Record successful attach only after attaching in SuspendAttach

[ngoldbaum]

SuspendAttach::drop increments ATTACH_COUNT before successfully attaching. This was missed during review of #6085 (ping @anuraaga). The fix is to only increment if we actually successfully attach.

This bug can lead to crashes if the attach happens during interpreter shutdown on pythons older than 3.13.8.

I hit this while working on Quansight/spinningjenny#14. See the workaround there for how I'm avoiding the problem. spinningjenny stores state in Rust TLS, and that pattern does not look like any existing tests in PyO3.

I updated the testing utilities we use in test_finalization.py to trigger this. It unfortunately requires a decent amount of complexity to ensure the race is triggered deterministically.

Adding the soundness label because I think the current behavior is unsound.

[codspeed-hq]

Merging this PR will degrade performance by 12.79%

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

❌ 3 regressed benchmarks
✅ 138 untouched benchmarks
⏩ 6 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
❌	into_u128_small	876.1 ns	1,041.4 ns	-15.87%
❌	into_u128_zero	816 ns	925.7 ns	-11.85%
❌	extract_bigint_small	1.4 µs	1.6 µs	-10.55%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing ngoldbaum:fix-shutdown-race (63cb82b) with main (fc4d08c)}

Footnotes

1. 6 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[ngoldbaum]

not needed since Rust 1.73

[ngoldbaum]

ping for a review here @davidhewitt

[davidhewitt]

Yikes, just when I thought we'd finally closed all these finalization bugs!

I think given we're hopefully about to ship 0.30 I'm not rushing to back port this, WDYT?

[ngoldbaum]

Yeah that's fine from my end.