[dinwwwh]

Requested by dinwwwh · project thread

What

build was a two-line function that only match called. Its two lines now sit in match, and its doc comment moves along with them.

Why

V8 spends about 15k instructions on each top-level function: once at import, and again the first time it compiles the function. I measured this with callgrind on 40 tiny functions, each called once, against one function called 40 times. Folding build in drops one function from every import, and one compile from the first request that builds a tree. The cost is about 70 bytes more for match to compile on any app's first request.

Measurements

Instruction counts from callgrind, node --single-threaded, averaged over hash seeds 1 and 2. The route sets are oRPC-style: 250 static RPC routes and 703 REST routes.

	before	after
min / min+gzip	5060 / 2108 B	5043 / 2100 B
import	2.177M	2.167M (−10k)
RPC 3 requests (static only)	327k	344k (+17k)
REST cold+3, GET	3.534M	3.502M (−0.9%)
REST cold+3, one request per method	5.115M	5.083M (−0.6%)
REST cold+10	6.074M	6.037M (−0.6%)

• REST: the router's whole cold+3 share (import, add and 3 requests) drops by about 0.5%.
• RPC (static only): it rises by about 0.2% (+7k), because match compiles slightly more code.
• Steady state: unchanged in every case (static, param, params2, catch-all, encoded, mixes).

Checks

• Tests pass with 100% coverage.
• The model fuzzers, including the late-add ones, find 0 diffs.
• Stryker finds 68 survivors, the same as on main.

---

Generated by Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[codspeed]

Merging this PR will degrade performance by 22.34%

2 warnings

⚠️ 1 benchmark is dominated by system calls

System calls cannot be simulated consistently and are left out of the measure. The reported value understates the real cost. Switch to the Walltime instrument to measure them.

Learn more about system calls

See affected benchmarks

⚠️ Base and head ran on different runner types

Some benchmarks with significant performance changes were compared across different runner types, so the changes might not come from the code. Run base and head on the same runner type, or use CodSpeed Macro Runners for more stable results.

Learn more about Macro Runners

Open the report in CodSpeed to investigate

❌ 1 regressed benchmark
✅ 28 untouched benchmarks
⏩ 2 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
❌	@openapi-spec/router	76 µs	97.8 µs	-22.34%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing claude/project-thread-m32ra9 (9fdf14a) with main (a2b7d67)}

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. ↩

[pullfrog]

✅ No new issues found.

Reviewed changes

• Inline build into match — the module-private two-line build(routes) helper (its sole caller was match) is folded into the routes.pending.length > 0 branch, with its doc comment carried over. The inlined expand(routes.root, routes.pending, 1, routes.unfiltered > 0) + routes.pending = [] are statement-for-statement identical to the removed body, so matching behavior is unchanged.

Verified: build( has no remaining references, expand is resolved in the same module, and the router suite passes (6 files / 73 tests).

  ^{｜ View workflow run ｜ Using deepseek-v4.1-flash (free via Pullfrog for OSS) ｜ 𝕏}

[dinwwwh]

CodSpeed's −22% on "match, 250 RPC-style procedures" comes from when V8 compiles match, not from slower matching. The router does the same work per request as before.

CodSpeed runs with --no-opt and measures one call after a few warm-up calls. match now has a few more bytes of bytecode, so Sparkplug's batch compile of it happens one call later: in the 8th call instead of the 7th. I replayed this under callgrind with CodSpeed's V8 flags, using the bench's routes and requests:

warm-up calls before the measured one	main	this PR
6	306k (compile lands here)	53k
7	48k	314k (compile lands here)
8	27.9k	46k
9, 12, 30	27.9k	27.9k

From the 9th call on, both versions take 27.9k instructions per call. With TurboFan on, the steady state is also unchanged: rpc 1589 vs 1589 and static 1644 vs 1645 instructions per request. This is the same Sparkplug shift as on #167 and #161. I'm merging on that basis.

---

Generated by Claude Code