- Shell 100%
| Filename | Latest commit message | Latest commit date |
|---|---|---|
|
All checks were successful
Publish Jellium Flatpak / publish (push) Successful in 2m2s
Upstream's CI already builds a Flatpak bundle for every commit on main. Take that, replace /app/bin/jellium-desktop with a launcher that points WAYLAND_DISPLAY at /run/flatpak/$WAYLAND_DISPLAY (the app's Wayland proxy overwrites flatpak's xdg-run symlink and leaves a dead socket, so every launch after the first hangs), sign and publish. JELLIUM_SOURCE_ID (bundle + launcher hashes) is published alongside; a run whose content is already live skips signing and pushing. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> |
||
| .forgejo/workflows | ||
| flatpak | ||
| README.md | ||
Jellium Flatpak builder
This Forgejo repository publishes the upstream
Jellium Desktop Flatpak to
the shared OSTree repository on the pages branch of
gmatte/flatpak. It does not
compile anything. Upstream's own CI builds a Flatpak bundle on every push to
main (the "non-Flathub bundle" nightly in their README). This workflow:
- downloads that nightly and checks that it was built from the commit the mirror gmatte/jellium-desktop points to;
- replaces
/app/bin/jellium-desktopwithflatpak/jellium-desktop-launcher.sh(see Wayland launcher below); - signs it, merges it into the existing repository (other apps' refs are
kept), and force-pushes the result to
pages.
Wayland launcher
Upstream's flatpak hangs on every launch after the first one under Wayland.
Its Wayland proxy claims the first free wayland-N name in
$XDG_RUNTIME_DIR. Inside the sandbox that is the app's persistent
xdg-run directory, where wayland-N is flatpak's symlink to the
compositor. The proxy replaces that symlink with its own socket and leaves
the socket behind on exit, even a clean one. On the next launch flatpak
cannot recreate the link (F: …/xdg-run/wayland-1 is not a symlink to "../../flatpak/wayland-1" as expected), the app connects to the dead
socket, and it freezes. It recovers only after a logout or reboot clears
/run/user, and reinstalling does not help.
The launcher points WAYLAND_DISPLAY at /run/flatpak/$WAYLAND_DISPLAY,
the compositor socket flatpak always provides, then runs the real binary.
Drop it once upstream fixes the proxy.
Publishing only new content
Each publish writes JELLIUM_SOURCE_ID next to the repository: the SHA-256
of the nightly bundle plus the SHA-256 of the launcher. A run whose ID
matches the published one stops before signing or pushing anything, so the
same build is never published twice. Run workflow with force publishes
anyway. JELLIUM_UPSTREAM_REVISION records the upstream commit.
Configure Forgejo before running
Add these Actions secrets to this repository:
FLATPAK_GPG_PRIVATE_KEY— contents of the dedicated GPG private-key file.FLATPAK_GPG_KEY_ID—D0E42090A407A10DE3CF99A0DF1B3BB7B059BD43.FLATPAK_DEPLOY_KEY— contents of the dedicated SSH private-key file.
Add the matching SSH public key as a write-enabled deploy key on
gmatte/flatpak:
ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIDOyw03DmvHTOPxpuzY4CpcRfUj/9TTC8p7a4ZHK4vCx gmatte-flatpak-pages-publish
In Forgejo, enable this on gmatte/flatpak under Settings → Deploy Keys.
The publisher push fails if the key is registered without Write access.
The workflow pins Forgejo's existing SSH host key and publishes only to the
pages branch. Do not expose these secrets to pull-request workflows. It
runs on the repository-scoped flatpak runner. Repacking needs no build
namespaces, so that runner's extra permissions are a leftover from when this
repository compiled the app.
Register a repository-level Actions runner for this repository and install
its registration token as the forgejo-jellium-runner-secret Kubernetes Secret
in namespace forgejo, with key registration-token. Do not use the
instance-wide runner token: this runner has extra container permissions and
must remain restricted to this builder repository.
Run a publish
Nothing polls. A run starts:
- When the mirror's
mainmoves. A webhook on the mirror dispatches the workflow (below). The mirror syncs every 8h, and Synchronize now on the mirror starts a run right away. If upstream's build for that commit has not finished, the run fails with a message saying so; run it again later. - On push to this repository's
main. - Manually with Publish Jellium Flatpak → Run workflow.
The flatpak-pages Kubernetes deployment serves the published branch.
Mirror webhook
On gmatte/jellium-desktop → Settings → Webhooks → Add webhook → Forgejo:
- Target URL:
https://forgejo.gmatte.xyz/api/v1/repos/gmatte/jellium-flatpak-builder/actions/workflows/build.yml/dispatches - HTTP method
POST, content typeapplication/json - Trigger on: Push events only, branch filter
main - Authorization header:
token <TOKEN>, a token withwrite:repositorythat can write to this repository
A mirror sync that brings new commits fires a push event. Its payload
carries "ref": "refs/heads/main" at the top level, which the dispatch API
reads as the ref to run this workflow on. The other fields are ignored, and
force defaults to false.
Local installation
Once the remote is live, add it and install the app:
flatpak remote-add --user --from gmatte \
https://flatpak.pages.forgejo.gmatte.xyz/gmatte.flatpakrepo
flatpak install --user gmatte net.nullsum.JelliumDesktop