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
test(release): boot a board built from the packages just published
0.21.0 was cut to fix a board that rendered raw message keys and had no
CSS, and shipped with only the first of the two fixes. A stacked pull
request merged into a base that had already merged, so its commits never
reached main. Nothing noticed: main was green, the tag was coherent, the
packages published, and the notes described a fix that was not in them. It
was caught by grepping main by hand before re-pinning the Vercel template.
The self-host template drifting three versions behind, serving the same
message-keys bug to anyone clicking "Use this template", was the same
shape. Every gate in this repository examines the repository. None of them
look at what a user downloads.
So the release pipeline now scaffolds from the create-meith it just
published, installs from the real registry rather than from anything in
this tree, and boots the result at both materialization depths — the same
three assertions the workspace smoke makes. It runs after the npm job for
that reason and `publish` waits on it, so a broken artefact stops the
release being announced. It cannot un-publish a bad version and is not
meant to; it makes the failure known in minutes rather than after someone
deploys it.
The boot and its assertions move to scripts/board-boot-check.mts so both
smokes share one definition of a board working, rather than the published
one drifting into a weaker check than the one it is meant to mirror.
0 commit comments