Problem
There are zero benchmarks across 129 Go files (grep -rn "func Benchmark" → no matches), despite a 100%-coverage gate and a determinism harness. Current performance characteristics had to be measured ad hoc for this audit (e.g. GitHub REST: 6.5 s compile, ~1.3 GB peak, 95.6 MB output; parse/resolve dominating the profile at 3.7 s of a 3.88 s compile). Regressions in compile time or allocations — e.g. from a parsing-library bump, to which the repo already pins behavioral coupling (cycles.go documents v1.24.0 specifics) — would land silently.
Proposed direction
go test -bench benchmarks over the golden petstore and a generated large spec (a corpus generator can live alongside irtest or in a script): compile wall/allocs, marshal, validate. Run them in the scheduled workflow from #77 with benchstat comparison, so the numbers above become tracked baselines instead of one-off measurements. Size: S.
Acceptance
go test -bench . ./... produces stable numbers for compile/marshal/validate on both corpora.
- The scheduled run records comparable output across commits.
Problem
There are zero benchmarks across 129 Go files (
grep -rn "func Benchmark"→ no matches), despite a 100%-coverage gate and a determinism harness. Current performance characteristics had to be measured ad hoc for this audit (e.g. GitHub REST: 6.5 s compile, ~1.3 GB peak, 95.6 MB output; parse/resolve dominating the profile at 3.7 s of a 3.88 s compile). Regressions in compile time or allocations — e.g. from a parsing-library bump, to which the repo already pins behavioral coupling (cycles.godocuments v1.24.0 specifics) — would land silently.Proposed direction
go test -benchbenchmarks over the golden petstore and a generated large spec (a corpus generator can live alongsideirtestor in a script): compile wall/allocs, marshal, validate. Run them in the scheduled workflow from #77 withbenchstatcomparison, so the numbers above become tracked baselines instead of one-off measurements. Size: S.Acceptance
go test -bench . ./...produces stable numbers for compile/marshal/validate on both corpora.