API guidelines

API consistency and stability are of great value; Therefore, API changes (e.g. signature changes, behavior changes, removals) will only be conducted if the added benefit is worth the effort of adapting existing code.

Because we are a visualization library, our primary output is the final visualization the user sees; therefore, the appearance of the figure is part of the API and any changes, either semantic or aesthetic, are backwards-incompatible API changes.

Add new API

Every new function, parameter and attribute that is not explicitly marked as private (i.e., starts with an underscore) becomes part of Matplotlib's public API. As discussed above, changing the existing API is cumbersome. Therefore, take particular care when adding new API:

• Mark helper functions and internal attributes as private by prefixing them with an underscore.
• Carefully think about good names for your functions and variables.
• Try to adopt patterns and naming conventions from existing parts of the Matplotlib API.
• Consider making as many arguments keyword-only as possible. See also API Evolution the Right Way -- Add Parameters Compatibly.

Add new rcParams

When adding a new rcParam, the following files must be updated:

1. :file:`lib/matplotlib/rcsetup.py` - Add a validator entry to the _validators dict and a corresponding _Param entry with default value and description.
2. :file:`lib/matplotlib/mpl-data/matplotlibrc` - Add a commented-out entry showing the default value.
3. :file:`lib/matplotlib/typing.py` - Add the key to the RcKeyType Literal so that it is recognized as a valid rcParam key.

Add or change colormaps, color sequences, and styles

Visual changes are considered an API break. Therefore, we generally do not modify existing colormaps, color sequences, or styles.

We put a high bar on adding new colormaps and styles to prevent excessively growing them. While the decision is case-by-case, evaluation criteria include:

• novelty: Does it support a new use case? e.g. slight variations of existing maps, sequences and styles are likely not accepted.
• usability and accessibility: Are colors of sequences sufficiently distinct? Has colorblindness been considered?
• evidence of wide spread usage: for example academic papers, industry blogs and whitepapers, or inclusion in other visualization libraries or domain specific tools
• open license: colormaps, sequences, and styles must have a BSD compatible license (see :ref:`license-discussion`)

Deprecate API

When deciding to deprecate API we carefully consider the balance between the advantages (clearer interfaces, better usability, less maintenance) and the disadvantages (users have to learn new API and have to modify existing code).

Tip

A rough estimate on the current usage of an API can be obtained by a GitHub code search. A good search pattern is typically [expression] language:Python NOT is:fork. [expression] may be a simple string, but often regular expressions are helpful to exclude incorrect matches. You can start simple and look at the search results, if there are too many incorrect matches, gradually refine your search criteria.

It can also be helpful to add NOT path:**/matplotlib/** NOT path:**/site-packages/** to exclude matches where the matplotlib codebase is checked into another repo, either as direct sources or as part of an environment.

Example: Calls of the method Figure.draw() could be matched using /\bfig(ure)?\.draw\(/. This expression employs a number of patterns:

• Add the opening bracket ( after the method name to only find method calls.
• Include a common object name if there are otherwise too many false positives. There are many draw() functions out there, but the ones we are looking for are likely called via fig.draw() or figure.draw().
• Use the word boundary marker \b to make sure your expression is not a matching as part of a longer word.

Link to the resulting GitHub search

API changes in Matplotlib have to be performed following the deprecation process below, except in very rare circumstances as deemed necessary by the development team. Generally API deprecation happens in two stages:

• introduce: warn users that the API will change
• expire: API is changed as described in the introduction period

This ensures that users are notified before the change will take effect and thus prevents unexpected breaking of code. Occasionally deprecations are marked as pending, which means that the deprecation will be introduced in a future release.

Rules

• Deprecations are targeted at the next :ref:`meso release <pr-milestones>` (e.g. 3.Y)
• Deprecated API is generally removed (expired) two point-releases after introduction of the deprecation. Longer deprecations can be imposed by core developers on a case-by-case basis to give more time for the transition
• The old API must remain fully functional during the deprecation period
• If alternatives to the deprecated API exist, they should be available during the deprecation period
• If in doubt, decisions about API changes are finally made by the API consistency lead developer.

Introduce deprecation

Deprecations are introduced to warn users that the API will change. The deprecation notice describes how the API will change. When alternatives to the deprecated API exist, they are also listed in the notice and decorators.

1. Create a :ref:`deprecation notice <api_whats_new>`

2. If possible, issue a ~matplotlib.MatplotlibDeprecationWarning when the deprecated API is used. There are a number of helper tools for this:

   • Use _api.warn_deprecated() for general deprecation warnings
   • Use the decorator @_api.deprecated to deprecate classes, functions, methods, or properties
   • Use @_api.deprecate_privatize_attribute to annotate deprecation of attributes while keeping the internal private version.
   • To warn on changes of the function signature, use the decorators @_api.delete_parameter, @_api.rename_parameter, and @_api.make_keyword_only

   All these helpers take a first parameter since, which should be set to the next point release, e.g. "3.x".

   You can use standard rst cross references in alternative.

3. Make appropriate changes to the type hints in the associated .pyi file. The general guideline is to match runtime reported behavior.

   • Items marked with @_api.deprecated or @_api.deprecate_privatize_attribute are generally kept during the expiry period, and thus no changes are needed on introduction.
   • Items decorated with @_api.rename_parameter or @_api.make_keyword_only report the new (post deprecation) signature at runtime, and thus should be updated on introduction.
   • Items decorated with @_api.delete_parameter should include a default value hint for the deleted parameter, even if it did not previously have one (e.g. param: <type> = ...).

Expire deprecation

The API changes described in the introduction notice are only implemented after the introduction period has expired.

1. Create a :ref:`deprecation announcement <api_whats_new>`. For the content, you can usually copy the deprecation notice and adapt it slightly.
2. Change the code functionality and remove any related deprecation warnings.
3. Make appropriate changes to the type hints in the associated .pyi file.
   • Items marked with @_api.deprecated or @_api.deprecate_privatize_attribute are to be removed on expiry.
   • Items decorated with @_api.rename_parameter or @_api.make_keyword_only will have been updated at introduction, and require no change now.
   • Items decorated with @_api.delete_parameter will need to be updated to the final signature, in the same way as the .py file signature is updated.
   • Any entries in :file:`ci/mypy-stubtest-allowlist.txt` which indicate a deprecation version should be double checked. In most cases this is not needed, though some items were never type hinted in the first place and were added to this file instead. For removed items that were not in the stub file, only deleting from the allowlist is required.

Pending deprecation

A pending deprecation is an announcement that a deprecation will be introduced in the future. By default, pending deprecations do not raise a warning to the user; however, pending deprecations are rendered in the documentation and listed in the release notes. Pending notices are primarily intended to give downstream library and tool developers time to adapt their code so that it does not raise a deprecation warning. This is because their users cannot act on warnings triggered by how the tools and libraries use Matplotlib. It's also possible to run Python in dev mode to raise PendingDeprecationWarning.

To mark a deprecation as pending, set the following parameters on the appropriate deprecation decorator: * the pending parameter is set to True * the removal parameter is left blank

When converting a pending deprecation to an introduced deprecation, update the decorator such that: * pending is set to False * since is set to the next meso release (3.Y+1) * removal is set to at least 2 meso releases after (3.Y+3) introduction.

Pending deprecations are documented in the :ref:`API change notes <api_whats_new>` in the same manner as introduced and expired deprecations. The notice should include pending deprecation in the title.

.. redirect-from:: /devel/coding_guide#new-features-and-api-changes

Announce new and deprecated API

When adding or changing the API in a backward in-compatible way, please add the appropriate :ref:`versioning directive <versioning-directives>` and document it in the :ref:`release notes <release-notes>` by adding an entry to the appropriate folder:

	versioning directive	announcement folder
new feature	.. versionadded:: 3.N	:file:`doc/release/next_whats_new/`
API change	.. versionchanged:: 3.N	:file:`doc/api/next_api_changes/[kind]`

When deprecating API, please add a notice as described in the :ref:`deprecation guidelines <deprecation-guidelines>` and summarized here:

stage		announcement folder
:ref:`introduce deprecation <intro-deprecation>`		:file:`doc/api/next_api_changes/deprecation`
:ref:`expire deprecation <expire-deprecation>`		:file:`doc/api/next_api_changes/[kind]`

Generally the introduction notices can be repurposed for the expiration notice as they are expected to be describing the same API changes and removals.

Versioning directives

When making a backward incompatible change, please add a versioning directive in the docstring. 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 micro release version is omitted and the directive should not be added to entire modules.

Release notes

For both change notes and what's new, please avoid using cross-references in section titles as it causes links to be confusing in the table of contents. Instead, ensure that a cross-reference is included in the descriptive text.

API change notes

What's new notes

Discourage API

We have API that we do not recommend anymore for new code, but that cannot be deprecated because its removal would be breaking backward-compatibility and too disruptive. In such a case we can formally discourage API. This can cover specific parameters, call patterns, whole methods etc.

To do so, add a note to the docstring

.. admonition:: Discouraged

   [description and suggested alternative]

You find several examples for good descriptions if you search the codebase for .. admonition:: Discouraged.

Additionally, if a whole function is discouraged, prefix the summary line with [*Discouraged*] so that it renders in the API overview like this

[Discouraged] Return the XAxis instance.