The repo's .gitattributes (inherited from upstream) routes certain paths
through git-lfs. Gitea's LFS store doesn't hold those blobs, so on
checkout the smudge filter tries to download them, 404s, and leaves git
in a state where subsequent 'git fetch' calls appear to succeed but
don't actually populate refs.
Run 89 was bitten by this: checkout 'succeeded' with an LFS smudge
fatal, then 'git fetch upstream develop' ran silently, 'git merge
--ff-only upstream/develop' failed because upstream/develop ref
didn't exist locally, and the workflow logged a misleading warning
blaming a divergence that wasn't there.
Setting GIT_LFS_SKIP_SMUDGE=1 keeps LFS pointers as-is. We don't need
image bytes to ff-merge and diff refs.
The Gitea pull-mirror of element-hq/element-x-android is slow to
populate its initial clone (~12 GB). Rather than block workflow
verification on the mirror landing, fetch straight from GitHub — the
runner has outbound access and GitHub isn't flaky. The mirror stays in
place as a fallback / LAN-cache for humans doing manual git fetches.
Daily cron at 12:00 UTC (plus manual dispatch) that:
1. Fetches from the Sulkta-Coop/element-x-upstream pull-mirror
2. Fast-forwards main to upstream/develop if it has advanced
3. Measures how many commits behind main the wallet branch is now
4. Posts a ping to the Infra Matrix room so we know a rebase is due
Uses the house-bot (Matrix) account for notifications; token lives in
the repo's MATRIX_HOUSE_BOT_TOKEN Actions secret.
Removed .github/workflows/* — upstream's 18 workflows are GitHub-specific
(GITHUB_TOKEN scopes, Firebase / Sonar / Sentry / Localazy secrets we
don't have, macOS runners, etc). They were triggering on every push and
failing immediately, flooding the runner log. We're not proposing these
back upstream — we're a fork that doesn't publish to Play/F-Droid, so
their CI isn't ours to run.
If we ever need to see upstream's workflow definitions for reference,
they're one click away on github.com/element-hq/element-x-android.
