[charliermarsh]

Summary

Formatter currently writes every element through &mut dyn Buffer, even when the destination is a vector. We now construct a formatter with direct references to the vector and formatting state when entering through VecBuffer::write_fmt:

Workload	Main	This PR	Time change
NumPy globals	29.72 µs	26.44 µs	−11.05%
Unicode pypinyin	107.40 µs	103.66 µs	−3.48%
Pydantic types	755.86 µs	739.81 µs	−2.12%
NumPy ctypeslib	406.22 µs	387.68 µs	−4.57%
Large dataset	1,662.46 µs	1,597.33 µs	−3.92%

This experiments with the dispatch part of Oxc #27559.

[astral-automations-bot]

ruff-ecosystem results

Formatter (stable)

✅ ecosystem check detected no format changes.

Formatter (preview)

✅ ecosystem check detected no format changes.

Formatter (stable) · Formatter (preview)

---

Results for 580f594 · Workflow run

[codspeed]

Merging this PR will improve performance by 4.77%

⚡ 2 improved benchmarks
✅ 58 untouched benchmarks
⏩ 238 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	CPU Simulation	formatter[large/dataset.py]	16.3 ms	15.5 ms	+4.93%
⚡	CPU Simulation	formatter[pydantic/types.py]	6.5 ms	6.2 ms	+4.6%

Tip

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

---

_{Comparing crmarsh/formatter-concrete-writer (580f594) with main (52cca7e)}

Footnotes

1. 238 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]

Nice. I suggest using Vec(&mut VecBuffer) over destructuring its fields. It keeps the intended isolation and avoids repeating logic.

[MichaReiser]

Beautiful 😆