3 people d94425b64f bench: router match/registration suite + BENCHMARKS.md (#3) ...

Prove the specificity-routing fork's performance on a representative
~70-route Hanzo cloud table (static, param, multi-param, wildcard).

- router_bench_test.go: match hot-path (app.next, 0-alloc), full dispatch,
  registration scaling (N=10/70/500/1000), and a Go 1.22 net/http.ServeMux
  routing + registration baseline, with precedence-correctness gates
  (static/param beat wildcard; fork agrees with ServeMux on every request).
- bench/: separate module (own go.mod, replace ../) that isolates the upstream
  gofiber/fiber v3.2.0 dependency — fork vs upstream match (per-route +
  aggregate sweep) and registration. Never pollutes the fork's own module.
- middleware/static: embed.FS serve benchmark (small 1 KiB + medium 64 KiB).
- BENCHMARKS.md: methodology (M1 Max, go1.26.4, -benchtime=1s -count=6,
  benchstat medians) and the real numbers.

Findings: the fork adds ZERO aggregate match cost (sweep 10.77us vs upstream
10.96us; per-route latency is a scan-order permutation, not a regression);
fiber routing is 0-alloc vs ServeMux's 1-5 allocs per dynamic route;
registration is O(n^2) (9.5ms at N=1000/method, <110us at N<=70) — the one
honest regression, and fixable with an index-based conflict check.

Co-authored-by: Hanzo Dev <[email protected]>

2026-07-11 05:06:59 -07:00

..

comparison_test.go

bench: router match/registration suite + BENCHMARKS.md (#3)

2026-07-11 05:06:59 -07:00

go.mod

bench: router match/registration suite + BENCHMARKS.md (#3)

2026-07-11 05:06:59 -07:00

go.sum

bench: router match/registration suite + BENCHMARKS.md (#3)

2026-07-11 05:06:59 -07:00