After a pull request's Build APK run succeeds, its APK is published as the
pr-<number> prerelease in Droid-Deck/DroidDeck-CI, named
DroidDeck-PR.<number>-<commit>.apk, with release notes taken from the pull
request. New commits to the same pull request replace its APK. Pull requests
from forks are resolved by head owner, branch and commit.
A pull request's build is signed only with the committed testkey, so every CI
build installs over the last; the publish checks the certificate against it
and refuses anything else.
Needs a DROIDDECK_CI_TOKEN secret: a fine-grained token with Contents
read/write on Droid-Deck/DroidDeck-CI. Without it the run skips with a warning.
The runtime carries Termux's proot 5.1.107-70 as a prebuilt binary, built from
termux/proot 4dba3afb (every file:line site in it matches that commit). The apk
now builds that same commit with the NDK, with three patches ported from
WinNative's feature/proot-enhancements, and prefers it over the runtime's copy:
- Tracees are hashed by pid, so each ptrace stop finds its tracee without
walking a list of every thread, and terminated tracees are swept only after
a termination.
- A clean absolute path three or more directories deep under the rootfs alone
has its parent opened once and taken as canonical when the kernel's path for
it is the one proot would build, instead of an lstat per component; the final
component of a call that does not follow it is no longer lstat'ed. Extensions
still see the parent's host path, and the f2fs workaround keeps the old walk.
- A clone3 thread keeps its inherited flags when clone_args cannot be read.
talloc is linked in statically from Samba's release tarball, so libproot.so
needs only libc and libdl. On an x86_64 Android 15 emulator the same build ran
a path and thread test suite with output identical to the unpatched commit
(fake id, fake root, link2symlink, with and without seccomp), and a stat-heavy
benchmark ran 25-27% faster, 31-36% with 1500 live threads.
droiddeck-gpu runs a program in a gamescope of its own on the app's compositor
(wayland-0), full screen over the desktop - the path the rail's play chips use,
where Vulkan works. droiddeck-desktop-gpu gives every Game/Emulator menu entry a
per-user copy through it (auto: while the desktop is on pixman), keeping 'Run in
a desktop window' as an action.
droiddeck-desktop loads /usr/local/lib/droiddeck-wlroots for the vulkan/gles2
renderers and falls back to pixman when labwc stops at start. The build stages
the wlroots release named by tools/wlroots/release.env, once published.
wlroots 0.20.2 (Arch's wlroots0.20, what the desktop's labwc links) gains an
allocator that exports Vulkan images as DMA-BUFs for a GPU node that is not a
DRM device - the Adreno KGSL stand-in - where gbm and the DRM lease path fail
('unable to create allocator'). linux-dmabuf skips its GEM import check on such
a node so client buffers are not all refused. Built by build-wlroots.yml.
Compositor: without a Wine desktop, a window that opens on top takes keyboard
focus (a gamescope started from the Linux desktop).
App name, app ID and code package: DroidDeck, com.droiddeck.launcher
(sources moved to com/droiddeck/launcher, the 68 JNI functions renamed
with them). The new app ID installs beside SteamDeck rather than over it.
Everywhere the name was the app's own: the sidebar title, the session
report, share zips, the shell title, wake-lock and log tags, User-Agents,
the Compose theme, the Download folder and switch files (droiddeck-env,
-osc, -driver, -tu-debug, -no-pad, -pad-log, -no-hud, -wlr-renderer; the
old names are not read), and inside the runtime (droiddeck-desktop,
Firefox's droiddeck.js, the shortcuts tag, the rumble socket, debug and
cache paths). Build scripts, CI artifact (droiddeck-apk), CMake project,
README and the progress log follow.
Kept on purpose: Valve's Steam Deck (the -steamdeck flag,
steamdeck_publicbeta/stable, 'Steam Deck mode', the steamos.cloud
package URLs) and names inside the hosted desktop package (steamdeck.png,
steamdeck-steam, steamdeck-bigpicture). 'Desktop installed' now tests for
labwc instead of the launcher the app stages itself. A committed
__pycache__ .pyc is removed.
The QAM offers Performance Overlay levels but nothing drew them: the
runtime has no mangoapp. CI now fetches Valve's mangohud-20260902.1 and
the five libraries mangoapp needs that the runtime lacks, at the versions
it was built against (fmt 10.2.0, spdlog 1.13.0, glfw 3.4, libtraceevent
1.8.2, libtracefs 1.8.0 from Valve's mash-20240705.2 snapshot), each
pinned by sha256 in tools/mangoapp/packages.txt. They are staged under
usr/local/lib/mangoapp with a wrapper on PATH; the runtime image is not
touched.
A second sink module: connects to the relay's socket (lazily, the relay comes up after the
daemon; a clocked null sink until then), sends the hello, maps the render ring it gets back and
keeps it topped up to the relay's target with a futex wait per burst. Every AAudio call, the
adaptive buffer and the route handling stay in the relay, outside proot. With DirectAudio on the
daemon loads this sink instead of module-aaudio-sink; the bundle carries both.
The inherited module is gone from app/src/main/assets/pulseaudio.tzst (74 → 73 modules); CI adds
the app's own at build time and refuses a bundle that already has one. BUNDLE_STAMP bumped so an
installed app re-unpacks.
Blocking-write design: the IO thread pulls one burst at a time from the sink and hands it to a
blocking AAudio output stream, whose buffer paces the thread; latency comes from the stream's
playback timestamp; with adaptive=1 the buffer grows a burst per underrun up to a cap. Same
module arguments as before, so the daemon's config line is unchanged. build.sh compiles it with
the NDK against the app's own libpulse* and a small config.h; a workflow checks it on every push.
Four of Armada's patches apply to 3.16.29 as they are (appid-focus fallback, ARM64 night mode,
ARM64 colour-temperature slider, P3 red); two more are ported by hand: realtime-priority Vulkan
queues on GAMESCOPE_FORCE_VULKAN_REALTIME=1 without CAP_SYS_NICE, and the gamepad cursor sprite
following the X pointer. build-gamescope.yml builds Arch Linux ARM's package with them in an
Arch container, checks every NEEDED library against the runtime's own list, and publishes the
binary as a release asset for the apk to stage over /usr/local/bin/gamescope.
Deck mode passes -steamdeck alone: -steamos3 opted the install into the steamdeck_stable
branch and started the restart loop Bannerlator hit on device (38996ea1). The SteamOS helpers
the client calls in that mode - the two Armada found it needs under steamos-polkit-helpers and
the three under /usr/bin - are staged from the apk as no-op stubs. The client branch on disk is
kept in step with the one forced on the command line, and an update that cannot install stops
the session after three restarts instead of looping for ever.
The Steam cog gets a Client branch row: public beta (every session so far) or Steam Deck
public beta, the channel Armada bootstraps from.
The client reads /sys/class/power_supply/BAT<n>/; Android's supply is called battery and laid
out differently, so the Quick Access Menu had no battery. BatteryComponent writes BAT0 and BAT1
from Android's battery API every five seconds - status, percentage, charge, energy, power and
time to empty - into a directory bound over /sys/class/power_supply.
A pull request is checked out at its own head and merged onto main by the job itself: clean,
and the merged result is what gets built; conflicting, and the files and every conflicting hunk
go into the run summary and the comment, and nothing is built. The Gradle log is kept, and a
failed build's compiler errors (paths shortened) and Gradle's own 'What went wrong' are quoted
back. One comment per pull request, rewritten on every push; fork pull requests, whose token is
read-only, get the same text in the run summary.
Every push to the pull request rebuilds it and cancels the build it replaces (the concurrency
group is already per ref, and a pull request's ref is its own); a draft waits until it is
marked ready. The apk comes back as the run's artifact.
The list spans three lines inside one string, and the case pattern looked for
the name between two spaces - so the last name on each line, followed by a
newline, could never match. libaaudio.so is one of those, and the DirectAudio
relay helper is the first library we ship that needs it.
Firefox decides whether it is online by dumping the routing table over
netlink, which Android denies to apps; it sees no default route, switches to
Work Offline and every page fails while the network is fine. A defaults file
staged into Firefox's pref directory turns that off, and drops the process
sandboxes that have nothing to set up under proot. Debug runs now log a name
lookup and a fetch over IPv4 and IPv6 from inside the session.
The panel was there all along, a bare strip at the top: the first package's
panel.conf described a panel it never named in [General] panels=, so
lxqt-panel built itself a default one with no plugins. panel.conf now travels
with the app like the rest of the desktop's defaults, and the old nameless
file on a device is replaced once. The nested labwc output was 1280x720 -
wlroots' default window - with the desktop upscaled from that; rc.xml now
carries the app's surface size and the autostart sets it again via wlr-randr.
Debug runs keep each program's full output under ~/.cache/steamdeck/.
The desktop came up with a wallpaper and nothing on it the second time. The
runtime directory outlives a session, and lxqt-panel and pcmanfm-qt are
single-instance programs that leave a lock and a socket there; finding last
session's, they decided they were already running and exited without a word.
The launcher now clears everything but the compositor's own socket, runs labwc
under dbus-run-session so the panel's menus and the file manager have the bus
they expect, and logs what the panel and the file manager say. The launcher
and labwc's defaults are staged by the app over the desktop package's copies,
so a fix here no longer costs a 426 MB re-download.
From Bannerlator's feat/linux-gamescope-runtime since this repo was cut:
- Session teardown (4509d788): proot was SIGKILLed, which never lets its
--kill-on-exit cleanup run; its tracees lived on with no tracer, every
seccomp-trapped syscall failing with ENOSYS, spinning at a full core for over
an hour. Now SIGTERM, a grace for proot's own cleanup, SIGKILL only if it
will not go, then a sweep of the tree it had — snapshotted first, each pid
checked against its start time.
- The Steam client's controller interposer is its own source
(fakeinput_steam.cpp, 28922e7c), separate from the one preloaded into Wine.
- The seeder runs on a beat so a prefix the client creates mid-session is
covered before its install scripts run (6ee86bff).
- GE-Proton and proton-cachyos as compatibility tools: the registrar
bannerlator-proton-extra and the compat script's adopt_extras (03867656).
- space.c in the session shim: a library's own free space, not the phone's.
The env-vs-args ordering bug (98d38931) does not apply here: nothing is added
after the script name.
It only checks the schemes an apk's own minSdk requires, so at 26 it reported
v1 as false while the JAR signature was sitting in META-INF the whole time.
The file check beside it says the same thing a second way.
AGP drops JAR signing at minSdk 24 and above and the DSL flag does not move it,
release build or not. zipalign then apksigner with the three schemes named
explicitly is the only version of this that can be checked.
AGP drops JAR signing from a debug build when minSdk is 24 or higher, whatever
the signing config says, so the check added with the testkey failed on its
first run. Nothing is minified in either variant; this only changes which one
CI hands over.
The session now belongs to a foreground service rather than the activity.
Android demotes a process as soon as its last visible activity goes and the
low-memory killer reaps the guest, so leaving Big Picture to answer a message
came back to a dead Steam; the service holds the process at perceptible
priority, a partial wake lock keeps the CPU from dropping the guest's threads,
and a high-performance WiFi lock keeps a backgrounded download from being
throttled to nothing. The notification is silent (IMPORTANCE_LOW), carries a
Stop action, and returns to the session when tapped. Back now puts the app away
and leaves Steam running; a swipe out of recents ends the session.
The compositor moved with it: it is process-wide, started once, and merely
swaps the Surface it presents into as the activity comes and goes.
Package is com.steamdeck.launcher, signed with the AOSP testkey (v1+v2+v3) so
every build installs over the last one, and CI asserts all three schemes.
The launcher icon is our own controller on a purple glow — the spirit of the
Steam tile the client entry uses, none of its geometry.
The repo is private, so each run is billed. Three caches now carry a rebuild:
Gradle's own through setup-gradle (dependencies, build cache, configuration
cache), the CMake/ninja tree keyed on the C sources and the NDK, and the two
cross-compiled preloads — the last of which also skips installing the cross
toolchain, since a run that changes only Kotlin has no reason to touch a C
compiler. Superseded pushes are cancelled, documentation pushes do not build,
and the apk is stored rather than recompressed.
The preload assertions moved from the cross binutils to readelf, which reads
any ELF whatever the host, so they keep running on a cache hit.
Credits name The412Banner and maxjivi05: the runtime under all of this is
Max's gamescope work from WinNative.
maxjivi05