Commit message guidelines: Difference between revisions

From Octave
Jump to navigation Jump to search
(→‎Examples: Update punctuation to GNU style)
 
(20 intermediate revisions by 8 users not shown)
Line 1: Line 1:
Our commit messages for Mercurial get automatically distilled into GNU Changelog entries. The GNU coding standards have [http://www.gnu.org/prep/standards/html_node/Style-of-Change-Logs.html 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.
Commit messages for [[Mercurial]] get automatically distilled into GNU Changelog entries. The GNU coding standards have [http://www.gnu.org/prep/standards/html_node/Style-of-Change-Logs.html some guidelines] for how to write Changelogs, and since Octave is a GNU project, we try to produce Changelogs in this style.


You can see how Mercurial will produce the Changelog-style output with the following command:
You can see how Mercurial will produce the Changelog-style output with the following command:


   hg log --style changelog
   hg log --template changelog


== Guidelines ==
For more options, see [https://www.mercurial-scm.org/repo/hg/help/log the Mercurial manual].
General structure of the commit message:
 
:: '''One-line description'''
== Examples ==
:: Empty line
 
:: '''Body of the commit message'''
<pre>
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.
</pre>
 
<pre>
Allow abbreviations for optimset and optimget (bug #38999).
 
* optimset.m, optimget.m: Handle abbreviated keys and warn for
ambiguous abbreviations. New tests.
</pre>
 
<pre>
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.
</pre>
 
<pre>
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.
</pre>


Each part is explained in detail below.
Here there is an example:
<pre>
<pre>
Give message id to undefined function error
tar, untar, unpack: Add support for BSD tar (bug #53695).


* pt-id.cc (tree_identifier::eval_undefined_error): used
* tar_is_bsd.m: New function.
  ::error_with_id instead of ::error to report undefined functions
* tar.m: Use it to determine how to run tar and parse command output.
  with id.
* unpack.m: Likewise.
</pre>
</pre>


=== One-line description ===  
=== One line examples ===
 
This examples are the rare cases where only one file is modified and the change is simple enough:
 
maint: Merge stable to default.
maint: Merge away accidental head.
maint: Strip trailing whitespace from source files.
maint: Update gnulib to latest changes.
doc: Grammarcheck documentation for 4.2 release.
pkg.m4: Update to latest version as released with pkg-config 0.29 (bug #48775).
uigetfile.m: Allow path names as input arg (bug #48828).
 
== Guidelines ==
 
The general structure of a commit message should be clear from the examples:
 
* After the first line, leave one blank line.
* Do not end the first line with a period (full stop).
* Keep it short, no longer than 80 characters.
* Add the bug number, e.g. <code>(bug #12345)</code>, where applicable.
* Use prefixes where applicable:
** <code>build:</code> for changes to the build system, for example autoconf or automake files.
** <code>doc:</code> for changes to the documentation.
** <code>gui:</code> for changes to the graphical user interface.
** <code>maint:</code> for reorganization of the sources that do not change the source. Regular merge commits are a prominent example.
** <code>test:</code> for changes to [[Tests]] only, e.g. new/removed BISTs, changed tolerances, etc.


The commit message has a one-line description. Keep it short, preferrably under 80 characters. If you're squashing a bug, the bug number you're squashing should be included in this one-line description, like so at the end: <tt>(bug #12345)</tt>. Omit final period (full stop).


Sometimes the one-line description is sufficient, for small changes. In that case, we use the following prefixes to give a description of what the small change is:
=== Wording ===


* maint: for reorganisation of the sources that do not change the source. Regular merge commits are a prominent example.
{| class="wikitable"
* doc: for changes to the documentation.
! bad
! good
|-
| style="color:darkred;" | "Added function"
| style="color:green;" | "New function"
|-
| style="color:darkred;" | "Changed to return retval"
| style="color:green;" | "Return retval"
|-
| style="color:darkred;" | "Fixed bug"
| style="color:green;" | Write '''what''' has changed.
|}


If your change only touches one file, then the filename of that file should 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.
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.


=== Body of the 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 commit 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.


Individual files you touched get their changes separately enumerated. If there a particular C or C++ function you changed, the general format is
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 (class1::function1): Made change foo.  
* file.cc (function1): Add an option. (function2): Add another option.
      (class2::function2): Made change bar.


Note the newline for describing the change to a different function. General GNU style guidelines can be followed here, so that similar changes to different files can be grouped.
Only the last file name component is typically needed, since most files have unique names across the entire repository. One notable exception are the {{codeline|module.mk}} files in every directory, they should include the complete directory and file name. For example


For style points, please also write "New function" instead of "Added function" or "Return retval" instead of "Changed to return retval".
* doc/interpreter/module.mk (dist_man_MANS): Include foo.1 in the list.


And please, never "Fixed bug" or similar. That doesn't add any specific information about what was changed.
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


The changelog info in the commit message should say what changed, not why. If you need to explain why, that info should probably go in comments in the code.
* oct-fftw.cc, oct-fftw.h (octave_fftw_version): New function.


For m-script and Fortran sources, the function name can be omitted if the m-script only contains one file. For changes outside of functions or classes, of course the parenthetical (function) or (class::function) specifiers can also be omitted.
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.


[[Category:Development]]
[[Category:Development]]

Latest revision as of 12:10, 14 March 2023

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.

You can see how Mercurial will produce the Changelog-style output with the following command:

 hg log --template changelog

For more options, see the Mercurial manual.

Examples[edit]

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 examples[edit]

This examples are the rare cases where only one file is modified and the change is simple enough:

maint: Merge stable to default.

maint: Merge away accidental head.

maint: Strip trailing whitespace from source files.

maint: Update gnulib to latest changes.

doc: Grammarcheck documentation for 4.2 release.

pkg.m4: Update to latest version as released with pkg-config 0.29 (bug #48775).

uigetfile.m: Allow path names as input arg (bug #48828).

Guidelines[edit]

The general structure of a commit message should be clear from the examples:

  • After the first line, leave one blank line.
  • Do not end the first line with a period (full stop).
  • Keep it short, no longer than 80 characters.
  • Add the bug number, e.g. (bug #12345), where applicable.
  • Use prefixes where applicable:
    • build: for changes to the build system, for example autoconf or automake files.
    • doc: for changes to the documentation.
    • gui: for changes to the graphical user interface.
    • maint: for reorganization of the sources that do not change the source. Regular merge commits are a prominent example.
    • test: for changes to Tests only, e.g. new/removed BISTs, changed tolerances, etc.


Wording[edit]

bad good
"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.

Body of the commit message[edit]

Each individual file changed by a commit 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.