gitpush version-gate suite does not exercise the non-strict call site at :2424
OPS-246 shipped the single-writer version gate in gitpush.sh v4.58/v4.59 and gitpush-bumporder.test.sh v1.2 proves it 8/8, but the suite exercises only the strict + website + Next.js branch at gitpush.sh:2367. The second call site at :2424 shares the same decision function, so the BEHAVIOUR is common to both and no defect is suspected - what is missing is EXERCISE. That distinction is the whole row: a shared function is an argument, not a measurement, and this class of bug (two writers to one version field, repeatedly declared fixed) is exactly where an argument has already failed. Ask: add one arm to gitpush-bumporder.test.sh that drives the non-strict call site through the same four cases - disk equal to HEAD bumps, disk above HEAD ships unbumped, disk below HEAD aborts, package.json absent from HEAD ships the disk value. Each arm forced red against v4.57 first, per the standing rule that a control unseen to fail is a green line rather than a control. Filed separately rather than holding OPS-246 open so the shipped work closes cleanly and the coverage gap stays visible instead of hiding inside a closed row.
Activity
-
Closed by aceb646. gitpush-bumporder.test.sh v1.4 adds arms [9]-[12] driving the non-strict call site at gitpush.sh:2539. 12 passed, 0 failed. [10]-[12] forced red against v4.57 recovered from a PINNED ref (8b591e7^) and RUN, not reasoned about: at this call site v4.57 absorbs a hand-set 2.5.0 into 2.5.1, pushes a version below HEAD with rc=0, and bumps a brand-new 0.1.0 to 0.1.1. [9] disk==HEAD is deliberately not forced red - no prior build can fail it - and is the anti-over-correction arm for this call site. No defect found at the second call site; the gap was exercise, not behaviour. Noted in the file: the two call sites are NOT identical - the strict one appends the bumped package.json to ONLY_PATHS_ARR and the non-strict one does not.