Elliott Sales de Andrade

> This format is at best rather difficult to write an accurate schema for, if not outright impossible depending on how pedantic you want to be. This makes it harder...

Looking at https://github.com/matplotlib/matplotlib/actions/runs/16535584290/job/46769299546, the diff is different this time. The star in the bottom left Axes is missing instead (and a few path names after it are renumbered).

This also breaks `click.open_file('-')`: ``` import click from click.testing import CliRunner from pytest import fixture @click.command() def main(): src = click.open_file('-').readlines() click.echo(f"Source: {src}") @fixture def runner(): return CliRunner() def test_program(runner):...

GTK would be fixed by #17791.

> #18693 fixed some of the problems for tk, but keyreleases are still wrong. > `gcf().canvas.mpl_connect("key_press_event", lambda event: print(event.key))` The question is whether key release is also fixed?

As an example, in DejaVu Sans, the letter A is glyph index 36: ``` >>> dejavu = FT2Font('lib/matplotlib/mpl-data/fonts/ttf/DejaVuSans.ttf') >>> dejavu.get_char_index(ord('A')) 36 ``` which has metrics like: ``` >>> g =...

Not going to wait for CI, because it's just going to fail, and then run again when merged, and fail there too...

Oh, and with respect to `TestStat/not_empty_dir`, subdirectories don't add to the link count either: ``` $ mkdir foo $ stat foo File: foo Size: 0 Blocks: 0 IO Block: 4096...

Oh, actually, it appears you don't even need to build from source, as wheels use system zlib, so a shorter reproducer: ``` $ podman run --rm -it fedora:rawhide # dnf...

> I've only done up to the formatters; please correct the default values in the rest. Please handle the rest of these.