Defer imports on the cold import path

[KRRT7]

Summary

Further reduce import attr and import attrs startup time by:

• Replacing the platform import with sys.implementation.name.
• Deferring copy imports until assoc(), evolve(), or Attribute.evolve() is used.

[codspeed-hq]

Merging this PR will improve performance by 44.45%

⚡ 2 improved benchmarks
✅ 15 untouched benchmarks
⏩ 2 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	WallTime	import attrs	127.3 ms	88 ms	+44.65%
⚡	WallTime	import attr	124.9 ms	86.5 ms	+44.26%

Tip

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

---

_{Comparing KRRT7:perf/defer-inspect-import-v2 (e685f6c) with main (ca0f045)}

Footnotes

1. 2 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. ↩

[KRRT7]

Merging this PR will improve performance by 46.76%

nice.

[hynek]

hmmm I think we'll have to try this again:

I'll switch main to wall time and merge it here.

[KRRT7]

it'll be interesting to see the result, but I don't think codspeed is a good tool for this, typically I use -X importtime and analyze the results.

[KRRT7]

Merging this PR will improve performance by 66.22%

never mind, I agree, I agree /hj

[hynek]

they have to be fucking kidding me

but NOW we're nice :P

+69.33%

[hynek]

Well, that was a morning well spent! +44.65% lfg