gh-124613: Deactivate perf support in tests if the jit is set

[pablogsal]

• Issue: test_perf_profiler fails on Linux (both AArch64 and x86) #124613

[pablogsal]

CC @brandtbucher @diegorusso

[pablogsal]

Hummm if we are going to make the JIT the default maybe we should only deactivate in YES and not in Yes-off @brandtbucher WDYT?

[brandtbucher]

Hummm if we are going to make the JIT the default maybe we should only deactivate in YES and not in Yes-off @brandtbucher WDYT?

I mean, a yes-off build can still have a JIT if the env var is set (ditto for unsetting it via an env var on yes builds). We really just want to deactivate the JIT, regardless of which mode it was built in or what env vars are set.

Check out test.support.without_optimizer, I think it's probably what you want here.

[brandtbucher]

Oh, whoops, wrong PR. I thought this was the one that's skipping perf tests.

[brandtbucher]

Yeah, we could just say that both yes and yes-off disable perf jit support. Your call... but yeah, that means we can't ship perf JIT support in the same build if we do that.

Perhaps we should just disable the JIT if perf support is activated at runtime? It requires a bit more surgery, but I think it's probably the best option.

[pablogsal]

Yeah, we could just say that both yes and yes-off disable perf jit support. Your call... but yeah, that means we can't ship perf JIT support in the same build if we do that.

No, we surely want perf in yes-off.

Perhaps we should just disable the JIT if perf support is activated at runtime? It requires a bit more surgery, but I think it's probably the best option.

Yeah we should use both things I think. Currently it's not switched to yes-off no?

[brandtbucher]

Currently it's not switched to yes-off no?

That's too many negatives for me to follow, haha.

What are you proposing, exactly? Disable the JIT when perf support is turned on at runtime? I believe this needs to cover PYTHON_PERF_JIT_SUPPORT, PYTHONPERFSUPPORT, -X perf, and -X perf_jit at startup, and sys.activate_stack_trampoline() and sys.deactivate_stack_trampoline() at runtime. Seems reasonable (maybe we warn if the JIT was activated with the env var to let the user know it's getting turned off?).

Also, the name PYTHON_PERF_JIT_SUPPORT is now more confusing than ever, haha.

[pablogsal]

What are you proposing, exactly? Disable the JIT when perf support is turned on at runtime?

Yes, but that in a separate PR

For now I have modified this PR to use the test.support.without_optimizer so this always runs without the JIT. Is there a way to do that with an env var instead? That way I can call the subprocess without the JIT always

[brandtbucher]

For now I have modified this PR to use the test.support.without_optimizer so this always runs without the JIT. Is there a way to do that with an env var instead? That way I can call the subprocess without the JIT always

PYTHON_JIT=0

[pablogsal]

For now I have modified this PR to use the test.support.without_optimizer so this always runs without the JIT. Is there a way to do that with an env var instead? That way I can call the subprocess without the JIT always

PYTHON_JIT=0

Ok, moved this PR to that