Commit message guidelines
Our commit messages for Mercurial get automatically distilled into GNU Changelog entries. The GNU coding standards have some guidelines for how to write Changelogs, and since Octave is a GNU project, we try to produce Changelogs in this style. However, certain things have to be adapted because the style in there is primarily for C sources, and because we are producing them from Mercurial commit messages.
You can see how Mercurial will produce the Changelog-style output with the following command:
hg log --style changelog
- Note that this command will print all changelogs to the screen, currently including all changelogs back to 2008 and approaching 200,0000 lines of text. You may use the followig command for a paged output:
hg log --style changelog | less
- Alternatively, you may save the changlogs to a text file. This will permit viewing and searching in a text editor for use in preparing your own commit messages:
hg log --style changelog >> octave_changelogs.log
General structure of a commit message:
- One-line description
- Empty line
- Body of the commit message
The commit message should start with a brief one-line description of what the
commit does. Keep it short, no longer than 80 characters. If you are working
on a bug or applying a patch, this one-line explanation should mention the bug
or patch number at the end like so:
... (bug #12345). Do not end
the first line with a period (full stop).
If your change only touches one file, then the name of that file can be the prefix of the one-line description. If it's a C++ or C file, the function or class that is being modified should be included in the parenthetical remark, as in the full body of the commit message.
In addition, there are a few prefixes for certain types of commits:
- maint: for reorganisation of the sources that do not change the source. Regular merge commits are a prominent example.
- doc: for changes to the documentation.
- build: for changes to the build system, for example autoconf or automake files.
If your change is small and only touches one file, then the one-line description may serve as the entire commit message.
Body of the commit message
If there is more than one file touched in different ways and the one-line description isn't enough to describe all changes, the commit message needs a full-body description.
Each individual file changed by a commmit must have its changes enumerated. For changes affecting specific C++ functions, each function name is listed in parentheses. For example
* file.cc (class1::function1): Add something. (function2, function3): Delete something else.
For changes affecting specific Octave built-ins, each built-in name is listed in parentheses with an "F" prefix, an implementation detail. For example
* data.cc (Fcolumns): Return columns.
When the same change is applied to a series of files, or to a set of functions in a single file, the file or function names may be grouped to shorten the commit message. For example:
* file1.cc, file2.cc, file3.cc, file4.cc: Include <sys/types.h>. * memory.cc (function1, function2, function3): Throw error if empty.
Each line of the commit message body should also be kept under 80 columns. The GNU standards recommend starting a new line for each parenthesized function, but if the line is short enough, we often avoid an extra newline. For example
* file.cc (function1): Add an option. (function2): Add another option.
Only the last file name component is typically needed, since most files have
unique names across the entire repository. One notable exception are the
module.mk files in every directory, they should include the
complete directory and file name. For example
* doc/interpreter/module.mk (dist_man_MANS): Include foo.1 in the list.
Avoid abbreviating or using shell globs or patterns when listing the names of files affected by a change, even when they have the same name with different file extensions. For example
* oct-fftw.cc, oct-fftw.h (octave_fftw_version): New function.
For m-file and Fortran sources, the function name can be omitted if the file contains only one function. For changes outside of functions or classes, of course the parenthetical (function) or (class::function) specifiers can also be omitted.
|"Added function"||"New function"|
|"Changed to return retval"||"Return retval"|
|"Fixed bug"||Write what has changed.|
The commit message should describe what was changed, not why it was changed. Any explanations should appear as comments in the code, particularly if there is something that might not be obvious to someone reading it later.
look for methods before constructors * symtab.cc (symbol_table::fcn_info::fcn_info_rep::find): Look for class methods before constructors, contrary to Matlab documentation. * test/ctor-vs-method: New directory of test classes. * test/test_ctor_vs_method.m: New file. * test/Makefile.am: Include ctor-vs-method/module.mk. (FCN_FILES): Include test_ctor_vs_method.m in the list.
allow abbreviations for optimset and optimget (bug #38999) * optimset.m, optimget.m: Handle abbreviated keys and warn for ambiguous abbreviations. New tests.
add format option to ticklabel (bug #34906) * graphics.cc: add new functions to support different input arguments to xyzticklabel. Add tests. * graphics.in.h: define set_xyzticklabel as external function
tag symbols in indexed assignments as variables (bug #39240) * pt-arg-list.cc (tree_argument_list::variable_names): Also return the symbol names from index expressions. * parser.tst: New test.
tar, untar, unpack: Add support for BSD tar (bug #53695) * tar_is_bsd.m: New function. * tar.m: Use it to determine how to run tar and parse command output. * unpack.m: Likewise.
One line commit examples
This examples are the rare cases where only one file is modified and the change is simple enough:
maint: merge away accidental head.
maint: Strip trailing whitespace from source files.
maint: Update gnulib to latest changes.
maint: Periodic merge of stable to default.
doc: grammarcheck documentation for 4.2 release.
pkg.m4: update to lastest version as released with pkg-config 0.29 (bug #48775)
uigetfile.m: allow path names as input arg (bug #48828)