Pull request guidelines#
Pull requests (PRs) on GitHub are the mechanism for contributing to Matplotlib's code and documentation.
It is recommended to check that your contribution complies with the following rules before submitting a pull request:
If your pull request addresses an issue, please use the title to describe the issue (e.g. "Add ability to plot timedeltas") and mention the issue number in the pull request description to ensure that a link is created to the original issue (e.g. "Closes #8869" or "Fixes #8869"). This will ensure the original issue mentioned is automatically closed when your PR is merged. See the GitHub documentation for more details.
Formatting should follow the recommendations of PEP8, as enforced by flake8. Matplotlib modifies PEP8 to extend the maximum line length to 88 characters. You can check flake8 compliance from the command line with
python -m pip install flake8 flake8 /path/to/module.py
or your editor may provide integration with it. Note that Matplotlib intentionally does not use the black auto-formatter (1), in particular due to its inability to understand the semantics of mathematical expressions (2, 3).
All public methods should have informative docstrings with sample usage when appropriate. Use the docstring standards.
For high-level plotting functions, consider adding a simple example either in the
Example
section of the docstring or the examples gallery.Changes (both new features and bugfixes) should have good test coverage. See Testing for more details.
Import the following modules using the standard scipy conventions:
import numpy as np import numpy.ma as ma import matplotlib as mpl import matplotlib.pyplot as plt import matplotlib.cbook as cbook import matplotlib.patches as mpatches
In general, Matplotlib modules should not import
rcParams
usingfrom matplotlib import rcParams
, but rather access it asmpl.rcParams
. This is because some modules are imported very early, before thercParams
singleton is constructed.If your change is a major new feature, add an entry to the
What's new
section by adding a new file indoc/users/next_whats_new
(seedoc/users/next_whats_new/README.rst
for more information).If you change the API in a backward-incompatible way, please document it in
doc/api/next_api_changes/behavior
, by adding a new file with the naming convention99999-ABC.rst
where the pull request number is followed by the contributor's initials. (seedoc/api/api_changes.rst
for more information)See below for additional points about Keyword argument processing, if applicable for your pull request.
Note
The current state of the Matplotlib code base is not compliant with all of these guidelines, but we expect that enforcing these constraints on all new contributions will move the overall code base quality in the right direction.
Summary for pull request reviewers#
Note
If you have commit rights, then you are trusted to use them. Please help review and merge PRs!
Be patient and kind with contributors.
Content topics:
Is the feature / bugfix reasonable?
Does the PR conform with the Coding guidelines?
Is the documentation (docstrings, examples, what's new, API changes) updated?
Is the change purely stylistic? Generally, such changes are discouraged when not part of other non-stylistic work because it obscures the git history of functional changes to the code. Reflowing a method or docstring as part of a larger refactor/rewrite is acceptable.
Organizational topics:
Make sure all automated tests pass.
The PR should target the main branch.
Tag with descriptive labels.
Set the milestone.
Keep an eye on the number of commits.
Approve if all of the above topics are handled.
Merge if a sufficient number of approvals is reached.
Detailed guidelines#
Documentation#
Every new feature should be documented. If it's a new module, don't forget to add a new rst file to the API docs.
Each high-level plotting function should have a small example in the
Examples
section of the docstring. This should be as simple as possible to demonstrate the method. More complex examples should go into a dedicated example file in theexamples
directory, which will be rendered to the examples gallery in the documentation.Build the docs and make sure all formatting warnings are addressed.
See Writing documentation for our documentation style guide.
New features and API changes#
When adding a major new feature or changing the API in a backward incompatible way, please document it by including a versioning directive in the docstring and adding an entry to the folder for either the what's new or API change notes.
for this addition |
include this directive |
create entry in this folder |
---|---|---|
new feature |
|
|
API change |
|
probably in |
The directives should be placed at the end of a description block. For example:
class Foo:
"""
This is the summary.
Followed by a longer description block.
Consisting of multiple lines and paragraphs.
.. versionadded:: 3.5
Parameters
----------
a : int
The first parameter.
b: bool, default: False
This was added later.
.. versionadded:: 3.6
"""
def set_b(b):
"""
Set b.
.. versionadded:: 3.6
Parameters
----------
b: bool
For classes and functions, the directive should be placed before the Parameters section. For parameters, the directive should be placed at the end of the parameter description. The patch release version is omitted and the directive should not be added to entire modules.
Labels#
If you have the rights to set labels, tag the PR with descriptive labels. See the list of labels.
If the PR makes changes to the wheel building Action, add the "Run cibuildwheel" label to enable testing wheels.
Milestones#
Set the milestone according to these rules:
New features and API changes are milestoned for the next minor release
v3.N.0
.Bugfixes, tests for released code, and docstring changes are milestoned for the next patch release
v3.N.M
.Documentation changes (all .rst files and examples) are milestoned
v3.N-doc
.
If multiple rules apply, choose the first matching from the above list.
Setting a milestone does not imply or guarantee that a PR will be merged for that release, but if it were to be merged what release it would be in.
All of these PRs should target the main branch. The milestone tag triggers an automatic backport for milestones which have a corresponding branch.
Merging#
Documentation and examples may be merged by the first reviewer. Use the threshold "is this better than it was?" as the review criteria.
For code changes (anything in
src
orlib
) at least two core developers (those with commit rights) should review all pull requests. If you are the first to review a PR and approve of the changes use the GitHub 'approve review' tool to mark it as such. If you are a subsequent reviewer please approve the review and if you think no more review is needed, merge the PR.Ensure that all API changes are documented in a file in one of the subdirectories of
doc/api/next_api_changes
, and significant new features have an entry indoc/user/whats_new
.If a PR already has a positive review, a core developer (e.g. the first reviewer, but not necessarily) may champion that PR for merging. In order to do so, they should ping all core devs both on GitHub and on the dev mailing list, and label the PR with the "Merge with single review?" label. Other core devs can then either review the PR and merge or reject it, or simply request that it gets a second review before being merged. If no one asks for such a second review within a week, the PR can then be merged on the basis of that single review.
A core dev should only champion one PR at a time and we should try to keep the flow of championed PRs reasonable.
Do not self merge, except for 'small' patches to un-break the CI or when another reviewer explicitly allows it (ex, "Approve modulo CI passing, may self merge when green").
Automated tests#
Whenever a pull request is created or updated, various automated test tools will run on all supported platforms and versions of Python.
Make sure the Linting, GitHub Actions, AppVeyor, CircleCI, and Azure pipelines are passing before merging (All checks are listed at the bottom of the GitHub page of your pull request). Here are some tips for finding the cause of the test failure:
If Linting fails, you have a code style issue, which will be listed as annotations on the pull request's diff.
If a GitHub Actions or AppVeyor run fails, search the log for
FAILURES
. The subsequent section will contain information on the failed tests.If CircleCI fails, likely you have some reStructuredText style issue in the docs. Search the CircleCI log for
WARNING
.If Azure pipelines fail with an image comparison error, you can find the images as artifacts of the Azure job:
Click Details on the check on the GitHub PR page.
Click View more details on Azure Pipelines to go to Azure.
On the overview page artifacts are listed in the section Related.
Codecov and CodeQL are currently for information only. Their failure is not necessarily a blocker.
tox is not used in the automated testing. It is supported for testing locally.
If you know only a subset of CIs need to be run, this can be controlled on individual commits by including the following substrings in commit messages:
[ci doc]
: restrict the CI to documentation checks. For when you only changed documentation.[skip circle]
: skip the documentation build check. For when you didn't change documentation.Unit tests can be turned off for individual platforms with
[skip actions]
: GitHub Actions[skip appveyor]
(must be in the first line of the commit): AppVeyor[skip azp]
: Azure Pipelines
[skip ci]
: skip all CIs. Use this only if you know your changes do not need to be tested at all, which is very rare.
Number of commits and squashing#
Squashing is case-by-case. The balance is between burden on the contributor, keeping a relatively clean history, and keeping a history usable for bisecting. The only time we are really strict about it is to eliminate binary files (ex multiple test image re-generations) and to remove upstream merges.
Do not let perfect be the enemy of the good, particularly for documentation or example PRs. If you find yourself making many small suggestions, either open a PR against the original branch, push changes to the contributor branch, or merge the PR and then open a new PR against upstream.
If you push to a contributor branch leave a comment explaining what you did, ex "I took the liberty of pushing a small clean-up PR to your branch, thanks for your work.". If you are going to make substantial changes to the code or intent of the PR please check with the contributor first.
Branches and backports#
Current branches#
The current active branches are
- main
The current development version. Future minor releases (v3.N.0) will be branched from this.
- v3.N.x
Maintenance branch for Matplotlib 3.N. Future patch releases will be branched from this.
- v3.N.M-doc
Documentation for the current release. On a patch release, this will be replaced by a properly named branch for the new release.
Branch selection for pull requests#
Generally, all pull requests should target the main branch.
Other branches are fed through automatic or manual. Directly targeting other branches is only rarely necessary for special maintenance work.
Backport strategy#
We will always backport to the patch release branch (v3.N.x):
critical bug fixes (segfault, failure to import, things that the user can not work around)
fixes for regressions against the last two releases.
Everything else (regressions against older releases, bugs/api inconsistencies the user can work around in their code) are on a case-by-case basis, should be low-risk, and need someone to advocate for and shepherd through the backport.
The only changes to be backported to the documentation branch (v3.N.M-doc)
are changes to doc
, examples
, or tutorials
.
Any changes to lib
or src
including docstring-only changes
should not be backported to this branch.
Automated backports#
We use MeeseeksDev bot to automatically backport merges to the correct
maintenance branch base on the milestone. To work properly the
milestone must be set before merging. If you have commit rights, the
bot can also be manually triggered after a merge by leaving a message
@meeseeksdev backport to BRANCH
on the PR. If there are conflicts
MeeseeksDev will inform you that the backport needs to be done
manually.
The target branch is configured by putting on-merge: backport to
TARGETBRANCH
in the milestone description on it's own line.
If the bot is not working as expected, please report issues to MeeseeksDev.
Manual backports#
When doing backports please copy the form used by MeeseeksDev,
Backport PR #XXXX: TITLE OF PR
. If you need to manually resolve
conflicts make note of them and how you resolved them in the commit
message.
We do a backport from main to v2.2.x assuming:
matplotlib
is a read-only remote branch of the matplotlib/matplotlib repo
The TARGET_SHA
is the hash of the merge commit you would like to
backport. This can be read off of the GitHub PR page (in the UI with
the merge notification) or through the git CLI tools.
Assuming that you already have a local branch v2.2.x
(if not, then
git checkout -b v2.2.x
), and that your remote pointing to
https://github.com/matplotlib/matplotlib
is called upstream
:
git fetch upstream
git checkout v2.2.x # or include -b if you don't already have this.
git reset --hard upstream/v2.2.x
git cherry-pick -m 1 TARGET_SHA
# resolve conflicts and commit if required
Files with conflicts can be listed by git status
,
and will have to be fixed by hand (search on >>>>>
). Once
the conflict is resolved, you will have to re-add the file(s) to the branch
and then continue the cherry pick:
git add lib/matplotlib/conflicted_file.py
git add lib/matplotlib/conflicted_file2.py
git cherry-pick --continue
Use your discretion to push directly to upstream or to open a PR; be
sure to push or PR against the v2.2.x
upstream branch, not main
!