logan
90f1239fba
release.yml fired on every push to main, so the trigger was "a PR was merged" and nothing else decided. That is a version per unit of *work* rather than per *shipment*: eight releases in twenty-two hours, v0.0.1 through v0.3.1, for one session -- each fanning out to four publishers on a runner with capacity 1, so roughly forty packaging jobs shipped three issues while ordinary PR CI queued behind them. pacman, Homebrew and Obtainium see every one. The push trigger is gone and workflow_dispatch, which was already there and already worked, is the whole mechanism. Nothing else had to change to batch releases, because semantic-release already reads every commit since the last tag: five fixes and two feats become one minor release with all seven in the notes. Release frequency was only ever how often this file fired. This is the rule index-artifact.yml states and is the other instance of: a job that mutates state which cannot be rebuilt in ten minutes is triggered deliberately, not by a push. A release here is a tag, a Gitea release, an Arch package, a Homebrew formula, a signed APK and desktop assets -- and an Android version going backwards costs the user their library. `dry_run` is what makes a manual trigger usable: the point of pulling a lever by hand is being able to look first, so the input runs semantic-release --dry-run -- the version and the notes, no tag, no release, no publishers. Anything but the literal string "true" releases for real, because a typo in a dispatch box must not silently turn a shipment into a green no-op. Two alternatives were considered and rejected, both recorded on the issue. A `beta` integration branch relocates the trigger rather than removing one: it needs a second protected branch carrying the same required checks, and it *adds* a full check + e2e run per batch on the very runner whose queue is the complaint. A schedule batches without anyone having to remember, but puts the decision back on a timer, which is the thing being removed. Closes #115
7.7 KiB
Releasing the Android APK
.gitea/workflows/android-apk.yml builds a signed arm64-v8a APK on
every v* tag and publishes it to Gitea's
generic package registry, which is readable without credentials —
which is what lets Obtainium poll a plain URL with no token. It also
attaches the same file to the Gitea release, which is what a person
looking at the release page downloads.
Tags are not pushed by hand any more, but releasing is a decision.
.gitea/workflows/release.yml reads the Conventional Commits since the
last tag, decides the version, and pushes the tag this workflow is keyed
on — so releasing the APK means running that workflow, not running
git tag. It has no push trigger: merging a fix: or feat: used to
be enough and produced a version per merged PR (issue #115). Run it with
dry_run first to see what the accumulated commits would ship. The
workflow_dispatch path below is a different thing and remains, for
rebuilding a tag that already exists.
A prerelease tag is skipped here, cleanly. This workflow triggers on
v*, which matches v0.4.0-beta.1, and it is the one where that would
hurt most: the APK goes to the credential-free generic registry that
Obtainium polls, and the versionCode maths below splits on dots — it
would read 1 out of 0-beta and produce a wrong number rather than a
failed build.
The 1.x installs cannot be upgraded to 0.0.x
Releases restarted at 0.0.1 when they became automatic (plan 017).
versionCode is computed as maj*10000 + min*100 + pat, so 0.0.1 is
1 against the 10300 an installed 1.3.0 build carries — and
Android refuses a downgrade outright, with
INSTALL_FAILED_VERSION_DOWNGRADE.
The only way through is adb uninstall app.yellowjacket (or the
launcher's own uninstall) before installing 0.0.1, and that takes the
device's library, playlists and play counts with it — the same loss the
signing key section below exists to prevent, arrived at from the other
direction. This was chosen deliberately over offsetting versionCode by
a constant, on the grounds that the honest number is worth one reinstall
while an offset is permanent. The workflow prints a warning whenever the
code it computes is below 10600.
https://git.ljones.me/api/packages/yonlu/generic/yellowjacket-android/latest/yellowjacket.apk
A versioned copy is kept alongside it at
…/yellowjacket-android/<version>/yellowjacket-<version>.apk.
The app on the device is app.yellowjacket. Its launcher activity is
com.wails.app.MainActivity — a different package, because that is the
Wails scaffold's Java package and renaming it would mean renaming its
source. Every am start needs the fully-qualified form.
The signing key is the thing you cannot lose
Android refuses to update an app whose signing certificate changed. There is no override and no recovery: the only way to install a build signed with a different key is to uninstall first, which takes the user's library, playlists and play counts with it. The key therefore outlives every other secret in this repo.
Two consequences are wired into the workflow rather than left to
discipline. It refuses to build when ANDROID_KEYSTORE_B64 is
absent, instead of falling through to Gradle's debug-keystore default —
a debug key differs between every machine and every CI runner, so a
build signed with one is un-updatable from the moment it is installed.
And it refuses to publish an APK whose certificate reads
CN=Android Debug, which is the same rule enforced one step later, on
the artifact rather than the configuration.
Creating it
keytool -genkeypair -v \
-keystore yellowjacket-release.jks \
-alias yellowjacket \
-keyalg RSA -keysize 2048 -validity 10000 \
-storepass '<a long random password>' \
-dname "CN=YellowJacket, O=Shadow-Puppet, C=GB"
Do not pass -keypass. keytool has produced PKCS12 keystores by
default since JDK 9 — the .jks extension does not change the format —
and PKCS12 cannot hold a key password distinct from the store password.
Given one it tells you so and ignores it:
Warning: Different store and key passwords not supported for PKCS12
KeyStores. Ignoring user-specified -keypass value.
So there is one password. Asking for a second is how someone sets a wrong value and then debugs Gradle at midnight.
Back the .jks up somewhere that is not this repository and not this
server. Record the certificate fingerprint the build prints
(Signer #1 certificate SHA-256 digest); if it ever changes, updates
have already broken.
The secrets
Repository → Settings → Actions → Secrets.
| Secret | Required | Notes |
|---|---|---|
ANDROID_KEYSTORE_B64 |
yes | base64 -w0 yellowjacket-release.jks |
ANDROID_KEYSTORE_PASSWORD |
yes | the -storepass above |
ANDROID_KEY_ALIAS |
no | defaults to yellowjacket |
ANDROID_KEY_PASSWORD |
no | defaults to the store password, and per the PKCS12 note it cannot differ |
PACKAGE_TOKEN |
already set | shared with arch-package.yml; publishes to the registry |
base64 -w0 yellowjacket-release.jks # paste as ANDROID_KEYSTORE_B64
Cutting a release
git tag v1.3.1
git push origin v1.3.1
That is the whole trigger. homebrew-formula.yml keys on the same tag,
so the desktop formula and the APK are cut together. The version code
Android orders releases by is derived from the tag — 1.3.1 → 10301,
monotonic as long as minor and patch stay below 100 — so do not
publish 1.100.0, and never move a tag that has already been built.
workflow_dispatch rebuilds without a new tag, taking an explicit
version input or falling back to the latest v* tag.
What the workflow checks before publishing
- the APK exists and is non-empty;
- it carries exactly one ABI (
native-code: 'arm64-v8a'). x86_64 Android cannot run this app at all —modernc.org/libcissues a rawlstatsyscall that Android's seccomp policy forbids, on every x86_64 device and not merely the emulator — so an x86_64 slice would be ~31 MB that runs nowhere, and its reappearance means someone put the ABI back inapp/build.gradlewithout knowing that; - its
versionCodeis the one derived from the tag; - it is not signed with the debug key.
The keystore is also opened with keytool -list before Gradle runs,
because Gradle only notices a bad password at
:app:validateSigningRelease — a minute of build time in — and reports
it as a missing file rather than a wrong password.
Caches, and the first run
The job mounts four cache volumes; on a cold runner the first build is slow and everything after it is not.
| Volume | Holds | Cold cost |
|---|---|---|
/cache/android-sdk |
SDK, platform, build-tools, NDK r26d | ~2 GB |
/cache/gradle |
GRADLE_USER_HOME — wrapper + AGP graph |
~700 MB |
/cache/tool |
the Go toolchain (shared with ci.yml) |
~200 MB |
/cache/pnpm-store |
pnpm store (shared with ci.yml) |
— |
Every path must be inside the runner's valid_volumes allowlist. One
that is not makes the job fail to start rather than silently skip
the mount.
The NDK is pinned to r26d (26.3.11579264). Newer NDKs have broken
the Wails Android build before; it is a version, not a floor.
Building one locally
make android # unsigned-ish: debug key, versionCode 1, 0.0.0
YJ_VERSION=1.3.1 YJ_VERSION_CODE=10301 \
ANDROID_KEYSTORE_FILE=$PWD/yellowjacket-release.jks \
ANDROID_KEYSTORE_PASSWORD=... ANDROID_KEY_ALIAS=yellowjacket \
make android # what CI produces
Running it is a separate tier — see
.pi/skills/yellowjacket-dev/references/android-tier.md, and read its
first section before you try, because a failing Android build looks
exactly like a working one.