Shrink call keyword capacity

[charliermarsh]

Summary

Collect call keywords in a Vec, then move them into an exactly sized ThinVec before retaining them in the AST. The ordinary conversion preserved the Vec's spare capacity, while constructing and shrinking a ThinVec directly added work to the hot parse loop. This keeps the original parsing path and performs one exact-capacity conversion at finalization.

This reduces retained AST memory from 956,062 bytes to 899,358 bytes (-5.93%) and full parsed memory from 1,208,698 bytes to 1,151,994 bytes on the existing five-file benchmark corpus. ty's normalized parsed_module memory also fell by 32,384 bytes across three identical runs.

The parser, AST, formatter, linter, and ruff_db test suites pass (4,254 tests), with no snapshot changes, and prek passes. This PR remains a draft while CodSpeed validates the revised construction strategy.

[MichaReiser]

are you taking over my unfinished PRs (feel free to do so, I haven't found any combination that's nice perf wise)

#26227

[charliermarsh]

No, I didn't see #26227

[codspeed-hq]

Merging this PR will improve performance by 6.39%

⚡ 1 improved benchmark
✅ 146 untouched benchmarks
⏩ 4 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	Memory	parser[large/dataset.py]	776.3 KB	729.7 KB	+6.39%

Tip

Curious why this is faster? Use the CodSpeed MCP and ask your agent.

---

_{Comparing charlie/reduce-ast-keyword-capacity (7c1aec5) with main (e4945bd)}

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. ↩

[MichaReiser]

Noooo, please fix it for me 😆 I already have way too many draft PRs

[charliermarsh]

Haha ok