gh-97922: Run the GC only on eval breaker

[pablogsal]

• Issue: Execute the garbage collector only on the eval breaker #97922

[arhadthedev]

A minor typo.

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @markshannon for commit 1634d14 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[markshannon]

Looks good, but let's run the buildbots before merging.

[brandtbucher]

This fixes the new test:

diff --git a/Lib/test/test_frame.py b/Lib/test/test_frame.py
index 4b86a60d2f..4b5bb7f94a 100644
--- a/Lib/test/test_frame.py+++ b/Lib/test/test_frame.py@@ -277,7 +277,7 @@ def callback(phase, info):
frame!
"""
nonlocal sneaky_frame_object
- sneaky_frame_object = sys._getframe().f_back+ sneaky_frame_object = sys._getframe().f_back.f_back
# We're done here:
gc.callbacks.remove(callback)

[brandtbucher]

Looks mostly good! Just a couple of questions regarding eval breaker recomputation.

Also, would it be better to run the GC after signals, etc.? I would guess that prompt signal handling, etc. would be more important than prompt GC. I have a slight preference towards preserving the current behavior for when signals, etc. are handled and just moving the GC around, rather than potentially changing when everything else is handled too.

[pablogsal]

Also, would it be better to run the GC after signals, etc.?

Indeed, it makes sense. I moved it after signals and pending calls but before gil switching

[brandtbucher]

Cool, one more fixup: