12 KiB
Git commit conventions
Each commit message should start with a directory or full file path prefix, so it was clear which part of codebase a commit affects. If a change affects one file, it's better to use path to a file. If it affects few files in a subdirectory, using subdirectory as a prefix is ok. For longish paths, it's acceptable to drop intermediate components, which still should provide good context of a change. It's also ok to drop file extensions.
Besides prefix, first line of a commit message should describe a change clearly and to the point, and be a grammatical sentence with final full stop. First line must fit within 72 characters. Examples of good first line of commit messages:
py/objstr: Add splitlines() method.
py: Rename FOO to BAR.
docs/machine: Fix typo in reset() description.
ports: Switch to use lib/foo instead of duplicated code.
After the first line add an empty line and in the following lines describe the change in a detail, if needed, with lines fitting within 75 characters (with an exception for long items like URLs which cannot be broken). Any change beyond 5 lines would likely require such detailed description.
To get good practical examples of good commits and their messages, browse
the git log
of the project.
When committing you must sign-off your commit by adding "Signed-off-by:"
line(s) at the end of the commit message, e.g. using git commit -s
. You
are then certifying and signing off against the following:
- That you wrote the change yourself, or took it from a project with a compatible license (in the latter case the commit message, and possibly source code should provide reference where the implementation was taken from and give credit to the original author, as required by the license).
- That you are allowed to release these changes to an open-source project (for example, changes done during paid work for a third party may require explicit approval from that third party).
- That you (or your employer) agree to release the changes under MicroPython's license, which is the MIT license. Note that you retain copyright for your changes (for smaller changes, the commit message conveys your copyright; if you make significant changes to a particular source module, you're welcome to add your name to the file header).
- Your contribution including commit message will be publicly and indefinitely available for anyone to access, including redistribution under the terms of the project's license.
- Your signature for all of the above, which is the "Signed-off-by" line, includes your full real name and a valid and active email address by which you can be contacted in the foreseeable future.
Code auto-formatting
Both C and Python code formatting are controlled for consistency across the
MicroPython codebase. C code is formatted using the tools/codeformat.py
script which uses uncrustify.
Python code is linted and formatted using
ruff & ruff format.
After making changes, and before committing, run tools/codeformat.py
to
reformat your C code and ruff format
for any Python code. Without
arguments this tool will reformat all source code (and may take some time
to run). Otherwise pass as arguments to the tool the files that changed,
and it will only reformat those.
uncrustify
Only uncrustify v0.71 or v0.72 can be used for MicroPython. Different uncrustify versions produce slightly different formatting, and the configuration file formats are often incompatible. v0.73 or newer will not work.
Depending on your operating system version, it may be possible to install a pre-compiled uncrustify version:
Ubuntu, Debian
Ubuntu versions 21.10 or 22.04LTS, and Debian versions bullseye or bookworm all include v0.72 so can be installed directly:
$ apt install uncrustify
Arch Linux
The current Arch uncrustify version is too new. There is an old Arch package for v0.72 that can be installed from the Arch Linux archive (more information). Use the IgnorePkg feature to prevent it re-updating.
Brew
This command may work, please raise a new Issue if it doesn't:
curl -L https://github.com/Homebrew/homebrew-core/raw/2b07d8192623365078a8b855a164ebcdf81494a6/Formula/uncrustify.rb > uncrustify.rb && brew install uncrustify.rb && rm uncrustify.rb
Automatic Pre-Commit Hooks
To have code formatting and commit message conventions automatically checked, a configuration file is provided for the pre-commit tool.
First install pre-commit
, either from your system package manager or via
pip
. When installing pre-commit
via pip, it is recommended to use a
virtual environment. Other sources, such as Brew are also available, see
the docs for details.
$ apt install pre-commit # Ubuntu, Debian
$ pacman -Sy python-precommit # Arch Linux
$ brew install pre-commit # Brew
$ pip install pre-commit # PyPI
Next, install uncrustify (see above). Other dependencies are managed by pre-commit automatically, but uncrustify needs to be installed and available on the PATH.
Then, inside the MicroPython repository, register the git hooks for pre-commit by running:
$ pre-commit install --hook-type pre-commit --hook-type commit-msg
pre-commit will now automatically run during git commit
for both code and
commit message formatting.
The same formatting checks will be run by CI for any Pull Request submitted to MicroPython. Pre-commit allows you to see any failure more quickly, and in many cases will automatically correct it in your local working copy.
To unregister pre-commit
from your MicroPython repository, run:
$ pre-commit uninstall --hook-type pre-commit --hook-type commit-msg
Tips:
- To skip pre-commit checks on a single commit, use
git commit -n
(for--no-verify
). - To ignore the pre-commit message format check temporarily, start the commit message subject line with "WIP" (for "Work In Progress").
Running pre-commit manually
Once pre-commit is installed as per the previous section it can be manually run against the MicroPython python codebase to update file formatting on demand, with either:
pre-commit run --all-files
to fix all files in the MicroPython codebasepre-commit run --file ./path/to/my/file
to fix just one filepre-commit run --file ./path/to/my/folder/*
to fix just one folder
Python code conventions
Python code follows PEP 8 and is auto-formatted using ruff format with a line-length of 99 characters.
Naming conventions:
- Module names are short and all lowercase; eg pyb, stm.
- Class names are CamelCase, with abbreviations all uppercase; eg I2C, not I2c.
- Function and method names are all lowercase with words separated by a single underscore as necessary to improve readability; eg mem_read.
- Constants are all uppercase with words separated by a single underscore; eg GPIO_IDR.
C code conventions
C code is auto-formatted using uncrustify
and the corresponding configuration file tools/uncrustify.cfg
, with a few
minor fix-ups applied by tools/codeformat.py
. When writing new C code please
adhere to the existing style and use tools/codeformat.py
to check any changes.
The main conventions, and things not enforceable via the auto-formatter, are
described below.
White space:
- Expand tabs to 4 spaces.
- Don't leave trailing whitespace at the end of a line.
- For control blocks (if, for, while), put 1 space between the keyword and the opening parenthesis.
- Put 1 space after a comma, and 1 space around operators.
Braces:
- Use braces for all blocks, even no-line and single-line pieces of code.
- Put opening braces on the end of the line it belongs to, not on a new line.
- For else-statements, put the else on the same line as the previous closing brace.
Header files:
- Header files should be protected from multiple inclusion with #if directives. See an existing header for naming convention.
Names:
- Use underscore_case, not camelCase for all names.
- Use CAPS_WITH_UNDERSCORE for enums and macros.
- When defining a type use underscore_case and put '_t' after it.
Integer types: MicroPython runs on 16, 32, and 64 bit machines, so it's important to use the correctly-sized (and signed) integer types. The general guidelines are:
- For most cases use mp_int_t for signed and mp_uint_t for unsigned integer values. These are guaranteed to be machine-word sized and therefore big enough to hold the value from a MicroPython small-int object.
- Use size_t for things that count bytes / sizes of objects.
- You can use int/uint, but remember that they may be 16-bits wide.
- If in doubt, use mp_int_t/mp_uint_t.
Comments:
- Be concise and only write comments for things that are not obvious.
- Use
//
prefix, NOT/* ... */
. No extra fluff.
Memory allocation:
- Use m_new, m_renew, m_del (and friends) to allocate and free heap memory. These macros are defined in py/misc.h.
Examples
Braces, spaces, names and comments:
#define TO_ADD (123)
// This function will always recurse indefinitely and is only used to show
// coding style
int foo_function(int x, int some_value) {
if (x < some_value) {
foo(some_value, x);
} else {
foo(x + TO_ADD, some_value - 1);
}
for (int my_counter = 0; my_counter < x; ++my_counter) {
}
}
Type declarations:
typedef struct _my_struct_t {
int member;
void *data;
} my_struct_t;
Documentation conventions
MicroPython generally follows CPython in documentation process and conventions. reStructuredText syntax is used for the documentation.
Specific conventions/suggestions:
- Use
*
markup to refer to arguments of a function, e.g.:
.. method:: poll.unregister(obj)
Unregister *obj* from polling.
- Use following syntax for cross-references/cross-links:
:func:`foo` - function foo in current module
:func:`module1.foo` - function foo in module "module1"
(similarly for other referent types)
:class:`Foo` - class Foo
:meth:`Class.method1` - method1 in Class
:meth:`~Class.method1` - method1 in Class, but rendered just as "method1()",
not "Class.method1()"
:meth:`title <method1>` - reference method1, but render as "title" (use only
if really needed)
:mod:`module1` - module module1
`symbol` - generic xref syntax which can replace any of the above in case
the xref is unambiguous. If there's ambiguity, there will be a warning
during docs generation, which need to be fixed using one of the syntaxes
above
- Cross-referencing arbitrary locations
.. _xref_target:
Normal non-indented text.
This is :ref:`reference <xref_target>`.
(If xref target is followed by section title, can be just
:ref:`xref_target`).
- Linking to external URL:
`link text <http://foo.com/...>`_
- Referencing builtin singleton objects:
``None``, ``True``, ``False``
- Use following syntax to create common description for more than one element:
.. function:: foo(x)
bar(y)
Description common to foo() and bar().
More detailed guides and quickrefs: