1 Like other projects, we also have some guidelines for our code. For
2 Git in general, a few rough rules are:
4 - Most importantly, we never say "It's in POSIX; we'll happily
5 ignore your needs should your system not conform to it."
6 We live in the real world.
8 - However, we often say "Let's stay away from that construct,
9 it's not even in POSIX".
11 - In spite of the above two rules, we sometimes say "Although
12 this is not in POSIX, it (is so convenient | makes the code
13 much more readable | has other good characteristics) and
14 practically all the platforms we care about support it, so
17 Again, we live in the real world, and it is sometimes a
18 judgement call, the decision based more on real world
19 constraints people face than what the paper standard says.
21 - Fixing style violations while working on a real change as a
22 preparatory clean-up step is good, but otherwise avoid useless code
23 churn for the sake of conforming to the style.
25 "Once it _is_ in the tree, it's not really worth the patch noise to
27 Cf. https://lore.kernel.org/all/20100126160632.3bdbe172.akpm@linux-foundation.org/
29 - Log messages to explain your changes are as important as the
30 changes themselves. Clearly written code and in-code comments
31 explain how the code works and what is assumed from the surrounding
32 context. The log messages explain what the changes wanted to
33 achieve and why the changes were necessary (more on this in the
34 accompanying SubmittingPatches document).
36 Make your code readable and sensible, and don't try to be clever.
38 As for more concrete guidelines, just imitate the existing code
39 (this is a good guideline, no matter which project you are
40 contributing to). It is always preferable to match the _local_
41 convention. New code added to Git suite is expected to match
42 the overall style of existing code. Modifications to existing
43 code are expected to match the style the surrounding code already
44 uses (even if it doesn't match the overall style of existing code).
46 But if you must have a list of rules, here are some language
47 specific ones. Note that Documentation/ToolsForGit.txt document
48 has a collection of tips to help you use some external tools
49 to conform to these guidelines.
51 For shell scripts specifically (not exhaustive):
53 - We use tabs for indentation.
55 - Case arms are indented at the same depth as case and esac lines,
67 - Redirection operators should be written with space before, but no
68 space after them. In other words, write 'echo test >"$file"'
69 instead of 'echo test> $file' or 'echo test > $file'. Note that
70 even though it is not required by POSIX to double-quote the
71 redirection target in a variable (as shown above), our code does so
72 because some versions of bash issue a warning without the quotes.
75 cat hello > world < universe
79 cat hello >world <universe
82 - We prefer $( ... ) for command substitution; unlike ``, it
83 properly nests. It should have been the way Bourne spelled
84 it from day one, but unfortunately isn't.
86 - If you want to find out if a command is available on the user's
87 $PATH, you should use 'type <command>', instead of 'which <command>'.
88 The output of 'which' is not machine parsable and its exit code
89 is not reliable across platforms.
91 - We use POSIX compliant parameter substitutions and avoid bashisms;
94 - We use ${parameter-word} and its [-=?+] siblings, and their
95 colon'ed "unset or null" form.
97 - We use ${parameter#word} and its [#%] siblings, and their
98 doubled "longest matching" form.
100 - No "Substring Expansion" ${parameter:offset:length}.
104 - No pattern replacement ${parameter/pattern/string}.
106 - We use Arithmetic Expansion $(( ... )).
108 - We do not use Process Substitution <(list) or >(list).
110 - Do not write control structures on a single line with semicolon.
111 "then" should be on the next line for if statements, and "do"
112 should be on the next line for "while" and "for".
115 if test -f hello; then
125 - If a command sequence joined with && or || or | spans multiple
126 lines, put each command on a separate line and put && and || and |
127 operators at the end of each line, rather than the start. This
128 means you don't need to use \ to join lines, since the above
129 operators imply the sequence isn't finished.
132 grep blob verify_pack_result \
133 | awk -f print_1.awk \
138 grep blob verify_pack_result |
143 - We prefer "test" over "[ ... ]".
145 - We do not write the noiseword "function" in front of shell
148 - We prefer a space between the function name and the parentheses,
149 and no space inside the parentheses. The opening "{" should also
160 - As to use of grep, stick to a subset of BRE (namely, no \{m,n\},
161 [::], [==], or [..]) for portability.
163 - We do not use \{m,n\};
165 - We do not use ? or + (which are \{0,1\} and \{1,\}
166 respectively in BRE) but that goes without saying as these
167 are ERE elements not BRE (note that \? and \+ are not even part
168 of BRE -- making them accessible from BRE is a GNU extension).
170 - Use Git's gettext wrappers in git-sh-i18n to make the user
171 interface translatable. See "Marking strings for translation" in
174 - We do not write our "test" command with "-a" and "-o" and use "&&"
175 or "||" to concatenate multiple "test" commands instead, because
176 the use of "-a/-o" is often error-prone. E.g.
178 test -n "$x" -a "$a" = "$b"
180 is buggy and breaks when $x is "=", but
182 test -n "$x" && test "$a" = "$b"
184 does not have such a problem.
186 - Even though "local" is not part of POSIX, we make heavy use of it
187 in our test suite. We do not use it in scripted Porcelains, and
188 hopefully nobody starts using "local" before they are reimplemented
191 - Use octal escape sequences (e.g. "\302\242"), not hexadecimal (e.g.
192 "\xc2\xa2") in printf format strings, since hexadecimal escape
193 sequences are not portable.
198 - We use tabs to indent, and interpret tabs as taking up to
201 - We try to keep to at most 80 characters per line.
203 - As a Git developer we assume you have a reasonably modern compiler
204 and we recommend you to enable the DEVELOPER makefile knob to
205 ensure your patch is clear of all compiler warnings we care about,
206 by e.g. "echo DEVELOPER=1 >>config.mak".
208 - We try to support a wide range of C compilers to compile Git with,
209 including old ones. As of Git v2.35.0 Git requires C99 (we check
210 "__STDC_VERSION__"). You should not use features from a newer C
211 standard, even if your compiler groks them.
213 New C99 features have been phased in gradually, if something's new
214 in C99 but not used yet don't assume that it's safe to use, some
215 compilers we target have only partial support for it. These are
216 considered safe to use:
218 . since around 2007 with 2b6854c863a, we have been using
219 initializer elements which are not computable at load time. E.g.:
221 const char *args[] = {"constant", variable, NULL};
223 . since early 2012 with e1327023ea, we have been using an enum
224 definition whose last element is followed by a comma. This, like
225 an array initializer that ends with a trailing comma, can be used
226 to reduce the patch noise when adding a new identifier at the end.
228 . since mid 2017 with cbc0f81d, we have been using designated
229 initializers for struct (e.g. "struct t v = { .val = 'a' };").
231 . since mid 2017 with 512f41cf, we have been using designated
232 initializers for array (e.g. "int array[10] = { [5] = 2 }").
234 . since early 2021 with 765dc168882, we have been using variadic
235 macros, mostly for printf-like trace and debug macros.
237 . since late 2021 with 44ba10d6, we have had variables declared in
238 the for loop "for (int i = 0; i < 10; i++)".
240 New C99 features that we cannot use yet:
242 . %z and %zu as a printf() argument for a size_t (the %z being for
243 the POSIX-specific ssize_t). Instead you should use
244 printf("%"PRIuMAX, (uintmax_t)v). These days the MSVC version we
245 rely on supports %z, but the C library used by MinGW does not.
247 . Shorthand like ".a.b = *c" in struct initializations is known to
248 trip up an older IBM XLC version, use ".a = { .b = *c }" instead.
249 See the 33665d98 (reftable: make assignments portable to AIX xlc
252 - Variables have to be declared at the beginning of the block, before
253 the first statement (i.e. -Wdeclaration-after-statement).
255 - NULL pointers shall be written as NULL, not as 0.
257 - When declaring pointers, the star sides with the variable
258 name, i.e. "char *string", not "char* string" or
259 "char * string". This makes it easier to understand code
260 like "char *string, c;".
262 - Use whitespace around operators and keywords, but not inside
263 parentheses and not around functions. So:
273 - Do not explicitly compare an integral value with constant 0 or '\0',
274 or a pointer value with constant NULL. For instance, to validate that
275 counted array <ptr, cnt> is initialized but has no elements, write:
278 BUG("empty array expected");
282 if (ptr == NULL || cnt != 0);
283 BUG("empty array expected");
285 - We avoid using braces unnecessarily. I.e.
291 is frowned upon. But there are a few exceptions:
293 - When the statement extends over a few lines (e.g., a while loop
294 with an embedded conditional, or a comment). E.g.:
305 * This one requires some explanation,
306 * so we're better off with braces to make
307 * it obvious that the indentation is correct.
312 - When there are multiple arms to a conditional and some of them
313 require braces, enclose even a single line block in braces for
324 - We try to avoid assignments in the condition of an "if" statement.
326 - Try to make your code understandable. You may put comments
327 in, but comments invariably tend to stale out when the code
328 they were describing changes. Often splitting a function
329 into two makes the intention of the code much clearer.
331 - Multi-line comments include their delimiters on separate lines from
336 * multi-line comment.
339 Note however that a comment that explains a translatable string to
340 translators uses a convention of starting with a magic token
341 "TRANSLATORS: ", e.g.
344 * TRANSLATORS: here is a comment that explains the string to
345 * be translated, that follows immediately after it.
347 _("Here is a translatable string explained by the above.");
349 - Double negation is often harder to understand than no negation
352 - There are two schools of thought when it comes to comparison,
353 especially inside a loop. Some people prefer to have the less stable
354 value on the left hand side and the more stable value on the right hand
355 side, e.g. if you have a loop that counts variable i down to the
358 while (i > lower_bound) {
363 Other people prefer to have the textual order of values match the
364 actual order of values in their comparison, so that they can
365 mentally draw a number line from left to right and place these
366 values in order, i.e.
368 while (lower_bound < i) {
373 Both are valid, and we use both. However, the more "stable" the
374 stable side becomes, the more we tend to prefer the former
375 (comparison with a constant, "i > 0", is an extreme example).
376 Just do not mix styles in the same part of the code and mimic
377 existing styles in the neighbourhood.
379 - There are two schools of thought when it comes to splitting a long
380 logical line into multiple lines. Some people push the second and
381 subsequent lines far enough to the right with tabs and align them:
383 if (the_beginning_of_a_very_long_expression_that_has_to ||
384 span_more_than_a_single_line_of ||
388 while other people prefer to align the second and the subsequent
389 lines with the column immediately inside the opening parenthesis,
390 with tabs and spaces, following our "tabstop is always a multiple
393 if (the_beginning_of_a_very_long_expression_that_has_to ||
394 span_more_than_a_single_line_of ||
398 Both are valid, and we use both. Again, just do not mix styles in
399 the same part of the code and mimic existing styles in the
402 - When splitting a long logical line, some people change line before
403 a binary operator, so that the result looks like a parse tree when
404 you turn your head 90-degrees counterclockwise:
406 if (the_beginning_of_a_very_long_expression_that_has_to
407 || span_more_than_a_single_line_of_the_source_text) {
409 while other people prefer to leave the operator at the end of the
412 if (the_beginning_of_a_very_long_expression_that_has_to ||
413 span_more_than_a_single_line_of_the_source_text) {
415 Both are valid, but we tend to use the latter more, unless the
416 expression gets fairly complex, in which case the former tends to
417 be easier to read. Again, just do not mix styles in the same part
418 of the code and mimic existing styles in the neighbourhood.
420 - When splitting a long logical line, with everything else being
421 equal, it is preferable to split after the operator at higher
422 level in the parse tree. That is, this is more preferable:
424 if (a_very_long_variable * that_is_used_in +
425 a_very_long_expression) {
430 if (a_very_long_variable *
431 that_is_used_in + a_very_long_expression) {
434 - Some clever tricks, like using the !! operator with arithmetic
435 constructs, can be extremely confusing to others. Avoid them,
436 unless there is a compelling reason to use them.
438 - Use the API. No, really. We have a strbuf (variable length
439 string), several arrays with the ALLOC_GROW() macro, a
440 string_list for sorted string lists, a hash map (mapping struct
441 objects) named "struct decorate", amongst other things.
443 - When you come up with an API, document its functions and structures
444 in the header file that exposes the API to its callers. Use what is
445 in "strbuf.h" as a model for the appropriate tone and level of
448 - The first #include in C files, except in platform specific compat/
449 implementations and sha1dc/, must be <git-compat-util.h>. This
450 header file insulates other header files and source files from
451 platform differences, like which system header files must be
452 included in what order, and what C preprocessor feature macros must
453 be defined to trigger certain features we expect out of the system.
454 A collorary to this is that C files should not directly include
455 system header files themselves.
457 There are some exceptions, because certain group of files that
458 implement an API all have to include the same header file that
459 defines the API and it is convenient to include <git-compat-util.h>
462 - the implementation of the built-in commands in the "builtin/"
463 directory that include "builtin.h" for the cmd_foo() prototype
466 - the test helper programs in the "t/helper/" directory that include
467 "t/helper/test-tool.h" for the cmd__foo() prototype definition,
469 - the xdiff implementation in the "xdiff/" directory that includes
470 "xdiff/xinclude.h" for the xdiff machinery internals,
472 - the unit test programs in "t/unit-tests/" directory that include
473 "t/unit-tests/test-lib.h" that gives them the unit-tests
476 - the source files that implement reftable in the "reftable/"
477 directory that include "reftable/system.h" for the reftable
480 are allowed to assume that they do not have to include
481 <git-compat-util.h> themselves, as it is included as the first
482 '#include' in these header files. These headers must be the first
483 header file to be "#include"d in them, though.
485 - A C file must directly include the header files that declare the
486 functions and the types it uses, except for the functions and types
487 that are made available to it by including one of the header files
488 it must include by the previous rule.
490 - If you are planning a new command, consider writing it in shell
491 or perl first, so that changes in semantics can be easily
492 changed and discussed. Many Git commands started out like
493 that, and a few are still scripts.
495 - Avoid introducing a new dependency into Git. This means you
496 usually should stay away from scripting languages not already
497 used in the Git core command set (unless your command is clearly
498 separate from it, such as an importer to convert random-scm-X
499 repositories to Git).
501 - When we pass <string, length> pair to functions, we should try to
502 pass them in that order.
504 - Use Git's gettext wrappers to make the user interface
505 translatable. See "Marking strings for translation" in po/README.
507 - Variables and functions local to a given source file should be marked
508 with "static". Variables that are visible to other source files
509 must be declared with "extern" in header files. However, function
510 declarations should not use "extern", as that is already the default.
512 - You can launch gdb around your program using the shorthand GIT_DEBUGGER.
513 Run `GIT_DEBUGGER=1 ./bin-wrappers/git foo` to simply use gdb as is, or
514 run `GIT_DEBUGGER="<debugger> <debugger-args>" ./bin-wrappers/git foo` to
515 use your own debugger and arguments. Example: `GIT_DEBUGGER="ddd --gdb"
516 ./bin-wrappers/git log` (See `wrap-for-bin.sh`.)
520 - Most of the C guidelines above apply.
522 - We try to support Perl 5.8.1 and later ("use Perl 5.008001").
524 - use strict and use warnings are strongly preferred.
526 - Don't overuse statement modifiers unless using them makes the
527 result easier to follow.
530 do_this() unless (condition);
531 ... do something else ...
533 is more readable than:
539 ... do something else ...
541 *only* when the condition is so rare that do_this() will be almost
544 - We try to avoid assignments inside "if ()" conditions.
546 - Learn and use Git.pm if you need that functionality.
550 - We follow PEP-8 (https://peps.python.org/pep-0008/).
552 - As a minimum, we aim to be compatible with Python 2.7.
554 - Where required libraries do not restrict us to Python 2, we try to
555 also be compatible with Python 3.1 and later.
560 We make a distinction between a Git command's primary output and
561 output which is merely chatty feedback (for instance, status
562 messages, running transcript, or progress display), as well as error
563 messages. Roughly speaking, a Git command's primary output is that
564 which one might want to capture to a file or send down a pipe; its
565 chatty output should not interfere with these use-cases.
567 As such, primary output should be sent to the standard output stream
568 (stdout), and chatty output should be sent to the standard error
569 stream (stderr). Examples of commands which produce primary output
570 include `git log`, `git show`, and `git branch --list` which generate
571 output on the stdout stream.
573 Not all Git commands have primary output; this is often true of
574 commands whose main function is to perform an action. Some action
575 commands are silent, whereas others are chatty. An example of a
576 chatty action commands is `git clone` with its "Cloning into
577 '<path>'..." and "Checking connectivity..." status messages which it
578 sends to the stderr stream.
580 Error messages from Git commands should always be sent to the stderr
586 - Do not end error messages with a full stop.
588 - Do not capitalize the first word, only because it is the first word
589 in the message ("unable to open %s", not "Unable to open %s"). But
590 "SHA-3 not supported" is fine, because the reason the first word is
591 capitalized is not because it is at the beginning of the sentence,
592 but because the word would be spelled in capital letters even when
593 it appeared in the middle of the sentence.
595 - Say what the error is first ("cannot open %s", not "%s: cannot open")
598 Externally Visible Names
600 - For configuration variable names, follow the existing convention:
602 . The section name indicates the affected subsystem.
604 . The subsection name, if any, indicates which of an unbounded set
605 of things to set the value for.
607 . The variable name describes the effect of tweaking this knob.
609 The section and variable names that consist of multiple words are
610 formed by concatenating the words without punctuation marks (e.g. `-`),
611 and are broken using bumpyCaps in documentation as a hint to the
614 When choosing the variable namespace, do not use variable name for
615 specifying possibly unbounded set of things, most notably anything
616 an end user can freely come up with (e.g. branch names). Instead,
617 use subsection names or variable values, like the existing variable
618 branch.<name>.description does.
621 Writing Documentation:
623 Most (if not all) of the documentation pages are written in the
624 AsciiDoc format in *.txt files (e.g. Documentation/git.txt), and
625 processed into HTML and manpages (e.g. git.html and git.1 in the
628 The documentation liberally mixes US and UK English (en_US/UK)
629 norms for spelling and grammar, which is somewhat unfortunate.
630 In an ideal world, it would have been better if it consistently
631 used only one and not the other, and we would have picked en_US
632 (if you wish to correct the English of some of the existing
633 documentation, please see the documentation-related advice in the
634 Documentation/SubmittingPatches file).
636 In order to ensure the documentation is inclusive, avoid assuming
637 that an unspecified example person is male or female, and think
638 twice before using "he", "him", "she", or "her". Here are some
639 tips to avoid use of gendered pronouns:
641 - Prefer succinctness and matter-of-factly describing functionality
642 in the abstract. E.g.
644 `--short`:: Emit output in the short-format.
646 and avoid something like these overly verbose alternatives:
648 `--short`:: Use this to emit output in the short-format.
649 `--short`:: You can use this to get output in the short-format.
650 `--short`:: A user who prefers shorter output could....
651 `--short`:: Should a person and/or program want shorter output, he
654 This practice often eliminates the need to involve human actors in
655 your description, but it is a good practice regardless of the
656 avoidance of gendered pronouns.
658 - When it becomes awkward to stick to this style, prefer "you" when
659 addressing the hypothetical user, and possibly "we" when
660 discussing how the program might react to the user. E.g.
662 You can use this option instead of `--xyz`, but we might remove
663 support for it in future versions.
665 while keeping in mind that you can probably be less verbose, e.g.
667 Use this instead of `--xyz`. This option might be removed in future
670 - If you still need to refer to an example person that is
671 third-person singular, you may resort to "singular they" to avoid
672 "he/she/him/her", e.g.
674 A contributor asks their upstream to pull from them.
676 Note that this sounds ungrammatical and unnatural to those who
677 learned that "they" is only used for third-person plural, e.g.
678 those who learn English as a second language in some parts of the
681 Every user-visible change should be reflected in the documentation.
682 The same general rule as for code applies -- imitate the existing
688 Literal parts (e.g. use of command-line options, command names,
689 branch names, URLs, pathnames (files and directories), configuration and
690 environment variables) must be typeset as verbatim (i.e. wrapped with
695 `http://git.example.com`
701 An environment variable must be prefixed with "$" only when referring to its
702 value and not when referring to the variable itself, in this case there is
703 nothing to add except the backticks:
704 `GIT_DIR` is specified
705 `$GIT_DIR/hooks/pre-receive`
707 Word phrases enclosed in `backtick characters` are rendered literally
708 and will not be further expanded. The use of `backticks` to achieve the
709 previous rule means that literal examples should not use AsciiDoc
716 Placeholders are spelled in lowercase and enclosed in
717 angle brackets surrounded by underscores:
721 If a placeholder has multiple words, they are separated by dashes:
723 _<template-directory>_
725 A placeholder is not enclosed in backticks, as it is not a literal.
727 When needed, use a distinctive identifier for placeholders, usually
728 made of a qualification and a type:
732 When literal and placeholders are mixed, each markup is applied for
733 each sub-entity. If they are stuck, a special markup, called
734 unconstrained formatting is required.
735 Unconstrained formating for placeholders is __<like-this>__
736 Unconstrained formatting for literal formatting is ++like this++
739 __<directory>__++/.git++
740 ++remote.++__<name>__++.mirror++
742 caveat: ++ unconstrained format is not verbatim and may expand
743 content. Use Asciidoc escapes inside them.
747 Syntax grammar is formatted neither as literal nor as placeholder.
749 A few commented examples follow to provide reference when writing or
750 modifying command usage strings and synopsis sections in the manual
753 Possibility of multiple occurrences is indicated by three dots:
755 (One or more of <file>.)
757 Optional parts are enclosed in square brackets:
759 (Zero or more of <file>.)
761 ++--exec-path++[++=++__<path>__]
762 (Option with an optional argument. Note that the "=" is inside the
766 (Zero or more of <patch>. Note that the dots are inside, not
767 outside the brackets.)
769 Multiple alternatives are indicated with vertical bars:
771 [`--utf8` | `--no-utf8`]
773 Use spacing around "|" token(s), but not immediately after opening or
774 before closing a [] or () pair:
775 Do: [`-q` | `--quiet`]
776 Don't: [`-q`|`--quiet`]
778 Don't use spacing around "|" tokens when they're used to separate the
779 alternate arguments of an option:
780 Do: ++--track++[++=++(`direct`|`inherit`)]`
781 Don't: ++--track++[++=++(`direct` | `inherit`)]
783 Parentheses are used for grouping:
784 [(_<rev>_ | _<range>_)...]
785 (Any number of either <rev> or <range>. Parens are needed to make
786 it clear that "..." pertains to both <rev> and <range>.)
788 [(`-p` _<parent>_)...]
789 (Any number of option -p, each with one <parent> argument.)
791 `git remote set-head` _<name>_ (`-a` | `-d` | _<branch>_)
792 (One and only one of "-a", "-d" or "<branch>" _must_ (no square
793 brackets) be provided.)
795 And a somewhat more contrived example:
796 `--diff-filter=[(A|C|D|M|R|T|U|X|B)...[*]]`
797 Here "=" is outside the brackets, because "--diff-filter=" is a
798 valid usage. "*" has its own pair of brackets, because it can
799 (optionally) be specified only when one or more of the letters is
803 Use 'git' (all lowercase) when talking about commands i.e. something
804 the user would type into a shell and use 'Git' (uppercase first letter)
805 when talking about the version control system and its properties.
807 If some place in the documentation needs to typeset a command usage
808 example with inline substitutions, it is fine to use +monospaced and
809 inline substituted text+ instead of `monospaced literal text`, and with
810 the former, the part that should not get substituted must be