Remove unused whitelist entries

[srittau]

No description provided.

[hauntsaninja]

Well, it's annoying that this didn't pan out nicely.

The typing.IO.closed one is probably because Travis is using an older patch release from Github Actions; for once the issue stubtest found was on CPython's side and the fix got backported. (Note one fix here is to pin the patch version; I'd rather not do that since our policy is to support the latest patch version and we aren't going to be good at updating our CI for each patch release)

The tkinter ones are also "real". Those genuinely don't exist (were removed in 3.6). I think it's possible to wind up with Python builds without _tkinter, in which case the whole module fails to import (so dependent whitelist entries wouldn't surface errors). Haven't checked, but seems possible that Github Actions' build doesn't have _tkinter.

Not fully sure what's going on with os. docs.python.org says:

The MFD_HUGE* flags are only available since Linux 4.14.

So maybe Travis is using an old kernel?

The site one is super weird. Maybe pypa/virtualenv#737

We can "fix" any thorns using the (pattern)? trick that allows whitelist entries to go unused.

[srittau]

Maybe we could just move all stubtests to GitHub? Then the Python version should be identical. Things should still get reported correctly.

[hauntsaninja]

Sure. Seems like the only downside is stubtest wouldn't check anything in CI for changes dependent on _tkinter, but that seems fine. tk stubs are some of the worst anyway...

[hauntsaninja]

It's also weird that Github Actions has issues with site for the Python versions earlier than 3.8...
Anyway, in the meantime, I opened #4217 with the whitelist entry removals that should pass existing CI.

[srittau]

Superseded by #4217.

[Akuli]

Seems like the only downside is stubtest wouldn't check anything in CI for changes dependent on _tkinter

How was this problem solved when the CI moved to github actions?

It's possible to run Python with _tkinter on github actions. I use xvfb-action for that, like this:

https://github.com/Akuli/porcupine/blob/440018f19409e99b22b70893f51f357611122f04/.github/workflows/main.yml#L34-L38

[hauntsaninja]

I'm sorry, I don't remember the specifics about what I was talking about here. (The "real" tkinter issues I mention were fixed in #4207)

Haven't checked, but seems possible that Github Actions' build doesn't have _tkinter.

It does seem to me that the Python build stubtest uses in Github Actions has _tkinter — are you seeing false negatives (or other issues) in CI?

Anyway, very happy that tk stubs are some of the worst anyway is no longer true. Thank you so much for that!

[Akuli]

Now that I think about it, stubtest has correctly pointed out many tkinter problems, so seems like it can at least import _tkinter. I think instantiating tkinter classes would fail though, because it's not ran inside xvfb.

Stubtest doesn't point out issues with attributes coming from __getattr__, and I think that's because it doesn't actually instantiate the classes that it imports. Most tkinter classes can be instantiated with no arguments, but again, that requires xvfb or a system with a display. But there are not many __getattr__s in tkinter.