Skip to content

release: prune the cache first, erofs and dbx checks for the versions update - #37

Merged
aledbf merged 4 commits into
mainfrom
release/20261103
Nov 3, 2026
Merged

aledbf merged 4 commits into
mainfrom
release/20261103

Conversation

@aledbf

@aledbf aledbf commented Nov 3, 2026

Copy link
Copy Markdown
Collaborator

What the release of the versions update needs, found on the way to it.

  • The builder prunes its cache before building, not after. A recipe renamed since the cache was stored left its old packages there: gst-plugins-bad 1.29.2 beside gst-plugins-bad-1.28 1.28.8, whose provides it outranks, failed snapshot's build on the builder and nowhere else. Pruning only came at the end of a release that had already succeeded, so that release could never come.
  • erofs does not log a killed reader's read as an error. -EINTR failed test:usb's journal check and showed in GNOME Logs, about 3 runs in 10, with nothing wrong with the filesystem: corruption is -EIO, and panics under dm-verity here. A kernel patch, to be sent upstream; other errors keep their message.
  • dbx current waits for LVFS's metadata. It ran early enough to pass without any.
  • The kernel rule says what already happens: a kernel newer than Arch's newest tag is built over that tag (kernel/Dockerfile already says what that leaves out).

Checked before merging: go test ./cmd/image, and task kernel:config and task kernel with the patch. The full gates run on main after the merge, because task release needs them at the merged commit.

🤖 Generated with Claude Code

aledbf and others added 4 commits November 2, 2026 21:21
A recipe renamed since the cache was stored left its old packages in it:
gst-plugins-bad 1.29.2 beside gst-plugins-bad-1.28 1.28.8, whose provides
it outranks, failed snapshot's build on the builder and nowhere else. The
cache was pruned only at the end of a release that had already succeeded,
so that release could never come. Prune after the pull, which also writes
the index with the builder's own key.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A process killed while a block it needs is being read gets -EINTR, which
erofs logged as a filesystem error. It failed test:usb's journal check and
showed in GNOME Logs, about 3 runs in 10, with nothing wrong: corruption is
-EIO, and panics under dm-verity here. Real errors keep their message.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
It ran early enough to pass with no metadata, when fwupd offers nothing
whatever the dbx, which is the case it exists to catch.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Construct adopts 7.3 before Arch has a tag for it. kernel/Dockerfile
already says what that leaves out and when the base moves; the rule and
the pin's comment said the same version, which is not always so.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@aledbf
aledbf merged commit fdc1349 into main Nov 3, 2026
5 checks passed
@aledbf
aledbf deleted the release/20261103 branch November 3, 2026 11:32
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.

1 participant