler function. See section Conversion Specifier Options, for a description of this data structure.

    The args is a vector of pointers to the arguments data. The number of arguments was determined by calling the argument information function provided by the user.

    Your handler function should return a value just like printf does: it should return the number of characters it has written, or a negative value to indicate an error.

    Data Type: printf_function
    This is the data type that a handler function should have.

    If you are going to use parse_printf_format in your application, you must also define a function to pass as the arginfo-function argument for each new conversion you install with register_printf_function.

    You have to define these functions with a prototype like:

    int function (const struct printf_info *info,
                        size_t n, int *argtypes)
    

    The return value from the function should be the number of arguments the conversion expects. The function should also fill in no more than n elements of the argtypes array with information about the types of each of these arguments. This information is encoded using the various `PA_' macros. (You will notice that this is the same calling convention parse_printf_format itself uses.)

    Data Type: printf_arginfo_function
    This type is used to describe functions that return information about the number and type of arguments used by a conversion specifier.

    printf Extension Example

    Here is an example showing how to define a printf handler function. This program defines a data structure called a Widget and defines the `%W' conversion to print information about Widget * arguments, including the pointer value and the name stored in the data structure. The `%W' conversion supports the minimum field width and left-justification options, but ignores everything else.

    #include <stdio.h>
    #include <stdlib.h>
    #include <printf.h>
    
    typedef struct
    {
      char *name;
    }
    Widget;
    
    int
    print_widget (FILE *stream,
                  const struct printf_info *info,
                  const void *const *args)
    {
      const Widget *w;
      char *buffer;
      int len;
    
      /* Format the output into a string. */
      w = *((const Widget **) (args[0]));
      len = asprintf (&buffer, "<Widget %p: %s>", w, w->name);
      if (len == -1)
        return -1;
    
      /* Pad to the minimum field width and print to the stream. */
      len = fprintf (stream, "%*s",
                     (info->left ? -info->width : info->width),
                     buffer);
    
      /* Clean up and return. */
      free (buffer);
      return len;
    }
    
    int
    print_widget_arginfo (const struct printf_info *info, size_t n,
                          int *argtypes)
    {
      /* We always take exactly one argument and this is a pointer to the
         structure.. */
      if (n > 0)
        argtypes[0] = PA_POINTER;
      return 1;
    }
    
    int
    main (void)
    {
      /* Make a widget to print. */
      Widget mywidget;
      mywidget.name = "mywidget";
    
      /* Register the print function for widgets. */
      register_printf_function ('W', print_widget, print_widget_arginfo);
    
      /* Now print the widget. */
      printf ("|%W|\n", &mywidget);
      printf ("|%35W|\n", &mywidget);
      printf ("|%-35W|\n", &mywidget);
    
      return 0;
    }
    

    The output produced by this program looks like:

    |<Widget 0xffeffb7c: mywidget>|
    |      <Widget 0xffeffb7c: mywidget>|
    |<Widget 0xffeffb7c: mywidget>      |
    

    Predefined printf Handlers

    The GNU libc also contains a concrete and useful application of the printf handler extension. There are two functions available which implement a special way to print floating-point numbers.

    Function: int printf_size (FILE *fp, const struct printf_info *info, const void *const *args)
    Print a given floating point number as for the format %f except that there is a postfix character indicating the divisor for the number to make this less than 1000. There are two possible divisors: powers of 1024 or powers of 1000. Which one is used depends on the format character specified while registered this handler. If the character is of lower case, 1024 is used. For upper case characters, 1000 is used.

    The postfix tag corresponds to bytes, kilobytes, megabytes, gigabytes, etc. The full table is:

    The default precision is 3, i.e., 1024 is printed with a lower-case format character as if it were %.3fk and will yield 1.000k.

    Due to the requirements of register_printf_function we must also provide the function which returns information about the arguments.

    Function: int printf_size_info (const struct printf_info *info, size_t n, int *argtypes)
    This function will return in argtypes the information about the used parameters in the way the vfprintf implementation expects it. The format always takes one argument.

    To use these functions both functions must be registered with a call like

    register_printf_function ('B', printf_size, printf_size_info);
    

    Here we register the functions to print numbers as powers of 1000 since the format character 'B' is an upper-case character. If we would additionally use 'b' in a line like

    register_printf_function ('b', printf_size, printf_size_info);
    

    we could also print using a power of 1024. Please note that all that is different in these two lines is the format specifier. The printf_size function knows about the difference between lower and upper case format specifiers.

    The use of 'B' and 'b' is no coincidence. Rather it is the preferred way to use this functionality since it is available on some other systems which also use format specifiers.

    Formatted Input

    The functions described in this section (scanf and related functions) provide facilities for formatted input analogous to the formatted output facilities. These functions provide a mechanism for reading arbitrary values under the control of a format string or template string.

    Formatted Input Basics

    Calls to scanf are superficially similar to calls to printf in that arbitrary arguments are read under the control of a template string. While the syntax of the conversion specifications in the template is very similar to that for printf, the interpretation of the template is oriented more towards free-format input and simple pattern matching, rather than fixed-field formatting. For example, most scanf conversions skip over any amount of "white space" (including spaces, tabs, and newlines) in the input file, and there is no concept of precision for the numeric input conversions as there is for the corresponding output conversions. Ordinarily, non-whitespace characters in the template are expected to match characters in the input stream exactly, but a matching failure is distinct from an input error on the stream.

    Another area of difference between scanf and printf is that you must remember to supply pointers rather than immediate values as the optional arguments to scanf; the values that are read are stored in the objects that the pointers point to. Even experienced programmers tend to forget this occasionally, so if your program is getting strange errors that seem to be related to scanf, you might want to double-check this.

    When a matching failure occurs, scanf returns immediately, leaving the first non-matching character as the next character to be read from the stream. The normal return value from scanf is the number of values that were assigned, so you can use this to determine if a matching error happened before all the expected values were read.

    The scanf function is typically used for things like reading in the contents of tables. For example, here is a function that uses scanf to initialize an array of double:

    void
    readarray (double *array, int n)
    {
      int i;
      for (i=0; i<n; i++)
        if (scanf (" %lf", &(array[i])) != 1)
          invalid_input_error ();
    }
    

    The formatted input functions are not used as frequently as the formatted output functions. Partly, this is because it takes some care to use them properly. Another reason is that it is difficult to recover from a matching error.

    If you are trying to read input that doesn't match a single, fixed pattern, you may be better off using a tool such as Flex to generate a lexical scanner, or Bison to generate a parser, rather than using scanf. For more information about these tools, see section `' in Flex: The Lexical Scanner Generator, and section `' in The Bison Reference Manual.

    Input Conversion Syntax

    A scanf template string is a string that contains ordinary multibyte characters interspersed with conversion specifications that start with `%'.

    Any whitespace character (as defined by the isspace function; see section Classification of Characters) in the template causes any number of whitespace characters in the input stream to be read and discarded. The whitespace characters that are matched need not be exactly the same whitespace characters that appear in the template string. For example, write ` , ' in the template to recognize a comma with optional whitespace before and after.

    Other characters in the template string that are not part of conversion specifications must match characters in the input stream exactly; if this is not the case, a matching failure occurs.

    The conversion specifications in a scanf template string have the general form:

    % flags width type conversion
    

    In more detail, an input conversion specification consists of an initial `%' character followed in sequence by:

    • An optional flag character `*', which says to ignore the text read for this specification. When scanf finds a conversion specification that uses this flag, it reads input as directed by the rest of the conversion specification, but it discards this input, does not use a pointer argument, and does not increment the count of successful assignments.
    • An optional flag character `a' (valid with string conversions only) which requests allocation of a buffer long enough to store the string in. (This is a GNU extension.) See section Dynamically Allocating String Conversions.
    • An optional decimal integer that specifies the maximum field width. Reading of characters from the input stream stops either when this maximum is reached or when a non-matching character is found, whichever happens first. Most conversions discard initial whitespace characters (those that don't are explicitly documented), and these discarded characters don't count towards the maximum field width. String input conversions store a null character to mark the end of the input; the maximum field width does not include this terminator.
    • An optional type modifier character. For example, you can specify a type modifier of `l' with integer conversions such as `%d' to specify that the argument is a pointer to a long int rather than a pointer to an int.
    • A character that specifies the conversion to be applied.

    The exact options that are permitted and how they are interpreted vary between the different conversion specifiers. See the descriptions of the individual conversions for information about the particular options that they allow.

    With the `-Wformat' option, the GNU C compiler checks calls to scanf and related functions. It examines the format string and verifies that the correct number and types of arguments are supplied. There is also a GNU C syntax to tell the compiler that a function you write uses a scanf-style format string. See section `Declaring Attributes of Functions' in Using GNU CC, for more information.

    Table of Input Conversions

    Here is a table that summarizes the various conversion specifications:

    `%d'
    Matches an optionally signed integer written in decimal. See section Numeric Input Conversions.
    `%i'
    Matches an optionally signed integer in any of the formats that the C language defines for specifying an integer constant. See section Numeric Input Conversions.
    `%o'
    Matches an unsigned integer written in octal radix. See section Numeric Input Conversions.
    `%u'
    Matches an unsigned integer written in decimal radix. See section Numeric Input Conversions.
    `%x', `%X'
    Matches an unsigned integer written in hexadecimal radix. See section Numeric Input Conversions.
    `%e', `%f', `%g', `%E', `%G'
    Matches an optionally signed floating-point number. See section Numeric Input Conversions.
    `%s'
    Matches a string containing only non-whitespace characters. See section String Input Conversions. The presence of the `l' modifier determines whether the output is stored as a wide character string or a multibyte string. If `%s' is used in a wide character function the string is converted as with multiple calls to wcrtomb into a multibyte string. This means that the buffer must provide room for MB_CUR_MAX bytes for each wide character read. In case `%ls' is used in a multibyte function the result is converted into wide characters as with multiple calls of mbrtowc before being stored in the user provided buffer.
    `%S'
    This is an alias for `%ls' which is supported for compatibility with the Unix standard.
    `%['
    Matches a string of characters that belong to a specified set. See section String Input Conversions. The presence of the `l' modifier determines whether the output is stored as a wide character string or a multibyte string. If `%[' is used in a wide character function the string is converted as with multiple calls to wcrtomb into a multibyte string. This means that the buffer must provide room for MB_CUR_MAX bytes for each wide character read. In case `%l[' is used in a multibyte function the result is converted into wide characters as with multiple calls of mbrtowc before being stored in the user provided buffer.
    `%c'
    Matches a string of one or more characters; the number of characters read is controlled by the maximum field width given for the conversion. See section String Input Conversions. If the `%c' is used in a wide stream function the read value is converted from a wide character to the corresponding multibyte character before storing it. Note that this conversion can produce more than one byte of output and therefore the provided buffer be large enough for up to MB_CUR_MAX bytes for each character. If `%lc' is used in a multibyte function the input is treated as a multibyte sequence (and not bytes) and the result is converted as with calls to mbrtowc.
    `%C'
    This is an alias for `%lc' which is supported for compatibility with the Unix standard.
    `%p'
    Matches a pointer value in the same implementation-defined format used by the `%p' output conversion for printf. See section Other Input Conversions.
    `%n'
    This conversion doesn't read any characters; it records the number of characters read so far by this call. See section Other Input Conversions.
    `%%'
    This matches a literal `%' character in the input stream. No corresponding argument is used. See section Other Input Conversions.

    If the syntax of a conversion specification is invalid, the behavior is undefined. If there aren't enough function arguments provided to supply addresses for all the conversion specifications in the template strings that perform assignments, or if the arguments are not of the correct types, the behavior is also undefined. On the other hand, extra arguments are simply ignored.

    Numeric Input Conversions

    This section describes the scanf conversions for reading numeric values.

    The `%d' conversion matches an optionally signed integer in decimal radix. The syntax that is recognized is the same as that for the strtol function (see section Parsing of Integers) with the value 10 for the base argument.

    The `%i' conversion matches an optionally signed integer in any of the formats that the C language defines for specifying an integer constant. The syntax that is recognized is the same as that for the strtol function (see section Parsing of Integers) with the value 0 for the base argument. (You can print integers in this syntax with printf by using the `#' flag character with the `%x', `%o', or `%d' conversion. See section Integer Conversions.)

    For example, any of the strings `10', `0xa', or `012' could be read in as integers under the `%i' conversion. Each of these specifies a number with decimal value 10.

    The `%o', `%u', and `%x' conversions match unsigned integers in octal, decimal, and hexadecimal radices, respectively. The syntax that is recognized is the same as that for the strtoul function (see section Parsing of Integers) with the appropriate value (8, 10, or 16) for the base argument.

    The `%X' conversion is identical to the `%x' conversion. They both permit either uppercase or lowercase letters to be used as digits.

    The default type of the corresponding argument for the %d and %i conversions is int *, and unsigned int * for the other integer conversions. You can use the following type modifiers to specify other sizes of integer:

    `hh'
    Specifies that the argument is a signed char * or unsigned char *. This modifier was introduced in ISO C99.
    `h'
    Specifies that the argument is a short int * or unsigned short int *.
    `j'
    Specifies that the argument is a intmax_t * or uintmax_t *. This modifier was introduced in ISO C99.
    `l'
    Specifies that the argument is a long int * or unsigned long int *. Two `l' characters is like the `L' modifier, below. If used with `%c' or `%s' the corresponding parameter is considered as a pointer to a wide character or wide character string respectively. This use of `l' was introduced in Amendment 1 to ISO C90.
    `ll'
    `L'
    `q'
    Specifies that the argument is a long long int * or unsigned long long int *. (The long long type is an extension supported by the GNU C compiler. For systems that don't provide extra-long integers, this is the same as long int.) The `q' modifier is another name for the same thing, which comes from 4.4 BSD; a long long int is sometimes called a "quad" int.
    `t'
    Specifies that the argument is a ptrdiff_t *. This modifier was introduced in ISO C99.
    `z'
    Specifies that the argument is a size_t *. This modifier was introduced in ISO C99.

    All of the `%e', `%f', `%g', `%E', and `%G' input conversions are interchangeable. They all match an optionally signed floating point number, in the same syntax as for the strtod function (see section Parsing of Floats).

    For the floating-point input conversions, the default argument type is float *. (This is different from the corresponding output conversions, where the default type is double; remember that float arguments to printf are converted to double by the default argument promotions, but float * arguments are not promoted to double *.) You can specify other sizes of float using these type modifiers:

    `l'
    Specifies that the argument is of type double *.
    `L'
    Specifies that the argument is of type long double *.

    For all the above number parsing formats there is an additional optional flag `''. When this flag is given the scanf function expects the number represented in the input string to be formatted according to the grouping rules of the currently selected locale (see section Generic Numeric Formatting Parameters).

    If the "C" or "POSIX" locale is selected there is no difference. But for a locale which specifies values for the appropriate fields in the locale the input must have the correct form in the input. Otherwise the longest prefix with a correct form is processed.

    String Input Conversions

    This section describes the scanf input conversions for reading string and character values: `%s', `%S', `%[', `%c', and `%C'.

    You have two options for how to receive the input from these conversions:

    • Provide a buffer to store it in. This is the default. You should provide an argument of type char * or wchar_t * (the latter of the `l' modifier is present). Warning: To make a robust program, you must make sure that the input (plus its terminating null) cannot possibly exceed the size of the buffer you provide. In general, the only way to do this is to specify a maximum field width one less than the buffer size. If you provide the buffer, always specify a maximum field width to prevent overflow.
    • Ask scanf to allocate a big enough buffer, by specifying the `a' flag character. This is a GNU extension. You should provide an argument of type char ** for the buffer address to be stored in. See section Dynamically Allocating String Conversions.

    The `%c' conversion is the simplest: it matches a fixed number of characters, always. The maximum field width says how many characters to read; if you don't specify the maximum, the default is 1. This conversion doesn't append a null character to the end of the text it reads. It also does not skip over initial whitespace characters. It reads precisely the next n characters, and fails if it cannot get that many. Since there is always a maximum field width with `%c' (whether specified, or 1 by default), you can always prevent overflow by making the buffer long enough.

    If the format is `%lc' or `%C' the function stores wide characters which are converted using the conversion determined at the time the stream was opened from the external byte stream. The number of bytes read from the medium is limited by MB_CUR_LEN * n but at most n wide character get stored in the output string.

    The `%s' conversion matches a string of non-whitespace characters. It skips and discards initial whitespace, but stops when it encounters more whitespace after having read something. It stores a null character at the end of the text that it reads.

    For example, reading the input:

     hello, world
    

    with the conversion `%10c' produces " hello, wo", but reading the same input with the conversion `%10s' produces "hello,".

    Warning: If you do not specify a field width for `%s', then the number of characters read is limited only by where the next whitespace character appears. This almost certainly means that invalid input can make your program crash--which is a bug.

    The `%ls' and `%S' format are handled just like `%s' except that the external byte sequence is converted using the conversion associated with the stream to wide characters with their own encoding. A width or precision specified with the format do not directly determine how many bytes are read from the stream since they measure wide characters. But an upper limit can be computed by multiplying the value of the width or precision by MB_CUR_MAX.

    To read in characters that belong to an arbitrary set of your choice, use the `%[' conversion. You specify the set between the `[' character and a following `]' character, using the same syntax used in regular expressions. As special cases:

    • A literal `]' character can be specified as the first character of the set.
    • An embedded `-' character (that is, one that is not the first or last character of the set) is used to specify a range of characters.
    • If a caret character `^' immediately follows the initial `[', then the set of allowed input characters is the everything except the characters listed.

    The `%[' conversion does not skip over initial whitespace characters.

    Here are some examples of `%[' conversions and what they mean:

    `%25[1234567890]'
    Matches a string of up to 25 digits.
    `%25[][]'
    Matches a string of up to 25 square brackets.
    `%25[^ \f\n\r\t\v]'
    Matches a string up to 25 characters long that doesn't contain any of the standard whitespace characters. This is slightly different from `%s', because if the input begins with a whitespace character, `%[' reports a matching failure while `%s' simply discards the initial whitespace.
    `%25[a-z]'
    Matches up to 25 lowercase characters.

    As for `%c' and `%s' the `%[' format is also modified to produce wide characters if the `l' modifier is present. All what is said about `%ls' above is true for `%l['.

    One more reminder: the `%s' and `%[' conversions are dangerous if you don't specify a maximum width or use the `a' flag, because input too long would overflow whatever buffer you have provided for it. No matter how long your buffer is, a user could supply input that is longer. A well-written program reports invalid input with a comprehensible error message, not with a crash.

    Dynamically Allocating String Conversions

    A GNU extension to formatted input lets you safely read a string with no maximum size. Using this feature, you don't supply a buffer; instead, scanf allocates a buffer big enough to hold the data and gives you its address. To use this feature, write `a' as a flag character, as in `%as' or `%a[0-9a-z]'.

    The pointer argument you supply for where to store the input should have type char **. The scanf function allocates a buffer and stores its address in the word that the argument points to. You should free the buffer with free when you no longer need it.

    Here is an example of using the `a' flag with the `%[...]' conversion specification to read a "variable assignment" of the form `variable = value'.

    {
      char *variable, *value;
    
      if (2 > scanf ("%a[a-zA-Z0-9] = %a[^\n]\n",
                     &variable, &value))
        {
          invalid_input_error ();
          return 0;
        }
    
      ...
    }
    

    Other Input Conversions

    This section describes the miscellaneous input conversions.

    The `%p' conversion is used to read a pointer value. It recognizes the same syntax used by the `%p' output conversion for printf (see section Other Output Conversions); that is, a hexadecimal number just as the `%x' conversion accepts. The corresponding argument should be of type void **; that is, the address of a place to store a pointer.

    The resulting pointer value is not guaranteed to be valid if it was not originally written during the same program execution that reads it in.

    The `%n' conversion produces the number of characters read so far by this call. The corresponding argument should be of type int *. This conversion works in the same way as the `%n' conversion for printf; see section Other Output Conversions, for an example.

    The `%n' conversion is the only mechanism for determining the success of literal matches or conversions with suppressed assignments. If the `%n' follows the locus of a matching failure, then no value is stored for it since scanf returns before processing the `%n'. If you store -1 in that argument slot before calling scanf, the presence of -1 after scanf indicates an error occurred before the `%n' was reached.

    Finally, the `%%' conversion matches a literal `%' character in the input stream, without using an argument. This conversion does not permit any flags, field width, or type modifier to be specified.

    Formatted Input Functions

    Here are the descriptions of the functions for performing formatted input. Prototypes for these functions are in the header file `stdio.h'.

    Function: int scanf (const char *template, ...)
    The scanf function reads formatted input from the stream stdin under the control of the template string template. The optional arguments are pointers to the places which receive the resulting values.

    The return value is normally the number of successful assignments. If an end-of-file condition is detected before any matches are performed, including matches against whitespace and literal characters in the template, then EOF is returned.

    Function: int wscanf (const wchar_t *template, ...)
    The wscanf function reads formatted input from the stream stdin under the control of the template string template. The optional arguments are pointers to the places which receive the resulting values.

    The return value is normally the number of successful assignments. If an end-of-file condition is detected before any matches are performed, including matches against whitespace and literal characters in the template, then WEOF is returned.

    Function: int fscanf (FILE *stream, const char *template, ...)
    This function is just like scanf, except that the input is read from the stream stream instead of stdin.

    Function: int fwscanf (FILE *stream, const wchar_t *template, ...)
    This function is just like wscanf, except that the input is read from the stream stream instead of stdin.

    Function: int sscanf (const char *s, const char *template, ...)
    This is like scanf, except that the characters are taken from the null-terminated string s instead of from a stream. Reaching the end of the string is treated as an end-of-file condition.

    The behavior of this function is undefined if copying takes place between objects that overlap--for example, if s is also given as an argument to receive a string read under control of the `%s', `%S', or `%[' conversion.

    Function: int swscanf (const wchar_t *ws, const char *template, ...)
    This is like wscanf, except that the characters are taken from the null-terminated string ws instead of from a stream. Reaching the end of the string is treated as an end-of-file condition.

    The behavior of this function is undefined if copying takes place between objects that overlap--for example, if ws is also given as an argument to receive a string read under control of the `%s', `%S', or `%[' conversion.

    Variable Arguments Input Functions

    The functions vscanf and friends are provided so that you can define your own variadic scanf-like functions that make use of the same internals as the built-in formatted output functions. These functions are analogous to the vprintf series of output functions. See section Variable Arguments Output Functions, for important information on how to use them.

    Portability Note: The functions listed in this section were introduced in ISO C99 and were before available as GNU extensions.

    Function: int vscanf (const char *template, va_list ap)
    This function is similar to scanf, but instead of taking a variable number of arguments directly, it takes an argument list pointer ap of type va_list (see section Variadic Functions).

    Function: int vwscanf (const wchar_t *template, va_list ap)
    This function is similar to wscanf, but instead of taking a variable number of arguments directly, it takes an argument list pointer ap of type va_list (see section Variadic Functions).

    Function: int vfscanf (FILE *stream, const char *template, va_list ap)
    This is the equivalent of fscanf with the variable argument list specified directly as for vscanf.

    Function: int vfwscanf (FILE *stream, const wchar_t *template, va_list ap)
    This is the equivalent of fwscanf with the variable argument list specified directly as for vwscanf.

    Function: int vsscanf (const char *s, const char *template, va_list ap)
    This is the equivalent of sscanf with the variable argument list specified directly as for vscanf.

    Function: int vswscanf (const wchar_t *s, const wchar_t *template, va_list ap)
    This is the equivalent of swscanf with the variable argument list specified directly as for vwscanf.

    In GNU C, there is a special construct you can use to let the compiler know that a function uses a scanf-style format string. Then it can check the number and types of arguments in each call to the function, and warn you when they do not match the format string. For details, See section `Declaring Attributes of Functions' in Using GNU CC.

    End-Of-File and Errors

    Many of the functions described in this chapter return the value of the macro EOF to indicate unsuccessful completion of the operation. Since EOF is used to report both end of file and random errors, it's often better to use the feof function to check explicitly for end of file and ferror to check for errors. These functions check indicators that are part of the internal state of the stream object, indicators set if the appropriate condition was detected by a previous I/O operation on that stream.

    Macro: int EOF
    This macro is an integer value that is returned by a number of narrow stream functions to indicate an end-of-file condition, or some other error situation. With the GNU library, EOF is -1. In other libraries, its value may be some other negative number.

    This symbol is declared in `stdio.h'.

    Macro: int WEOF
    This macro is an integer value that is returned by a number of wide stream functions to indicate an end-of-file condition, or some other error situation. With the GNU library, WEOF is -1. In other libraries, its value may be some other negative number.

    This symbol is declared in `wchar.h'.

    Function: int feof (FILE *stream)
    The feof function returns nonzero if and only if the end-of-file indicator for the stream stream is set.

    This symbol is declared in `stdio.h'.

    Function: int feof_unlocked (FILE *stream)
    The feof_unlocked function is equivalent to the feof function except that it does not implicitly lock the stream.

    This function is a GNU extension.

    This symbol is declared in `stdio.h'.

    Function: int ferror (FILE *stream)
    The ferror function returns nonzero if and only if the error indicator for the stream stream is set, indicating that an error has occurred on a previous operation on the stream.

    This symbol is declared in `stdio.h'.

    Function: int ferror_unlocked (FILE *stream)
    The ferror_unlocked function is equivalent to the ferror function except that it does not implicitly lock the stream.

    This function is a GNU extension.

    This symbol is declared in `stdio.h'.

    In addition to setting the error indicator associated with the stream, the functions that operate on streams also set errno in the same way as the corresponding low-level functions that operate on file descriptors. For example, all of the functions that perform output to a stream--such as fputc, printf, and fflush---are implemented in terms of write, and all of the errno error conditions defined for write are meaningful for these functions. For more information about the descriptor-level I/O functions, see section Low-Level Input/Output.

    Recovering from errors

    You may explicitly clear the error and EOF flags with the clearerr function.

    Function: void clearerr (FILE *stream)
    This function clears the end-of-file and error indicators for the stream stream.

    The file positioning functions (see section File Positioning) also clear the end-of-file indicator for the stream.

    Function: void clearerr_unlocked (FILE *stream)
    The clearerr_unlocked function is equivalent to the clearerr function except that it does not implicitly lock the stream.

    This function is a GNU extension.

    Note that it is not correct to just clear the error flag and retry a failed stream operation. After a failed write, any number of characters since the last buffer flush may have been committed to the file, while some buffered data may have been discarded. Merely retrying can thus cause lost or repeated data.

    A failed read may leave the file pointer in an inappropriate position for a second try. In both cases, you should seek to a known position before retrying.

    Most errors that can happen are not recoverable -- a second try will always fail again in the same way. So usually it is best to give up and report the error to the user, rather than install complicated recovery logic.

    One important exception is EINTR (see section Primitives Interrupted by Signals). Many stream I/O implementations will treat it as an ordinary error, which can be quite inconvenient. You can avoid this hassle by installing all signals with the SA_RESTART flag.

    For similar reasons, setting nonblocking I/O on a stream's file descriptor is not usually advisable.

    Text and Binary Streams

    The GNU system and other POSIX-compatible operating systems organize all files as uniform sequences of characters. However, some other systems make a distinction between files containing text and files containing binary data, and the input and output facilities of ISO C provide for this distinction. This section tells you how to write programs portable to such systems.

    When you open a stream, you can specify either a text stream or a binary stream. You indicate that you want a binary stream by specifying the `b' modifier in the opentype argument to fopen; see section Opening Streams. Without this option, fopen opens the file as a text stream.

    Text and binary streams differ in several ways:

    • The data read from a text stream is divided into lines which are terminated by newline ('\n') characters, while a binary stream is simply a long series of characters. A text stream might on some systems fail to handle lines more than 254 characters long (including the terminating newline character).
    • On some systems, text files can contain only printing characters, horizontal tab characters, and newlines, and so text streams may not support other characters. However, binary streams can handle any character value.
    • Space characters that are written immediately preceding a newline character in a text stream may disappear when the file is read in again.
    • More generally, there need not be a one-to-one mapping between characters that are read from or written to a text stream, and the characters in the actual file.

    Since a binary stream is always more capable and more predictable than a text stream, you might wonder what purpose text streams serve. Why not simply always use binary streams? The answer is that on these operating systems, text and binary streams use different file formats, and the only way to read or write "an ordinary file of text" that can work with other text-oriented programs is through a text stream.

    In the GNU library, and on all POSIX systems, there is no difference between text streams and binary streams. When you open a stream, you get the same kind of stream regardless of whether you ask for binary. This stream can handle any file content, and has none of the restrictions that text streams sometimes have.

    File Positioning

    The file position of a stream describes where in the file the stream is currently reading or writing. I/O on the stream advances the file position through the file. In the GNU system, the file position is represented as an integer, which counts the number of bytes from the beginning of the file. See section File Position.

    During I/O to an ordinary disk file, you can change the file position whenever you wish, so as to read or write any portion of the file. Some other kinds of files may also permit this. Files which support changing the file position are sometimes referred to as random-access files.

    You can use the functions in this section to examine or modify the file position indicator associated with a stream. The symbols listed below are declared in the header file `stdio.h'.

    Function: long int ftell (FILE *stream)
    This function returns the current file position of the stream stream.

    This function can fail if the stream doesn't support file positioning, or if the file position can't be represented in a long int, and possibly for other reasons as well. If a failure occurs, a value of -1 is returned.

    Function: off_t ftello (FILE *stream)
    The ftello function is similar to ftell, except that it returns a value of type off_t. Systems which support this type use it to describe all file positions, unlike the POSIX specification which uses a long int. The two are not necessarily the same size. Therefore, using ftell can lead to problems if the implementation is written on top of a POSIX compliant low-level I/O implementation, and using ftello is preferable whenever it is available.

    If this function fails it returns (off_t) -1. This can happen due to missing support for file positioning or internal errors. Otherwise the return value is the current file position.

    The function is an extension defined in the Unix Single Specification version 2.

    When the sources are compiled with _FILE_OFFSET_BITS == 64 on a 32 bit system this function is in fact ftello64. I.e., the LFS interface transparently replaces the old interface.

    Function: off64_t ftello64 (FILE *stream)
    This function is similar to ftello with the only difference that the return value is of type off64_t. This also requires that the stream stream was opened using either fopen64, freopen64, or tmpfile64 since otherwise the underlying file operations to position the file pointer beyond the @math{2^31} bytes limit might fail.

    If the sources are compiled with _FILE_OFFSET_BITS == 64 on a 32 bits machine this function is available under the name ftello and so transparently replaces the old interface.

    Function: int fseek (FILE *stream, long int offset, int whence)
    The fseek function is used to change the file position of the stream stream. The value of whence must be one of the constants SEEK_SET, SEEK_CUR, or SEEK_END, to indicate whether the offset is relative to the beginning of the file, the current file position, or the end of the file, respectively.

    This function returns a value of zero if the operation was successful, and a nonzero value to indicate failure. A successful call also clears the end-of-file indicator of stream and discards any characters that were "pushed back" by the use of ungetc.

    fseek either flushes any buffered output before setting the file position or else remembers it so it will be written later in its proper place in the file.

    Function: int fseeko (FILE *stream, off_t offset, int whence)
    This function is similar to fseek but it corrects a problem with fseek in a system with POSIX types. Using a value of type long int for the offset is not compatible with POSIX. fseeko uses the correct type off_t for the offset parameter.

    For this reason it is a good idea to prefer ftello whenever it is available since its functionality is (if different at all) closer the underlying definition.

    The functionality and return value is the same as for fseek.

    The function is an extension defined in the Unix Single Specification version 2.

    When the sources are compiled with _FILE_OFFSET_BITS == 64 on a 32 bit system this function is in fact fseeko64. I.e., the LFS interface transparently replaces the old interface.

    Function: int fseeko64 (FILE *stream, off64_t offset, int whence)
    This function is similar to fseeko with the only difference that the offset parameter is of type off64_t. This also requires that the stream stream was opened using either fopen64, freopen64, or tmpfile64 since otherwise the underlying file operations to position the file pointer beyond the @math{2^31} bytes limit might fail.

    If the sources are compiled with _FILE_OFFSET_BITS == 64 on a 32 bits machine this function is available under the name fseeko and so transparently replaces the old interface.

    Portability Note: In non-POSIX systems, ftell, ftello, fseek and fseeko might work reliably only on binary streams. See section Text and Binary Streams.

    The following symbolic constants are defined for use as the whence argument to fseek. They are also used with the lseek function (see section Input and Output Primitives) and to specify offsets for file locks (see section Control Operations on Files).

    Macro: int SEEK_SET
    This is an integer constant which, when used as the whence argument to the fseek or fseeko function, specifies that the offset provided is relative to the beginning of the file.

    Macro: int SEEK_CUR
    This is an integer constant which, when used as the whence argument to the fseek or fseeko function, specifies that the offset provided is relative to the current file position.

    Macro: int SEEK_END
    This is an integer constant which, when used as the whence argument to the fseek or fseeko function, specifies that the offset provided is relative to the end of the file.

    Function: void rewind (FILE *stream)
    The rewind function positions the stream stream at the beginning of the file. It is equivalent to calling fseek or fseeko on the stream with an offset argument of 0L and a whence argument of SEEK_SET, except that the return value is discarded and the error indicator for the stream is reset.

    These three aliases for the `SEEK_...' constants exist for the sake of compatibility with older BSD systems. They are defined in two different header files: `fcntl.h' and `sys/file.h'.

    L_SET
    An alias for SEEK_SET.
    L_INCR
    An alias for SEEK_CUR.
    L_XTND
    An alias for SEEK_END.

    Portable File-Position Functions

    On the GNU system, the file position is truly a character count. You can specify any character count value as an argument to fseek or fseeko and get reliable results for any random access file. However, some ISO C systems do not represent file positions in this way.

    On some systems where text streams truly differ from binary streams, it is impossible to represent the file position of a text stream as a count of characters from the beginning of the file. For example, the file position on some systems must encode both a record offset within the file, and a character offset within the record.

    As a consequence, if you want your programs to be portable to these systems, you must observe certain rules:

    • The value returned from ftell on a text stream has no predictable relationship to the number of characters you have read so far. The only thing you can rely on is that you can use it subsequently as the offset argument to fseek or fseeko to move back to the same file position.
    • In a call to fseek or fseeko on a text stream, either the offset must be zero, or whence must be SEEK_SET and and the offset must be the result of an earlier call to ftell on the same stream.
    • The value of the file position indicator of a text stream is undefined while there are characters that have been pushed back with ungetc that haven't been read or discarded. See section Unreading.

    But even if you observe these rules, you may still have trouble for long files, because ftell and fseek use a long int value to represent the file position. This type may not have room to encode all the file positions in a large file. Using the ftello and fseeko functions might help here since the off_t type is expected to be able to hold all file position values but this still does not help to handle additional information which must be associated with a file position.

    So if you do want to support systems with peculiar encodings for the file positions, it is better to use the functions fgetpos and fsetpos instead. These functions represent the file position using the data type fpos_t, whose internal representation varies from system to system.

    These symbols are declared in the header file `stdio.h'.

    Data Type: fpos_t
    This is the type of an object that can encode information about the file position of a stream, for use by the functions fgetpos and fsetpos.

    In the GNU system, fpos_t is equivalent to off_t or long int. In other systems, it might have a different internal representation.

    When compiling with _FILE_OFFSET_BITS == 64 on a 32 bit machine this type is in fact equivalent to off64_t since the LFS interface transparently replaced the old interface.

    Data Type: fpos64_t
    This is the type of an object that can encode information about the file position of a stream, for use by the functions fgetpos64 and fsetpos64.

    In the GNU system, fpos64_t is equivalent to off64_t or long long int. In other systems, it might have a different internal representation.

    Function: int fgetpos (FILE *stream, fpos_t *position)
    This function stores the value of the file position indicator for the stream stream in the fpos_t object pointed to by position. If successful, fgetpos returns zero; otherwise it returns a nonzero value and stores an implementation-defined positive value in errno.

    When the sources are compiled with _FILE_OFFSET_BITS == 64 on a 32 bit system the function is in fact fgetpos64. I.e., the LFS interface transparently replaced the old interface.

    Function: int fgetpos64 (FILE *stream, fpos64_t *position)
    This function is similar to fgetpos but the file position is returned in a variable of type fpos64_t to which position points.

    If the sources are compiled with _FILE_OFFSET_BITS == 64 on a 32 bits machine this function is available under the name fgetpos and so transparently replaces the old interface.

    Function: int fsetpos (FILE *stream, const fpos_t *position)
    This function sets the file position indicator for the stream stream to the position position, which must have been set by a previous call to fgetpos on the same stream. If successful, fsetpos clears the end-of-file indicator on the stream, discards any characters that were "pushed back" by the use of ungetc, and returns a value of zero. Otherwise, fsetpos returns a nonzero value and stores an implementation-defined positive value in errno.

    When the sources are compiled with _FILE_OFFSET_BITS == 64 on a 32 bit system the function is in fact fsetpos64. I.e., the LFS interface transparently replaced the old interface.

    Function: int fsetpos64 (FILE *stream, const fpos64_t *position)
    This function is similar to fsetpos but the file position used for positioning is provided in a variable of type fpos64_t to which position points.

    If the sources are compiled with _FILE_OFFSET_BITS == 64 on a 32 bits machine this function is available under the name fsetpos and so transparently replaces the old interface.

    Stream Buffering

    Characters that are written to a stream are normally accumulated and transmitted asynchronously to the file in a block, instead of appearing as soon as they are output by the application program. Similarly, streams often retrieve input from the host environment in blocks rather than on a character-by-character basis. This is called buffering.

    If you are writing programs that do interactive input and output using streams, you need to understand how buffering works when you design the user interface to your program. Otherwise, you might find that output (such as progress or prompt messages) doesn't appear when you intended it to, or displays some other unexpected behavior.

    This section deals only with controlling when characters are transmitted between the stream and the file or device, and not with how things like echoing, flow control, and the like are handled on specific classes of devices. For information on common control operations on terminal devices, see section Low-Level Terminal Interface.

    You can bypass the stream buffering facilities altogether by using the low-level input and output functions that operate on file descriptors instead. See section Low-Level Input/Output.

    Buffering Concepts

    There are three different kinds of buffering strategies:

    • Characters written to or read from an unbuffered stream are transmitted individually to or from the file as soon as possible.
    • Characters written to a line buffered stream are transmitted to the file in blocks when a newline character is encountered.
    • Characters written to or read from a fully buffered stream are transmitted to or from the file in blocks of arbitrary size.

    Newly opened streams are normally fully buffered, with one exception: a stream connected to an interactive device such as a terminal is initially line buffered. See section Controlling Which Kind of Buffering, for information on how to select a different kind of buffering. Usually the automatic selection gives you the most convenient kind of buffering for the file or device you open.

    The use of line buffering for interactive devices implies that output messages ending in a newline will appear immediately--which is usually what you want. Output that doesn't end in a newline might or might not show up immediately, so if you want them to appear immediately, you should flush buffered output explicitly with fflush, as described in section Flushing Buffers.

    Flushing Buffers

    Flushing output on a buffered stream means transmitting all accumulated characters to the file. There are many circumstances when buffered output on a stream is flushed automatically:

    • When you try to do output and the output buffer is full.
    • When the stream is closed. See section Closing Streams.
    • When the program terminates by calling exit. See section Normal Termination.
    • When a newline is written, if the stream is line buffered.
    • Whenever an input operation on any stream actually reads data from its file.

    If you want to flush the buffered output at another time, call fflush, which is declared in the header file `stdio.h'.

    Function: int fflush (FILE *stream)
    This function causes any buffered output on stream to be delivered to the file. If stream is a null pointer, then fflush causes buffered output on all open output streams to be flushed.

    This function returns EOF if a write error occurs, or zero otherwise.

    Function: int fflush_unlocked (FILE *stream)
    The fflush_unlocked function is equivalent to the fflush function except that it does not implicitly lock the stream.

    The fflush function can be used to flush all streams currently opened. While this is useful in some situations it does often more than necessary since it might be done in situations when terminal input is required and the program wants to be sure that all output is visible on the terminal. But this means that only line buffered streams have to be flushed. Solaris introduced a function especially for this. It was always available in the GNU C library in some form but never officially exported.

    Function: void _flushlbf (void)
    The _flushlbf function flushes all line buffered streams currently opened.

    This function is declared in the `stdio_ext.h' header.

    Compatibility Note: Some brain-damaged operating systems have been known to be so thoroughly fixated on line-oriented input and output that flushing a line buffered stream causes a newline to be written! Fortunately, this "feature" seems to be becoming less common. You do not need to worry about this in the GNU system.

    In some situations it might be useful to not flush the output pending for a stream but instead simply forget it. If transmission is costly and the output is not needed anymore this is valid reasoning. In this situation a non-standard function introduced in Solaris and available in the GNU C library can be used.

    Function: void __fpurge (FILE *stream)
    The __fpurge function causes the buffer of the stream stream to be emptied. If the stream is currently in read mode all input in the buffer is lost. If the stream is in output mode the buffered output is not written to the device (or whatever other underlying storage) and the buffer the cleared.

    This function is declared in `stdio_ext.h'.

    Controlling Which Kind of Buffering

    After opening a stream (but before any other operations have been performed on it), you can explicitly specify what kind of buffering you want it to have using the setvbuf function.

    The facilities listed in this section are declared in the header file `stdio.h'.

    Function: int setvbuf (FILE *stream, char *buf, int mode, size_t size)
    This function is used to specify that the stream stream should have the buffering mode mode, which can be either _IOFBF (for full buffering), _IOLBF (for line buffering), or _IONBF (for unbuffered input/output).

    If you specify a null pointer as the buf argument, then setvbuf allocates a buffer itself using malloc. This buffer will be freed when you close the stream.

    Otherwise, buf should be a character array that can hold at least size characters. You should not free the space for this array as long as the stream remains open and this array remains its buffer. You should usually either allocate it statically, or malloc (see section Unconstrained Allocation) the buffer. Using an automatic array is not a good idea unless you close the file before exiting the block that declares the array.

    While the array remains a stream buffer, the stream I/O functions will use the buffer for their internal purposes. You shouldn't try to access the values in the array directly while the stream is using it for buffering.

    The setvbuf function returns zero on success, or a nonzero value if the value of mode is not valid or if the request could not be honored.

    Macro: int _IOFBF
    The value of this macro is an integer constant expression that can be used as the mode argument to the setvbuf function to specify that the stream should be fully buffered.

    Macro: int _IOLBF
    The value of this macro is an integer constant expression that can be used as the mode argument to the setvbuf function to specify that the stream should be line buffered.

    Macro: int _IONBF
    The value of this macro is an integer constant expression that can be used as the mode argument to the setvbuf function to specify that the stream should be unbuffered.

    Macro: int BUFSIZ
    The value of this macro is an integer constant expression that is good to use for the size argument to setvbuf. This value is guaranteed to be at least 256.

    The value of BUFSIZ is chosen on each system so as to make stream I/O efficient. So it is a good idea to use BUFSIZ as the size for the buffer when you call setvbuf.

    Actually, you can get an even better value to use for the buffer size by means of the fstat system call: it is found in the st_blksize field of the file attributes. See section The meaning of the File Attributes.

    Sometimes people also use BUFSIZ as the allocation size of buffers used for related purposes, such as strings used to receive a line of input with fgets (see section Character Input). There is no particular reason to use BUFSIZ for this instead of any other integer, except that it might lead to doing I/O in chunks of an efficient size.

    Function: void setbuf (FILE *stream, char *buf)
    If buf is a null pointer, the effect of this function is equivalent to calling setvbuf with a mode argument of _IONBF. Otherwise, it is equivalent to calling setvbuf with buf, and a mode of _IOFBF and a size argument of BUFSIZ.

    The setbuf function is provided for compatibility with old code; use setvbuf in all new programs.

    Function: void setbuffer (FILE *stream, char *buf, size_t size)
    If buf is a null pointer, this function makes stream unbuffered. Otherwise, it makes stream fully buffered using buf as the buffer. The size argument specifies the length of buf.

    This function is provided for compatibility with old BSD code. Use setvbuf instead.

    Function: void setlinebuf (FILE *stream)
    This function makes stream be line buffered, and allocates the buffer for you.

    This function is provided for compatibility with old BSD code. Use setvbuf instead.

    It is possible to query whether a given stream is line buffered or not using a non-standard function introduced in Solaris and available in the GNU C library.

    Function: int __flbf (FILE *stream)
    The __flbf function will return a nonzero value in case the stream stream is line buffered. Otherwise the return value is zero.

    This function is declared in the `stdio_ext.h' header.

    Two more extensions allow to determine the size of the buffer and how much of it is used. These functions were also introduced in Solaris.

    Function: size_t __fbufsize (FILE *stream)
    The __fbufsize function return the size of the buffer in the stream stream. This value can be used to optimize the use of the stream.

    This function is declared in the `stdio_ext.h' header.

    Function: size_t __fpending (FILE *stream) The __fpending
    function returns the number of bytes currently in the output buffer. For wide-oriented stream the measuring unit is wide characters. This function should not be used on buffers in read mode or opened read-only.

    This function is declared in the `stdio_ext.h' header.

    Other Kinds of Streams

    The GNU library provides ways for you to define additional kinds of streams that do not necessarily correspond to an open file.

    One such type of stream takes input from or writes output to a string. These kinds of streams are used internally to implement the sprintf and sscanf functions. You can also create such a stream explicitly, using the functions described in section String Streams.

    More generally, you can define streams that do input/output to arbitrary objects using functions supplied by your program. This protocol is discussed in section Programming Your Own Custom Streams.

    Portability Note: The facilities described in this section are specific to GNU. Other systems or C implementations might or might not provide equivalent functionality.

    String Streams

    The fmemopen and open_memstream functions allow you to do I/O to a string or memory buffer. These facilities are declared in `stdio.h'.

    Function: FILE * fmemopen (void *buf, size_t size, const char *opentype)
    This function opens a stream that allows the access specified by the opentype argument, that reads from or writes to the buffer specified by the argument buf. This array must be at least size bytes long.

    If you specify a null pointer as the buf argument, fmemopen dynamically allocates an array size bytes long (as with malloc; see section Unconstrained Allocation). This is really only useful if you are going to write things to the buffer and then read them back in again, because you have no way of actually getting a pointer to the buffer (for this, try open_memstream, below). The buffer is freed when the stream is open.

    The argument opentype is the same as in fopen (see section Opening Streams). If the opentype specifies append mode, then the initial file position is set to the first null character in the buffer. Otherwise the initial file position is at the beginning of the buffer.

    When a stream open for writing is flushed or closed, a null character (zero byte) is written at the end of the buffer if it fits. You should add an extra byte to the size argument to account for this. Attempts to write more than size bytes to the buffer result in an error.

    For a stream open for reading, null characters (zero bytes) in the buffer do not count as "end of file". Read operations indicate end of file only when the file position advances past size bytes. So, if you want to read characters from a null-terminated string, you should supply the length of the string as the size argument.

    Here is an example of using fmemopen to create a stream for reading from a string:

    #include <stdio.h>
    
    static char buffer[] = "foobar";
    
    int
    main (void)
    {
      int ch;
      FILE *stream;
    
      stream = fmemopen (buffer, strlen (buffer), "r");
      while ((ch = fgetc (stream)) != EOF)
        printf ("Got %c\n", ch);
      fclose (stream);
    
      return 0;
    }
    

    This program produces the following output:

    Got f
    Got o
    Got o
    Got b
    Got a
    Got r
    

    Function: FILE * open_memstream (char **ptr, size_t *sizeloc)
    This function opens a stream for writing to a buffer. The buffer is allocated dynamically (as with malloc; see section Unconstrained Allocation) and grown as necessary.

    When the stream is closed with fclose or flushed with fflush, the locations ptr and sizeloc are updated to contain the pointer to the buffer and its size. The values thus stored remain valid only as long as no further output on the stream takes place. If you do more output, you must flush the stream again to store new values before you use them again.

    A null character is written at the end of the buffer. This null character is not included in the size value stored at sizeloc.

    You can move the stream's file position with fseek or fseeko (see section File Positioning). Moving the file position past the end of the data already written fills the intervening space with zeroes.

    Here is an example of using open_memstream:

    #include <stdio.h>
    
    int
    main (void)
    {
      char *bp;
      size_t size;
      FILE *stream;
    
      stream = open_memstream (&bp, &size);
      fprintf (stream, "hello");
      fflush (stream);
      printf ("buf = `%s', size = %d\n", bp, size);
      fprintf (stream, ", world");
      fclose (stream);
      printf ("buf = `%s', size = %d\n", bp, size);
    
      return 0;
    }
    

    This program produces the following output:

    buf = `hello', size = 5
    buf = `hello, world', size = 12
    

    Obstack Streams

    You can open an output stream that puts it data in an obstack. See section Obstacks.

    Function: FILE * open_obstack_stream (struct obstack *obstack)
    This function opens a stream for writing data into the obstack obstack. This starts an object in the obstack and makes it grow as data is written (see section Growing Objects).

    Calling fflush on this stream updates the current size of the object to match the amount of data that has been written. After a call to fflush, you can examine the object temporarily.

    You can move the file position of an obstack stream with fseek or fseeko (see section File Positioning). Moving the file position past the end of the data written fills the intervening space with zeros.

    To make the object permanent, update the obstack with fflush, and then use obstack_finish to finalize the object and get its address. The following write to the stream starts a new object in the obstack, and later writes add to that object until you do another fflush and obstack_finish.

    But how do you find out how long the object is? You can get the length in bytes by calling obstack_object_size (see section Status of an Obstack), or you can null-terminate the object like this:

    obstack_1grow (obstack, 0);
    

    Whichever one you do, you must do it before calling obstack_finish. (You can do both if you wish.)

    Here is a sample function that uses open_obstack_stream:

    char *
    make_message_string (const char *a, int b)
    {
      FILE *stream = open_obstack_stream (&message_obstack);
      output_task (stream);
      fprintf (stream, ": ");
      fprintf (stream, a, b);
      fprintf (stream, "\n");
      fclose (stream);
      obstack_1grow (&message_obstack, 0);
      return obstack_finish (&message_obstack);
    }
    

    Programming Your Own Custom Streams

    This section describes how you can make a stream that gets input from an arbitrary data source or writes output to an arbitrary data sink programmed by you. We call these custom streams. The functions and types described here are all GNU extensions.

    Custom Streams and Cookies

    Inside every custom stream is a special object called the cookie. This is an object supplied by you which records where to fetch or store the data read or written. It is up to you to define a data type to use for the cookie. The stream functions in the library never refer directly to its contents, and they don't even know what the type is; they record its address with type void *.

    To implement a custom stream, you must specify how to fetch or store the data in the specified place. You do this by defining hook functions to read, write, change "file position", and close the stream. All four of these functions will be passed the stream's cookie so they can tell where to fetch or store the data. The library functions don't know what's inside the cookie, but your functions will know.

    When you create a custom stream, you must specify the cookie pointer, and also the four hook functions stored in a structure of type cookie_io_functions_t.

    These facilities are declared in `stdio.h'.

    Data Type: cookie_io_functions_t
    This is a structure type that holds the functions that define the communications protocol between the stream and its cookie. It has the following members:

    cookie_read_function_t *read
    This is the function that reads data from the cookie. If the value is a null pointer instead of a function, then read operations on this stream always return EOF.
    cookie_write_function_t *write
    This is the function that writes data to the cookie. If the value is a null pointer instead of a function, then data written to the stream is discarded.
    cookie_seek_function_t *seek
    This is the function that performs the equivalent of file positioning on the cookie. If the value is a null pointer instead of a function, calls to fseek or fseeko on this stream can only seek to locations within the buffer; any attempt to seek outside the buffer will return an ESPIPE error.
    cookie_close_function_t *close
    This function performs any appropriate cleanup on the cookie when closing the stream. If the value is a null pointer instead of a function, nothing special is done to close the cookie when the stream is closed.

    Function: FILE * fopencookie (void *cookie, const char *opentype, cookie_io_functions_t io-functions)
    This function actually creates the stream for communicating with the cookie using the functions in the io-functions argument. The opentype argument is interpreted as for fopen; see section Opening Streams. (But note that the "truncate on open" option is ignored.) The new stream is fully buffered.

    The fopencookie function returns the newly created stream, or a null pointer in case of an error.

    Custom Stream Hook Functions

    Here are more details on how you should define the four hook functions that a custom stream needs.

    You should define the function to read data from the cookie as:

    ssize_t reader (void *cookie, char *buffer, size_t size)
    

    This is very similar to the read function; see section Input and Output Primitives. Your function should transfer up to size bytes into the buffer, and return the number of bytes read, or zero to indicate end-of-file. You can return a value of -1 to indicate an error.

    You should define the function to write data to the cookie as:

    ssize_t writer (void *cookie, const char *buffer, size_t size)
    

    This is very similar to the write function; see section Input and Output Primitives. Your function should transfer up to size bytes from the buffer, and return the number of bytes written. You can return a value of -1 to indicate an error.

    You should define the function to perform seek operations on the cookie as:

    int seeker (void *cookie, fpos_t *position, int whence)
    

    For this function, the position and whence arguments are interpreted as for fgetpos; see section Portable File-Position Functions. In the GNU library, fpos_t is equivalent to off_t or long int, and simply represents the number of bytes from the beginning of the file.

    After doing the seek operation, your function should store the resulting file position relative to the beginning of the file in position. Your function should return a value of 0 on success and -1 to indicate an error.

    You should define the function to do cleanup operations on the cookie appropriate for closing the stream as:

    int cleaner (void *cookie)
    

    Your function should return -1 to indicate an error, and 0 otherwise.

    Data Type: cookie_read_function
    This is the data type that the read function for a custom stream should have. If you declare the function as shown above, this is the type it will have.

    Data Type: cookie_write_function
    The data type of the write function for a custom stream.

    Data Type: cookie_seek_function
    The data type of the seek function for a custom stream.

    Data Type: cookie_close_function
    The data type of the close function for a custom stream.

    Formatted Messages

    On systems which are based on System V messages of programs (especially the system tools) are printed in a strict form using the fmtmsg function. The uniformity sometimes helps the user to interpret messages and the strictness tests of the fmtmsg function ensure that the programmer follows some minimal requirements.

    Printing Formatted Messages

    Messages can be printed to standard error and/or to the console. To select the destination the programmer can use the following two values, bitwise OR combined if wanted, for the classification parameter of fmtmsg:

    MM_PRINT
    Display the message in standard error.
    MM_CONSOLE
    Display the message on the system console.

    The erroneous piece of the system can be signalled by exactly one of the following values which also is bitwise ORed with the classification parameter to fmtmsg:

    MM_HARD
    The source of the condition is some hardware.
    MM_SOFT
    The source of the condition is some software.
    MM_FIRM
    The source of the condition is some firmware.

    A third component of the classification parameter to fmtmsg can describe the part of the system which detects the problem. This is done by using exactly one of the following values:

    MM_APPL
    The erroneous condition is detected by the application.
    MM_UTIL
    The erroneous condition is detected by a utility.
    MM_OPSYS
    The erroneous condition is detected by the operating system.

    A last component of classification can signal the results of this message. Exactly one of the following values can be used:

    MM_RECOVER
    It is a recoverable error.
    MM_NRECOV
    It is a non-recoverable error.

    Function: int fmtmsg (long int classification, const char *label, int severity, const char *text, const char *action, const char *tag)
    Display a message described by its parameters on the device(s) specified in the classification parameter. The label parameter identifies the source of the message. The string should consist of two colon separated parts where the first part has not more than 10 and the second part not more than 14 characters. The text parameter describes the condition of the error, the action parameter possible steps to recover from the error and the tag parameter is a reference to the online documentation where more information can be found. It should contain the label value and a unique identification number.

    Each of the parameters can be a special value which means this value is to be omitted. The symbolic names for these values are:

    MM_NULLLBL
    Ignore label parameter.
    MM_NULLSEV
    Ignore severity parameter.
    MM_NULLMC
    Ignore classification parameter. This implies that nothing is actually printed.
    MM_NULLTXT
    Ignore text parameter.
    MM_NULLACT
    Ignore action parameter.
    MM_NULLTAG
    Ignore tag parameter.

    There is another way certain fields can be omitted from the output to standard error. This is described below in the description of environment variables influencing the behaviour.

    The severity parameter can have one of the values in the following table:

    MM_NOSEV
    Nothing is printed, this value is the same as MM_NULLSEV.
    MM_HALT
    This value is printed as HALT.
    MM_ERROR
    This value is printed as ERROR.
    MM_WARNING
    This value is printed as WARNING.
    MM_INFO
    This value is printed as INFO.

    The numeric value of these five macros are between 0 and 4. Using the environment variable SEV_LEVEL or using the addseverity function one can add more severity levels with their corresponding string to print. This is described below (see section Adding Severity Classes).

    If no parameter is ignored the output looks like this:

    label: severity-string: text
    TO FIX: action tag
    

    The colons, new line characters and the TO FIX string are inserted if necessary, i.e., if the corresponding parameter is not ignored.

    This function is specified in the X/Open Portability Guide. It is also available on all systems derived from System V.

    The function returns the value MM_OK if no error occurred. If only the printing to standard error failed, it returns MM_NOMSG. If printing to the console fails, it returns MM_NOCON. If nothing is printed MM_NOTOK is returned. Among situations where all outputs fail this last value is also returned if a parameter value is incorrect.

    There are two environment variables which influence the behaviour of fmtmsg. The first is MSGVERB. It is used to control the output actually happening on standard error (not the console output). Each of the five fields can explicitly be enabled. To do this the user has to put the MSGVERB variable with a format like the following in the environment before calling the fmtmsg function the first time:

    MSGVERB=keyword[:keyword[:...]]
    

    Valid keywords are label, severity, text, action, and tag. If the environment variable is not given or is the empty string, a not supported keyword is given or the value is somehow else invalid, no part of the message is masked out.

    The second environment variable which influences the behaviour of fmtmsg is SEV_LEVEL. This variable and the change in the behaviour of fmtmsg is not specified in the X/Open Portability Guide. It is available in System V systems, though. It can be used to introduce new severity levels. By default, only the five severity levels described above are available. Any other numeric value would make fmtmsg print nothing.

    If the user puts SEV_LEVEL with a format like

    SEV_LEVEL=[description[:description[:...]]]
    

    in the environment of the process before the first call to fmtmsg, where description has a value of the form

    severity-keyword,level,printstring
    

    The severity-keyword part is not used by fmtmsg but it has to be present. The level part is a string representation of a number. The numeric value must be a number greater than 4. This value must be used in the severity parameter of fmtmsg to select this class. It is not possible to overwrite any of the predefined classes. The printstring is the string printed when a message of this class is processed by fmtmsg (see above, fmtsmg does not print the numeric value but instead the string representation).

    Adding Severity Classes

    There is another possibility to introduce severity classes besides using the environment variable SEV_LEVEL. This simplifies the task of introducing new classes in a running program. One could use the setenv or putenv function to set the environment variable, but this is toilsome.

    Function: int addseverity (int severity, const char *string)
    This function allows the introduction of new severity classes which can be addressed by the severity parameter of the fmtmsg function. The severity parameter of addseverity must match the value for the parameter with the same name of fmtmsg, and string is the string printed in the actual messages instead of the numeric value.

    If string is NULL the severity class with the numeric value according to severity is removed.

    It is not possible to overwrite or remove one of the default severity classes. All calls to addseverity with severity set to one of the values for the default classes will fail.

    The return value is MM_OK if the task was successfully performed. If the return value is MM_NOTOK something went wrong. This could mean that no more memory is available or a class is not available when it has to be removed.

    This function is not specified in the X/Open Portability Guide although the fmtsmg function is. It is available on System V systems.

    How to use fmtmsg and addseverity

    Here is a simple example program to illustrate the use of the both functions described in this section.

    #include <fmtmsg.h>
    
    int
    main (void)
    {
      addseverity (5, "NOTE2");
      fmtmsg (MM_PRINT, "only1field", MM_INFO, "text2", "action2", "tag2");
      fmtmsg (MM_PRINT, "UX:cat", 5, "invalid syntax", "refer to manual",
              "UX:cat:001");
      fmtmsg (MM_PRINT, "label:foo", 6, "text", "action", "tag");
      return 0;
    }
    

    The second call to fmtmsg illustrates a use of this function as it usually occurs on System V systems, which heavily use this function. It seems worthwhile to give a short explanation here of how this system works on System V. The value of the label field (UX:cat) says that the error occured in the Unix program cat. The explanation of the error follows and the value for the action parameter is "refer to manual". One could be more specific here, if necessary. The tag field contains, as proposed above, the value of the string given for the label parameter, and additionally a unique ID (001 in this case). For a GNU environment this string could contain a reference to the corresponding node in the Info page for the program.

    Running this program without specifying the MSGVERB and SEV_LEVEL function produces the following output:

    UX:cat: NOTE2: invalid syntax
    TO FIX: refer to manual UX:cat:001
    

    We see the different fields of the message and how the extra glue (the colons and the TO FIX string) are printed. But only one of the three calls to fmtmsg produced output. The first call does not print anything because the label parameter is not in the correct form. The string must contain two fields, separated by a colon (see section Printing Formatted Messages). The third fmtmsg call produced no output since the class with the numeric value 6 is not defined. Although a class with numeric value 5 is also not defined by default, the call to addseverity introduces it and the second call to fmtmsg produces the above output.

    When we change the environment of the program to contain SEV_LEVEL=XXX,6,NOTE when running it we get a different result:

    UX:cat: NOTE2: invalid syntax
    TO FIX: refer to manual UX:cat:001
    label:foo: NOTE: text
    TO FIX: action tag
    

    Now the third call to fmtmsg produced some output and we see how the string NOTE from the environment variable appears in the message.

    Now we can reduce the output by specifying which fields we are interested in. If we additionally set the environment variable MSGVERB to the value severity:label:action we get the following output:

    UX:cat: NOTE2
    TO FIX: refer to manual
    label:foo: NOTE
    TO FIX: action
    

    I.e., the output produced by the text and the tag parameters to fmtmsg vanished. Please also note that now there is no colon after the NOTE and NOTE2 strings in the output. This is not necessary since there is no more output on this line because the text is missing.

    Low-Level Input/Output

    This chapter describes functions for performing low-level input/output operations on file descriptors. These functions include the primitives for the higher-level I/O functions described in section Input/Output on Streams, as well as functions for performing low-level control operations for which there are no equivalents on streams.

    Stream-level I/O is more flexible and usually more convenient; therefore, programmers generally use the descriptor-level functions only when necessary. These are some of the usual reasons:

    • For reading binary files in large chunks.
    • For reading an entire file into core before parsing it.
    • To perform operations other than data transfer, which can only be done with a descriptor. (You can use fileno to get the descriptor corresponding to a stream.)
    • To pass descriptors to a child process. (The child can create its own stream to use a descriptor that it inherits, but cannot inherit a stream directly.)

    Opening and Closing Files

    This section describes the primitives for opening and closing files using file descriptors. The open and creat functions are declared in the header file `fcntl.h', while close is declared in `unistd.h'.

    Function: int open (const char *filename, int flags[, mode_t mode])
    The open function creates and returns a new file descriptor for the file named by filename. Initially, the file position indicator for the file is at the beginning of the file. The argument mode is used only when a file is created, but it doesn't hurt to supply the argument in any case.

    The flags argument controls how the file is to be opened. This is a bit mask; you create the value by the bitwise OR of the appropriate parameters (using the `|' operator in C). See section File Status Flags, for the parameters available.

    The normal return value from open is a non-negative integer file descriptor. In the case of an error, a value of @math{-1} is returned instead. In addition to the usual file name errors (see section File Name Errors), the following errno error conditions are defined for this function:

    EACCES
    The file exists but is not readable/writeable as requested by the flags argument, the file does not exist and the directory is unwriteable so it cannot be created.
    EEXIST
    Both O_CREAT and O_EXCL are set, and the named file already exists.
    EINTR
    The open operation was interrupted by a signal. See section Primitives Interrupted by Signals.
    EISDIR
    The flags argument specified write access, and the file is a directory.
    EMFILE
    The process has too many files open. The maximum number of file descriptors is controlled by the RLIMIT_NOFILE resource limit; see section Limiting Resource Usage.
    ENFILE
    The entire system, or perhaps the file system which contains the directory, cannot support any additional open files at the moment. (This problem cannot happen on the GNU system.)
    ENOENT
    The named file does not exist, and O_CREAT is not specified.
    ENOSPC
    The directory or file system that would contain the new file cannot be extended, because there is no disk space left.
    ENXIO
    O_NONBLOCK and O_WRONLY are both set in the flags argument, the file named by filename is a FIFO (see section Pipes and FIFOs), and no process has the file open for reading.
    EROFS
    The file resides on a read-only file system and any of O_WRONLY, O_RDWR, and O_TRUNC are set in the flags argument, or O_CREAT is set and the file does not already exist.

    If on a 32 bit machine the sources are translated with _FILE_OFFSET_BITS == 64 the function open returns a file descriptor opened in the large file mode which enables the file handling functions to use files up to @math{2^63} bytes in size and offset from @math{-2^63} to @math{2^63}. This happens transparently for the user since all of the lowlevel file handling functions are equally replaced.

    This function is a cancellation point in multi-threaded programs. This is a problem if the thread allocates some resources (like memory, file descriptors, semaphores or whatever) at the time open is called. If the thread gets cancelled these resources stay allocated until the program ends. To avoid this calls to open should be protected using cancellation handlers.

    The open function is the underlying primitive for the fopen and freopen functions, that create streams.

    Function: int open64 (const char *filename, int flags[, mode_t mode])
    This function is similar to open. It returns a file descriptor which can be used to access the file named by filename. The only difference is that on 32 bit systems the file is opened in the large file mode. I.e., file length and file offsets can exceed 31 bits.

    When the sources are translated with _FILE_OFFSET_BITS == 64 this function is actually available under the name open. I.e., the new, extended API using 64 bit file sizes and offsets transparently replaces the old API.

    Obsolete function: int creat (const char *filename, mode_t mode)
    This function is obsolete. The call:

    creat (filename, mode)
    

    is equivalent to:

    open (filename, O_WRONLY | O_CREAT | O_TRUNC, mode)
    

    If on a 32 bit machine the sources are translated with _FILE_OFFSET_BITS == 64 the function creat returns a file descriptor opened in the large file mode which enables the file handling functions to use files up to @math{2^63} in size and offset from @math{-2^63} to @math{2^63}. This happens transparently for the user since all of the lowlevel file handling functions are equally replaced.

    Obsolete function: int creat64 (const char *filename, mode_t mode)
    This function is similar to creat. It returns a file descriptor which can be used to access the file named by filename. The only the difference is that on 32 bit systems the file is opened in the large file mode. I.e., file length and file offsets can exceed 31 bits.

    To use this file descriptor one must not use the normal operations but instead the counterparts named *64, e.g., read64.

    When the sources are translated with _FILE_OFFSET_BITS == 64 this function is actually available under the name open. I.e., the new, extended API using 64 bit file sizes and offsets transparently replaces the old API.

    Function: int close (int filedes)
    The function close closes the file descriptor filedes. Closing a file has the following consequences:

    • The file descriptor is deallocated.
    • Any record locks owned by the process on the file are unlocked.
    • When all file descriptors associated with a pipe or FIFO have been closed, any unread data is discarded.

    This function is a cancellation point in multi-threaded programs. This is a problem if the thread allocates some resources (like memory, file descriptors, semaphores or whatever) at the time close is called. If the thread gets cancelled these resources stay allocated until the program ends. To avoid this, calls to close should be protected using cancellation handlers.

    The normal return value from close is @math{0}; a value of @math{-1} is returned in case of failure. The following errno error conditions are defined for this function:

    EBADF
    The filedes argument is not a valid file descriptor.
    EINTR
    The close call was interrupted by a signal. See section Primitives Interrupted by Signals. Here is an example of how to handle EINTR properly:
    TEMP_FAILURE_RETRY (close (desc));
    
    ENOSPC
    EIO
    EDQUOT
    When the file is accessed by NFS, these errors from write can sometimes not be detected until close. See section Input and Output Primitives, for details on their meaning.

    Please note that there is no separate close64 function. This is not necessary since this function does not determine nor depend on the mode of the file. The kernel which performs the close operation knows which mode the descriptor is used for and can handle this situation.

    To close a stream, call fclose (see section Closing Streams) instead of trying to close its underlying file descriptor with close. This flushes any buffered output and updates the stream object to indicate that it is closed.

    Input and Output Primitives

    This section describes the functions for performing primitive input and output operations on file descriptors: read, write, and lseek. These functions are declared in the header file `unistd.h'.

    Data Type: ssize_t
    This data type is used to represent the sizes of blocks that can be read or written in a single operation. It is similar to size_t, but must be a signed type.

    Function: ssize_t read (int filedes, void *buffer, size_t size)
    The read function reads up to size bytes from the file with descriptor filedes, storing the results in the buffer. (This is not necessarily a character string, and no terminating null character is added.)

    The return value is the number of bytes actually read. This might be less than size; for example, if there aren't that many bytes left in the file or if there aren't that many bytes immediately available. The exact behavior depends on what kind of file it is. Note that reading less than size bytes is not an error.

    A value of zero indicates end-of-file (except if the value of the size argument is also zero). This is not considered an error. If you keep calling read while at end-of-file, it will keep returning zero and doing nothing else.

    If read returns at least one character, there is no way you can tell whether end-of-file was reached. But if you did reach the end, the next read will return zero.

    In case of an error, read returns @math{-1}. The following errno error conditions are defined for this function:

    EAGAIN
    Normally, when no input is immediately available, read waits for some input. But if the O_NONBLOCK flag is set for the file (see section File Status Flags), read returns immediately without reading any data, and reports this error. Compatibility Note: Most versions of BSD Unix use a different error code for this: EWOULDBLOCK. In the GNU library, EWOULDBLOCK is an alias for EAGAIN, so it doesn't matter which name you use. On some systems, reading a large amount of data from a character special file can also fail with EAGAIN if the kernel cannot find enough physical memory to lock down the user's pages. This is limited to devices that transfer with direct memory access into the user's memory, which means it does not include terminals, since they always use separate buffers inside the kernel. This problem never happens in the GNU system. Any condition that could result in EAGAIN can instead result in a successful read which returns fewer bytes than requested. Calling read again immediately would result in EAGAIN.
    EBADF
    The filedes argument is not a valid file descriptor, or is not open for reading.
    EINTR
    read was interrupted by a signal while it was waiting for input. See section Primitives Interrupted by Signals. A signal will not necessary cause read to return EINTR; it may instead result in a successful read which returns fewer bytes than requested.
    EIO
    For many devices, and for disk files, this error code indicates a hardware error. EIO also occurs when a background process tries to read from the controlling terminal, and the normal action of stopping the process by sending it a SIGTTIN signal isn't working. This might happen if the signal is being blocked or ignored, or because the process group is orphaned. See section Job Control, for more information about job control, and section Signal Handling, for information about signals.

    Please note that there is no function named read64. This is not necessary since this function does not directly modify or handle the possibly wide file offset. Since the kernel handles this state internally, the read function can be used for all cases.

    This function is a cancellation point in multi-threaded programs. This is a problem if the thread allocates some resources (like memory, file descriptors, semaphores or whatever) at the time read is called. If the thread gets cancelled these resources stay allocated until the program ends. To avoid this, calls to read should be protected using cancellation handlers.

    The read function is the underlying primitive for all of the functions that read from streams, such as fgetc.

    Function: ssize_t pread (int filedes, void *buffer, size_t size, off_t offset)
    The pread function is similar to the read function. The first three arguments are identical, and the return values and error codes also correspond.

    The difference is the fourth argument and its handling. The data block is not read from the current position of the file descriptor filedes. Instead the data is read from the file starting at position offset. The position of the file descriptor itself is not affected by the operation. The value is the same as before the call.

    When the source file is compiled with _FILE_OFFSET_BITS == 64 the pread function is in fact pread64 and the type off_t has 64 bits, which makes it possible to handle files up to @math{2^63} bytes in length.

    The return value of pread describes the number of bytes read. In the error case it returns @math{-1} like read does and the error codes are also the same, with these additions:

    EINVAL
    The value given for offset is negative and therefore illegal.
    ESPIPE
    The file descriptor filedes is associate with a pipe or a FIFO and this device does not allow positioning of the file pointer.

    The function is an extension defined in the Unix Single Specification version 2.

    Function: ssize_t pread64 (int filedes, void *buffer, size_t size, off64_t offset)
    This function is similar to the pread function. The difference is that the offset parameter is of type off64_t instead of off_t which makes it possible on 32 bit machines to address files larger than @math{2^31} bytes and up to @math{2^63} bytes. The file descriptor filedes must be opened using open64 since otherwise the large offsets possible with off64_t will lead to errors with a descriptor in small file mode.

    When the source file is compiled with _FILE_OFFSET_BITS == 64 on a 32 bit machine this function is actually available under the name pread and so transparently replaces the 32 bit interface.

    Function: ssize_t write (int filedes, const void *buffer, size_t size)
    The write function writes up to size bytes from buffer to the file with descriptor filedes. The data in buffer is not necessarily a character string and a null character is output like any other character.

    The return value is the number of bytes actually written. This may be size, but can always be smaller. Your program should always call write in a loop, iterating until all the data is written.

    Once write returns, the data is enqueued to be written and can be read back right away, but it is not necessarily written out to permanent storage immediately. You can use fsync when you need to be sure your data has been permanently stored before continuing. (It is more efficient for the system to batch up consecutive writes and do them all at once when convenient. Normally they will always be written to disk within a minute or less.) Modern systems provide another function fdatasync which guarantees integrity only for the file data and is therefore faster. You can use the O_FSYNC open mode to make write always store the data to disk before returning; see section I/O Operating Modes.

    In the case of an error, write returns @math{-1}. The following errno error conditions are defined for this function:

    EAGAIN
    Normally, write blocks until the write operation is complete. But if the O_NONBLOCK flag is set for the file (see section Control Operations on Files), it returns immediately without writing any data and reports this error. An example of a situation that might cause the process to block on output is writing to a terminal device that supports flow control, where output has been suspended by receipt of a STOP character. Compatibility Note: Most versions of BSD Unix use a different error code for this: EWOULDBLOCK. In the GNU library, EWOULDBLOCK is an alias for EAGAIN, so it doesn't matter which name you use. On some systems, writing a large amount of data from a character special file can also fail with EAGAIN if the kernel cannot find enough physical memory to lock down the user's pages. This is limited to devices that transfer with direct memory access into the user's memory, which means it does not include terminals, since they always use separate buffers inside the kernel. This problem does not arise in the GNU system.
    EBADF
    The filedes argument is not a valid file descriptor, or is not open for writing.
    EFBIG
    The size of the file would become larger than the implementation can support.
    EINTR
    The write operation was interrupted by a signal while it was blocked waiting for completion. A signal will not necessarily cause write to return EINTR; it may instead result in a successful write which writes fewer bytes than requested. See section Primitives Interrupted by Signals.
    EIO
    For many devices, and for disk files, this error code indicates a hardware error.
    ENOSPC
    The device containing the file is full.
    EPIPE
    This error is returned when you try to write to a pipe or FIFO that isn't open for reading by any process. When this happens, a SIGPIPE signal is also sent to the process; see section Signal Handling.

    Unless you have arranged to prevent EINTR failures, you should check errno after each failing call to write, and if the error was EINTR, you should simply repeat the call. See section Primitives Interrupted by Signals. The easy way to do this is with the macro TEMP_FAILURE_RETRY, as follows:

    nbytes = TEMP_FAILURE_RETRY (write (desc, buffer, count));
    

    Please note that there is no function named write64. This is not necessary since this function does not directly modify or handle the possibly wide file offset. Since the kernel handles this state internally the write function can be used for all cases.

    This function is a cancellation point in multi-threaded programs. This is a problem if the thread allocates some resources (like memory, file descriptors, semaphores or whatever) at the time write is called. If the thread gets cancelled these resources stay allocated until the program ends. To avoid this, calls to write should be protected using cancellation handlers.

    The write function is the underlying primitive for all of the functions that write to streams, such as fputc.

    Function: ssize_t pwrite (int filedes, const void *buffer, size_t size, off_t offset)
    The pwrite function is similar to the write function. The first three arguments are identical, and the return values and error codes also correspond.

    The difference is the fourth argument and its handling. The data block is not written to the current position of the file descriptor filedes. Instead the data is written to the file starting at position offset. The position of the file descriptor itself is not affected by the operation. The value is the same as before the call.

    When the source file is compiled with _FILE_OFFSET_BITS == 64 the pwrite function is in fact pwrite64 and the type off_t has 64 bits, which makes it possible to handle files up to @math{2^63} bytes in length.

    The return value of pwrite describes the number of written bytes. In the error case it returns @math{-1} like write does and the error codes are also the same, with these additions:

    EINVAL
    The value given for offset is negative and therefore illegal.
    ESPIPE
    The file descriptor filedes is associated with a pipe or a FIFO and this device does not allow positioning of the file pointer.

    The function is an extension defined in the Unix Single Specification version 2.

    Function: ssize_t pwrite64 (int filedes, const void *buffer, size_t size, off64_t offset)
    This function is similar to the pwrite function. The difference is that the offset parameter is of type off64_t instead of off_t which makes it possible on 32 bit machines to address files larger than @math{2^31} bytes and up to @math{2^63} bytes. The file descriptor filedes must be opened using open64 since otherwise the large offsets possible with off64_t will lead to errors with a descriptor in small file mode.

    When the source file is compiled using _FILE_OFFSET_BITS == 64 on a 32 bit machine this function is actually available under the name pwrite and so transparently replaces the 32 bit interface.

    Setting the File Position of a Descriptor

    Just as you can set the file position of a stream with fseek, you can set the file position of a descriptor with lseek. This specifies the position in the file for the next read or write operation. See section File Positioning, for more information on the file position and what it means.

    To read the current file position value from a descriptor, use lseek (desc, 0, SEEK_CUR).

    Function: off_t lseek (int filedes, off_t offset, int whence)
    The lseek function is used to change the file position of the file with descriptor filedes.

    The whence argument specifies how the offset should be interpreted, in the same way as for the fseek function, and it must be one of the symbolic constants SEEK_SET, SEEK_CUR, or SEEK_END.

    SEEK_SET
    Specifies that whence is a count of characters from the beginning of the file.
    SEEK_CUR
    Specifies that whence is a count of characters from the current file position. This count may be positive or negative.
    SEEK_END
    Specifies that whence is a count of characters from the end of the file. A negative count specifies a position within the current extent of the file; a positive count specifies a position past the current end. If you set the position past the current end, and actually write data, you will extend the file with zeros up to that position.

    The return value from lseek is normally the resulting file position, measured in bytes from the beginning of the file. You can use this feature together with SEEK_CUR to read the current file position.

    If you want to append to the file, setting the file position to the current end of file with SEEK_END is not sufficient. Another process may write more data after you seek but before you write, extending the file so the position you write onto clobbers their data. Instead, use the O_APPEND operating mode; see section I/O Operating Modes.

    You can set the file position past the current end of the file. This does not by itself make the file longer; lseek never changes the file. But subsequent output at that position will extend the file. Characters between the previous end of file and the new position are filled with zeros. Extending the file in this way can create a "hole": the blocks of zeros are not actually allocated on disk, so the file takes up less space than it appears to; it is then called a "sparse file".

    If the file position cannot be changed, or the operation is in some way invalid, lseek returns a value of @math{-1}. The following errno error conditions are defined for this function:

    EBADF
    The filedes is not a valid file descriptor.
    EINVAL
    The whence argument value is not valid, or the resulting file offset is not valid. A file offset is invalid.
    ESPIPE
    The filedes corresponds to an object that cannot be positioned, such as a pipe, FIFO or terminal device. (POSIX.1 specifies this error only for pipes and FIFOs, but in the GNU system, you always get ESPIPE if the object is not seekable.)

    When the source file is compiled with _FILE_OFFSET_BITS == 64 the lseek function is in fact lseek64 and the type off_t has 64 bits which makes it possible to handle files up to @math{2^63} bytes in length.

    This function is a cancellation point in multi-threaded programs. This is a problem if the thread allocates some resources (like memory, file descriptors, semaphores or whatever) at the time lseek is called. If the thread gets cancelled these resources stay allocated until the program ends. To avoid this calls to lseek should be protected using cancellation handlers.

    The lseek function is the underlying primitive for the fseek, fseeko, ftell, ftello and rewind functions, which operate on streams instead of file descriptors.

    Function: off64_t lseek64 (int filedes, off64_t offset, int whence)
    This function is similar to the lseek function. The difference is that the offset parameter is of type off64_t instead of off_t which makes it possible on 32 bit machines to address files larger than @math{2^31} bytes and up to @math{2^63} bytes. The file descriptor filedes must be opened using open64 since otherwise the large offsets possible with off64_t will lead to errors with a descriptor in small file mode.

    When the source file is compiled with _FILE_OFFSET_BITS == 64 on a 32 bits machine this function is actually available under the name lseek and so transparently replaces the 32 bit interface.

    You can have multiple descriptors for the same file if you open the file more than once, or if you duplicate a descriptor with dup. Descriptors that come from separate calls to open have independent file positions; using lseek on one descriptor has no effect on the other. For example,

    {
      int d1, d2;
      char buf[4];
      d1 = open ("foo", O_RDONLY);
      d2 = open ("foo", O_RDONLY);
      lseek (d1, 1024, SEEK_SET);
      read (d2, buf, 4);
    }
    

    will read the first four characters of the file `foo'. (The error-checking code necessary for a real program has been omitted here for brevity.)

    By contrast, descriptors made by duplication share a common file position with the original descriptor that was duplicated. Anything which alters the file position of one of the duplicates, including reading or writing data, affects all of them alike. Thus, for example,

    {
      int d1, d2, d3;
      char buf1[4], buf2[4];
      d1 = open ("foo", O_RDONLY);
      d2 = dup (d1);
      d3 = dup (d2);
      lseek (d3, 1024, SEEK_SET);
      read (d1, buf1, 4);
      read (d2, buf2, 4);
    }
    

    will read four characters starting with the 1024'th character of `foo', and then four more characters starting with the 1028'th character.

    Data Type: off_t
    This is an arithmetic data type used to represent file sizes. In the GNU system, this is equivalent to fpos_t or long int.

    If the source is compiled with _FILE_OFFSET_BITS == 64 this type is transparently replaced by off64_t.

    Data Type: off64_t
    This type is used similar to off_t. The difference is that even on 32 bit machines, where the off_t type would have 32 bits, off64_t has 64 bits and so is able to address files up to @math{2^63} bytes in length.

    When compiling with _FILE_OFFSET_BITS == 64 this type is available under the name off_t.

    These aliases for the `SEEK_...' constants exist for the sake of compatibility with older BSD systems. They are defined in two different header files: `fcntl.h' and `sys/file.h'.

    L_SET
    An alias for SEEK_SET.
    L_INCR
    An alias for SEEK_CUR.
    L_XTND
    An alias for SEEK_END.

    Descriptors and Streams

    Given an open file descriptor, you can create a stream for it with the fdopen function. You can get the underlying file descriptor for an existing stream with the fileno function. These functions are declared in the header file `stdio.h'.

    Function: FILE * fdopen (int filedes, const char *opentype)
    The fdopen function returns a new stream for the file descriptor filedes.

    The opentype argument is interpreted in the same way as for the fopen function (see section Opening Streams), except that the `b' option is not permitted; this is because GNU makes no distinction between text and binary files. Also, "w" and "w+" do not cause truncation of the file; these have an effect only when opening a file, and in this case the file has already been opened. You must make sure that the opentype argument matches the actual mode of the open file descriptor.

    The return value is the new stream. If the stream cannot be created (for example, if the modes for the file indicated by the file descriptor do not permit the access specified by the opentype argument), a null pointer is returned instead.

    In some other systems, fdopen may fail to detect that the modes for file descriptor do not permit the access specified by opentype. The GNU C library always checks for this.

    For an example showing the use of the fdopen function, see section Creating a Pipe.

    Function: int fileno (FILE *stream)
    This function returns the file descriptor associated with the stream stream. If an error is detected (for example, if the stream is not valid) or if stream does not do I/O to a file, fileno returns @math{-1}.

    Function: int fileno_unlocked (FILE *stream)
    The fileno_unlocked function is equivalent to the fileno function except that it does not implicitly lock the stream if the state is FSETLOCKING_INTERNAL.

    This function is a GNU extension.

    There are also symbolic constants defined in `unistd.h' for the file descriptors belonging to the standard streams stdin, stdout, and stderr; see section Standard Streams.

    STDIN_FILENO
    This macro has value 0, which is the file descriptor for standard input.
    STDOUT_FILENO
    This macro has value 1, which is the file descriptor for standard output.
    STDERR_FILENO
    This macro has value 2, which is the file descriptor for standard error output.

    Dangers of Mixing Streams and Descriptors

    You can have multiple file descriptors and streams (let's call both streams and descriptors "channels" for short) connected to the same file, but you must take care to avoid confusion between channels. There are two cases to consider: linked channels that share a single file position value, and independent channels that have their own file positions.

    It's best to use just one channel in your program for actual data transfer to any given file, except when all the access is for input. For example, if you open a pipe (something you can only do at the file descriptor level), either do all I/O with the descriptor, or construct a stream from the descriptor with fdopen and then do all I/O with the stream.

    Linked Channels

    Channels that come from a single opening share the same file position; we call them linked channels. Linked channels result when you make a stream from a descriptor using fdopen, when you get a descriptor from a stream with fileno, when you copy a descriptor with dup or dup2, and when descriptors are inherited during fork. For files that don't support random access, such as terminals and pipes, all channels are effectively linked. On random-access files, all append-type output streams are effectively linked to each other.

    If you have been using a stream for I/O, and you want to do I/O using another channel (either a stream or a descriptor) that is linked to it, you must first clean up the stream that you have been using. See section Cleaning Streams.

    Terminating a process, or executing a new program in the process, destroys all the streams in the process. If descriptors linked to these streams persist in other processes, their file positions become undefined as a result. To prevent this, you must clean up the streams before destroying them.

    Independent Channels

    When you open channels (streams or descriptors) separately on a seekable file, each channel has its own file position. These are called independent channels.

    The system handles each channel independently. Most of the time, this is quite predictable and natural (especially for input): each channel can read or write sequentially at its own place in the file. However, if some of the channels are streams, you must take these precautions:

    • You should clean an output stream after use, before doing anything else that might read or write from the same part of the file.
    • You should clean an input stream before reading data that may have been modified using an independent channel. Otherwise, you might read obsolete data that had been in the stream's buffer.

    If you do output to one channel at the end of the file, this will certainly leave the other independent channels positioned somewhere before the new end. You cannot reliably set their file positions to the new end of file before writing, because the file can always be extended by another process between when you set the file position and when you write the data. Instead, use an append-type descriptor or stream; they always output at the current end of the file. In order to make the end-of-file position accurate, you must clean the output channel you were using, if it is a stream.

    It's impossible for two channels to have separate file pointers for a file that doesn't support random access. Thus, channels for reading or writing such files are always linked, never independent. Append-type channels are also always linked. For these channels, follow the rules for linked channels; see section Linked Channels.

    Cleaning Streams

    On the GNU system, you can clean up any stream with fclean:

    Function: int fclean (FILE *stream)
    Clean up the stream stream so that its buffer is empty. If stream is doing output, force it out. If stream is doing input, give the data in the buffer back to the system, arranging to reread it.

    On other systems, you can use fflush to clean a stream in most cases.

    You can skip the fclean or fflush if you know the stream is already clean. A stream is clean whenever its buffer is empty. For example, an unbuffered stream is always clean. An input stream that is at end-of-file is clean. A line-buffered stream is clean when the last character output was a newline.

    There is one case in which cleaning a stream is impossible on most systems. This is when the stream is doing input from a file that is not random-access. Such streams typically read ahead, and when the file is not random access, there is no way to give back the excess data already read. When an input stream reads from a random-access file, fflush does clean the stream, but leaves the file pointer at an unpredictable place; you must set the file pointer before doing any further I/O. On the GNU system, using fclean avoids both of these problems.

    Closing an output-only stream also does fflush, so this is a valid way of cleaning an output stream. On the GNU system, closing an input stream does fclean.

    You need not clean a stream before using its descriptor for control operations such as setting terminal modes; these operations don't affect the file position and are not affected by it. You can use any descriptor for these operations, and all channels are affected simultaneously. However, text already "output" to a stream but still buffered by the stream will be subject to the new terminal modes when subsequently flushed. To make sure "past" output is covered by the terminal settings that were in effect at the time, flush the output streams for that terminal before setting the modes. See section Terminal Modes.

    Fast Scatter-Gather I/O

    Some applications may need to read or write data to multiple buffers, which are separated in memory. Although this can be done easily enough with multiple calls to read and write, it is inefficent because there is overhead associated with each kernel call.

    Instead, many platforms provide special high-speed primitives to perform these scatter-gather operations in a single kernel call. The GNU C library will provide an emulation on any system that lacks these primitives, so they are not a portability threat. They are defined in sys/uio.h.

    These functions are controlled with arrays of iovec structures, which describe the location and size of each buffer.

    Data Type: struct iovec

    The iovec structure describes a buffer. It contains two fields:

    void *iov_base
    Contains the address of a buffer.
    size_t iov_len
    Contains the length of the buffer.

    Function: ssize_t readv (int filedes, const struct iovec *vector, int count)

    The readv function reads data from filedes and scatters it into the buffers described in vector, which is taken to be count structures long. As each buffer is filled, data is sent to the next.

    Note that readv is not guaranteed to fill all the buffers. It may stop at any point, for the same reasons read would.

    The return value is a count of bytes (not buffers) read, @math{0} indicating end-of-file, or @math{-1} indicating an error. The possible errors are the same as in read.

    Function: ssize_t writev (int filedes, const struct iovec *vector, int count)

    The writev function gathers data from the buffers described in vector, which is taken to be count structures long, and writes them to filedes. As each buffer is written, it moves on to the next.

    Like readv, writev may stop midstream under the same conditions write would.

    The return value is a count of bytes written, or @math{-1} indicating an error. The possible errors are the same as in write.

    Note that if the buffers are small (under about 1kB), high-level streams may be easier to use than these functions. However, readv and writev are more efficient when the individual buffers themselves (as opposed to the total output), are large. In that case, a high-level stream would not be able to cache the data effectively.

    Memory-mapped I/O

    On modern operating systems, it is possible to mmap (pronounced "em-map") a file to a region of memory. When this is done, the file can be accessed just like an array in the program.

    This is more efficent than read or write, as only the regions of the file that a program actually accesses are loaded. Accesses to not-yet-loaded parts of the mmapped region are handled in the same way as swapped out pages.

    Since mmapped pages can be stored back to their file when physical memory is low, it is possible to mmap files orders of magnitude larger than both the physical memory and swap space. The only limit is address space. The theoretical limit is 4GB on a 32-bit machine - however, the actual limit will be smaller since some areas will be reserved for other purposes. If the LFS interface is used the file size on 32-bit systems is not limited to 2GB (offsets are signed which reduces the addressable area of 4GB by half); the full 64-bit are available.

    Memory mapping only works on entire pages of memory. Thus, addresses for mapping must be page-aligned, and length values will be rounded up. To determine the size of a page the machine uses one should use

    size_t page_size = (size_t) sysconf (_SC_PAGESIZE);
    

    These functions are declared in `sys/mman.h'.

    Function: void * mmap (void *address, size_t length,int protect, int flags, int filedes, off_t offset)

    The mmap function creates a new mapping, connected to bytes (offset) to (offset + length) in the file open on filedes.

    address gives a preferred starting address for the mapping. NULL expresses no preference. Any previous mapping at that address is automatically removed. The address you give may still be changed, unless you use the MAP_FIXED flag.

    protect contains flags that control what kind of access is permitted. They include PROT_READ, PROT_WRITE, and PROT_EXEC, which permit reading, writing, and execution, respectively. Inappropriate access will cause a segfault (see section Program Error Signals).

    Note that most hardware designs cannot support write permission without read permission, and many do not distinguish read and execute permission. Thus, you may receive wider permissions than you ask for, and mappings of write-only files may be denied even if you do not use PROT_READ.

    flags contains flags that control the nature of the map. One of MAP_SHARED or MAP_PRIVATE must be specified.

    They include:

    MAP_PRIVATE
    This specifies that writes to the region should never be written back to the attached file. Instead, a copy is made for the process, and the region will be swapped normally if memory runs low. No other process will see the changes. Since private mappings effectively revert to ordinary memory when written to, you must have enough virtual memory for a copy of the entire mmapped region if you use this mode with PROT_WRITE.
    MAP_SHARED
    This specifies that writes to the region will be written back to the file. Changes made will be shared immediately with other processes mmaping the same file. Note that actual writing may take place at any time. You need to use msync, described below, if it is important that other processes using conventional I/O get a consistent view of the file.
    MAP_FIXED
    This forces the system to use the exact mapping address specified in address and fail if it can't.
    MAP_ANONYMOUS
    MAP_ANON
    This flag tells the system to create an anonymous mapping, not connected to a file. filedes and off are ignored, and the region is initialized with zeros. Anonymous maps are used as the basic primitive to extend the heap on some systems. They are also useful to share data between multiple tasks without creating a file. On some systems using private anonymous mmaps is more efficient than using malloc for large blocks. This is not an issue with the GNU C library, as the included malloc automatically uses mmap where appropriate.

    mmap returns the address of the new mapping, or @math{-1} for an error.

    Possible errors include:

    EINVAL
    Either address was unusable, or inconsistent flags were given.
    EACCES
    filedes was not open for the type of access specified in protect.
    ENOMEM
    Either there is not enough memory for the operation, or the process is out of address space.
    ENODEV
    This file is of a type that doesn't support mapping.
    ENOEXEC
    The file is on a filesystem that doesn't support mapping.

    Function: void * mmap64 (void *address, size_t length,int protect, int flags, int filedes, off64_t offset)
    The mmap64 function is equivalent to the mmap function but the offset parameter is of type off64_t. On 32-bit systems this allows the file associated with the filedes descriptor to be larger than 2GB. filedes must be a descriptor returned from a call to open64 or fopen64 and freopen64 where the descriptor is retrieved with fileno.

    When the sources are translated with _FILE_OFFSET_BITS == 64 this function is actually available under the name mmap. I.e., the new, extended API using 64 bit file sizes and offsets transparently replaces the old API.

    Function: int munmap (void *addr, size_t length)

    munmap removes any memory maps from (addr) to (addr + length). length should be the length of the mapping.

    It is safe to unmap multiple mappings in one command, or include unmapped space in the range. It is also possible to unmap only part of an existing mapping. However, only entire pages can be removed. If length is not an even number of pages, it will be rounded up.

    It returns @math{0} for success and @math{-1} for an error.

    One error is possible:

    EINVAL
    The memory range given was outside the user mmap range or wasn't page aligned.

    Function: int msync (void *address, size_t length, int flags)

    When using shared mappings, the kernel can write the file at any time before the mapping is removed. To be certain data has actually been written to the file and will be accessible to non-memory-mapped I/O, it is necessary to use this function.

    It operates on the region address to (address + length). It may be used on part of a mapping or multiple mappings, however the region given should not contain any unmapped space.

    flags can contain some options:

    MS_SYNC
    This flag makes sure the data is actually written to disk. Normally msync only makes sure that accesses to a file with conventional I/O reflect the recent changes.
    MS_ASYNC
    This tells msync to begin the synchronization, but not to wait for it to complete.

    msync returns @math{0} for success and @math{-1} for error. Errors include:

    EINVAL
    An invalid region was given, or the flags were invalid.
    EFAULT
    There is no existing mapping in at least part of the given region.

    Function: void * mremap (void *address, size_t length, size_t new_length, int flag)

    This function can be used to change the size of an existing memory area. address and length must cover a region entirely mapped in the same mmap statement. A new mapping with the same characteristics will be returned with the length new_length.

    One option is possible, MREMAP_MAYMOVE. If it is given in flags, the system may remove the existing mapping and create a new one of the desired length in another location.

    The address of the resulting mapping is returned, or @math{-1}. Possible error codes include:

    EFAULT
    There is no existing mapping in at least part of the original region, or the region covers two or more distinct mappings.
    EINVAL
    The address given is misaligned or inappropriate.
    EAGAIN
    The region has pages locked, and if extended it would exceed the process's resource limit for locked pages. See section Limiting Resource Usage.
    ENOMEM
    The region is private writeable, and insufficent virtual memory is available to extend it. Also, this error will occur if MREMAP_MAYMOVE is not given and the extension would collide with another mapped region.

    This function is only available on a few systems. Except for performing optional optimizations one should not rely on this function.

    Not all file descriptors may be mapped. Sockets, pipes, and most devices only allow sequential access and do not fit into the mapping abstraction. In addition, some regular files may not be mmapable, and older kernels may not support mapping at all. Thus, programs using mmap should have a fallback method to use should it fail. See section `Mmap' in GNU Coding Standards.

    Waiting for Input or Output

    Sometimes a program needs to accept input on multiple input channels whenever input arrives. For example, some workstations may have devices such as a digitizing tablet, function button box, or dial box that are connected via normal asynchronous serial interfaces; good user interface style requires responding immediately to input on any device. Another example is a program that acts as a server to several other processes via pipes or sockets.

    You cannot normally use read for this purpose, because this blocks the program until input is available on one particular file descriptor; input on other channels won't wake it up. You could set nonblocking mode and poll each file descriptor in turn, but this is very inefficient.

    A better solution is to use the select function. This blocks the program until input or output is ready on a specified set of file descriptors, or until a timer expires, whichever comes first. This facility is declared in the header file `sys/types.h'.

    In the case of a server socket (see section Listening for Connections), we say that "input" is available when there are pending connections that could be accepted (see section Accepting Connections). accept for server sockets blocks and interacts with select just as read does for normal input.

    The file descriptor sets for the select function are specified as fd_set objects. Here is the description of the data type and some macros for manipulating these objects.

    Data Type: fd_set
    The fd_set data type represents file descriptor sets for the select function. It is actually a bit array.

    Macro: int FD_SETSIZE
    The value of this macro is the maximum number of file descriptors that a fd_set object can hold information about. On systems with a fixed maximum number, FD_SETSIZE is at least that number. On some systems, including GNU, there is no absolute limit on the number of descriptors open, but this macro still has a constant value which controls the number of bits in an fd_set; if you get a file descriptor with a value as high as FD_SETSIZE, you cannot put that descriptor into an fd_set.

    Macro: void FD_ZERO (fd_set *set)
    This macro initializes the file descriptor set set to be the empty set.

    Macro: void FD_SET (int filedes, fd_set *set)
    This macro adds filedes to the file descriptor set set.

    Macro: void FD_CLR (int filedes, fd_set *set)
    This macro removes filedes from the file descriptor set set.

    Macro: int FD_ISSET (int filedes, fd_set *set)
    This macro returns a nonzero value (true) if filedes is a member of the file descriptor set set, and zero (false) otherwise.

    Next, here is the description of the select function itself.

    Function: int select (int nfds, fd_set *read-fds, fd_set *write-fds, fd_set *except-fds, struct timeval *timeout)
    The select function blocks the calling process until there is activity on any of the specified sets of file descriptors, or until the timeout period has expired.

    The file descriptors specified by the read-fds argument are checked to see if they are ready for reading; the write-fds file descriptors are checked to see if they are ready for writing; and the except-fds file descriptors are checked for exceptional conditions. You can pass a null pointer for any of these arguments if you are not interested in checking for that kind of condition.

    A file descriptor is considered ready for reading if it is not at end of file. A server socket is considered ready for reading if there is a pending connection which can be accepted with accept; see section Accepting Connections. A client socket is ready for writing when its connection is fully established; see section Making a Connection.

    "Exceptional conditions" does not mean errors--errors are reported immediately when an erroneous system call is executed, and do not constitute a state of the descriptor. Rather, they include conditions such as the presence of an urgent message on a socket. (See section Sockets, for information on urgent messages.)

    The select function checks only the first nfds file descriptors. The usual thing is to pass FD_SETSIZE as the value of this argument.

    The timeout specifies the maximum time to wait. If you pass a null pointer for this argument, it means to block indefinitely until one of the file descriptors is ready. Otherwise, you should provide the time in struct timeval format; see section High-Resolution Calendar. Specify zero as the time (a struct timeval containing all zeros) if you want to find out which descriptors are ready without waiting if none are ready.

    The normal return value from select is the total number of ready file descriptors in all of the sets. Each of the argument sets is overwritten with information about the descriptors that are ready for the corresponding operation. Thus, to see if a particular descriptor desc has input, use FD_ISSET (desc, read-fds) after select returns.

    If select returns because the timeout period expires, it returns a value of zero.

    Any signal will cause select to return immediately. So if your program uses signals, you can't rely on select to keep waiting for the full time specified. If you want to be sure of waiting for a particular amount of time, you must check for EINTR and repeat the select with a newly calculated timeou