GeezerVison

Having matplotlib.style be a package with the entire implementation in
matplotlib.style.core is a bit overkill, and also makes it slightly
awkward that USER_LIBRARY_PATHS is effectively a public API (clearly
intended as so, per the comment, even though we may do it differently
nowadays...) but only available in matplotlib.style.core, whereas
everything else is re-exported by matplotlib.style.

Prepare to flatten the implementation by deprecating
matplotlib.style.core and reexporting USER_LIBRARY_PATHS in
matplotlib.style.  Once the deprecation elapses, we'll be able to
move the implementation into a plain matplotlib/style.py module.

Instead of a fixed-size buffer, a dynamic buffer, and all the tracking
for them, just use a standard C++ container, which will clean up for
itself automatically.

Fixed incomplete deletion of all images that have passed tests before upload

Simplify RendererAgg::draw_markers buffers

Nested gridspecs are explained in
`galleries/examples/subplots_axes_and_figures/gridspec_nested.py`, and
layout is more generally explained in
`galleries/users_explain/axes/arranging_axes.py` and
`galleries/users_explain/axes/constrainedlayout_guide.py`

Closes matplotlib#25800

DOC: Remove last userdemo example

This type seems to cover the intent more clearly than `std::pair<bool,
...>`.

deques are exactly suited for the task at hand, and should be (here,
marginally) more efficient than repeatedly slicing off the start of the
list.

Also drop unneeded reference to old_frame_seq (the iterator already
keeps that data alive).

Fix argument types in examples and tests

…atplotlib#29481)

* Possible fix for issue matplotlib#13919

* fix flake warnings

* Resolve suggested changes

* Add new validators in stub file

* make stub file arguments same as runtime

* Add distinction for x and y axis in grid line options

* fix flake8

* Add What's New note

* Extend `_val_or_rc` to support multiply names
`_val_or_rc` now accept multiple rc names and return val or the first non-None value in rcParams. Returns last rc name if all other are None. Also, simplified code in `Tick` for grid lines creatation

* Fix linting for and _validate_linestyle_or_None to mypy allowlist

* Remove validate linestyle functions from stubtest allow list

* Revert change to just grid.major/minor distinction in rcParams

* Update What's New note and reduced example

* Add testing for `grid.major/minor.*` in rcParams

* fix indentation and linting

* Fix example description

* Fix spelling

* Fix type

* Fix formatting
Removed outer brackets

* `validate_color_or_None` private and fix argument names

* Fix validator name in stub file as well

* correct validators for grid.*.color keys

* Revert change in mypy-stubtest-allowlist.txt

---------

Co-authored-by: Tim Hoffmann <[email protected]>

Use collections.deque to store animation cache data.

Prepare to turn matplotlib.style into a plain module.

- Use a simpler deterministic mapping of tex font names to pdf embedding
  names.
- Only resolve the required attributes when needed (in _embedTeXFont),
  which avoids e.g. having to carry around and worry about attributes
  with different names (e.g. "encoding" vs. "encodingfile").

Simplify dviFontInfo layout in backend pdf.

To deep copy an object without calling deepcopy on the object itself,
create a new object of the correct class and iterate calling deepcopy on
its __dict__.

Closes matplotlib#29157 without relying on private CPython methods.
Does not fix the other issue with TransformNode.__copy__.

Co-authored-by: Serhiy Storchaka <[email protected]>

Co-authored-by: Elliott Sales de Andrade <[email protected]>

without calling copy.copy

[Doc]: fix bug in release notes for matplotlib v3.5.0 and v3.7.0

The subgridspec to be considered should be the one containing both the
axes and the colorbar, not the sub-subgridspec of just the axes.

…etaclass

Implement Path.__deepcopy__ avoiding infinite recursion

This should only end up on the nightly wheel upload for now.

Also, re-enable testing in places where Pillow wheels were previously
missing, but are now available.

Closes matplotlib#29688

Objects without a parent are not necessarily cleaned up in `PyQt5/6`
when their reference count reaches zero, and must be explicitly cleaned
up with `deleteLater()`

This prevents the notifier firing after the signal handling was supposed
to have been reset to its previous state.

Rather than have both `bakend_bases._allow_interrupt()` and
`backend_qt._allow_interrupt_qt()` hold a reference to the notifier, we
pass it to  the backend-specific `handle_signint()` function for
cleanup.

Note the approach to cleaning up the notifier with `.deleteLater()`
followed by `sendPostedEvents()` is the documented workaround for when
immediate deletion is desired:

https://doc.qt.io/qt-6/qobject.html#deleteLater

This ensures the object is still deleted up even if the same event loop
does not run again.

* Adding a decorator and Refactoring functions

* Fixing Ruff Errors

* Update scale.pyi

* Adding new line to the end of scale.pyi

* Update in docstring

* Fixing Handle Function

* Support optional axis in scales

Updated my refactor based on the feedbacks received

* Fixing ruff error

* change in parameters and in decorator

* parameter fix

* minor change in pyi

* Update lib/matplotlib/scale.py

Co-authored-by: Tim Hoffmann <[email protected]>

* Update lib/matplotlib/scale.py

Co-authored-by: Tim Hoffmann <[email protected]>

* Update lib/matplotlib/scale.py

Co-authored-by: Tim Hoffmann <[email protected]>

* Update lib/matplotlib/scale.pyi

Co-authored-by: Tim Hoffmann <[email protected]>

* Updating self and axis in pyi

* returning scale_factory to default

* Ruff checks

* description fix

* Update lib/matplotlib/scale.pyi

Co-authored-by: Tim Hoffmann <[email protected]>

* Undoing Unrelated Modifications

* fixing mypy tests

* Update lib/matplotlib/scale.pyi

Co-authored-by: Tim Hoffmann <[email protected]>

* keyword-argument suggestion

Co-authored-by: Elliott Sales de Andrade <[email protected]>

* kwargs pop before function call

Co-authored-by: Elliott Sales de Andrade <[email protected]>

---------

Co-authored-by: Tim Hoffmann <[email protected]>
Co-authored-by: Elliott Sales de Andrade <[email protected]>

This also has the side-effect of casting torch Tensors to NumPy arrays,
which fixes matplotlib#30216. Since `stem` is made up of `plot` and `[hv]lines`
whic already do this cast, this just moves it up one level which
prevents doing it twice.

Make the difference between the two methods clearer (only the latter
takes clipping into account).

Tweak docstrings of get_window_extent/get_tightbbox.

Clean up mypy & ruff config

Check that stem input is 1D

Fix figure legend when drawing stackplots

Don't expose private styles in style.available

This was originally for i686 on Fedora, but is now applicable to WASM,
which is 32-bit. The older implementation doesn't OOM.

Use old stride_windows implementation on 32-bit builds

This allows checking that there are no _new_ failures, without
committing the new figures to the repo until the branch is complete.