Git commit messages format
==========================

The principle of the git commit log is to document at least the
*what*; in English.  That is redundant with the commit diff, yes.  But
that redundancy does help in understanding the commit diff better.  If
appropriate, the commit log can also document the *why*, but only if
it does so by respecting the format of the commit log.  The reason why
we are so strict about the format is that the commit log is later
parsed by a tool to build a ChangeLog, which we want to stay compliant
with the GNU ChangeLog format.

So here is the format we are talking about.

The first line of a git commit message should start at column 1, with
no space. That line should be a short summary of the purpose of the
commit.  If the commit relates to a bug filed into bugzilla, the line
should begin with the bug (or Problem Report) number, followed by a
white space; e.g:

Bug <number-of-the-bug> This is a great commit

The line in its entirety should not be longer than 50 characters.

The next line should be an empty line, with no spaces.

Subsequent lines can be a free form introductory text that should
start column 0.  The introductory text can have an arbitrary number of
lines.  No line in that text should start with the sequence
<white-space>*.  That is, no line in that text should start with a
sequence of white spaces followed by the start character (*).

If there was an introductory text, then the next line should be an
empty line, with no spaces.

The subsequent lines should have the form of the Body of a GNU ChangeLog
entry, i.e:

	* file1.c (func1): Changed foo in this function.
	(func2): Changed blah in that function
	* file2.c (func_foo): Changed something here.

Note that before the '*', there is a tab that is 8 spaces long.  Also
note that right after the '*', there is a space.

An example of commit message would be:

~=~
Bug 123456 Add super feature

The super feature requires modifying function_bleh to make it call
function_foo instead of function_bar.  As function_bar is no more
used, this patch removes it.

	* file1.c (function_foo): Define new static function.
	* file2.c (function_bar): Removed static function.
	* file3.c (function_bleh): Modified this function to call
	function_foo, rather than function function_bar.
~=~

Note how, in the ChangeLog part of the commit log, each function
modification is mentioned, by referring to the name of the function in
parenthesis.  The length of a line should not exceed 72 characters.
The description of what happens to the function should be succinct.
Just describe the "what".

The "how" should be described by comments in the code change itself,
so there is no need to describe in the ChangeLog part of the commit
log..  The "why" and "general spirit" of the change should be
described in the introductory text that precedes the ChangeLog part.
So, again, no need to add to the ChangeLog part.

For files that contain no function definitions, the ChangeLog looks
like:

~=~ Bug 123456 Shorten compilation lines

	* configure.ac: Shorten compilation lines by regrouping
	PKG_CHECK_MODULES calls.
	* tests/Makefile.am: Adjust this.
~=~

Another one could be:

~=~
Bug 123456 Shorten compilation lines

Blah blah, this is an introductory text explaining the purpose of this
commit.  It can contain whatever character I want.  It just cannot
contain a line that starts with white spaces immediately followed by
the star character.

	* configure.ac: Shorten compilation lines by regrouping
	PKG_CHECK_MODULES calls.
	* tests/Makefile.am: Adjust this.
~=~

When it's needed, the introductory text is very important.  Please
take time to explain the current status of the code (before your
patch) and why it was in the need of your patch.  In other words, take
time to explain the problem you are trying to solve.  If the problem
is explained in a bug in the bugzilla, please try to explain it again,
using your words.  Just linking to the bugzilla is generally not
enough.  And then, yes, refer to the bugzilla.

Then explain how your changes address the issue that you've just
described.

In other words, the introductory text should tell a story.  So please
be generous :-)

And then the ChangeLog part of the commit log is another exercise.
This one has to be succinct.  Every single function or global variable
or type changed (or added/removed) has to be mentioned explicitly by
name as shown in one of the examples above.

Writing the ChangeLog part of the commit log can seem tedious, but
it's an excellent way to perform an auto-review of your work.  You'd
be surprised by the number of issues that you catch in the process.

Also, please keep in mind that reviewers of the code are going to do a
function-by-function and line-by-line review of your changes, so
please, take the time to do so as well.  This is a great way to give
great quality to your code.

We encourage you to look at the existing commit logs or at the
ChangeLog file for inspiration.

Happy Hacking!