👍 in principle (5.12 in from 2019).

For pyside2 there are no wheel on pypi for Python 3.11, but there are conda packages. That is the thing messing about right now. I guess it is not that hard to fix, but seems a bit annoying that pyside2/py3.11 is the only combination not working...

we use a full conda stack on appeveyor so we could test pyside2 + py311 there.

Seem like the following versions of pyside2 are used on GA (for Linux)

3.9 minimum

PySide2-5.15.1-5.15.1-cp35.cp36.cp37.cp38.cp39-abi3-manylinux1_x86_64.whl

3.9 and 3.10

PySide2-5.15.2.1-5.15.2-cp35.cp36.cp37.cp38.cp39.cp310-abi3-manylinux1_x86_64.whl

3.11

PySide2-5.13.2-5.13.2-cp35.cp36.cp37-abi3-manylinux1_x86_64.whl

Not so obvious why... Note that 3.11 is using a version for 3.5-3.7. The _to_int conversions in relation to | must be kept because of this.

More recent versions of pyside2 have an explicit <3.11, so 5.13.2 is likely just the most recent version to not have an upper bound on python version...

pyside6 is available for py 3.11, though

< insert note about the importance of being able to adjust pins post-build and the ground truth of the dependencies living outside of the build artifact >

The outstanding question here is which versions of which bindings we are testing so we are sure that what we are claiming in the minimum actually works.

This is from the minver test section:

So I guess that it is far to say that we test with 5.12 if nothing else (although for pyside2 we actually cannot).

I have rebased this. As noted, we already test with 5.12 as the minimum version.

This is a follow-up to #24919 where the version was bumped in the CI, but any effects were ignored.