You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
-perf.commit must equal the measured tree's HEAD (refuse fabricated or dirty commits) #12
-perf.commit is recorded verbatim into perf.json and perf-table.md. On 2026-11-12 a full 40-character SHA reconstructed from a 7-character prefix was passed, so the table would have recorded a commit nobody measured.
Proposal (rbtest, whenever the module-root guard applies):
With -perf.commit: resolve HEAD of the tree at <root> (git -C <root> rev-parse HEAD); refuse unless the flag equals that 40-character SHA exactly, naming both values.
Dirty tree (git status --porcelain non-empty) with the flag: refuse, listing the dirty paths. Exception: only bench/go.mod / bench/go.sum dirty (consumers bump refbench locally; those are the harness).
Without the flag: fill in HEAD, suffixed -dirty when the tree is dirty.
No git available: refuse only if the flag was given.
The git lookup is replaceable for tests; README documents the flag.
Prior art: priorart on this repo (exit 1) lists only perf-related entries #1, #4, #5, #6, #7, #9, #10; I read their titles, none concerns validating the recorded commit.
Prior art: priorart "perf.commit must equal HEAD sha" at 2026-11-12 10:43 CET reported prior art; checked: read titles of #1,#4,#5,#6,#7,#9,#10: none validates the recorded commit
Activity
rbtest: harden the -perf.commit check after review
rbtest: reach the 100% coverage gate
rbtest: -perf.commit must equal the measured tree's HEAD (#13)
bench: bump go-scikit/refbench to v0.3.0 (#62)