drgn/CONTRIBUTING.rst

Contributing
============

Thanks for your interest in drgn! See below for how to build, test, code, and
submit changes for drgn.

Building
--------

The easiest way to develop drgn is by building and running it locally. Please
build with warnings enabled. Install the dependencies from the `installation
instructions <README.rst#from-source>`_, then run:

.. code-block:: console

    $ git clone https://github.com/osandov/drgn.git
    $ cd drgn
    $ CONFIGURE_FLAGS="--enable-compiler-warnings=error" python3 setup.py build_ext -i
    $ python3 -m drgn --help

Drgn can build, run, and pass its test suite on Python 3.6 or later. However,
many of the tools used as part of the development workflow do not support Python
versions once they have reached their end-of-life. Thus, your main drgn
development environment should use a Python version which is actively supported
upstream. In particular, the drgn development workflow no longer supported on
Python 3.6.

Testing
-------

Tests should be added for all features and bug fixes.

drgn's test suite can be run with:

.. code-block:: console

    $ python3 setup.py test

To run Linux kernel helper tests in a virtual machine on all supported kernels,
add ``-K``. See `vmtest <vmtest/README.rst>`_ for more details.

Tests can also be run manually with `unittest
<https://docs.python.org/3/library/unittest.html#command-line-interface>`_
after building locally:

.. code-block:: console

    $ python3 -m unittest discover -v

To run Linux kernel helper tests on the running kernel, this must be run as
root, and debug information for the running kernel must be available.

pre-commit
----------

Several linters and checks are run on every pull request. If you'd like to run
them locally prior to submission, you can install `pre-commit
<https://pre-commit.com/>`_:

.. code-block:: console

    $ pip install pre-commit

Then, you can either install the checks as Git hooks so that they're run when
creating a commit:

.. code-block:: console

    $ pre-commit install --install-hooks

Or you can run them manually:

.. code-block:: console

    $ pre-commit run --all-files

Please remember that these pre-commit hooks do not support Python 3.6; they
require a Python major version which is actively supported upstream.

Coding Guidelines
-----------------

* Core functionality should be implemented in ``libdrgn`` and exposed to Python
  via the `C extension <libdrgn/python>`_. Only the CLI and helpers should be
  in pure Python.

C
^

drgn is written in C11. C code in drgn mostly follows the `Linux kernel coding
style <https://www.kernel.org/doc/html/latest/process/coding-style.html>`_ with
some slightly more modern preferences:

* Variables should be declared as close as possible to where they are used (as
  opposed to the C89 style of declaring everything at the top of a function).

  * As an exception, if a function has a local ``struct drgn_error *err``, it
    should usually be declared at the top of the function. (This is because
    must functions have such a variable, and it adds noise to have it in the
    middle of the function.)

* Scope guards and the `cleanup attribute
  <https://gcc.gnu.org/onlinedocs/gcc/Common-Variable-Attributes.html#index-cleanup-variable-attribute>`_
  should be used liberally.
* ``//``-style comments are preferred over ``/* */``.

  * As an exception, Doxygen comments should use ``/** */``.

  For example:

  .. code-block:: c

      /** Good example. */
      struct drgn_error *my_func(struct drgn_program *prog, size_t n)
      {
              struct drgn_error *err;

              _cleanup_free_ void *buf = malloc(n);
              if (!buf)
                      return &drgn_enomem;
              // 0xffff0000 is a nice address.
              err = drgn_program_read_memory(prog, buf, 0xffff0000, n, false);
              if (err)
                      return err;
              ...
              return NULL;
      }

  NOT:

  .. code-block:: c

      /* BAD example. */
      struct drgn_error *my_func(struct drgn_program *prog, size_t n) {
              struct drgn_error *err;
              void *buf;

              buf = malloc(n);
              if (!buf) {
                      return &drgn_enomem;
              }
              /* 0xffff0000 is a nice address. */
              err = drgn_program_read_memory(prog, buf, 0xffff0000, n, false);
              if (err) goto out;
              ...
              err = NULL;
      out:
              free(buf);
              return err;
      }

A few other guidelines/conventions:

* Constants should be defined as enums or ``static const`` variables rather
  than macros.
* Functions that can fail should return a ``struct drgn_error *`` (and return
  their result via an out parameter if necessary).
* Out parameters should be named ``ret`` (or suffixed with ``_ret`` if there
  are multiple) and be the last parameter(s) of the function.
* Functions that initialize an already allocated structure should be suffixed
  with ``_init`` and take the structure to initialize as the first argument,
  e.g., ``struct drgn_error *foo_init(struct foo *foo, int foo_flags)``.
* The matching function to deinitialize a structure should be suffixed with
  ``_deinit``, e.g., ``void foo_deinit(struct foo *foo)``. If possible, the
  definition should be placed directly after the definition of ``_init`` so
  that it is easier to visually verify that everything is cleaned up.
* Functions that allocate and initialize a structure should be suffixed with
  ``_create`` and either return the structure as an out parameter (e.g.,
  ``struct drgn_error *foo_create(int foo_flags, struct foo **ret)``) or as the
  return value if they can only fail with an out-of-memory error (e.g.,
  ``struct foo *foo_create(int foo_flags)``).
* The matching function to free an allocated structure should be suffixed with
  ``_destroy``, e.g., ``void foo_destroy(struct foo *foo)``. If possible, the
  definition should be placed directly after the definition of ``_create``.
  ``_destroy`` should usually allow a ``NULL`` argument, just like ``free()``.
* Functions that return a result in a ``struct drgn_object *`` parameter should
  only modify the object if the function succeeds.

drgn assumes some `implementation-defined behavior
<https://gcc.gnu.org/onlinedocs/gcc/C-Implementation.html>`_ for sanity:

* Signed integers are represented with two's complement.
* Bitwise operators on signed integers operate on the two's complement
  representation.
* Right shift of a signed integer type is arithmetic.
* Conversion to a signed integer type is modular.
* Casting between pointers and integers does not change the bit representation.

Python
^^^^^^

Python code in drgn should be compatible with Python 3.6 and newer.

Python code is formatted with `Black <https://github.com/psf/black>`_ and
`isort <https://github.com/PyCQA/isort>`_.

Type hints are required everywhere (including helpers and the C extension),
except in tests.

Linux Kernel Helpers
^^^^^^^^^^^^^^^^^^^^

Linux kernel helpers should work on all `supported kernels
<https://drgn.readthedocs.io/en/latest/support_matrix.html#linux-kernel-versions>`_
if possible. This may require handling changes between kernel releases.

* Do NOT check the kernel version number to do this; Linux distributions often
  backport changes without updating the version number. Instead, use the
  presence or absence of variables, types, structure members, etc.
* Optimize for the latest kernel release, and follow "easier to ask for
  forgiveness than permission" (`EAFP
  <https://docs.python.org/3/glossary.html#term-EAFP>`_). For example, assume
  that a structure member from the latest release exists and catch the
  exception if it doesn't.
* Reference the diverging commit and version number in the format ``Linux
  kernel commit $abbreviated_commit_hash "$commit_subject" (in
  v$kernel_version)``.

  For example:

  .. code-block:: python3

      # Since Linux kernel commit 2f064a59a11f ("sched: Change
      # task_struct::state") (in v5.14), the task state is named "__state".
      # Before that, it is named "state".
      try:
          return task.__state
      except AttributeError:
          return task.state

  NOT:

  .. code-block:: python3

      # BAD
      if hasattr(task, "state"):
          return task.state
      else:
          return task.__state

* Document the expected C types of arguments and return values. For example:

  .. code-block:: python3

      def cgroup_parent(cgrp: Object) -> Object:
          """
          Return the parent cgroup of the given cgroup if it exists, ``NULL``
          otherwise.

          :param cgrp: ``struct cgroup *``
          :return: ``struct cgroup *``
          """
          ...

Submitting PRs
--------------

Pull requests and issues are always welcome. Feel free to start a discussion
with a prototype.

Signing Off
^^^^^^^^^^^

All commits must be signed off (i.e., ``Signed-off-by: Jane Doe
<janedoe@example.org>``) as per the `Developer Certificate of Origin
<https://developercertificate.org/>`_. ``git commit -s`` can do this for you.

Separating Changes
^^^^^^^^^^^^^^^^^^

Each logical change should be a separate commit. For example, if a PR adds new
functionality to the core library and a new helper that uses the new
functionality, the core change and the helper should be separate commits. This
makes code review much easier.

Each commit should build, pass tests, follow coding guidelines, and run
correctly. (In other words, within a PR, later commits often build on top of
earlier commits, but later commits shouldn't need to "fix" earlier commits.)
This makes it easier to track down problems with tools like ``git bisect``
which may check out any commit in the middle of a PR.

Commit Messages
^^^^^^^^^^^^^^^

The template for a good commit message is:

.. code-block:: none

    One line summary

    Longer explanation including more details, background, and/or
    motivation.

    Signed-off-by: Jane Doe <janedoe@example.org>

See `this post <https://chris.beams.io/posts/git-commit/>`_ for more
information about writing good commit messages.