In modern C, header files are crucial tools that must be designed and used correctly. They allow the compiler to cross-check independently compiled parts of a program.
Headers declare types, functions, macros etc that are needed by the consumers of a set of facilities. All the code that uses any of those facilities includes the header. All the code that defines those facilities includes the header. This allows the compiler to check that the uses and definitions match.
There are a number of guidelines to follow when creating and using header files in a C project:
If a header file is included multiple times in a translation unit (TU), it should not break builds.
If you need the facilities declared in a header file, you should not have to include any other headers explicitly.
You should not be able to remove any information from a header without causing builds to fail.
Include What You Use (IWYU)
Of more concern to C++ than C, but nevertheless important in C too.
If the code in a TU (call it
code.c) directly uses the features
declared by a header (call it
#include "headerA.h" directly, even if the TU includes another
header (call it
"headerB.h") that happens, at the moment, to include
Occasionally, there might be good enough reasons to break one or more of these guidelines, but you should both be aware that you are breaking the rule and be aware of the consequences of doing so before you break it.
If a particular header file is included more than once in a translation unit (TU),
there should not be any compilation problems.
This is termed 'idempotence'; your headers should be idempotent.
Think how difficult life would be if you had to ensure that
#include <stdio.h> was only included once.
There are two ways to achieve idempotence: header guards and the
#pragma once directive.
Header guards are simple and reliable and conform to the C standard. The first non-comment lines in a header file should be of the form:
#ifndef UNIQUE_ID_FOR_HEADER #define UNIQUE_ID_FOR_HEADER
The last non-comment line should be
#endif, optionally with a comment after it:
#endif /* UNIQUE_ID_FOR_HEADER */
All the operational code, including other
#include directives, should
be between these lines.
Each name must be unique. Often, a name scheme such as
HEADER_H_INCLUDED is used.
Some older code uses a symbol defined as the header guard (e.g.
#ifndef BUFSIZ in
<stdio.h>), but it is not as reliable as a unique name.
One option would be to use a generated MD5 (or other) hash for the header guard name. You should avoid emulating the schemes used by system headers which frequently use names reserved to the implementation — names starting with an underscore followed by either another underscore or an upper-case letter.
Alternatively, some compilers support the
#pragma once directive which has the same effect as the three lines shown for header guards.
The compilers which support
#pragma once include MS Visual Studio and GCC and Clang.
However, if portability is a concern, it is better to use header guards, or use both.
Modern compilers (those supporting C89 or later) are required to ignore, without comment, pragmas that they do not recognize ('Any such pragma that is not recognized by the implementation is ignored') but old versions of GCC were not so indulgent.
Modern headers should be self-contained, which means that a program that needs to use the facilities defined by
header.h can include that header (
#include "header.h") and not worry about whether other headers need to be included first.
Historically, this has been a mildly contentious subject.
Once upon another millennium, the AT&T Indian Hill C Style and Coding Standards stated:
Header files should not be nested. The prologue for a header file should, therefore, describe what other headers need to be
#included for the header to be functional. In extreme cases, where a large number of header files are to be included in several different source files, it is acceptable to put all common
#includes in one include file.
This is the antithesis of self-containment.
However, since then, opinion has tended in the opposite direction.
If a source file needs to use the facilities declared by a header
header.h, the programmer should be able to write:
and (subject only to having the correct search paths set on the command
line), any necessary pre-requisite headers will be included by
header.h without needing any further headers added to the source file.
This provides better modularity for the source code. It also protects the source from the "guess why this header was added" conundrum that arises after the code has been modified and hacked for a decade or two.
The NASA Goddard Space Flight Center (GSFC) coding standards for C is one of the more modern standards — but is now a little hard to track down. It states that headers should be self-contained. It also provides a simple way to ensure that headers are self-contained: the implementation file for the header should include the header as the first header. If it is not self-contained, that code will not compile.
The rationale given by GSFC includes:
§2.1.1 Header include rationale
This standard requires a unit’s header to contain
#includestatements for all other headers required by the unit header. Placing
#includefor the unit header first in the unit body allows the compiler to verify that the header contains all required
An alternate design, not permitted by this standard, allows no
#includestatements in headers; all
#includesare done in the body files. Unit header files then must contain #ifdef statements that check that the required headers are included in the proper order.
One advantage of the alternate design is that the
#includelist in the body file is exactly the dependency list needed in a makefile, and this list is checked by the compiler. With the standard design, a tool must be used to generate the dependency list. However, all of the branch recommended development environments provide such a tool.
A major disadvantage of the alternate design is that if a unit’s required header list changes, each file that uses that unit must be edited to update the
#includestatement list. Also, the required header list for a compiler library unit may be different on different targets.
Another disadvantage of the alternate design is that compiler library header files, and other third party files, must be modified to add the required
Thus, self-containment means that:
header.hneeds a new nested header
extra.h, you do not have to check every source file that uses
header.hto see whether you need to add
header.hno longer needs to include a specific header
notneeded.h, you do not have to check every source file that uses
header.hto see whether you can safely remove
notneeded.h(but see Include what you use.
See Linking against a static
library for a script
chkhdr that can be used to test idempotence and self-containment of
a header file.
Headers are a crucial consistency checking mechanism, but they should be as small as possible. In particular, that means that a header should not include other headers just because the implementation file will need the other headers. A header should contain only those headers necessary for a consumer of the services described.
For example, a project header should not include
<stdio.h> unless one
of the function interfaces uses the type
FILE * (or one of the other
types defined solely in
If an interface uses
size_t, the smallest header that suffices is
Obviously, if another header that defines
size_t is included, there is
no need to include
If the headers are minimal, then it keeps the compilation time to a minimum too.
It is possible to devise headers whose sole purpose is to include a lot
of other headers.
These seldom turn out to be a good idea in the long run because few
source files will actually need all the facilities described by all the
For example, a
<standard-c.h> could be devised that includes all the
standard C headers — with care since some headers are not always
However, very few programs actually use the facilities of
Google's Include What You Use project, or IWYU, ensures source files include all headers used in the code.
Suppose a source file
source.c includes a header
arbitrary.h which in turn coincidentally includes
freeloader.h, but the source file also explicitly and independently uses the facilities from
freeloader.h. All is well to start with. Then one day
arbitrary.h is changed so its clients no longer need the facilities of
source.c stops compiling — because it didn't meet the IWYU criteria. Because the code in
source.c explicitly used the facilities of
freeloader.h, it should have included what it uses — there should have been an explicit
#include "freeloader.h" in the source too. (Idempotency would have ensured there wasn't a problem.)
The IWYU philosophy maximizes the probability that code continues to compile even with reasonable changes made to interfaces. Clearly, if your code calls a function that is subsequently removed from the published interface, no amount of preparation can prevent changes becoming necessary. This is why changes to APIs are avoided when possible, and why there are deprecation cycles over multiple releases, etc.
This is a particular problem in C++ because standard headers are allowed to include each other. Source file
file.cpp could include one header
header1.h that on one platform includes another header
file.cpp might turn out to use the facilities of
header2.h as well. This wouldn't be a problem initially - the code would compile because
header2.h. On another platform, or an upgrade of the current platform,
header1.h could be revised so it no longer includes
header2.h, and then
file.cpp would stop compiling as a result.
IWYU would spot the problem and recommend that
header2.h be included directly in
file.cpp. This would ensure it continues to compile. Analogous considerations apply to C code too.
The C standard says that there is very little difference between the
#include <header.h> and
#include "header.h" notations.
#include <header.h>] searches a sequence of implementation-defined places for a header identified uniquely by the specified sequence between the
>delimiters, and causes the replacement of that directive by the entire contents of the header. How the places are specified or the header identified is implementation-defined.
#include "header.h"] causes the replacement of that directive by the entire contents of the source file identified by the specified sequence between the
"…"delimiters. The named source file is searched for in an implementation-defined manner. If this search is not supported, or if the search fails, the directive is reprocessed as if it read [
#include <header.h>] …
So, the double quoted form may look in more places than the angle-bracketed form. The standard specifies by example that the standard headers should be included in angle-brackets, even though the compilation works if you use double quotes instead. Similarly, standards such as POSIX use the angle-bracketed format — and you should too. Reserve double-quoted headers for headers defined by the project. For externally-defined headers (including headers from other projects your project relies on), the angle-bracket notation is most appropriate.
Note that there should be a space between
#include and the header,
even though the compilers will accept no space there.
Spaces are cheap.
A number of projects use a notation such as:
#include <openssl/ssl.h> #include <sys/stat.h> #include <linux/kernel.h>
You should consider whether to use that namespace control in your
project (it is quite probably a good idea).
You should steer clear of the names used by existing projects (in
linux would be bad choices).
If you use this, your code should be careful and consistent in the use of the notation.
Do not use
#include "../include/header.h" notation.
Header files should seldom if ever define variables. Although you will keep global variables to a minimum, if you need a global variable, you will declare it in a header, and define it in one suitable source file, and that source file will include the header to cross-check the declaration and definition, and all source files that use the variable will use the header to declare it.
Corollary: you will not declare global variables in a source file — a source file will only contain definitions.
Header files should seldom declare
static functions, with the notable
static inline functions which will be defined in headers
if the function is needed in more than one source file.
externever useful in C99?
externto share variables between source files?
"../include/header.h"for a header?