* fix(session): stop re-compressing game memory every 30 s when the game reads it straight back
On devices with 13 GB or less, the low-memory page-out skipped its
read-back backoff whenever memory was nearly exhausted. When the game
was actively using what was compressed, it read it straight back, so
the session pushed and re-read gigabytes every 30 seconds, pinning all
cores and freezing the game for 5-10 s each time. The read-back check
now applies on that path too: an effective page-out still repeats every
30 s, an ineffective one backs off as on the normal path.
* zRAM: keep the read-back backoff on the nearly-out-of-memory path
* zRAM: retry a skipped page-out after 30 s when nearly out of memory
* fix(steam): reconcile tool dependencies by supported platform
* fix(steam): refresh tool metadata and clear stale GE mappings
* fix(steam): clear any Proton mapping from Linux-only tools
* fix(steam): watch folders the client creates after the helper starts
* Watch each missing directory level as it appears
* Make the watcher level test wait on inotify readiness
* fix(gog): install scripts' support files reach the prefix
A goggame-<id>.script can copy files into the prefix with supportData
actions: default INI files, Plugins.txt and other settings that mods such
as one-click editions need. Only setRegistry actions were applied, so
those games started without their configuration and ran as the base game.
droiddeck-store-launch now applies supportData after the registry, once
per script revision: copies from {app} and {supportDir} into {userdocs}
and {localappdata}, matching paths without regard to case. Links are
copied as links, paths outside those folders are refused, and existing
files are replaced only when the action asks for it.
* Keep GOG support data copies inside the game and prefix folders
* Copy linked GOG support sources as links and keep a player's own links
The session's TU_DEBUG=sysmem is meant for the Steam client and
Chromium-based apps. Flatpak and Snap apps inherited it, so emulators
and other programs installed that way rendered through system memory.
droiddeck-game-env now holds the Chromium check next to game_gpu, and a
small droiddeck-game-gpu wrapper applies both to any program. The
AppImage, Flatpak and Snap launchers use it, so the rule lives in one
place.
The session sets TU_DEBUG=sysmem for the Steam client's interface and hands
games their own value in DROIDDECK_GAME_TU_DEBUG, which droiddeck-game-env
applies to Steam launches (#1188). droiddeck-appimage-run never did, so
emulators and other AppImage programs rendered every pass through system
memory instead of Turnip's per render pass choice.
Apply the game value in droiddeck-appimage-run the same way: an empty
value unsets TU_DEBUG, any other value replaces it. Chromium-based apps,
like the Steam client's own interface, keep the session's setting.
fix(proton): a Valve Proton picked for a game in Steam stays on the ARM64 build made from it
Steam lists Valve's own Proton Experimental and Proton 11.0 next to the
DroidDeck tools. They are not runnable here, so a per-game pick of one was
replaced with the default. That used to be Proton Experimental ARM64, so the
pick looked kept; since an installed DroidDeck-Proton became the default, the
game is moved to DroidDeck-Proton instead.
When Steam records one of those tools for a title the sync already knows
(a title it mapped or one with a recorded pick), the title now gets the
DroidDeck tool built from the same Valve depot, stops following the default
and the choice is mirrored into the app's per-game setting. The same applies
to Steam's default. Titles nobody picked anything for still follow the
default, and Valve tools with no ARM64 build still fall back to it.
Mesa 26.0.8, which the Ubuntu runtime builds, carries the backport of
"zink: create views for samplers lazily", which dropped the depth/stencil
aspect check in init_sampler_view(). On drivers with
needs_zs_shader_swizzle (Imagination) every colour view with a constant-1
alpha, such as RGBX/BGRX textures, then got the red-only depth swizzle
view, so Steam's interface and games showed red as R=G=B.
Carry Icenowy Zheng's upstream fix d55418dae7 (Mesa MR 41923), which
restores the aspect check. It applies on top of patch 0004.
When a Build APK run finishes, Build validation finds its pull request with
commits/{sha}/pulls, which GitHub answers with an empty list for a commit that
only exists on a fork. The validator then took the build as superseded, so the
"Build APK" status of every fork pull request stayed at "awaiting APK
compilation" after a successful build. It now falls back to the open pull
request for the run's head owner and branch, still requiring the head commit
to match.
* ci: cache the pinned Turnip archive and retry its download
"Prepare graphics driver variants" fetched the same checksum-pinned Banners-Turnip
zip from GitHub on every run with a single 120 s attempt, so a 502 or a read
timeout failed whole PR builds three times today. The archive is now cached
between runs (it is still checked against its SHA-256 after a restore), and a
cold download is retried on timeouts, 429 and 5xx with a growing pause.
* ci: keep PowerVR's verified downloads between runs
tools/powervr/fetch.py already keeps checksum-verified downloads in
~/.cache/droiddeck-build, but CI never saved that folder, so every run fetched
the PowerVR components again and a read timeout failed "Build the add-on
packs". The folder joins the existing pack-inputs cache, keyed on
release.json too.
* fix(session): give the runtime a machine ID so D-Bus programs start
The Ubuntu runtime ships an empty /etc/machine-id for systemd to fill on first
boot, and nothing fills it in a DroidDeck session. libdbus aborts when it needs
the ID ("D-Bus library appears to be incorrectly set up"), so melonDS closed at
once from the desktop and GTK Flatpaks started outside the session bus failed.
The session now writes a fresh ID once when the file is empty or malformed and
never changes a valid one.
* fix(session): check the whole machine ID file and reject the all-zero ID
fix(desktop): catalog AppImages start from the desktop menu through the shipped runtime
RPCS3 from the desktop menu exited at once: its own AppImage runtime runs
dwarfsextract from a memfd, which PRoot cannot execute (EACCES), and it has no
fallback. Emulator rail launches already avoid this by starting arm64 images
through the shipped uruntime (#1191); the menu entries the catalog writes did not.
The rule now lives in one launcher, droiddeck-appimage-exec, used by the session
and by the catalog's menu entries. Entries written by older versions are moved
to it with the existing menu migration.
* fix(proton): an installed DroidDeck-Proton replaces a Proton Experimental default nobody picked, and added shortcuts follow the default
* test: an owned shortcut left on the previous default follows a default change
* gamescope: a confine is relative only while it holds the X pointer
Patch 0128 made every hidden-cursor confine relative. Games that hide the
system cursor, draw their own and confine the pointer to their window then
took the host lock, and a direct touch drag became camera motion with no
button held, so cards and items could not be dragged.
A confine now goes relative once the X pointer stays away from the seat
position while input moves, as when a game recentres it or clips it to a
small rectangle, and returns to absolute when the pointer follows again or
the confine ends. Locks and the warp follow rule are unchanged.
* gamescope: a constraint without a hint is relative only while the game holds the X pointer
Xwayland turns a hidden-cursor grab into a lock with no cursor position hint, so the confine-only rule never applied to games that draw their own cursor (Stardew Valley, Balatro). A hinted constraint is relative at once, as upstream. An unhinted lock or confine goes relative once the X pointer stops following the seat's motion, which a recentring or clipping game does and a drawn-cursor game does not.
* gamescope: a pointer constraint is relative only while the game holds the X pointer
Stardew Valley's grab with a hidden cursor becomes an Xwayland lock with a cursor position hint, which upstream made relative at once. Without relative mode that lock still follows gamescope's relative motion and touches still place the pointer, so the hold rule now decides for every constraint, hinted or not.
* gamescope: a hinted lock starts relative and is released once the X pointer follows
Releasing a hinted lock only through the hold rule let a recentring game's first touch swipe press the button. A hinted lock now starts relative as upstream and is released after 30 moving frames on which the X pointer follows the motion gamescope sends, counted from the relative motion while the seat is locked.
* gamescope: count away frames within one half-second window
* gamescope: a hint that arrives with the game's first warp starts the lock relative
* fix(input): end the touch press when a program locks the pointer mid-swipe
A game that only shows its pointer lock after its first warp took the first
swipe as an absolute drag, with the button held until the finger lifted.
The first locked move now ends the forwarded touch before sending deltas.
* gamescope: a new focus window or constraint starts its own pointer hold
A held state from one constrained window carried over to the next one
when focus moved before a reset frame. Track the input focus window and
the constraint, and restart the hold when either changes.
* gamescope: read the pointer constraint under the wlserver lock
GetCursorConstraint asserts the lock is held, so the hold tracker
aborted as soon as a program constrained the pointer.
* fix(input): a mid-swipe pointer lock ends the touch but keeps a held button
The first locked move cancelled the whole direct-touch hold, releasing a
right button held by a long press. It now cancels only the forwarded touch;
the held button is released when the finger lifts, as before.
* fix(input): extra fingers during a locked swipe stay out of the cancelled touch
Once a locked swipe has ended the forwarded touch, a second finger restarted
it with a new touch-down while the swipe carried on as relative motion. Extra
fingers are now ignored until the swipe ends; the final lift still releases a
held button.
* fix(input): a swipe that ended its touch stays closed until the finger lifts
* perf: answer Steam's missing-file checks on FUSE libraries from the directory cache
Before creating and committing each file, the Steam client checks whether it
already exists under its exact, lower-cased and folded spellings. On an SD or
USB library reached through Android's FUSE layer every one of those misses is a
round trip to MediaProvider (Brawlhalla: about 65,000 misses, 92 of the 161 s
of preallocation and 60 of the 93 s of commit).
The client already caches the listings of its steamapps directories on FUSE and
keeps them exact through its own changes, inotify and a short TTL. access() now
answers a name that is absent from a live listing of its directory without
asking the filesystem. Directory paths match case-insensitively only when the
library's filesystem folds case, and names compare case-insensitively, so an
existing file is never reported missing.
* fix: reuse a listing for a differently cased folder only when the whole path folds case
Checking that steamapps alone folds case let a listing of one folder answer
for a different folder below it whose name differs only in case. The fill
now compares the listed folder with its fully case-swapped path below
steamapps, so a listing is reused across case only when the filesystem
folds every level.
* fix(dircache): match a folded listing only below the same prefix and skip it during a tracked change
* fix(linuxfs): a folder renamed or removed above a cached listing drops that listing
* fix(linuxfs): watch the folders above a cached listing so a rename or removal by any process drops it
* fix(linuxfs): also watch the folders above steamapps, and stop a stuck probe from hanging the tests
* fix(linuxfs): a listing under a symlink or an unwatchable folder no longer answers absent lookups
* fix(linuxfs): a symlink check that cannot be watched also stops absent answers
* fix(linuxfs): a differently cased lookup reuses a listing only for the same folder