[3.13] gh-133441: Fix STORE_ATTR_WITH_HINT bytecode

[vstinner]

Deoptimize if the dict is a dict subclass.

• Issue: 3.13: Abort from failed assertion in _PyEval_EvalFrameDefault #133441

[vstinner]

cc @ZeroIntensity @sobolevn

[ZeroIntensity]

It looks like the offending code is on main as well.

[vstinner]

It looks like the offending code is on main as well.

I failed to write code to trigger the bug on main, so I'm not sure that main is affected.

[Fidget-Spinner]

This should be in test_opcache not test_dict IMO.

[vstinner]

I moved the test to test_opcache, but I'm not sure if TestInstanceDict is a good home for such test.

[Fidget-Spinner]

Thanks!

[vstinner]

@Fidget-Spinner: Do you think that the main branch is also affected? I'm unable to trigger the bug in the main branch, but the code looks the same. I would prefer to not deoptimize just for an hypothetical case if it cannot occur in practice. The code in the main branch is a little bit different.

[Fidget-Spinner]

@vstinner did you try wrapping the repro in a loop? Sometimes that might trigger it. I mean an actual Python for loop, not python -m test -F

[vstinner]

@vstinner did you try wrapping the repro in a loop? Sometimes that might trigger it. I mean an actual Python for loop, not python -m test -F

I just tried to put the reproducer in a loop: I still cannot reproduce the failure on the main branch.

[ZeroIntensity]

LGTM as well. If 3.14 and 3.15 aren't affected, someone should add the test case.

[vstinner]

Merged, thanks.