[3.13] gh-145008: Maintain stack consistency in CALL_BOUND_METHOD_EXACT_ARGS

[Fidget-Spinner]

The problem is that 3.13 actually doesn't guarantee stack consistency always between uops. This manifests as a bug that seems like stack corruption. In 3.14 and 3.15, this isn't a problem because we guarantee stack consistency between uops always by flushing or otherwise. The bandaid for 3.13 is to move all deopts to before we modify the stack.

@marc-hb could you please try the patch and check if it fixes your code? I tried on my end and it does, just need a second checker.

Can't test this easily, as it depends on specific recursion limits and interaction of specialization.

• Issue: Recursion check in 3.13.8 in CALL_BOUND_METHOD_GENERAL leaves stack in inconsistent state #145008

[marc-hb]

I cherry-picked the single commit 58c2e83 from this PR on top of v3.13.12 and I can confirm that this solves my issue #145008 and makes that infinite recursion test pass.

(Not like this should matter much but I also ran the other, non-recursive tests in the suite while at it and nothing went wrong)

[encukou]

This looks fine to me but I don't know enough about the JIT to review in depth.

[Fidget-Spinner]

JIT tests are failing, no clue why. However, since 3.13 JIT is experimental and pretty much no one uses it, I'll ignore it for now. @encukou this isn't a JIT failure but an actual interpreter crash.