Since the "fix" has been merged into the conda-forge feedstock, do we want to have the workaround enabled only for specific versions of pyside bindings; or is there minimal downside to the workaround that that's not needed?

I did a check, the issue is also present in conda-forge's build of PySide2 5.15.15.
The issue is not present in Qt's official build PySide2 5.15.2.1.

refcnt leak introduced:
pyside/pyside2-setup@821480b

the following comment says that Qt defaults to limited-api builds (and therefore their builds aren't affected by the leak introduced):
pyside/pyside2-setup@032cb10

I couldn't find where pyside2-feedstock was configuring for non-limited-api build though.

Given that conda-forge has patched PySide2 to build for Python 3.11, and seem willing (conda-forge/pyside2-feedstock#246) to push it to Python 3.12, what is pyqtgraph's policy regarding support of PySide2 going to be?

do we want to have the workaround enabled only for specific versions of pyside bindings;

For PySide6, we can don't apply the workaround. conda-forge PySide6 6.7.3 through PySide6 6.8.2 build 0 will have the leak.

For PySide2, maybe we can apply the workaround? If PySide2 is a non-preferred binding for pyqtgraph, I guess penalizing it is not really an issue? i.e. better to be less efficient than having to worry about whether a leak is present for non-official builds.

Other places where the leak occurs:

• Ubuntu 22.04 / Python 3.10.12 / PySide2 5.15.2
• Ubuntu 24.04 / Python 3.12.3 / PySide2 5.15.13

For reference, pyqtgraph first started taking a memory view to QByteArray in this commit. (2021 August)
287ec9f

PySide2 5.15.2 had already been released for some time.

Thanks @pijyoi !