[gh-117657] _Py_MergeZeroLocalRefcount isn't loading ob_ref_shared with strong enough semantics #118111
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
In the nogil-integration branch there's some intermittent TSAN warnings like these:
https://gist.github.com/DinoV/52b7131cb149b9ebb2e965970c4ae9e8
It looks like we're racing on the loading of
ob_ref_shared
in_Py_MergeZeroLocalRefcount
because we're doing a relaxed load, leading to the object potentially being freed while another thread holds a reference to it. This is occurring because with a lock-free read from the inline values there's no other synchronization points with the thread which owns the object.This just changes the read to be acquire.