The minimum test is failing because we are building pillow from source with out webp, we probably should add a skip check on that test (I have a vague memory of someone else reporting the same failure, but not sure where).

Alternatively, we could bump the minimum version of Pillow if we wanted to ensure wheels are available.

Looks like the first version to supply a 3.9 wheel is pillow 8.0.0 (released Oct 2020)

By NEP 29/SPEC 0 standards (24 months) would be pillow 8.1 or thereabouts (and later for expected mpl 3.8 release date)

webp was here: #24711 (Py 3.12) so it may still make sense to have it fail gracefully (people may build on strange platforms etc).

By Pillow's docs, they do not support pillow <8.0 on py3.9 so I think I'll start there. I don't disagree that a graceful fallback/skip/xfail is a better solution long term, just lean towards following upstream when it comes to python support.

If it still fails, I'll add the skips here, but if it passes, I think that can be done in an alternate PR (either on its own or in #24711)

Other potential bump to minimum supported versions (picked out because the minimum version currently specified is not installable as a wheel on 3.9):

• kiwisolver: bump from 1.0.1 (rel Oct 2017) to 1.3.1 (rel Nov 2020, 1.3.0 has py3.9 wheel, but was marked as yanked on pypi)

Appveyor may need to be 3.10, waiting for the queue to see if the problems on 3.11 persist

Failure comes from GTK4-warnings: https://gitlab.gnome.org/GNOME/pygobject/-/issues/494

It looks likes we already filter warnings on that test and just need to add ImportWarning as well (which I am going to assume is the next step up from the DeprecationWarning that gtk4 was triggering?)

Lets do numpy in it's own PR.

Did the Azure tests stop working with this PR or was it just around this PR (or am I missing something)?

NEP 29 was superseded by the scientific python ecosystem coordination guideline SPEC 0 — Minimum Supported Versions. Where NEP had a 42 month support windows (3.5 years) for each Python version, SPEC 0 recommends a 3 year support window.

If we were to follow SPEC 0, than Python 3.9 support can already be dropped, since it was released October 5th, 2020.

3.8 is already released; we aren't going to change the supported Python version in it.

Not dropping Python 3.8 support for the existing Matplotlib 3.8.x releases of course, but it can be dropped on the current master branch (towards the 3.9.x releases).

(which coincidentally is very fitting, requiring Python 3.9 for Matplotlib 3.9)