aio-libs / multidict / 35471524033

87%

push

github

Use release/acquire ordering for the retired-table reader count (#1481)

<!-- Thank you for your contribution! -->

## What do these changes do?

Fixes a ThreadSanitizer-flagged data race on the free-threaded build:
`keys->num_readers`, which gates whether a retired hash table is safe
to free, was incremented, decremented, and checked entirely with
relaxed atomics. Relaxed ops give no happens-before guarantee, so a
lock-free reader's in-progress field reads were not formally ordered
before a concurrent free of the same table. This adds `_acquire`/
`_release` atomic helpers to `atomic_helpers.h` and uses release on
the reader-exit decrement, acquire on the drain's free check. The
increment stays relaxed, and the coarse `num_active_readers` gate
keeps its existing seq_cst ordering, which it needs for an unrelated
(Dekker-style) reason spelled out in the surrounding comment.

## Are there changes in behavior for the user?

No. This only strengthens internal memory ordering on the
free-threaded build's lock-free read path; there is no change to any
public API or observable behavior.

## Is it a substantial burden for the maintainers to support this?

No. It adds two small, narrowly-scoped helper functions matching the
existing style of `atomic_helpers.h` and updates three call sites plus
their design comment.

## Related issue number

N/A

## Checklist

- [x] I think the code is well written
- [x] Unit tests for the changes exist
- [x] Documentation reflects the changes
- [ ] If you provide code modification, please add yourself to
`CONTRIBUTORS.txt`
- [x] Add a new news fragment into the `CHANGES/` folder
- [x] `make doc-spelling` passes and any new technical words are added
to `docs/spelling_wordlist.txt`

<details>
<summary>Agent run details (optional, for reviewers)</summary>

Found via the ThreadSanitizer CI job on an unrelated PR
(aio-libs/multidict#1480), which caught this pre-existing race on
master.

Tests: full suite passed on both the default C... (continued)

756 of 1512 branches covered (50.0%)

Branch coverage included in aggregate %.

6237 of 6505 relevant lines covered (95.88%)

1.92 hits per line