Introspection: use typing_extensions.Buffer instead of collections.abc.Buffer (py312+)

[jorenham]

collections.abc.Buffer was added in Python 3.12 (docs), but PyO3 also supports older versions. By using typing_extensions.Buffer instead, this will no longer lead to typing issues. Note that within .pyi stubs typing_extensions is always available, even if not installed, because typing_extensions is bundled as part of the stdlib stubs through typeshed: https://github.com/python/typeshed/blob/main/stdlib/typing_extensions.pyi

[jonasdedden]

@jorenham just as a datapoint, I'm no PyO3 maintainer: In recently merged PR #6362, there was a bit of back and forth of how to deal with typing features not available to all Python versions.

In that PR, things converged to let it render, depending on Python version, to the variant with maximum portability or compatibility. In this specific instance, a disjoint_base pointer was used internally, but was rendered to typing.disjoint_base vs. typing_extensions.disjoint_base, where the former is only emitted in codebases targeting 3.15+.

Similarly, one could introduce a similar rule here, i.e. emit collections.abc.Buffer for Python 3.12+, and typing_extensions.Buffer for <3.12.

This is because using typing_extensions.Buffer would trigger ruff UP035 (deprecated-import, pyupgrade), so I would consider collections.abc.Buffer to be the more canonical variant of Buffer.

Verified empirically with ruff:

# test_buffer.py
from typing_extensions import Buffer

$ uvx ruff check --select UP035 --target-version py312 test_buffer.py --diff
-from typing_extensions import Buffer
+from collections.abc import Buffer
Would fix 1 error.

[jorenham]

Similarly, one could introduce a similar rule here, i.e. emit collections.abc.Buffer for Python 3.12+, and typing_extensions.Buffer for <3.12.

This is because using typing_extensions.Buffer would trigger ruff UP035 (deprecated-import, pyupgrade), so I would consider collections.abc.Buffer to be the more canonical variant of Buffer.

Ah yea, that's indeed a problem. Doing it as you suggest crossed my mind, but to be honest I was just being lazy and went with the easy solution :p

[Tpt]

Thank you! Just a test issue and some ideas where push backs are more than welcome

[codspeed-hq]

Merging this PR will improve performance by 12.77%

⚠️ 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

⚡ 2 improved benchmarks
✅ 139 untouched benchmarks
⏩ 6 skipped benchmarks¹

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
⚡	into_biguint_zero	1.4 µs	1.2 µs	+13.49%
⚡	into_biguint_small	1.5 µs	1.4 µs	+12.06%

Tip

Curious why performance improved? Comment @codspeedbot explain why performance improved on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing jorenham:introspection/typing_extensions.Buffer (20edb7e) with main (360f899)}

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. ↩