Skip to content

Vary the release matrix by feed, not by target - #3

Merged
jetm merged 1 commit into
mainfrom
release-matrix-by-feed
Sep 15, 2026
Merged

jetm merged 1 commit into
mainfrom
release-matrix-by-feed

Conversation

@jetm

@jetm jetm commented Sep 15, 2026

Copy link
Copy Markdown
Contributor

Problem

The 0.1.0 release published qemux86-64 and then failed on imx93-frdm with
that extension version is already taken. Both rows pointed at 2024/next, and an
extension version is global within a distro release and channel, so the second row
was a duplicate publish rather than a second target.

Solution

Vary release and channel instead. Connect distributes a published extension to the
per-target feeds itself - avocado-ext-cli and avocado-ext-microclaw both ship
architecture-specific binaries, both publish only a qemux86-64 row, and both are
present in imx93-frdm-ext as well as qemux86-64-ext - so the target input
names the target used to package, not the set of devices that can install the result.

This also fixes coverage the per-target matrix never had: 0.1.0 reached 2024/next
only, leaving 2026/next without the extension entirely.

Verification

Tag 0.1.1 published successfully to both feeds:

publish (qemux86-64, 2024, next): success
publish (qemux86-64, 2026, next): success

Reviewer notes

  • The version moves to 0.1.1 because 0.1.0 is taken in 2024/next and a collision
    is refused rather than overwritten, so re-tagging 0.1.0 would fail the 2024 leg
    again.
  • supported_targets still declares qemux86-64 and imx93-frdm. That list is
    what the extension supports; it is not the publish matrix, and the two should not
    be kept in step.
  • Tag 0.1.1 already points at this commit, which is not yet reachable from main -
    main is protected, so the fix went out as a tag before it could land here.
    Merging this reconciles them.

The 0.1.0 release published qemux86-64 successfully and then failed on
imx93-frdm with "that extension version is already taken". Both rows
targeted 2024/next, and an extension version is global within a distro
release and channel, so the second row was a duplicate publish rather
than a second target.

Per-target rows buy nothing to lose. Connect distributes a published
extension to the per-target feeds itself: avocado-ext-cli and
avocado-ext-microclaw both ship architecture-specific binaries, both
publish only a qemux86-64 row, and both are present in imx93-frdm-ext as
well as qemux86-64-ext. So the target input names the target used to
package, not the set of devices that can install the result.

Varying release and channel instead also covers what the per-target
matrix never did: 0.1.0 only ever reached 2024/next, leaving 2026/next
without the extension at all. Both feeds now get every release.

The version moves to 0.1.1 because 0.1.0 is taken in 2024/next and the
publish is refused on a collision rather than overwriting. Re-tagging
0.1.0 would fail the 2024 leg for the same reason it failed the first
time.

Signed-off-by: Javier Tia <javier@peridio.com>
@jetm
jetm requested a review from nicksinas September 15, 2026 19:47
@jetm
jetm merged commit 9cad591 into main Sep 15, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants