Fix alru_cache crash with deferred annotations on Python 3.14

[rodrigobnogueira]

What do these changes do?

Fix alru_cache crashing at decoration time on Python 3.14 when an annotation refers to a name defined after the decorated function. PEP 649 makes annotations lazy, and copying fn.__annotations__ eagerly forces their evaluation; functools.update_wrapper copies the lazy __annotate__ function on 3.14 instead, so both wrapper classes now do the same, keeping the __annotations__ copy as the fallback for 3.13 and older. Laziness now matches functools.lru_cache: inspect.get_annotations() resolves once the name exists and raises NameError only on access when it never does.

Are there changes in behavior for the user?

Only on Python 3.14: the wrapper no longer exposes an eagerly evaluated __annotations__ attribute; annotations are reachable through inspect.get_annotations() and typing.get_type_hints(), matching functools.lru_cache there. No changes on 3.13 and older.

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

No.

Related issue number

Fixes #773

Checklist

• I think the code is well written
• Unit tests for the changes exist
• Documentation reflects the changes
• Add a new news fragment into the CHANGES folder (N/A; CHANGES.rst is updated in release PRs in this repo)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.39%. Comparing base (b5717c7) to head (26f7eda).
⚠️ Report is 5 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #781      +/-   ##
==========================================
+ Coverage   97.78%   98.39%   +0.60%     
==========================================
  Files          16       17       +1     
  Lines        1176     1246      +70     
  Branches       61       63       +2     
==========================================
+ Hits         1150     1226      +76     
+ Misses         23       17       -6     
  Partials        3        3              

Flag	Coverage Δ
unit	97.35% <97.43%> (+0.49%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[codspeed-hq]

Merging this PR will not alter performance

✅ 63 untouched benchmarks
⏩ 4 skipped benchmarks¹

---

_{Comparing fix/deferred-annotations-py314 (26f7eda) with master (b5717c7)²}

Footnotes

1. 4 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. ↩

2. No successful run was found on master (593715a) during the generation of this report, so b5717c7 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩

[Vizonex]

@rodrigobnogueira I would recommend using sys.version_info as a faster workaround for this problem as this is 3.14+ related, this way this part of the code doesn't trigger a performance regression.

            if sys.version_info >= (3, 14): 
                self.__annotate__ = fn.__annotate__  # type: ignore[attr-defined]

[rodrigobnogueira]

Done in 73906f3. Both copy sites now branch on sys.version_info >= (3, 14); older interpreters do the plain __annotations__ copy with no try/except on a missing __annotate__. The narrow try/except stays inside the 3.14 branch for wrapped partials, which the partial test covers. The type: ignore[attr-defined] became unused under the gate and was removed. Suite green on 3.12 and 3.14, mypy clean on both.

[Vizonex]

I like this change. Thank you for this now there shouldn't be a performance regressions with older versions of python.

[Dreamsorcerer]

Should really make both comments concise like this. I'd also drop the prefix so there's less maintenance. We can see the version check below.

[rodrigobnogueira]

Done in 26f7eda — both comments trimmed, prefix dropped.