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
0.1.0 named the specification by path. That was true inside the tarball it was
published from and false for a consumer: a project naming openkal-kit alongside
any implementation reached openkal two ways --- by path from the kit, by version
from the implementation --- and the build refused, correctly, because two routes
to one package are two copies of it.
⚠️ 0.1.0 IS REMOVED RATHER THAN SUPERSEDED. A range requirement would take the
larger version and never see it, but `= 0.1.0` would, and there is no graph in
which that version does anything but fail. A version that cannot be used is not
a version a consumer should be able to select.
0.1.1 is published from the tag `kit-0.1.1`, not from a version of the
specification. The specification did not change, and releasing it to repair a
package beside it would have required five implementations to move so that an
in-step check could be satisfied about a contract that had not moved. The
precedent for a version key that differs from its tag is already here:
grpc-plugin 1.83.0 comes from grpc-m's v1.83.0-4.
The publisher asserted the archive's own `*/kit/mcpp.toml` before uploading ---
that it declares 0.1.1 and names the specification by version --- because a tag
placed on the wrong commit produces an archive that downloads, verifies against
its hash, and still carries the defect the release exists to remove.
Co-authored-by: speak-agent <x.d2learn.org@gmail.com>
0 commit comments