fix segfaults on tests in debug builds for PyPy

[davidhewitt]

Change Summary

This PR unifies the recursion guard limit between serialization and validation to 255 (thanks @adamreichold for the suggestion to move to a single limit).

I also add pytest.skip to a few tests which fail on PyPy with debug builds. This is because Rust debug builds don't optimize their stack space, so we hit trivial stack overflows.

Finally I add a test-debug job for PyPy 3.9 - can bump to PyPy 3.10 once the upstream PyO3 bug is fixed.

Related issue number

Ref #659
This should avoid crashes on debug builds with PyPy. (Doesn't close the full issue, which also exposes a bug in PyO3.)

Checklist

• Unit tests for the changes exist
• Documentation reflects the changes where applicable
• Pydantic tests pass with this pydantic-core (except for expected changes)
• My PR is ready to review, please add a comment including the phrase "please review" to assign reviewers

[codecov]

Codecov Report

Merging #744 (7e8a88d) into main (d08d269) will not change coverage.
The diff coverage is 100.00%.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #744   +/-   ##
=======================================
  Coverage   93.63%   93.63%           
=======================================
  Files          99       99           
  Lines       13895    13895           
  Branches       25       25           
=======================================
  Hits        13010    13010           
  Misses        879      879           
  Partials        6        6           

Impacted Files	Coverage Δ
src/lib.rs	100.00% <100.00%> (ø)
src/recursion_guard.rs	95.65% <100.00%> (ø)
src/serializers/extra.rs	99.10% <100.00%> (-0.01%)	⬇️
src/validators/definitions.rs	88.49% <100.00%> (ø)

---

Continue to review full report in Codecov by Sentry.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d08d269...7e8a88d. Read the comment docs.

[codspeed-hq]

CodSpeed Performance Report

Merging #744 will improve performances by 12.51%

_{Comparing dh/pypy-segfaults (7e8a88d) with main (4527b80)}

Summary

🔥 2 improvements
✅ 124 untouched benchmarks

Benchmarks breakdown

	Benchmark	main	dh/pypy-segfaults	Change
🔥	test_complete_core_json	2.8 ms	2.6 ms	+10.13%
🔥	test_small_class_core_model	51.2 µs	45.5 µs	+12.51%

[mgorny]

Thank you! I can confirm that with this patch added, the tests no longer crash for me.

[adriangb]

One thing I noticed the other day is that this limit gets hit if you just nest a bunch of function validators, e.g.

schema = int_schema()
for _ in range(1000):
    schema = no_info_after_validator_function(lambda x: x, schema)

Not something we need to fix here, but just food for thought.

[davidhewitt]

Yes, I think the best thing to consider would be that we make the recursion limit configurable. This is just a best-effort thing to protect users from segfaults on stack overflow, so if they think they have more stack space they could increase our limit safely.

Either we could make our limit configurable, or potentially we could use Py_EnterRecursiveCall to hang off Python's recursion limit. (I think that's 1000 by default, though will include Python calls. If we make our stack frames small enough that'll be fine.)