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
The workspace has 15 bench files and 61 benches, and no CI job runs any of them. There is
a mise run bench task, but neither ci.yml nor extended.yml invokes it.
Two consequences:
The benches can rot silently. Nothing compiles them, so an API change breaks them and no gate
notices until someone runs the task by hand.
Minimum fix is a compile-only check (cargo bench --no-run) in the fast lane, which is cheap and
catches the rot. Whether benches should additionally assert a regression threshold — and where
those numbers would come from on preemptible shared runners — is the harder question and the
reason this is its own issue rather than part of #435.
docs/testing.md explicitly places benches outside what it governs, so this is not blocked on the
testing corpus pass.
The workspace has 15 bench files and 61 benches, and no CI job runs any of them. There is
a
mise run benchtask, but neitherci.ymlnorextended.ymlinvokes it.Two consequences:
notices until someone runs the task by hand.
gamut-cmmbenchnumbers were the cited evidence for filing gamut-cmm: fixed-point/integer + SIMD evaluation paths (6.6x lcms2 gap) #423. Evidence that no gate reproduces is evidence
with a short shelf life.
Minimum fix is a compile-only check (
cargo bench --no-run) in thefastlane, which is cheap andcatches the rot. Whether benches should additionally assert a regression threshold — and where
those numbers would come from on preemptible shared runners — is the harder question and the
reason this is its own issue rather than part of #435.
docs/testing.mdexplicitly places benches outside what it governs, so this is not blocked on thetesting corpus pass.
Refs #435