[3.11] GH-93516: Backport GH-93769

[markshannon]

Reduces the overhead of dispatching in DO_TRACING considerably.

My informal measures on the test program in #93516 (comment) show that this reduces execution time by about 10%

@pablogsal Should I regenerate the ABI for this PR?

• Issue: Severe performance degradation for tracing under 3.11 #93516

[arhadthedev]

Probably the issue should be renamed to ... Backport GH-93769 for a clickable number in https://github.com/python/cpython/commits/main (though, it's just a nitpick).

[markshannon]

@pablogsal I'd like to get this merged, as it will be needed for the backport of #94371

[pablogsal]

Looks good to me 👍

[pablogsal]

@pablogsal Should I regenerate the ABI for this PR?

I can try to do it for you. Otherwise you can follow these instructions:

@pablogsal Should I regenerate the ABI for this PR?

[pablogsal]

I cannot push to the PR, but you can copy this commit into yours if you want:

pablogsal@8380dd0

[markshannon]

Hmm, I followed the instructions, yet a0ced71 is clearly different to pablogsal@8380dd0

[markshannon]

I see, it is because of the assert changing the line numbers.
Why does an application binary interface include line numbers?

[pablogsal]

Why does an application binary interface include line numbers?

Those are not taken into account when checking it. They just come from the DWARF if I am not mistaken

[pablogsal]

LGTM !