![[image of the Head of a GNU]](/graphics/gnu-head-sm.jpg)
(jpeg 7k)
(jpeg 21k)
no gifs due to patent problems
(jpeg 7k)
(jpeg 21k)
no gifs due to patent problems
The GNU Coding Standards were written by Richard Stallman and other GNU Project volunteers. Their purpose is to make the GNU system clean, consistent, and easy to install. This document can also be read as a guide to writing portable, robust and reliable programs. It focuses on programs written in C, but many of the rules and principles are useful even if you write in another programming language. The rules often state reasons for writing in a certain way.
Corrections or suggestions for this document should be sent to gnu@gnu.org. If you make a suggestion, please include a suggested new wording for it; our time is limited. We prefer a context diff to the `standards.texi' or `make-stds.texi' files, but if you don't have those files, please mail your suggestion anyway.
This release of the GNU Coding Standards was last updated June 24, 1999.
This node discusses how you can make sure that GNU software remains unencumbered.
Don't in any circumstances refer to Unix source code for or during your work on GNU! (Or to any other proprietary programs.)
If you have a vague recollection of the internals of a Unix program, this does not absolutely mean you can't write an imitation of it, but do try to organize the imitation internally along different lines, because this is likely to make the details of the Unix version irrelevant and dissimilar to your results.
For example, Unix utilities were generally optimized to minimize memory use; if you go for speed instead, your program will be very different. You could keep the entire input file in core and scan it there instead of using stdio. Use a smarter algorithm discovered more recently than the Unix program. Eliminate use of temporary files. Do it in one pass instead of two (we did this in the assembler).
Or, on the contrary, emphasize simplicity instead of speed. For some applications, the speed of today's computers makes simpler algorithms adequate.
Or go for generality. For example, Unix programs often have static tables or fixed-size strings, which make for arbitrary limits; use dynamic allocation instead. Make sure your program handles NULs and other funny characters in the input files. Add a programming language for extensibility and write part of the program in that language.
Or turn some parts of the program into independently usable libraries. Or use a simple garbage collector instead of tracking precisely when to free memory, or use a new GNU facility such as obstacks.
If the program you are working on is copyrighted by the Free Software Foundation, then when someone else sends you a piece of code to add to the program, we need legal papers to use it--just as we asked you to sign papers initially. Each person who makes a nontrivial contribution to a program must sign some sort of legal papers in order for us to have clear title to the program; the main author alone is not enough.
So, before adding in any contributions from other people, please tell us, so we can arrange to get the papers. Then wait until we tell you that we have received the signed papers, before you actually use the contribution.
This applies both before you release the program and afterward. If you receive diffs to fix a bug, and they make significant changes, we need legal papers for that change.
This also applies to comments and documentation files. For copyright law, comments and code are just text. Copyright applies to all kinds of text, so we need legal papers for all kinds.
We know it is frustrating to ask for legal papers; it's frustrating for us as well. But if you don't wait, you are going out on a limb--for example, what if the contributor's employer won't sign a disclaimer? You might have to take that code out again!
You don't need papers for changes of a few lines here or there, since they are not significant for copyright purposes. Also, you don't need papers if all you get from the suggestion is some ideas, not actual code which you use. For example, if someone send you one implementation, but you write a different implementation of the same idea, you don't need to get papers.
The very worst thing is if you forget to tell us about the other contributor. We could be very embarrassed in court some day as a result.
We have more detailed advice for maintainers of programs; if you have reached the stage of actually maintaining a program for GNU (whether released or not), please ask us for a copy.
This node discusses some of the issues you should take into account when designing your program.
With occasional exceptions, utility programs and libraries for GNU should be upward compatible with those in Berkeley Unix, and upward compatible with ANSI C if ANSI C specifies their behavior, and upward compatible with POSIX if POSIX specifies their behavior.
When these standards conflict, it is useful to offer compatibility modes for each of them.
ANSI C and POSIX prohibit many kinds of extensions. Feel free to make the extensions anyway, and include a `--ansi', `--posix', or `--compatible' option to turn them off. However, if the extension has a significant chance of breaking any real programs or scripts, then it is not really upward compatible. Try to redesign its interface.
Many GNU programs suppress extensions that conflict with POSIX if the
environment variable POSIXLY_CORRECT is defined (even if it is
defined with a null value). Please make your program recognize this
variable if appropriate.
When a feature is used only by users (not by programs or command
files), and it is done poorly in Unix, feel free to replace it
completely with something totally different and better. (For example,
vi is replaced with Emacs.) But it is nice to offer a compatible
feature as well. (There is a free vi clone, so we offer it.)
Additional useful features not in Berkeley Unix are welcome.
Many GNU facilities that already exist support a number of convenient extensions over the comparable Unix facilities. Whether to use these extensions in implementing your program is a difficult question.
On the one hand, using the extensions can make a cleaner program. On the other hand, people will not be able to build the program unless the other GNU tools are available. This might cause the program to work on fewer kinds of machines.
With some extensions, it might be easy to provide both alternatives.
For example, you can define functions with a "keyword" INLINE
and define that as a macro to expand into either inline or
nothing, depending on the compiler.
In general, perhaps it is best not to use the extensions if you can straightforwardly do without them, but to use the extensions if they are a big improvement.
An exception to this rule are the large, established programs (such as Emacs) which run on a great variety of systems. Such programs would be broken by use of GNU extensions.
Another exception is for programs that are used as part of compilation: anything that must be compiled with other compilers in order to bootstrap the GNU compilation facilities. If these require the GNU compiler, then no one can compile them without having them installed already. That would be no good.
Do not ever use the "trigraph" feature of ANSI C.
ANSI C is widespread enough now that it is ok to write new programs that use ANSI C features (and therefore will not work in non-ANSI compilers). And if a program is already written in ANSI C, there's no need to convert it to support non-ANSI compilers.
If you don't know non-ANSI C, there's no need to learn it; just write in ANSI C.
However, it is easy to support non-ANSI compilers in most programs, so you might still consider doing so when you write a program. And if a program you are maintaining has such support, you should try to keep it working.
To support pre-ANSI C, instead of writing function definitions in ANSI prototype form,
int foo (int x, int y) ...
write the definition in pre-ANSI style like this,
int
foo (x, y)
int x, y;
...
and use a separate declaration to specify the argument prototype:
int foo (int, int);
You need such a declaration anyway, in a header file, to get the benefit of ANSI C prototypes in all the files where the function is called. And once you have the declaration, you normally lose nothing by writing the function definition in the pre-ANSI style.
This technique does not work for integer types narrower than int.
If you think of an argument as being of a type narrower than int,
declare it as int instead.
There are a few special cases where this technique is hard to use. For
example, if a function argument needs to hold the system type
dev_t, you run into trouble, because dev_t is shorter than
int on some machines; but you cannot use int instead,
because dev_t is wider than int on some machines. There
is no type you can safely use on all machines in a non-ANSI
definition. The only way to support non-ANSI C and pass such an
argument is to check the width of dev_t using Autoconf and choose
the argument type accordingly. This may not be worth the trouble.
Using a language other than C is like using a non-standard feature: it will cause trouble for users. Even if GCC supports the other language, users may find it inconvenient to have to install the compiler for that other language in order to build your program. For example, if you write your program in C++, people will have to install the C++ compiler in order to compile your program. Thus, it is better if you write in C.
But there are three situations when there is no disadvantage in using some other language:
C has one other advantage over C++ and other compiled languages: more people know C, so more people will find it easy to read and modify the program if it is written in C.
This node describes how to write robust software. It also describes general standards for error messages, the command line interface, and how libraries should behave.
Avoid arbitrary limits on the length or number of any data structure, including file names, lines, files, and symbols, by allocating all data structures dynamically. In most Unix utilities, "long lines are silently truncated". This is not acceptable in a GNU utility.
Utilities reading files should not drop NUL characters, or any other nonprinting characters including those with codes above 0177. The only sensible exceptions would be utilities specifically intended for interface to certain types of terminals or printers that can't handle those characters. Whenever possible, try to make programs work properly with sequences of bytes that represent multibyte characters, using encodings such as UTF-8 and others.
Check every system call for an error return, unless you know you wish to
ignore errors. Include the system error text (from perror or
equivalent) in every error message resulting from a failing
system call, as well as the name of the file if any and the name of the
utility. Just "cannot open foo.c" or "stat failed" is not
sufficient.
Check every call to malloc or realloc to see if it
returned zero. Check realloc even if you are making the block
smaller; in a system that rounds block sizes to a power of 2,
realloc may get a different block if you ask for less space.
In Unix, realloc can destroy the storage block if it returns
zero. GNU realloc does not have this bug: if it fails, the
original block is unchanged. Feel free to assume the bug is fixed. If
you wish to run your program on Unix, and wish to avoid lossage in this
case, you can use the GNU malloc.
You must expect free to alter the contents of the block that was
freed. Anything you want to fetch from the block, you must fetch before
calling free.
If malloc fails in a noninteractive program, make that a fatal
error. In an interactive program (one that reads commands from the
user), it is better to abort the command and return to the command
reader loop. This allows the user to kill other processes to free up
virtual memory, and then try the command again.
Use getopt_long to decode arguments, unless the argument syntax
makes this unreasonable.
When static storage is to be written in during program execution, use explicit C code to initialize it. Reserve C initialized declarations for data that will not be changed.
Try to avoid low-level interfaces to obscure Unix data structures (such
as file directories, utmp, or the layout of kernel memory), since these
are less likely to work compatibly. If you need to find all the files
in a directory, use readdir or some other high-level interface.
These are supported compatibly by GNU.
The preferred signal handling facilities are the BSD variant of
signal, and the POSIX sigaction function; the
alternative USG signal interface is an inferior design.
Nowadays, using the POSIX signal functions may be the easiest way
to make a program portable. If you use signal, then on GNU/Linux
systems running GNU libc version 1, you should include
`bsd/signal.h' instead of `signal.h', so as to get BSD
behavior. It is up to you whether to support systems where
signal has only the USG behavior, or give up on them.
In error checks that detect "impossible" conditions, just abort. There is usually no point in printing any message. These checks indicate the existence of bugs. Whoever wants to fix the bugs will have to read the source code and run a debugger. So explain the problem with comments in the source. The relevant data will be in variables, which are easy to examine with the debugger, so there is no point moving them elsewhere.
Do not use a count of errors as the exit status for a program. That does not work, because exit status values are limited to 8 bits (0 through 255). A single run of the program might have 256 errors; if you try to return 256 as the exit status, the parent process will see 0 as the status, and it will appear that the program succeeded.
If you make temporary files, check the TMPDIR environment
variable; if that variable is defined, use the specified directory
instead of `/tmp'.
Try to make library functions reentrant. If they need to do dynamic
storage allocation, at least try to avoid any nonreentrancy aside from
that of malloc itself.
Here are certain name conventions for libraries, to avoid name conflicts.
Choose a name prefix for the library, more than two characters long. All external function and variable names should start with this prefix. In addition, there should only be one of these in any given library member. This usually means putting each one in a separate source file.
An exception can be made when two external symbols are always used together, so that no reasonable program could use one without the other; then they can both go in the same file.
External symbols that are not documented entry points for the user should have names beginning with `_'. They should also contain the chosen name prefix for the library, to prevent collisions with other libraries. These can go in the same files with user entry points if you like.
Static functions and variables can be used as you like and need not fit any naming convention.
Error messages from compilers should look like this:
source-file-name:lineno: message
If you want to mention the column number, use this format:
source-file-name:lineno:column: message
Line numbers should start from 1 at the beginning of the file, and column numbers should start from 1 at the beginning of the line. (Both of these conventions are chosen for compatibility.) Calculate column numbers assuming that space and all ASCII printing characters have equal width, and assuming tab stops every 8 columns.
Error messages from other noninteractive programs should look like this:
program:source-file-name:lineno: message
when there is an appropriate source file, or like this:
program: message
when there is no relevant source file.
If you want to mention the column number, use this format:
program:source-file-name:lineno:column: message
In an interactive program (one that is reading commands from a terminal), it is better not to include the program name in an error message. The place to indicate which program is running is in the prompt or with the screen layout. (When the same program runs with input from a source other than a terminal, it is not interactive and would do best to print error messages using the noninteractive style.)
The string message should not begin with a capital letter when it follows a program name and/or file name. Also, it should not end with a period.
Error messages from interactive programs, and other messages such as usage messages, should start with a capital letter. But they should not end with a period.
Please don't make the behavior of a utility depend on the name used to invoke it. It is useful sometimes to make a link to a utility with a different name, and that should not change what it does.
Instead, use a run time option or a compilation switch or both to select among the alternate behaviors.
Likewise, please don't make the behavior of the program depend on the type of output device it is used with. Device independence is an important principle of the system's design; do not compromise it merely to save someone from typing an option now and then. (Variation in error message syntax when using a terminal is ok, because that is a side issue that people do not depend on.)
If you think one behavior is most useful when the output is to a terminal, and another is most useful when the output is a file or a pipe, then it is usually best to make the default behavior the one that is useful with output to a terminal, and have an option for the other behavior.
Compatibility requires certain programs to depend on the type of output
device. It would be disastrous if ls or sh did not do so
in the way all users expect. In some of these cases, we supplement the
program with a preferred alternate version that does not depend on the
output device type. For example, we provide a dir program much
like ls except that its default output format is always
multi-column format.
It is a good idea to follow the POSIX guidelines for the
command-line options of a program. The easiest way to do this is to use
getopt to parse them. Note that the GNU version of getopt
will normally permit options anywhere among the arguments unless the
special argument `--' is used. This is not what POSIX
specifies; it is a GNU extension.
Please define long-named options that are equivalent to the
single-letter Unix-style options. We hope to make GNU more user
friendly this way. This is easy to do with the GNU function
getopt_long.
One of the advantages of long-named options is that they can be consistent from program to program. For example, users should be able to expect the "verbose" option of any GNU program which has one, to be spelled precisely `--verbose'. To achieve this uniformity, look at the table of common long-option names when you choose the option names for your program (see section Table of Long Options).
It is usually a good idea for file names given as ordinary arguments to be input files only; any output files would be specified using options (preferably `-o' or `--output'). Even if you allow an output file name as an ordinary argument for compatibility, try to provide an option as another way to specify it. This will lead to more consistency among GNU utilities, and fewer idiosyncracies for users to remember.
All programs should support two standard options: `--version' and `--help'.
--version
GNU Emacs 19.30The program's name should be a constant string; don't compute it from
argv[0]. The idea is to state the standard or canonical
name for the program, not its file name. There are other ways to find
out the precise file name where a command is found in PATH.
If the program is a subsidiary part of a larger package, mention the
package name in parentheses, like this:
emacsserver (GNU Emacs) 19.30If the package has a version number which is different from this program's version number, you can mention the package version number just before the close-parenthesis. If you need to mention the version numbers of libraries which are distributed separately from the package which contains this program, you can do so by printing an additional line of version info for each library you want to mention. Use the same format for these lines as for the first line. Please do not mention all of the libraries that the program uses "just for completeness"---that would produce a lot of unhelpful clutter. Please mention library version numbers only if you find in practice that they are very important to you in debugging. The following line, after the version number line or lines, should be a copyright notice. If more than one copyright notice is called for, put each on a separate line. Next should follow a brief statement that the program is free software, and that users are free to copy and change it on certain conditions. If the program is covered by the GNU GPL, say so here. Also mention that there is no warranty, to the extent permitted by law. It is ok to finish the output with a list of the major authors of the program, as a way of giving credit. Here's an example of output that follows these rules:
GNU Emacs 19.34.5 Copyright (C) 1996 Free Software Foundation, Inc. GNU Emacs comes with NO WARRANTY, to the extent permitted by law. You may redistribute copies of GNU Emacs under the terms of the GNU General Public License. For more information about these matters, see the files named COPYING.You should adapt this to your program, of course, filling in the proper year, copyright holder, name of program, and the references to distribution terms, and changing the rest of the wording as necessary. This copyright notice only needs to mention the most recent year in which changes were made--there's no need to list the years for previous versions' changes. You don't have to mention the name of the program in these notices, if that is inconvenient, since it appeared in the first line.
--help
Report bugs to mailing-address.
Here is a table of long options used by GNU programs. It is surely incomplete, but we aim to list all the options that a new program might want to be compatible with. If you use names not already in the table, please send gnu@gnu.org a list of them, with their meanings, so we can update the table.
tar.
du, ls, nm, stty, uname,
and unexpand.
diff.
ls.
etags, tee, time;
`-r' in tar.
cp.
shar.
m4.
diff.
gawk.
recode.
wdiff.
ptx.
wdiff.
ctags.
shar.
tac.
cpio and diff.
shar.
cpio and tar.
head and tail.
ptx.
head, split, and tail.
etags.
tar.
chgrp and chown.
ls.
recode.
su;
`-x' in GDB.
tar.
gawk.
tar and shar.
tar.
tar.
diff.
gawk.
ptx, recode, and wdiff;
`-W copyright' in gawk.
who.
du.
tar and cpio.
shar.
ctags.
touch.
m4;
`-t' in Bison.
m4.
ctags.
tar.
chgrp, chown, cpio, du,
ls, and tar.
du.
recode.
look.
tar.
csplit.
ls, it
means to show directories themselves rather than their contents. In
rm and ln, it means to not treat links to directories
specially.
strip.
strip.
diff.
csplit.
wdiff.
wdiff.
diff.
xargs.
makeinfo.
m4.
ls.
tar.
xargs.
unshar.
diff.
sed.
nm.
cpio;
`-x' in tar.
finger.
su.
m4.
info, gawk, Make, mt, and tar;
`-n' in sed;
`-r' in touch.
gawk.
ls.
tar.
makeinfo.
ptx.
tail.
makeinfo.
cp, ln, mv, and rm.
shar.
ls, time, and ptx.
m4.
ptx.
tar.
ul.
recode.
install.
tar and shar.
m4.
objdump and recode
who.
shar.
ls.
who.
diff.
ls;
`-x' in recode.
diff.
ls.
diff.
look and ptx;
`-i' in diff and wdiff.
ptx.
etags.
tee.
diff.
diff.
tar.
etags;
`-I' in m4.
tar.
expand.
diff.
ls.
cp, ln, mv, rm;
`-e' in m4;
`-p' in xargs;
`-w' in tar.
shar.
csplit.
du and ls.
etags.
wdiff.
shar.
split.
split, head, and tail.
cpio.
gawk.
cpio;
`-l' in recode.
tar.
ls.
su.
ptx.
hello and uname.
cpio.
xargs.
xargs.
xargs.
xargs.
who.
who.
diff.
shar.
install, mkdir, and mkfifo.
tar.
tar.
m4.
shar.
shar.
shar.
wdiff.
touch.
etags.
wdiff.
cp.
wdiff.
shar.
gprof.
etags.
nm.
makeinfo.
gprof.
gprof.
shar.
makeinfo.
emacsclient.
info.
uname.
cpio.
objdump.
xargs.
cat.
cat.
nm.
cpio and ls.
tar.
tar, cp, and du.
ptx.
gprof.
gprof.
getopt, fdlist, fdmount,
fdmountd, and fdumount.
shar.
rm.
unshar.
install.
diff.
makeinfo.
mkdir and rmdir.
ul.
cpio.
finger.
cpio and tar.
gawk.
m4.
csplit.
tar and cp.
su.
cpio.
tar.
tar.
diff.
cmp.
nm.
nm.
wdiff.
ed.
shar.
shar
ls.
diff.
gawk.
tar.
tar.
chgrp, chown, cp, ls, diff,
and rm.
makeinfo.
ptx.
tac and etags.
uname.
m4.
objdump.
cpio.
xargs.
diff.
cpio.
ls and nm.
diff.
ptx.
tar.
tar.
stty.
ptx.
du.
tac.
recode to chose files or pipes for sequencing passes.
su.
cat.
diff.
cat.
diff.
cat.
ls.
ls.
gawk.
tar.
diff.
unshar.
shar.
cat.
wdiff.
wdiff.
tar and diff to specify which file within
a directory to start processing with.
wdiff.
shar.
recode.
install.
strip.
strip.
shar.
cp, ln, mv.
csplit.
gprof.
du.
ln.
objdump.
m4.
uname.
expand and unexpand.
ls.
tput and ul.
`-t' in wdiff.
diff.
shar.
ls and touch.
tar.
du.
ranlib, and recode.
m4.
hello;
`-W traditional' in gawk;
`-G' in ed, m4, and ptx.
ctags.
ctags.
ptx.
tar.
cpio.
m4.
nm.
cp, ctags, mv, tar.
gawk; same as `--help'.
shar.
shar.
tar.
cp, ln, mv.
ctags.
tar.
shar.
ls and ptx.
ptx.
who.
gprof.
If it typically uses just a few meg of memory, don't bother making any effort to reduce memory usage. For example, if it is impractical for other reasons to operate on files more than a few meg long, it is reasonable to read entire input files into core to operate on them.
However, for programs such as cat or tail, that can
usefully operate on very large files, it is important to avoid using a
technique that would artificially limit the size of files it can handle.
If a program works by lines and could be applied to arbitrary
user-supplied input files, it should keep only a line in memory, because
this is not very hard and users will want to be able to operate on input
files that are bigger than will fit in core all at once.
If your program creates complicated data structures, just make them in
core and give a fatal error if malloc returns zero.
This node provides advice on how best to use the C language when writing GNU software.
It is important to put the open-brace that starts the body of a C function in column zero, and avoid putting any other open-brace or open-parenthesis or open-bracket in column zero. Several tools look for open-braces in column zero to find the beginnings of C functions. These tools will not work on code not formatted that way.
It is also important for function definitions to start the name of the function in column zero. This helps people to search for function definitions, and may also help certain tools recognize them. Thus, the proper format is this:
static char *
concat (s1, s2) /* Name starts in column zero here */
char *s1, *s2;
{ /* Open brace in column zero here */
...
}
or, if you want to use ANSI C, format the definition like this:
static char *
concat (char *s1, char *s2)
{
...
}
In ANSI C, if the arguments don't fit nicely on one line, split it like this:
int
lots_of_args (int an_integer, long a_long, short a_short,
double a_double, float a_float)
...
For the body of the function, we prefer code formatted like this:
if (x < foo (y, z))
haha = bar[4] + 5;
else
{
while (z)
{
haha += foo (z, z);
z--;
}
return ++x + bar ();
}
We find it easier to read a program when it has spaces before the open-parentheses and after the commas. Especially after the commas.
When you split an expression into multiple lines, split it before an operator, not after one. Here is the right way:
if (foo_this_is_long && bar > win (x, y, z)
&& remaining_condition)
Try to avoid having two operators of different precedence at the same level of indentation. For example, don't write this:
mode = (inmode[j] == VOIDmode
|| GET_MODE_SIZE (outmode[j]) > GET_MODE_SIZE (inmode[j])
? outmode[j] : inmode[j]);
Instead, use extra parentheses so that the indentation shows the nesting:
mode = ((inmode[j] == VOIDmode
|| (GET_MODE_SIZE (outmode[j]) > GET_MODE_SIZE (inmode[j])))
? outmode[j] : inmode[j]);
Insert extra parentheses so that Emacs will indent the code properly. For example, the following indentation looks nice if you do it by hand, but Emacs would mess it up:
v = rup->ru_utime.tv_sec*1000 + rup->ru_utime.tv_usec/1000
+ rup->ru_stime.tv_sec*1000 + rup->ru_stime.tv_usec/1000;
But adding a set of parentheses solves the problem:
v = (rup->ru_utime.tv_sec*1000 + rup->ru_utime.tv_usec/1000
+ rup->ru_stime.tv_sec*1000 + rup->ru_stime.tv_usec/1000);
Format do-while statements like this:
do
{
a = foo (a);
}
while (a > 0);
Please use formfeed characters (control-L) to divide the program into pages at logical places (but not within a function). It does not matter just how long the pages are, since they do not have to fit on a printed page. The formfeeds should appear alone on lines by themselves.
Every program should start with a comment saying briefly what it is for. Example: `fmt - filter for simple filling of text'.
Please write the comments in a GNU program in English, because English is the one language that nearly all programmers in all countries can read. If you do not write English well, please write comments in English as well as you can, then ask other people to help rewrite them. If you can't write comments in English, please find someone to work with you and translate your comments into English.
Please put a comment on each function saying what the function does,
what sorts of arguments it gets, and what the possible values of
arguments mean and are used for. It is not necessary to duplicate in
words the meaning of the C argument declarations, if a C type is being
used in its customary fashion. If there is anything nonstandard about
its use (such as an argument of type char * which is really the
address of the second character of a string, not the first), or any
possible values that would not work the way one would expect (such as,
that strings containing newlines are not guaranteed to work), be sure
to say so.
Also explain the significance of the return value, if there is one.
Please put two spaces after the end of a sentence in your comments, so that the Emacs sentence commands will work. Also, please write complete sentences and capitalize the first word. If a lower-case identifier comes at the beginning of a sentence, don't capitalize it! Changing the spelling makes it a different identifier. If you don't like starting a sentence with a lower case letter, write the sentence differently (e.g., "The identifier lower-case is ...").
The comment on a function is much clearer if you use the argument names to speak about the argument values. The variable name itself should be lower case, but write it in upper case when you are speaking about the value rather than the variable itself. Thus, "the inode number NODE_NUM" rather than "an inode".
There is usually no purpose in restating the name of the function in the comment before it, because the reader can see that for himself. There might be an exception when the comment is so long that the function itself would be off the bottom of the screen.
There should be a comment on each static variable as well, like this:
/* Nonzero means truncate lines in the display; zero means continue them. */ int truncate_lines;
Every `#endif' should have a comment, except in the case of short conditionals (just a few lines) that are not nested. The comment should state the condition of the conditional that is ending, including its sense. `#else' should have a comment describing the condition and sense of the code that follows. For example:
#ifdef foo ... #else /* not foo */ ... #endif /* not foo */ #ifdef foo ... #endif /* foo */
but, by contrast, write the comments this way for a `#ifndef':
#ifndef foo ... #else /* foo */ ... #endif /* foo */ #ifndef foo ... #endif /* not foo */
Please explicitly declare all arguments to functions.
Don't omit them just because they are ints.
Declarations of external functions and functions to appear later in the
source file should all go in one place near the beginning of the file
(somewhere before the first function definition in the file), or else
should go in a header file. Don't put extern declarations inside
functions.
It used to be common practice to use the same local variables (with
names like tem) over and over for different values within one
function. Instead of doing this, it is better declare a separate local
variable for each distinct purpose, and give it a name which is
meaningful. This not only makes programs easier to understand, it also
facilitates optimization by good compilers. You can also move the
declaration of each local variable into the smallest scope that includes
all its uses. This makes the program even cleaner.
Don't use local variables or parameters that shadow global identifiers.
Don't declare multiple variables in one declaration that spans lines. Start a new declaration on each line, instead. For example, instead of this:
int foo,
bar;
write either this:
int foo, bar;
or this:
int foo; int bar;
(If they are global variables, each should have a comment preceding it anyway.)
When you have an if-else statement nested in another
if statement, always put braces around the if-else.
Thus, never write like this:
if (foo)
if (bar)
win ();
else
lose ();
always like this:
if (foo)
{
if (bar)
win ();
else
lose ();
}
If you have an if statement nested inside of an else
statement, either write else if on one line, like this,
if (foo) ... else if (bar) ...
with its then-part indented like the preceding then-part,
or write the nested if within braces like this:
if (foo)
...
else
{
if (bar)
...
}
Don't declare both a structure tag and variables or typedefs in the same declaration. Instead, declare the structure tag separately and then use it to declare the variables or typedefs.
Try to avoid assignments inside if-conditions. For example,
don't write this:
if ((foo = (char *) malloc (sizeof *foo)) == 0)
fatal ("virtual memory exhausted");
instead, write this:
foo = (char *) malloc (sizeof *foo);
if (foo == 0)
fatal ("virtual memory exhausted");
Don't make the program ugly to placate lint. Please don't insert any
casts to void. Zero without a cast is perfectly fine as a null
pointer constant, except when calling a varargs function.
The names of global variables and functions in a program serve as comments of a sort. So don't choose terse names--instead, look for names that give useful information about the meaning of the variable or function. In a GNU program, names should be English, like other comments.
Local variable names can be shorter, because they are used only within one context, where (presumably) comments explain their purpose.
Try to limit your use of abbreviations in symbol names. It is ok to make a few abbreviations, explain what they mean, and then use them frequently, but don't use lots of obscure abbreviations.
Please use underscores to separate words in a name, so that the Emacs
word commands can be useful within them. Stick to lower case; reserve
upper case for macros and enum constants, and for name-prefixes
that follow a uniform convention.
For example, you should use names like ignore_space_change_flag;
don't use names like iCantReadThis.
Variables that indicate whether command-line options have been specified should be named after the meaning of the option, not after the option-letter. A comment should state both the exact meaning of the option and its letter. For example,
/* Ignore changes in horizontal whitespace (-b). */ int ignore_space_change_flag;
When you want to define names with constant integer values, use
enum rather than `#define'. GDB knows about enumeration
constants.
Use file names of 14 characters or less, to avoid creating gratuitous
problems on older System V systems. You can use the program
doschk to test for this. doschk also tests for potential
name conflicts if the files were loaded onto an MS-DOS file
system--something you may or may not care about.
In the Unix world, "portability" refers to porting to different Unix versions. For a GNU program, this kind of portability is desirable, but not paramount.
The primary purpose of GNU software is to run on top of the GNU kernel, compiled with the GNU C compiler, on various types of CPU. The amount and kinds of variation among GNU systems on different CPUs will be comparable to the variation among Linux-based GNU systems or among BSD systems today. So the kinds of portability that are absolutely necessary are quite limited.
But many users do run GNU software on non-GNU Unix or Unix-like systems. So supporting a variety of Unix-like systems is desirable, although not paramount.
The easiest way to achieve portability to most Unix-like systems is to use Autoconf. It's unlikely that your program needs to know more information about the host platform than Autoconf can provide, simply because most of the programs that need such knowledge have already been written.
Avoid using the format of semi-internal data bases (e.g., directories)
when there is a higher-level alternative (readdir).
As for systems that are not like Unix, such as MSDOS, Windows, the Macintosh, VMS, and MVS, supporting them is often a lot of work. When that is the case, it is better to spend your time adding features that will be useful on GNU and GNU/Linux, rather than on supporting other incompatible systems.
Even GNU systems will differ because of differences among CPU
types--for example, difference in byte ordering and alignment
requirements. It is absolutely essential to handle these differences.
However, don't make any effort to cater to the possibility that an
int will be less than 32 bits. We don't support 16-bit machines
in GNU.
Don't assume that the address of an int object is also the
address of its least-significant byte. This is false on big-endian
machines. Thus, don't make the following mistake:
int c; ... while ((c = getchar()) != EOF) write(file_descriptor, &c, 1);
When calling functions, you need not worry about the difference between pointers of various types, or between pointers and integers. On most machines, there's no difference anyway. As for the few machines where there is a difference, all of them support ANSI C, so you can use prototypes (conditionalized to be active only in ANSI C) to make the code work on those systems.
In certain cases, it is ok to pass integer and pointer arguments
indiscriminately to the same function, and use no prototype on any
system. For example, many GNU programs have error-reporting functions
that pass their arguments along to printf and friends:
error (s, a1, a2, a3)
char *s;
char *a1, *a2, *a3;
{
fprintf (stderr, "error: ");
fprintf (stderr, s, a1, a2, a3);
}
In practice, this works on all machines, since a pointer is generally the widest possible kind of argument, and it is much simpler than any "correct" alternative. Be sure not to use a prototype for such functions.
However, avoid casting pointers to integers unless you really need to. Outside of special situations, such casts greatly reduce portability, and in most programs they are easy to avoid. In the cases where casting pointers to integers is essential--such as, a Lisp interpreter which stores type information as well as an address in one word--it is ok to do it, but you'll have to make explicit provisions to handle different word sizes.
C implementations differ substantially. ANSI C reduces but does not eliminate the incompatibilities; meanwhile, many users wish to compile GNU software with pre-ANSI compilers. This chapter gives recommendations for how to use the more or less standard C library functions to avoid unnecessary loss of portability.
sprintf. It returns the number of
characters written on some systems, but not on all systems.
main should be declared to return type int. It should
terminate either by calling exit or by returning the integer
status code; make sure it cannot ever return an undefined value.
malloc or
realloc.
Most GNU programs use those functions just once, in functions
conventionally named xmalloc and xrealloc. These
functions call malloc and realloc, respectively, and
check the results.
Because xmalloc and xrealloc are defined in your program,
you can declare them in other files without any risk of type conflict.
On most systems, int is the same length as a pointer; thus, the
calls to malloc and realloc work fine. For the few
exceptional systems (mostly 64-bit machines), you can use
conditionalized declarations of malloc and
realloc---or put these declarations in configuration files
specific to those systems.
strcpy strncpy strcat strncat strlen strcmp strncmp strchr strrchrThe copy and concatenate functions work fine without a declaration as long as you don't use their values. Using their values without a declaration fails on systems where the width of a pointer differs from the width of
int, and perhaps in other cases. It is trivial to
avoid using their values, so do that.
The compare functions and strlen work fine without a declaration
on most systems, possibly all the ones that GNU software runs on.
You may find it necessary to declare them conditionally on a
few systems.
The search functions must be declared to return char *. Luckily,
there is no variation in the data type they return. But there is
variation in their names. Some systems give these functions the names
index and rindex; other systems use the names
strchr and strrchr. Some systems support both pairs of
names, but neither pair works on all systems.
You should pick a single pair of names and use it throughout your
program. (Nowadays, it is better to choose strchr and
strrchr for new programs, since those are the standard ANSI
names.) Declare both of those names as functions returning char
*. On systems which don't support those names, define them as macros
in terms of the other pair. For example, here is what to put at the
beginning of your file (or in a header) if you want to use the names
strchr and strrchr throughout:
#ifndef HAVE_STRCHR #define strchr index #endif #ifndef HAVE_STRRCHR #define strrchr rindex #endif char *strchr (); char *strrchr ();
Here we assume that HAVE_STRCHR and HAVE_STRRCHR are
macros defined in systems where the corresponding functions exist.
One way to get them properly defined is to use Autoconf.
GNU has a library called GNU gettext that makes it easy to translate the messages in a program into various languages. You should use this library in every program. Use English for the messages as they appear in the program, and let gettext provide the way to translate them into other languages.
Using GNU gettext involves putting a call to the gettext macro
around each string that might need translation--like this:
printf (gettext ("Processing file `%s'..."));
This permits GNU gettext to replace the string "Processing file
`%s'..." with a translated version.
Once a program uses gettext, please make a point of writing calls to
gettext when you add new strings that call for translation.
Using GNU gettext in a package involves specifying a text domain name for the package. The text domain name is used to separate the translations for this package from the translations for other packages. Normally, the text domain name should be the same as the name of the package--for example, `fileutils' for the GNU file utilities.
To enable gettext to work well, avoid writing code that makes assumptions about the structure of words or sentences. When you want the precise text of a sentence to vary depending on the data, use two or more alternative string constants each containing a complete sentences, rather than inserting conditionalized words or phrases into a single sentence framework.
Here is an example of what not to do:
printf ("%d file%s processed", nfiles,
nfiles != 1 ? "s" : "");
The problem with that example is that it assumes that plurals are made by adding `s'. If you apply gettext to the format string, like this,
printf (gettext ("%d file%s processed"), nfiles,
nfiles != 1 ? "s" : "");
the message can use different words, but it will still be forced to use `s' for the plural. Here is a better way:
printf ((nfiles != 1 ? "%d files processed"
: "%d file processed"),
nfiles);
This way, you can apply gettext to each of the two strings independently:
printf ((nfiles != 1 ? gettext ("%d files processed")
: gettext ("%d file processed")),
nfiles);
This can be any method of forming the plural of the word for "file", and also handles languages that require agreement in the word for "processed".
A similar problem appears at the level of sentence structure with this code:
printf ("# Implicit rule search has%s been done.\n",
f->tried_implicit ? "" : " not");
Adding gettext calls to this code cannot give correct results for
all languages, because negation in some languages requires adding words
at more than one place in the sentence. By contrast, adding
gettext calls does the job straightfowardly if the code starts
out like this:
printf (f->tried_implicit
? "# Implicit rule search has been done.\n",
: "# Implicit rule search has not been done.\n");
Don't assume that mmap either works on all files or fails
for all files. It may work on some files and fail on others.
The proper way to use mmap is to try it on the specific file for
which you want to use it--and if mmap doesn't work, fall back on
doing the job in another way using read and write.
The reason this precaution is needed is that the GNU kernel (the HURD)
provides a user-extensible file system, in which there can be many
different kinds of "ordinary files." Many of them support
mmap, but some do not. It is important to make programs handle
all these kinds of files.
The preferred way to document part of the GNU system is to write a
manual in the Texinfo formatting language. This makes it possible to
produce a good quality formatted book, using TeX, and to generate an
Info file. It is also possible to generate HTML output from Texinfo
source. See the Texinfo manual, either the hardcopy, or the on-line
version available through info or the Emacs Info subsystem
(C-h i).
Programmers often find it most natural to structure the documentation following the structure of the implementation, which they know. But this structure is not necessarily good for explaining how to use the program; it may be irrelevant and confusing for a user.
At every level, from the sentences in a paragraph to the grouping of topics into separate manuals, the right way to structure documentation is according to the concepts and questions that a user will have in mind when reading it. Sometimes this structure of ideas matches the structure of the implementation of the software being documented--but often they are different. Often the most important part of learning to write good documentation is learning to notice when you are structuring the documentation like the implementation, and think about better alternatives.
For example, each program in the GNU system probably ought to be documented in one manual; but this does not mean each program should have its own manual. That would be following the structure of the implementation, rather than the structure that helps the user understand.
Instead, each manual should cover a coherent topic. For example,
instead of a manual for diff and a manual for diff3, we
have one manual for "comparison of files" which covers both of those
programs, as well as cmp. By documenting these programs
together, we can make the whole subject clearer.
The manual which discusses a program should document all of the program's command-line options and all of its commands. It should give examples of their use. But don't organize the manual as a list of features. Instead, organize it logically, by subtopics. Address the questions that a user will ask when thinking about the job that the program does.
In general, a GNU manual should serve both as tutorial and reference. It should be set up for convenient access to each topic through Info, and for reading straight through (appendixes aside). A GNU manual should give a good introduction to a beginner reading through from the start, and should also provide all the details that hackers want. The Bison manual is a good example of this--please take a look at it to see what we mean.
That is not as hard as it first sounds. Arrange each chapter as a logical breakdown of its topic, but order the sections, and write their text, so that reading the chapter straight through makes sense. Do likewise when structuring the book into chapters, and when structuring a section into paragraphs. The watchword is, at each point, address the most fundamental and important issue raised by the preceding text.
If necessary, add extra chapters at the beginning of the manual which are purely tutorial and cover the basics of the subject. These provide the framework for a beginner to understand the rest of the manual. The Bison manual provides a good example of how to do this.
Don't use Unix man pages as a model for how to write GNU documentation; most of them are terse, badly structured, and give inadequate explanation of the underlying concepts. (There are, of course exceptions.) Also Unix man pages use a particular format which is different from what we use in GNU manuals.
Please include an email address in the manual for where to report bugs in the manual.
Please do not use the term "pathname" that is used in Unix documentation; use "file name" (two words) instead. We use the term "path" only for search paths, which are lists of directory names.
Please do not use the term "illegal" to refer to erroneous input to a computer program. Please use "invalid" for this, and reserve the term "illegal" for violations of law.
The title page of the manual should state the version of the programs or packages documented in the manual. The Top node of the manual should also contain this information. If the manual is changing more frequently than or independent of the program, also state a version number for the manual in both of these places.
Each program documented in the manual should have a node named `program Invocation' or `Invoking program'. This node (together with its subnodes, if any) should describe the program's command line arguments and how to run it (the sort of information people would look in a man page for). Start with an `@example' containing a template for all the options and arguments that the program uses.
Alternatively, put a menu item in some menu whose item name fits one of the above patterns. This identifies the node which that item points to as the node for this purpose, regardless of the node's actual name.
There will be automatic features for specifying a program name and quickly reading just this part of its manual.
If one manual describes several programs, it should have such a node for each program described.
If the manual contains a copy of the GNU GPL or GNU LGPL, or if it contains chapters that make political or personal statements, please copy the distribution terms of the GNU Emacs Manual, and adapt it by modifying appropriately the list of special chapters that may not be modified or deleted.
If the manual does not contain any such chapters, then imitate the simpler distribution terms of the Texinfo manual.
In addition to its manual, the package should have a file named `NEWS' which contains a list of user-visible changes worth mentioning. In each new release, add items to the front of the file and identify the version they pertain to. Don't discard old items; leave them in the file after the newer items. This way, a user upgrading from any previous version can see what is new.
If the `NEWS' file gets very long, move some of the older items into a file named `ONEWS' and put a note at the end referring the user to that file.
Keep a change log to describe all the changes made to program source files. The purpose of this is so that people investigating bugs in the future will know about the changes that might have introduced the bug. Often a new bug can be found by looking at what was recently changed. More importantly, change logs can help you eliminate conceptual inconsistencies between different parts of a program, by giving you a history of how the conflicting concepts arose and who they came from.
You can think of the change log as a conceptual "undo list" which explains how earlier versions were different from the current version. People can see the current version; they don't need the change log to tell them what is in it. What they want from a change log is a clear explanation of how the earlier version differed.
The change log file is normally called `ChangeLog' and covers an entire directory. Each directory can have its own change log, or a directory can use the change log of its parent directory--it's up to you.
Another alternative is to record change log information with a version
control system such as RCS or CVS. This can be converted automatically
to a `ChangeLog' file using rcs2log; in Emacs, the command
C-x v a (vc-update-change-log) does the job.
There's no need to describe the full purpose of the changes or how they work together. If you think that a change calls for explanation, you're probably right. Please do explain it--but please put the explanation in comments in the code, where people will see it whenever they see the code. For example, "New function" is enough for the change log when you add a function, because there should be a comment before the function definition to explain what it does.
However, sometimes it is useful to write one line to describe the overall purpose of a batch of changes.
The easiest way to add an entry to `ChangeLog' is with the Emacs command M-x add-change-log-entry. An entry should have an asterisk, the name of the changed file, and then in parentheses the name of the changed functions, variables or whatever, followed by a colon. Then describe the changes you made to that function or variable.
Here are some examples of change log entries:
* register.el (insert-register): Return nil. (jump-to-register): Likewise. * sort.el (sort-subr): Return nil. * tex-mode.el (tex-bibtex-file, tex-file, tex-region): Restart the tex shell if process is gone or stopped. (tex-shell-running): New function. * expr.c (store_one_arg): Round size up for move_block_to_reg. (expand_call): Round up when emitting USE insns. * stmt.c (assign_parms): Round size up for move_block_from_reg.
It's important to name the changed function or variable in full. Don't abbreviate function or variable names, and don't combine them. Subsequent maintainers will often search for a function name to find all the change log entries that pertain to it; if you abbreviate the name, they won't find it when they search.
For example, some people are tempted to abbreviate groups of function
names by writing `* register.el ({insert,jump-to}-register)';
this is not a good idea, since searching for jump-to-register or
insert-register would not find that entry.
Separate unrelated change log entries with blank lines. When two entries represent parts of the same change, so that they work together, then don't put blank lines between them. Then you can omit the file name and the asterisk when successive entries are in the same file.
Certain simple kinds of changes don't need much detail in the change log.
When you change the calling sequence of a function in a simple fashion, and you change all the callers of the function, there is no need to make individual entries for all the callers that you changed. Just write in the entry for the function being called, "All callers changed."
* keyboard.c (Fcommand_execute): New arg SPECIAL. All callers changed.
When you change just comments or doc strings, it is enough to write an entry for the file, without mentioning the functions. Just "Doc fixes" is enough for the change log.
There's no need to make change log entries for documentation files. This is because documentation is not susceptible to bugs that are hard to fix. Documentation does not consist of parts that must interact in a precisely engineered fashion. To correct an error, you need not know the history of the erroneous passage; it is enough to compare what the documentation says with the way the program actually works.
C programs often contain compile-time #if conditionals. Many
changes are conditional; sometimes you add a new definition which is
entirely contained in a conditional. It is very useful to indicate in
the change log the conditions for which the change applies.
Our convention for indicating conditional changes is to use square brackets around the name of the condition.
Here is a simple example, describing a change which is conditional but does not have a function or entity name associated with it:
* xterm.c [SOLARIS2]: Include string.h.
Here is an entry describing a new definition which is entirely
conditional. This new definition for the macro FRAME_WINDOW_P is
used only when HAVE_X_WINDOWS is defined:
* frame.h [HAVE_X_WINDOWS] (FRAME_WINDOW_P): Macro defined.
Here is an entry for a change within the function init_display,
whose definition as a whole is unconditional, but the changes themselves
are contained in a `#ifdef HAVE_LIBNCURSES' conditional:
* dispnew.c (init_display) [HAVE_LIBNCURSES]: If X, call tgetent.
Here is an entry for a change that takes affect only when a certain macro is not defined:
(gethostname) [!HAVE_SOCKETS]: Replace with winsock version.
In the GNU project, man pages are secondary. It is not necessary or expected for every GNU program to have a man page, but some of them do. It's your choice whether to include a man page in your program.
When you make this decision, consider that supporting a man page requires continual effort each time the program is changed. The time you spend on the man page is time taken away from more useful work.
For a simple program which changes little, updating the man page may be a small job. Then there is little reason not to include a man page, if you have one.
For a large program that changes a great deal, updating a man page may be a substantial burden. If a user offers to donate a man page, you may find this gift costly to accept. It may be better to refuse the man page unless the same person agrees to take full responsibility for maintaining it--so that you can wash your hands of it entirely. If this volunteer later ceases to do the job, then don't feel obliged to pick it up yourself; it may be better to withdraw the man page from the distribution until someone else agrees to update it.
When a program changes only a little, you may feel that the discrepancies are small enough that the man page remains useful without updating. If so, put a prominent note near the beginning of the man page explaining that you don't maintain it and that the Texinfo manual is more authoritative. The note should say how to access the Texinfo documentation.
There may be non-free books or documentation files that describe the program you are documenting.
It is ok to use these documents for reference, just as the author of a new algebra textbook can read other books on algebra. A large portion of any non-fiction book consists of facts, in this case facts about how a certain program works, and these facts are necessarily the same for everyone who writes about the subject. But be careful not to copy your outline structure, wording, tables or examples from preexisting non-free documentation. Copying from free documentation may be ok; please check with the FSF about the individual case.
Making a release is more than just bundling up your source files in a tar file and putting it up for FTP. You should set up your software so that it can be configured to run on a variety of systems. Your Makefile should conform to the GNU standards described below, and your directory layout should also conform to the standards discussed below. Doing so makes it easy to include your package into the larger framework of all GNU software.
Each GNU distribution should come with a shell script named
configure. This script is given arguments which describe the
kind of machine and system you want to compile the program for.
The configure script must record the configuration options so
that they affect compilation.
One way to do this is to make a link from a standard name such as `config.h' to the proper configuration file for the chosen system. If you use this technique, the distribution should not contain a file named `config.h'. This is so that people won't be able to build the program without configuring it first.
Another thing that configure can do is to edit the Makefile. If
you do this, the distribution should not contain a file named
`Makefile'. Instead, it should include a file `Makefile.in' which
contains the input used for editing. Once again, this is so that people
won't be able to build the program without configuring it first.
If configure does write the `Makefile', then `Makefile'
should have a target named `Makefile' which causes configure
to be rerun, setting up the same configuration that was set up last
time. The files that configure reads should be listed as
dependencies of `Makefile'.
All the files which are output from the configure script should
have comments at the beginning explaining that they were generated
automatically using configure. This is so that users won't think
of trying to edit them by hand.
The configure script should write a file named `config.status'
which describes which configuration options were specified when the
program was last configured. This file should be a shell script which,
if run, will recreate the same configuration.
The configure script should accept an option of the form
`--srcdir=dirname' to specify the directory where sources are found
(if it is not the current directory). This makes it possible to build
the program in a separate directory, so that the actual source directory
is not modified.
If the user does not specify `--srcdir', then configure should
check both `.' and `..' to see if it can find the sources. If
it finds the sources in one of these places, it should use them from
there. Otherwise, it should report that it cannot find the sources, and
should exit with nonzero status.
Usually the easy way to support `--srcdir' is by editing a
definition of VPATH into the Makefile. Some rules may need to
refer explicitly to the specified source directory. To make this
possible, configure can add to the Makefile a variable named
srcdir whose value is precisely the specified directory.
The configure script should also take an argument which specifies the
type of system to build the program for. This argument should look like
this:
cpu-company-system
For example, a Sun 3 might be `m68k-sun-sunos4.1'.
The configure script needs to be able to decode all plausible
alternatives for how to describe a machine. Thus, `sun3-sunos4.1'
would be a valid alias. For many programs, `vax-dec-ultrix' would
be an alias for `vax-dec-bsd', simply because the differences
between Ultrix and BSD are rarely noticeable, but a few programs
might need to distinguish them.
There is a shell script called `config.sub' that you can use as a subroutine to validate system types and canonicalize aliases.
Other options are permitted to specify in more detail the software or hardware present on the machine, and include or exclude optional parts of the package:
All configure scripts should accept all of these "detail"
options, whether or not they make any difference to the particular
package at hand. In particular, they should accept any option that
starts with `--with-' or `--enable-'. This is so users will
be able to configure an entire GNU source tree at once with a single set
of options.
You will note that the categories `--with-' and `--enable-' are narrow: they do not provide a place for any sort of option you might think of. That is deliberate. We want to limit the possible configuration options in GNU software. We do not want GNU programs to have idiosyncratic configuration options.
Packages that perform part of the compilation process may support cross-compilation.
In such a case, the host and target machines for the program may be
different. The configure script should normally treat the
specified type of system as both the host and the target, thus producing
a program which works for the same type of machine that it runs on.
The way to build a cross-compiler, cross-assembler, or what have you, is
to specify the option `--host=hosttype' when running
configure. This specifies the host system without changing the
type of target system. The syntax for hosttype is the same as
described above.
Bootstrapping a cross-compiler requires compiling it on a machine other than the host it will run on. Compilation packages accept a configuration option `--build=hosttype' for specifying the configuration on which you will compile them, in case that is different from the host.
Programs for which cross-operation is not meaningful need not accept the `--host' option, because configuring an entire operating system for cross-operation is not a meaningful thing.
Some programs have ways of configuring themselves automatically. If
your program is set up to do this, your configure script can simply
ignore most of its arguments.
@lowersections
This section describes conventions for writing the Makefiles for GNU programs. Using Automake will help you write a Makefile that follows these conventions.
Every Makefile should contain this line:
SHELL = /bin/sh
to avoid trouble on systems where the SHELL variable might be
inherited from the environment. (This is never a problem with GNU
make.)
Different make programs have incompatible suffix lists and
implicit rules, and this sometimes creates confusion or misbehavior. So
it is a good idea to set the suffix list explicitly using only the
suffixes you need in the particular Makefile, like this:
.SUFFIXES: .SUFFIXES: .c .o
The first line clears out the suffix list, the second introduces all suffixes which may be subject to implicit rules in this Makefile.
Don't assume that `.' is in the path for command execution. When you need to run programs that are a part of your package during the make, please make sure that it uses `./' if the program is built as part of the make or `$(srcdir)/' if the file is an unchanging part of the source code. Without one of these prefixes, the current search path is used.
The distinction between `./' (the build directory) and `$(srcdir)/' (the source directory) is important because users can build in a separate directory using the `--srcdir' option to `configure'. A rule of the form:
foo.1 : foo.man sedscript
sed -e sedscript foo.man > foo.1
will fail when the build directory is not the source directory, because `foo.man' and `sedscript' are in the the source directory.
When using GNU make, relying on `VPATH' to find the source
file will work in the case where there is a single dependency file,
since the make automatic variable `$<' will represent the
source file wherever it is. (Many versions of make set `$<'
only in implicit rules.) A Makefile target like
foo.o : bar.c
$(CC) -I. -I$(srcdir) $(CFLAGS) -c bar.c -o foo.o
should instead be written as
foo.o : bar.c
$(CC) -I. -I$(srcdir) $(CFLAGS) -c $< -o $@
in order to allow `VPATH' to work correctly. When the target has multiple dependencies, using an explicit `$(srcdir)' is the easiest way to make the rule work well. For example, the target above for `foo.1' is best written as:
foo.1 : foo.man sedscript
sed -e $(srcdir)/sedscript $(srcdir)/foo.man > $@
GNU distributions usually contain some files which are not source files--for example, Info files, and the output from Autoconf, Automake, Bison or Flex. Since these files normally appear in the source directory, they should always appear in the source directory, not in the build directory. So Makefile rules to update them should put the updated files in the source directory.
However, if a file does not appear in the distribution, then the Makefile should not put it in the source directory, because building a program in ordinary circumstances should not modify the source directory in any way.
Try to make the build and installation targets, at least (and all their
subtargets) work correctly with a parallel make.
Write the Makefile commands (and any shell scripts, such as
configure) to run in sh, not in csh. Don't use any
special features of ksh or bash.
The configure script and the Makefile rules for building and
installation should not use any utilities directly except these:
cat cmp cp diff echo egrep expr false grep install-info ln ls mkdir mv pwd rm rmdir sed sleep sort tar test touch true
The compression program gzip can be used in the dist rule.
Stick to the generally supported options for these programs. For example, don't use `mkdir -p', convenient as it may be, because most systems don't support it.
It is a good idea to avoid creating symbolic links in makefiles, since a few systems don't support them.
The Makefile rules for building and installation can also use compilers
and related programs, but should do so via make variables so that the
user can substitute alternatives. Here are some of the programs we
mean:
ar bison cc flex install ld ldconfig lex make makeinfo ranlib texi2dvi yacc
Use the following make variables to run those programs:
$(AR) $(BISON) $(CC) $(FLEX) $(INSTALL) $(LD) $(LDCONFIG) $(LEX) $(MAKE) $(MAKEINFO) $(RANLIB) $(TEXI2DVI) $(YACC)
When you use ranlib or ldconfig, you should make sure
nothing bad happens if the system does not have the program in question.
Arrange to ignore an error from that command, and print a message before
the command to tell the user that failure of this command does not mean
a problem. (The Autoconf `AC_PROG_RANLIB' macro can help with
this.)
If you use symbolic links, you should implement a fallback for systems that don't have symbolic links.
Additional utilities that can be used via Make variables are:
chgrp chmod chown mknod
It is ok to use other utilities in Makefile portions (or scripts) intended only for particular systems where you know those utilities exist.
Makefiles should provide variables for overriding certain commands, options, and so on.
In particular, you should run most utility programs via variables.
Thus, if you use Bison, have a variable named BISON whose default
value is set with `BISON = bison', and refer to it with
$(BISON) whenever you need to use Bison.
File management utilities such as ln, rm, mv, and
so on, need not be referred to through variables in this way, since users
don't need to replace them with other programs.
Each program-name variable should come with an options variable that is
used to supply options to the program. Append `FLAGS' to the
program-name variable name to get the options variable name--for
example, BISONFLAGS. (The names CFLAGS for the C
compiler, YFLAGS for yacc, and LFLAGS for lex, are
exceptions to this rule, but we keep them because they are standard.)
Use CPPFLAGS in any compilation command that runs the
preprocessor, and use LDFLAGS in any compilation command that
does linking as well as in any direct use of ld.
If there are C compiler options that must be used for proper
compilation of certain files, do not include them in CFLAGS.
Users expect to be able to specify CFLAGS freely themselves.
Instead, arrange to pass the necessary options to the C compiler
independently of CFLAGS, by writing them explicitly in the
compilation commands or by defining an implicit rule, like this:
CFLAGS = -g
ALL_CFLAGS = -I. $(CFLAGS)
.c.o:
$(CC) -c $(CPPFLAGS) $(ALL_CFLAGS) $<
Do include the `-g' option in CFLAGS, because that is not
required for proper compilation. You can consider it a default
that is only recommended. If the package is set up so that it is
compiled with GCC by default, then you might as well include `-O'
in the default value of CFLAGS as well.
Put CFLAGS last in the compilation command, after other variables
containing compiler options, so the user can use CFLAGS to
override the others.
CFLAGS should be used in every invocation of the C compiler,
both those which do compilation and those which do linking.
Every Makefile should define the variable INSTALL, which is the
basic command for installing a file into the system.
Every Makefile should also define the variables INSTALL_PROGRAM
and INSTALL_DATA. (The default for each of these should be
$(INSTALL).) Then it should use those variables as the commands
for actual installation, for executables and nonexecutables
respectively. Use these variables as follows:
$(INSTALL_PROGRAM) foo $(bindir)/foo $(INSTALL_DATA) libfoo.a $(libdir)/libfoo.a
Optionally, you may prepend the value of DESTDIR to the target
filename. Doing this allows the installer to create a snapshot of the
installation to be copied onto the real target filesystem later. Do not
set the value of DESTDIR in your Makefile, and do not include it
in any installed files. With support for DESTDIR, the above
examples become:
$(INSTALL_PROGRAM) foo $(DESTDIR)$(bindir)/foo $(INSTALL_DATA) libfoo.a $(DESTDIR)$(libdir)/libfoo.a
Always use a file name, not a directory name, as the second argument of the installation commands. Use a separate command for each file to be installed.
Installation directories should always be named by variables, so it is easy to install in a nonstandard place. The standard names for these variables are described below. They are based on a standard filesystem layout; variants of it are used in SVR4, 4.4BSD, Linux, Ultrix v4, and other modern operating systems.
These two variables set the root for the installation. All the other installation directories should be subdirectories of one of these two, and nothing should be directly installed into these two directories.
prefix should be `/usr/local'.
When building the complete GNU system, the prefix will be empty and
`/usr' will be a symbolic link to `/'.
(If you are using Autoconf, write it as `@prefix@'.)
Running `make install' with a different value of prefix
from the one used to build the program should not recompile
the program.
exec_prefix should
be $(prefix).
(If you are using Autoconf, write it as `@exec_prefix@'.)
Generally, $(exec_prefix) is used for directories that contain
machine-specific files (such as executables and subroutine libraries),
while $(prefix) is used directly for other directories.
Running `make install' with a different value of exec_prefix
from the one used to build the program should not recompile the
program.
Executable programs are installed in one of the following directories.
Data files used by the program during its execution are divided into categories in two ways.
This makes for six different possibilities. However, we want to discourage the use of architecture-dependent files, aside from object files and libraries. It is much cleaner to make other data files architecture-independent, and it is generally not hard.
Therefore, here are the variables Makefiles should use to specify directories:
libdir should normally be
`/usr/local/lib', but write it as `$(exec_prefix)/lib'.
(If you are using Autoconf, write it as `@libdir@'.)
lispdir='${datadir}/emacs/site-lisp'
AC_SUBST(lispdir)
includedir and one
specified by oldincludedir.
oldincludedir is empty. If it is, they should not try to use
it; they should cancel the second installation of the header files.
A package should not replace an existing header in this directory unless
the header came from the same package. Thus, if your Foo package
provides a header file `foo.h', then it should install the header
file in the oldincludedir directory if either (1) there is no
`foo.h' there or (2) the `foo.h' that exists came from the Foo
package.
To tell whether `foo.h' came from the Foo package, put a magic
string in the file--part of a comment--and grep for that string.
Unix-style man pages are installed in one of the following:
And finally, you should set the following variable:
configure shell script.
(If you are using Autconf, use `srcdir = @srcdir@'.)
For example:
# Common prefix for installation directories. # NOTE: This directory must exist when you start the install. prefix = /usr/local exec_prefix = $(prefix) # Where to put the executable for the command `gcc'. bindir = $(exec_prefix)/bin # Where to put the directories used by the compiler. libexecdir = $(exec_prefix)/libexec # Where to put the Info files. infodir = $(prefix)/info
If your program installs a large number of files into one of the
standard user-specified directories, it might be useful to group them
into a subdirectory particular to that program. If you do this, you
should write the install rule to create these subdirectories.
Do not expect the user to include the subdirectory name in the value of any of the variables listed above. The idea of having a uniform set of variable names for installation directories is to enable the user to specify the exact same values for several different GNU packages. In order for this to be useful, all the packages must be designed so that they will work sensibly when the user does so.
All GNU programs should have the following targets in their Makefiles:
install-strip target to do that.
If possible, write the install target rule so that it does not
modify anything in the directory where the program was built, provided
`make all' has just been done. This is convenient for building the
program under one user name and installing it under another.
The commands should create all the directories in which files are to be
installed, if they don't already exist. This includes the directories
specified as the values of the variables prefix and
exec_prefix, as well as all subdirectories that are needed.
One way to do this is by means of an installdirs target
as described below.
Use `-' before any command for installing a man page, so that
make will ignore any errors. This is in case there are systems
that don't have the Unix man page documentation system installed.
The way to install Info files is to copy them into `$(infodir)'
with $(INSTALL_DATA) (see section Variables for Specifying Commands), and then run
the install-info program if it is present. install-info
is a program that edits the Info `dir' file to add or update the
menu entry for the given Info file; it is part of the Texinfo package.
Here is a sample rule to install an Info file:
$(DESTDIR)$(infodir)/foo.info: foo.info
$(POST_INSTALL)
# There may be a newer info file in . than in srcdir.
-if test -f foo.info; then d=.; \
else d=$(srcdir); fi; \
$(INSTALL_DATA) $$d/foo.info $(DESTDIR)$@; \
# Run install-info only if it exists.
# Use `if' instead of just prepending `-' to the
# line so we notice real errors from install-info.
# We use `$(SHELL) -c' because some shells do not
# fail gracefully when there is an unknown command.
if $(SHELL) -c 'install-info --version' \
>/dev/null 2>&1; then \
install-info --dir-file=$(DESTDIR)$(infodir)/dir \
$(DESTDIR)$(infodir)/foo.info; \
else true; fi
When writing the install target, you must classify all the
commands into three categories: normal ones, pre-installation
commands and post-installation commands. See section Install Command Categories.
install, but strip the executable files while installing
them. In many cases, the definition of this target can be very simple:
install-strip:
$(MAKE) INSTALL_PROGRAM='$(INSTALL_PROGRAM) -s' \
install
Normally we do not recommend stripping an executable unless you are sure
the program has no bugs. However, it can be reasonable to install a
stripped executable for actual execution while saving the unstripped
executable elsewhere in case there is a bug.
distclean, plus more: C source files produced by
Bison, tags tables, Info files, and so on.
The reason we say "almost everything" is that running the command
`make maintainer-clean' should not delete `configure' even if
`configure' can be remade using a rule in the Makefile. More generally,
`make maintainer-clean' should not delete anything that needs to
exist in order to run `configure' and then begin to build the
program. This is the only exception; maintainer-clean should
delete everything else that can be rebuilt.
The `maintainer-clean' target is intended to be used by a maintainer of
the package, not by ordinary users. You may need special tools to
reconstruct some of the files that `make maintainer-clean' deletes.
Since these files are normally included in the distribution, we don't
take care to make them easy to reconstruct. If you find you need to
unpack the full distribution again, don't blame us.
To help make users aware of this, the commands for the special
maintainer-clean target should start with these two:
@echo 'This command is intended for maintainers to use; it' @echo 'deletes files that may need special tools to rebuild.'
info: foo.info
foo.info: foo.texi chap1.texi chap2.texi
$(MAKEINFO) $(srcdir)/foo.texi
You must define the variable MAKEINFO in the Makefile. It should
run the makeinfo program, which is part of the Texinfo
distribution.
Normally a GNU distribution comes with Info files, and that means the
Info files are present in the source directory. Therefore, the Make
rule for an info file should update it in the source directory. When
users build the package, ordinarily Make will not update the Info files
because they will already be up to date.
dvi: foo.dvi
foo.dvi: foo.texi chap1.texi chap2.texi
$(TEXI2DVI) $(srcdir)/foo.texi
You must define the variable TEXI2DVI in the Makefile. It should
run the program texi2dvi, which is part of the Texinfo
distribution.(1) Alternatively,
write just the dependencies, and allow GNU make to provide the command.
ln or cp to install the proper files in it, and
then tar that subdirectory.
Compress the tar file file with gzip. For example, the actual
distribution file for GCC version 1.40 is called `gcc-1.40.tar.gz'.
The dist target should explicitly depend on all non-source files
that are in the distribution, to make sure they are up to date in the
distribution.
See section Making Releases.
The following targets are suggested as conventional names, for programs in which they are useful.
installcheck
installdirs
# Make sure all installation directories (e.g. $(bindir))
# actually exist by making them if necessary.
installdirs: mkinstalldirs
$(srcdir)/mkinstalldirs $(bindir) $(datadir) \
$(libdir) $(infodir) \
$(mandir)
This rule should not modify the directories where compilation is done.
It should do nothing but create installation directories.
When writing the install target, you must classify all the
commands into three categories: normal ones, pre-installation
commands and post-installation commands.
Normal commands move files into their proper places, and set their modes. They may not alter any files except the ones that come entirely from the package they belong to.
Pre-installation and post-installation commands may alter other files; in particular, they can edit global configuration files or data bases.
Pre-installation commands are typically executed before the normal commands, and post-installation commands are typically run after the normal commands.
The most common use for a post-installation command is to run
install-info. This cannot be done with a normal command, since
it alters a file (the Info directory) which does not come entirely and
solely from the package being installed. It is a post-installation
command because it needs to be done after the normal command which
installs the package's Info files.
Most programs don't need any pre-installation commands, but we have the feature just in case it is needed.
To classify the commands in the install rule into these three
categories, insert category lines among them. A category line
specifies the category for the commands that follow.
A category line consists of a tab and a reference to a special Make variable, plus an optional comment at the end. There are three variables you can use, one for each category; the variable name specifies the category. Category lines are no-ops in ordinary execution because these three Make variables are normally undefined (and you should not define them in the makefile).
Here are the three possible category lines, each with a comment that explains what it means:
$(PRE_INSTALL) # Pre-install commands follow.
$(POST_INSTALL) # Post-install commands follow.
$(NORMAL_INSTALL) # Normal commands follow.
If you don't use a category line at the beginning of the install
rule, all the commands are classified as normal until the first category
line. If you don't use any category lines, all the commands are
classified as normal.
These are the category lines for uninstall:
$(PRE_UNINSTALL) # Pre-uninstall commands follow.
$(POST_UNINSTALL) # Post-uninstall commands follow.
$(NORMAL_UNINSTALL) # Normal commands follow.
Typically, a pre-uninstall command would be used for deleting entries from the Info directory.
If the install or uninstall target has any dependencies
which act as subroutines of installation, then you should start
each dependency's commands with a category line, and start the
main target's commands with a category line also. This way, you can
ensure that each command is placed in the right category regardless of
which of the dependencies actually run.
Pre-installation and post-installation commands should not run any programs except for these:
[ basename bash cat chgrp chmod chown cmp cp dd diff echo egrep expand expr false fgrep find getopt grep gunzip gzip hostname install install-info kill ldconfig ln ls md5sum mkdir mkfifo mknod mv printenv pwd rm rmdir sed sort tee test touch true uname xargs yes
The reason for distinguishing the commands in this way is for the sake of making binary packages. Typically a binary package contains all the executables and other files that need to be installed, and has its own method of installing them--so it does not need to run the normal installation commands. But installing the binary package does need to execute the pre-installation and post-installation commands.
Programs to build binary packages work by extracting the pre-installation and post-installation commands. Here is one way of extracting the pre-installation commands:
make -n install -o all \
PRE_INSTALL=pre-install \
POST_INSTALL=post-install \
NORMAL_INSTALL=normal-install \
| gawk -f pre-install.awk
where the file `pre-install.awk' could contain this:
$0 ~ /^\t[ \t]*(normal_install|post_install)[ \t]*$/ {on = 0}
on {print $0}
$0 ~ /^\t[ \t]*pre_install[ \t]*$/ {on = 1}
The resulting file of pre-installation commands is executed as a shell script as part of installing the binary package. @raisesections
Package the distribution of Foo version 69.96 up in a gzipped tar
file with the name `foo-69.96.tar.gz'. It should unpack into a
subdirectory named `foo-69.96'.
Building and installing the program should never modify any of the files contained in the distribution. This means that all the files that form part of the program in any way must be classified into source files and non-source files. Source files are written by humans and never changed automatically; non-source files are produced from source files by programs under the control of the Makefile.
The distribution should contain a file named `README' which gives the name of the package, and a general description of what it does. It is also good to explain the purpose of each of the first-level subdirectories in the package, if there are any. The `README' file should either state the version number of the package, or refer to where in the package it can be found.
The `README' file should refer to the file `INSTALL', which should contain an explanation of the installation procedure.
The `README' file should also refer to the file which contains the copying conditions. The GNU GPL, if used, should be in a file called `COPYING'. If the GNU LGPL is used, it should be in a file called `COPYING.LIB'.
Naturally, all the source files must be in the distribution. It is okay
to include non-source files in the distribution, provided they are
up-to-date and machine-independent, so that building the distribution
normally will never modify them. We commonly include non-source files
produced by Bison, lex, TeX, and makeinfo; this helps avoid
unnecessary dependencies between our distributions, so that users can
install whichever packages they want to install.
Non-source files that might actually be modified by building and installing the program should never be included in the distribution. So if you do distribute non-source files, always make sure they are up to date when you make a new distribution.
Make sure that the directory into which the distribution unpacks (as
well as any subdirectories) are all world-writable (octal mode 777).
This is so that old versions of tar which preserve the
ownership and permissions of the files from the tar archive will be
able to extract all the files even if the user is unprivileged.
Make sure that all the files in the distribution are world-readable.
Make sure that no file name in the distribution is more than 14 characters long. Likewise, no file created by building the program should have a name longer than 14 characters. The reason for this is that some systems adhere to a foolish interpretation of the POSIX standard, and refuse to open a longer name, rather than truncating as they did in the past.
Don't include any symbolic links in the distribution itself. If the tar file contains symbolic links, then people cannot even unpack it on systems that don't support symbolic links. Also, don't use multiple names for one file in different directories, because certain file systems cannot handle this and that prevents unpacking the distribution.
Try to make sure that all the file names will be unique on MS-DOS. A name on MS-DOS consists of up to 8 characters, optionally followed by a period and up to three characters. MS-DOS will truncate extra characters both before and after the period. Thus, `foobarhacker.c' and `foobarhacker.o' are not ambiguous; they are truncated to `foobarha.c' and `foobarha.o', which are distinct.
Include in your distribution a copy of the `texinfo.tex' you used to test print any `*.texinfo' or `*.texi' files.
Likewise, if your program uses small GNU software packages like regex, getopt, obstack, or termcap, include them in the distribution file. Leaving them out would make the distribution file a little smaller at the expense of possible inconvenience to a user who doesn't know what other files to get.
A GNU program should not recommend use of any non-free program. We can't stop some people from writing proprietary programs, or stop other people from using them. But we can and should avoid helping to advertise them to new customers.
Sometimes it is important to mention how to build your package on top of some non-free operating system or other non-free base package. In such cases, please mention the name of the non-free package or system in the briefest possible way. Don't include any references for where to find more information about the proprietary program. The goal should be that people already using the proprietary program will get the advice they need about how to use your free program, while people who don't already use the proprietary program will not see anything to encourage them to take an interest in it.
Likewise, a GNU package should not refer the user to any non-free documentation for free software. The need for free documentation to go with free software is now a major focus of the GNU project; to show that we are serious about the need for free documentation, we must not undermine our position by recommending use of documentation that isn't free.
texi2dvi uses TeX to do the real work
of formatting. TeX is not distributed with Texinfo.
This document was generated on 25 June 1999 using the texi2html translator version 1.54.
FSF & GNU inquiries & questions to gnu@gnu.org. Other ways to contact the FSF.
Comments on these web pages to webmasters@gnu.org, send other questions to gnu@gnu.org.
Copyright notice above.