0.36.2 2026-10-06
Release-process fix found by auditing the v0.36.1 deploy rather than trusting it.
Fixed
- Production reported the wrong version: the box ran the new code while showing the old tag. The release flow is push, CI green, tag, release, so the deploy triggered by the master push runs before the tag exists, and
git describe --tags --abbrev=0therefore resolved to the previous tag. Verified against a real deploy log, not inferred: the run for v0.36.1 printedBuilding calionauta/gogogo v0.36.0 (commit e4a2206), and the live box served0.36.0for a commit taggedv0.36.1minutes after the deploy started.
Two coupled defects, both fixed:
deploy.ymldid not trigger on tags at all. Addedtags: ["v*"](nested underpush; GitHub does not path-filter tag pushes, so a tag always deploys).- Even had it triggered, the deploy job's
if:allowed onlyrefs/heads/masterorworkflow_dispatch. A tag push setsrefs/tags/vX.Y.Z, so the job would have been skipped and the tag-triggered deploy would have built nothing.
The extra restart is the deliberate price of not tagging before CI has passed; tagging first would avoid it but would put a red CI on the tag, which is worse. Proven end-to-end: a tag-triggered deploy for v0.36.1 now logs Building calionauta/gogogo v0.36.1 and the live box reports 0.36.1.
docs/deploy.mdasserted "the badge on the running box matches the tag that produced it", which was false for exactly the ordering above. Corrected with the observed evidence and the rationale for the second deploy.
Verification
- Empirical, not assumed: re-triggered the deploy against the existing
v0.36.1tag and read the build log (Building v0.36.1) and the live page (0.36.1) back. The fix was confirmed by observing the corrected behaviour, not by reading the YAML. make ci-localgreen end-to-end;make site-checkgreen