* fix(audio): restart host after stalled wake recovery
* Serialize audio host restarts and make the stall counter race-free
* fix(audio): an overtaken restart leaves the socket alone and a sink guard only acts on its own host
---------
Co-authored-by: Kurt Himebauch <136133082+xXJSONDeruloXx@users.noreply.github.com>
* fix(input): read a USB mouse directly where Android won't capture it, as in REDMAGIC cast mode
REDMAGIC's cast mode shows the session on a virtual display while its own
mirror activity keeps focus on the phone's display. Android only grants
pointer capture on the focused display, so the session got a cursor that
stopped at the screen edge and the camera stopped turning in mouse-look games.
When capture is wanted but not granted within 500 ms, the session claims the
mouse's USB HID boot interface instead, reads its reports with the layout from
its report descriptor (boot protocol if that can't be read), and sends motion,
five buttons and the wheel the way captured motion goes. The mouse goes back
to Android whenever capture is no longer wanted.
* feat(input): hold the middle button still to open the session menu while the mouse is read over USB
While the session reads the mouse over USB, Android has no cursor, so the
mouse can't reach REDMAGIC's cast-mode panel or anything else outside the
session. Holding the middle button for two seconds without moving opens the
session menu, which hands the mouse back to Android until the menu closes.
Quick middle clicks and middle-button drags still reach the game.
Relative motion sent to a locked pointer now hides the session's own cursor,
as pad pointer motion already does. A desktop keeps its grab while the menu
is open, so a click made with Android's cursor on the way back showed the
session cursor, which then stayed where the click was.
* fix(input): give a USB mouse back to Android when the app starts after being killed while reading it
A process killed while it reads the mouse, by a crash or an update, leaves
the mouse's interface with no driver, so Android has no mouse until it is
replugged. At startup, any permitted USB boot mouse that can be claimed
without force has no driver, and releasing it reattaches Android's.
* fix(input): only sessions shown off the phone's screen fall back to reading the USB mouse
Android grants pointer capture on the focused display, which the phone's own
screen is, so a session there never needs the USB mouse. Checking the display
keeps a slow capture grant from ever asking for the mouse's USB access there.
* fix(input): reject out-of-range HID report sizes and counts, and release USB buttons when the mouse is lost
A report descriptor can give any 32-bit report count or size, which drove the
field loop for billions of iterations and could overflow bit offsets into a
negative index on the reader thread. Items outside 32 bits and 4096 fields are
rejected, so such a mouse falls back to the boot protocol.
A mouse that stops responding while it is read over USB now releases the
buttons it held and cancels a pending menu hold, through a separate onLost
callback, since the session's own release path only runs while the grab is
still active.
* test(input): an oversized report count is rejected even when X and Y follow it
The earlier case had no X or Y fields, so it returned null without the count
check too. A 4097-field item before valid X and Y is now rejected, and a
4096-field one still parses.
* fix(input): Auto relative mouse sees a mouse read over USB, and re-checks when it is lost
The Auto relative-mouse decision from #2305 reads mouseAttached() on the
relative-mouse thread, which counts a mouse read over USB, so the grab's
state is now volatile. Losing that mouse re-runs the decision, as removing
an Android mouse does.
* fix(input): set up the USB mouse on its reader thread, select report protocol, and retry a failed permission request
Opening the receiver and its descriptor and protocol transfers now run on the
reader thread, so an unresponsive receiver can't stall the session for up to
two seconds. A release during setup lets the thread clean up as it already
did. A mouse that fails setup isn't tried again until it is replugged, since
releasing it re-adds Android's mouse, which would start another attempt.
A parsed report descriptor now selects report protocol first: HID 1.11
7.2.6 says a host must not assume the protocol a device is in.
A permission request that throws no longer marks the mouse as asked.
* fix(input): a cancelled or overlapping USB reader can't claim the mouse or mark it failed
A new reader waits for the previous one to let go of the interface, so a
quick release and grab, like toggling the session menu, no longer fails the
claim. A reader released before or during setup doesn't claim the interface
or keeps going. Failing to open or claim leaves the mouse with Android without
marking it failed, since nothing re-adds it to start another attempt; only a
failure after claiming, or a lost mouse, marks it, and only for the current
reader.
If SET_PROTOCOL(report) fails, setup tries boot protocol before giving up,
so the decoder always matches the protocol the mouse was set to.
* fix(input): release can't slip between the USB reader's cancellation check and its claim
The reader checked `running` and then claimed the interface, so a release in
between still let it claim and start setup. The check and claim now share a
lock with release, so the reader either never claims or claims first and sees
the release before setup.
* fix(input): hand the mouse back to Android while the WoW chat bar is on top
appUiOwnsPointer() already shows Android's pointer for the on-top chat bar,
but pointer capture, and now the USB mouse, stayed on, so that pointer
couldn't reach the bar. The bar on top now ends capture and USB reading like
the session menus do, and its changes re-run the capture decision.
* fix(game-env): keep DroidDeck's DLL overrides when a game sets its own
* fix(game-env): keep the game row when the merged DLL overrides pass 8192
---------
Co-authored-by: Kurt Himebauch <136133082+xXJSONDeruloXx@users.noreply.github.com>
* fix(dual-screen): the bottom-screen terminal keeps its shell when another app covers that screen
* fix(session): sweep the kept terminal shell's process tree when the session ends
* fix(graphics): fall back from unsupported auto display drivers
* refactor(graphics): drop the unused Auto recommendation lookup
* fix(graphics): repair automatic display fallback build
* test(wayland): keep Vulkan caller harness linkable
---------
Co-authored-by: Kurt Himebauch <136133082+xXJSONDeruloXx@users.noreply.github.com>
* feat(performance): add idle zram compression toggle
* Gate the idle compression toggle on memory cap support and refresh counters first
* fix(zram): retain launcher reclaim when idle compression is off
* test(memory): verify launcher-only reclaim signals
* fix(performance): keep game reclaim when focus is unknown
* fix(performance): say when to turn idle compression off and that it applies on the next session
---------
Co-authored-by: Kurt Himebauch <136133082+xXJSONDeruloXx@users.noreply.github.com>
* fix(fex): the x86 runtime is built from files inside it so FEX finds its libraries
* fix(fex): the x86 runtime keeps the permissions of the directories it is built from
* fix(fex): rebuilding the x86 runtime refreshes the Steam client links and removes read-only trees
* fix(fex): rebuilding the x86 runtime removes old trees that contain unreadable directories
* fix(fex): isolate generated rootfs files
* fix: refresh the session loading tips and drop stale translations
* fix: add tips from common support questions
* fix(session): restore translated loading tips and fix the string that broke resource merging
* fix(tips): list every store the Stores tab signs in to
* fix(shortcuts): move the accessibility service into an optional companion so Play Protect stops blocking the install
* fix(shortcuts): bind only same-signer relays, recheck enablement per call, drop timed-out keys
* fix: Remote Play gets the Steam Deck controller
Steam's Remote Play client reads the controller itself from its own
streaming_client process. The input preload only exposed the virtual
Deck controller (and the DualShock 4 / DualSense pads that share its
code) to the process named steam, so in Steam Deck controller mode
streaming_client was refused /dev/hidraw16 and the stream got no input:
streaming_clien: deck: /dev/hidraw16 refused: not the Steam client
streaming_client now sees the same Deck, DualShock 4 and DualSense pads
as Steam. Each opener gets its own report stream from the read-only
ring, so the two processes do not compete for input. The Steam
Controller paths stay Steam-only: the paired controller's feature and
output slot is serialised only within one process.
Fixes#1860
* fix: Remote Play also gets the Steam Controller and touch controls
streaming_client now opens the Steam Controller pads (the app's
emulated one and a paired controller) and Steam's touch controller,
like the Steam Deck pad.
A paired controller's feature request and output slots, and the touch
controller's state, are shared with the app and written under a
seqlock that expects one writer. Their writers now also take a flock
on a lock file in the hook directory, so Steam and streaming_client
take turns. touch_state() was a template holding its own static mutex,
so each caller had a separate one; it now shares a single mutex.
The pad type and touch plug watchers run in streaming_client too, so
it gets the udev events for mode switches and replugs; those only reach
the process that injects them. Poking /dev watchers and moving event0
on the shared filesystem stay with Steam, so they happen once.
* fix: the shared slot lock fails safe and never waits forever
If the lock file cannot be opened or locked, ProcessSlotLock used to
carry on unlocked. Steam alone is still a single writer then, exactly
as before streaming_client could open these devices, so it carries on;
streaming_client now leaves the paired controller's feature and output
slots and the touch state alone instead (feature reads fail with EIO).
The first failure is logged.
The wait for the lock is now bounded: a session freeze can stop the
process holding it, and writers wait with controller_mutex held, so an
unbounded flock could stall all input in the other process. Feature
reads wait up to their 500 ms reply timeout, outputs 100 ms and touch
state 20 ms. When the deadline passes the lock is held by the other
process, so neither proceeds; the operation is skipped and logged.
* fix: Steam no longer turns motion off under its Remote Play client
Steam and streaming_client both set the Deck pad's IMU mode, and both
wrote the one word Android reads to decide whether to feed the gyro.
Steam keeps toggling its own mode during a stream, so it could switch
motion off while the Remote Play layout needed it, and the client never
re-sent an unchanged mode.
Each process now has its own word. Android reads motion while either
wants it, and ignores the Remote Play word once that process is gone.
* fix: Remote Play decides motion while Steam has set no mode
Until Steam published its IMU mode, Android reported the mode as unknown
without reading the Remote Play client's word, so a Remote Play request
to stop motion had no effect. The word that has been set now decides,
and the mode stays unknown only while neither process has set one.
The rule moves into ImuModes so it can be tested on the JVM.
---------
Co-authored-by: Kurt Himebauch <136133082+xXJSONDeruloXx@users.noreply.github.com>
* feat(compositor): pen samples drive the pointer like a mouse
Input record 13 carries the pen's state (in proximity, tip, right and
middle button) and its position in input space as 24.8 fixed point.
Hover moves the pointer, the tip holds the left button, and a record
without the in-proximity bit releases every button. Under a pointer
lock only the buttons pass. A button pressed or released mid-stroke
keeps the implicit grab on the window the stroke started in, unless
that window closed.
* feat(input): a pen moves the pointer while hovering, clicks with its tip and right-clicks with its side button
In Mouse mode (the default) a pen's hover moves the pointer, the tip
holds the left button and the side button right-clicks; Touch keeps it a
finger. The side button can send a middle click or nothing instead, and
a Bluetooth pen key counts as the side button while the pen is near.
With glasses or a program's pointer lock the pen drives the touchpad.
A hover exit followed within 50 ms by more hover or a touch keeps the
pen near, so a pen at the edge of its hover range does not flicker out,
and a hover that sends nothing for 2 s leaves proximity. On the desktop
a pen that comes near releases KWin's grab of a hardware mouse, so the
pen moves KWin's pointer. Whatever cancels touches also ends the pen,
and stylus devices no longer count as hardware mice.
* feat(input): pen mode and side button settings in the Controls tab and the session drawer
A Pen group follows the touch settings once the phone has a stylus
device or a pen event was seen, which is remembered as penSeen and kept
by a settings reset. It holds Pen (Mouse, Touch) and Side button (Right
click, Middle click, Off); Side button is hidden in Touch. Before
Android 14 a touchscreen reports the stylus source without a pen, so
there it counts only with its first pen event. The strings are
translated into all 32 languages.
* docs: describe pen input and its limits
---------
Co-authored-by: alien2003 <11610652+alien2003@users.noreply.github.com>
droiddeckctl input pen hover, tap, stroke and button register a virtual
direct pen with `adb shell uinput -` and play the gesture through it, so
Android's input reader makes the same stylus events a built-in pen does,
with pressure, tilt, distance, the eraser and both side buttons. The app
is not involved, which lets pen handling be tested without a pen.
Positions are fractions of the screen as shown, placed inside the
physical frame of the internal display viewport in dumpsys input: the
frame is smaller than the panel when `wm size` sets another aspect
ratio, and Android drops pen events outside it. The rotation and size
come from the same viewport. The values are checked before anything
runs on the device.
The script waits 1 s after registering, since uinput reads it in one go
and cannot wait for the device, and does not reset uinput's time base
after that, which would drop the first samples. Contact sends a pressure
of at least 1, because Android takes a pen at pressure 0 for a hovering
one, and a lift first moves to where it lifts, because Android's UP
repeats the last contact. A pen held still moves by one unit without
leaving the physical frame so that it keeps reporting, since the kernel
drops frames that change nothing.
Co-authored-by: alien2003 <11610652+alien2003@users.noreply.github.com>
* Head tracking for the XREAL One, One Pro, One Pro+ and Aura
These glasses use the same USB Ethernet link and 128-byte IMU report as the
1S. XrealModel now has one model per product: the 1S keeps its tested
display-mode control, and the others switch 2D and 3D only with GET_INPUT
(0x2821) and SET_INPUT (0x2822). Display-mode set (0x275f), brightness and
the dimmer are never sent to them. The Aura's IMU reads x right, y up, z back
(about +9.8 m/s2 on y at rest), so its axes are not flipped like the rest of
the family. Recovery-mode product ids (one below each) stay excluded.
Signed-off-by: Mike Haye <84657517+mhaye20@users.noreply.github.com>
* Describe the wider XREAL support in the glasses panel and docs
The panel's help text and docs/smart-glasses-support.md named only the 1S.
They now list the One, One Pro, One Pro+ and Aura, say what was tested on
the Aura (head tracking and 2D/3D switching), and note that these models get
2D and SBS only, without brightness, dimming or display-mode set commands.
Translations still describe the 1S only.
Signed-off-by: Mike Haye <84657517+mhaye20@users.noreply.github.com>
* Docs: XREAL support is no longer 1S only
The paragraph under the support matrix still said the One and One Pro were excluded.
Signed-off-by: Mike Haye <84657517+mhaye20@users.noreply.github.com>
---------
Signed-off-by: Mike Haye <84657517+mhaye20@users.noreply.github.com>
* feat(theme): AMOLED black, a theme with a pure black page and panels
* feat(theme): rename to True black and let the user pick its highlight color
* fix(theme): True black leaves the page unlit, with no art wash or corner glow
* feat(xbox): stage the PlayFab Game Saves save-folder proxy from the Nexus runtime
Xbox titles that use PlayFab Game Saves, such as Forza Horizon 6, stop with
"Initialization Failed" E:0-17 right after sign-in. The SDK's default provider
needs Gaming Runtime save APIs that xgameruntime does not have, and its
in-process provider needs a save folder the title never passes on Windows.
xodus-nexus already builds a proxy for this into the runtime archive we pin
(scripts/runtime-compat, MIT): it keeps the title's SDK as
nexus_pgs_native.dll, sets ForceUseInprocGameSaves and passes the
PersistentLocalStorage PlayFabGameSaves folder. runtime.py now takes it and
its license from the archive, checked against runtime.lock.json, and
droiddeck-xbox stages it before each launch when the game's
PlayFabGameSave.dll has exactly the SDK's 40 exports. A game update that
brings a new SDK replaces the kept copy; anything unknown is left alone.
Unlike Nexus, no DLL override is set for the two DLLs: Wine has no builtin of
either, and forcing them native-only made Forza Horizon 6 fail with E:0-0-4.
* fix(xbox): check the whole PlayFab export table and never overwrite an unknown kept SDK
The SDK and proxy checks now also require the export address table to hold
exactly 40 functions, so a DLL with the same names plus an ordinal-only export
is no longer taken for the SDK. Before copying a new SDK to
nexus_pgs_native.dll, an existing file there must itself be a regular file
holding the SDK; an unknown file or a link is left as it is.
* docs(phantom): how Android counts guest processes and why detaching cannot escape it
* docs(phantom): clarify probe permissions and batch capture
* docs(phantom): qualify probe precondition by SDK and size the limit device-wide
* docs(phantom): keep phantom discovery on for the 12L probe and raise the limit instead
* docs(phantom): warn about the CPU kill, raise the limit before the flag and restore both after the probe
* docs(phantom): restore the monitor flag before the phantom limit after the probe
* docs(phantom): restore the flag before the limit whether each is put or deleted
* fix(frontend): game file export no longer fails before the game library folder exists
* fix(frontend): previous automatic library check accepts a library folder not created yet
* fix(drivers): never load a KGSL-only display driver on a GPU without KGSL, and report the drivers actually in use
* fix(drivers): gate KGSL-only drivers on the KGSL kernel node, not Adreno detection
* fix(ui): show dialogs on the second-screen menu without a separate platform window
* fix(ui): track presentation alert window for controller events
* test(ui): assert presentation and platform dialog window semantics
* test(ui): target the presentation popup for outside taps
* test(ui): keep presentation popup tap tests off the auto-advancing idle loop
The outside-tap dialog tests hang in the shared unit-test JVM when the
compose clock auto-advances; step frames explicitly instead.
zink_rebind_all_images() rebuilds a sampler view once per backing-object
change, through the rebind_count guard. A second slot bound to the same
view then skipped its descriptor refresh and kept the old VkImageView.
The Mesa patch now refreshes every slot whose cached view no longer
matches the view's current surface.
Follow-up to #2107 (CodeRabbit review). Tested on a Pixel 10 Pro Fold
(PowerVR): debug Mesa 26.0.8 with the patch, Steam web helper threaded,
8 fast Home-shelf scroll cycles, no web helper restart, thumbnails intact.
Opening a folder by touch and then pressing an arrow left the highlight on the page behind: focus starts outside the page, so the folder's first request was turned to the page's own entry tile. The folder now keeps asking until focus is inside it.
* fix(apps): closing a folder while renaming returns focus to the folder tile
* fix(apps): keep the folder tile focused while a folder closed from its name field fades out
* fix(apps): a fresh press ends the focus hold after a folder closes
---------
Co-authored-by: James O'Hara <james@jamess-Mac-Mini-317.local>
nativeSendSceneInput took bare integers (3, 4, 6, 11, 12) and the
wheel step (120) and pointer fixed-point scale (256) were repeated
inline. BottomDesktopPanel kept its own private copies of three of
them. Move them to WaylandCompositor next to the native method and
use them at every call site. No behavior change.
Co-authored-by: Dennis Zabolotny <15849186+zdennisz@users.noreply.github.com>
The 8BitDo Ultimate Mobile registers a mouse interface next to its gamepad.
On REDMAGIC the back gesture has no virtual-hard-key flag, so with that
mouse attached it was sent to the game as a right-click. A mouse that
shares its USB id with an attached controller no longer counts.
screencap without -d prints a warning before the PNG when it finds more
than one display, so droiddeckctl screenshot failed on foldables and
dual-screen phones. Without --display, screenshot and record now use the
physical display behind logical display 0 when there is more than one.
Co-authored-by: alien2003 <11610652+alien2003@users.noreply.github.com>
Game thread hints only ran on the Odin 3 (CQ8725S). Phones with the SM8750 have the same core layout and power HAL, so turn them on there too.
Co-authored-by: Tapuuk <177422350+Tapuuk@users.noreply.github.com>
Back replays the history stack, so a round trip between drives leaves one
entry per hop: internal storage, Linux, internal storage, then Back lands
on Linux again and leaving takes one press per hop. Two roots in a row
means the user hopped rather than descended into a folder, so Back leaves
instead. Returning to a folder from a drive root still works as before.
Co-authored-by: Dantalian <Dantalian-dev@users.noreply.github.com>
* fix(game-env): keep x64 system DLLs that only x64 code loads in ARM64 prefixes
The foreign-architecture repair treated every plain x64 DLL in system32 of
an ARM64 prefix as broken and swapped in the ARM64 builtin. ARM64EC games
load x64 DLLs fine, so this broke native-only DLLs such as the Khronos
vulkan-1.dll loader (vulkan-1=n leaves Wine nothing to load) and native
runtimes that games depend on.
An x64 DLL is now foreign only when Proton's own ARM64 programs import it,
found from the import tables of the Proton build's ARM64 executables.
Prefixes the old rule already repaired get their native copy back. This
replaces the Visual C++ runtime name list.
* fix(game-env): keep the old rule when Proton's programs cannot be read, restore the newest backup across gaps
* fix(game-env): number new backups above the highest one, treat truncated import tables as unreadable
* fix(game-env): treat an import directory outside every section as unreadable
* fix(game-env): treat truncated data directories and section tables as unreadable
* perf(game-env): index foreign DLL backups per directory
* fix(game-env): count only ARM64 Proton programs as roots, cover x64 component DLLs across launches
---------
Co-authored-by: xXJsonDeruloXx <danielhimebauch@gmail.com>
* fix(stores): log why a GOG cloud sync failed and stage uploads in the cache folder
* fix(stores): strip URL credentials and bound the error body in GOG cloud failure details
* fix(stores): send Galaxy's communication-service User-Agent to GOG cloud storage
GOG cloud storage now answers every list with HTTP 400 invalid_version for the bare
"GOG Galaxy" value, so uploads and downloads failed for every game. List, HEAD, GET and
PUT now identify as GOGGalaxyCommunicationService/2.0.13.27 like gogdl and Lutris, and
PUT carries the same value as X-Object-Meta-User-Agent.
* fix(stores): bound cloud list errors and verify test requests
* fix(stores): preserve cloud error classification
* fix(game-env): turn off Proton's tabtip unless the Steam keyboard is enabled
Proton's tabtip only opens Steam's keyboard with SteamDeck=1, but it always
runs UI Automation queries inside the focused game, and those calls crash
games that answer them (Godot 4.5 AccessKit). Game launches now add
tabtip.exe=d unless SteamDeck=1 or the user's overrides name tabtip.
* fix(game-env): match WINEDLLOVERRIDES modules split by spaces and tabs
* fix(game-env): count only tabtip.exe entries with a load order as the user's own setting
* fix(game-env): count *tabtip.exe and tabtip.exe.dll overrides as the user's own setting
* test(game-env): cover keyboard overrides during sync fallback
* fix(input): stop controller rumble when the game that started it is gone
A crashed game never sends the stop for a held rumble, so the keepalive re-armed it every 1.5 s until the session ended. A held effect now stops on the next keepalive once the process that sent it or the Steam game it started in has exited. The fake controller also stops its rumble when it's closed or the playing effect is erased, like the kernel does.
* test: read the test's own pid from /proc/self
* fix(input): track who plays rumble on each slot, check the game when a hold starts
The process that plays an effect now owns it, and only the handle currently playing on a slot sends the stop when it closes or erases its effect. A held effect records whether a game is running when it starts, and keeps checking that its sender still exists even when its start time couldn't be read.
* fix(input): only cache a running game, not its absence
A new hold claimed right after a no-game result reused it, so a game that started within the cache window was never watched.
* fix(input): treat a vanished sender of our own user as gone
Guest processes run as the app's user, so the app can always see them in /proc. If one is missing at the first lookup it already exited; only senders of another user (root-spawned guests) are left unchecked.
* fix(input): a stop only ends rumble the same process started
The fake controller's playing-handle tracking is per process, so another process closing or erasing its older effect still sent a slot-wide stop. The app now ignores a stop whose sender isn't the one behind the slot's current effect; stops from an unknown sender work as before.
---------
Co-authored-by: Tapuuk <177422350+Tapuuk@users.noreply.github.com>
* fix(ci): a test build's files outlive the catalog that names them, and a merged PR's build lasts until Preview carries it
- A new push's PR release keeps the previous push's files until the catalog names the new
ones, then five minutes more, so a download started from the old catalog still finds its APK.
The app also reloads the catalog and retries once when a test build's download fails.
- Closing a PR without merging still deletes its test build. A merged PR's build stays, and
stays listed, until the newest Preview build was built from a main holding the merge, or
three days pass (docs-only merges build no Preview). Main's build, the PR cleanup and the
sweep share one script for this.
* fix(ci): name the current push's APK in the catalog and prune nothing it still names
* fix(ci): only a Preview with every APK carries a merge or is offered
* fix(ci): publish no preview while none has every APK
* test(ci): build_catalog publishes no preview while its APKs are uploading
* fix(venus): DXVK runs on Mali GPUs without dual-source blending
DXVK requires dualSrcBlend for D3D11 and skips any GPU without it, so on
Mali through Venus no adapter was found. The Mali compatibility layer now
reports the feature and maxFragmentDualSrcAttachments 1, keeps it out of
the driver's device, turns fragment outputs decorated Index 1 into Private
variables (shader modules and inline stage code), and maps SRC1 blend
factors to source alpha in pipelines and vkCmdSetColorBlendEquationEXT.
Only active when the driver lacks the feature.
Adds a mock loader and driver test for the layer, and fetches the
layer's pinned headers for it in CI.
* build(venus): pin the Mali component with dual-source blending
* fix(agent): one-state composer with model and usage on top; reading back stays detached while the agent streams
- The composer no longer has a compact reading bar that a first tap expands: the editor is
always the editor, and the model picker and context ring move to the top beside the menu.
- The pet chat drops its DeeDee title so that top bar gets the whole width.
- Scrolling back stops following new output; new entries and streamed text only scroll the
list while it is at the bottom, and sending a message resumes following.
- When the editor slides back in or the keyboard rises, a list at the bottom keeps its last
lines above the editor, even mid-drag.
* fix(agent): a revise while reading back shows the editor it focuses
* feat(agent): dictate a message, and attach what the running game shows
- With nothing typed, the editor's send button dictates instead, through Android's speech
recognizer; what is heard follows the text already there. It asks for the microphone the
first time.
- While a game session is open, a camera button attaches the game's current frame. It is
copied from the game's own surface, as the bottom screen's screenshots are, so the chat or
drawer over it is not in the picture.
* feat(agent): a notification when a turn ends unseen, and replies that end in choices offer them as buttons
- A turn that ends while neither the Agent page nor the pet is in view posts a notification:
finished, failed or asking, with the reply's first line. Opening the Agent page clears it.
- Claude Code's question tool is not available in print mode, so the brief asks the agent to
end a reply with a choices block instead. The chat shows its options as buttons, and the one
picked is sent as the next message.
* feat(agent): a subagent's work shows under the call that started it
Claude Code tags a subagent's messages with the call that started it. Its text, thinking and
streamed deltas used to show as the main agent's reply, and its tool calls mixed into the
turn. Now only its tool calls are kept, stored with their parent, and shown inside the
parent's card: a live line with the step count and the latest call, and every call when
the card is opened. Its token counts no longer overwrite the context reading.
* feat(agent): copy a reply or a tool's text, and open a turn's changes in one place
- A turn's footer has a copy button for everything the agent said in it, and tool input and
output boxes have one each.
- The files-changed chip opens every edit the turn made, file by file.
* feat(agent): search conversations in History
A field above the list keeps the conversations whose title, messages or replies hold the
text, ignoring case. Chat files are read off the main thread, a quarter second after the
last keystroke.
* fix(agent): sending while reading back jumps to the latest
* feat(agent): manage Claude Code's MCP servers and skills from the Agent menu
A new panel lists the MCP servers claude mcp list reports, adds one for every project from a
URL (over HTTP) or a start command (over stdio), and removes it; connectors from the
claude.ai account are listed but managed there. It also lists the skills in
~/.claude/skills with their descriptions, and deletes one after a confirming second tap.
* fix(agent): review follow-ups for the Agent tab refinements
- Choices only come from one fenced block that ends the reply.
- The Agent page counts as on screen only while its activity is started, so a turn ending
during a game still notifies.
- Picking a choice or Go jumps to the latest like sending does.
- Dictation stops when the chat is hidden or the keyboard sends; stopping it by hand keeps
the recognizer until its final result.
- A failed JPEG encode is not attached.
- The copied reply is read from the turn each time.
- MCP start commands keep their quoted arguments; the CLI is read beside its 90 s deadline;
deleting a skill runs off the main thread and removes links without following them.
* fix(agent): scroll the shrinking list after layout, not during measure
* fix(agent): fence-aware choices parsing; dictation stops overwriting edited drafts
* fix(agent): ignore choices blocks nested in a code fence
* fix(agent): treat tilde fences like backtick fences when finding choices
* test(agent): cover a tilde-fenced choices block
* fix(epic): install the Epic Games Launcher from its Linux .deb
Epic no longer serves the Linux AppImage: the installer URL answers 404
errors.com.epicgames.launcher.app_direct_url_not_found (30003), so Install
failed for everyone. Epic now ships EpicGamesLauncher-x86_64.deb
(0.113.0+g2164614).
- The installer downloads the .deb, checks its ar magic and the CDN ETag,
and droiddeck-epic unpacks its data.tar.* member straight from the
archive.
- The .deb links WebKitGTK 4.1, JavaScriptCoreGTK 4.1 and GTK 3, which the
Steam Runtime 4 image does not ship. 32 Debian 13 amd64 packages, pinned
by SHA-256 and size (60.2 MB download, 201 MB unpacked), are unpacked
into the launcher's desktop/ folder (shared libraries only) and put on
its LD_LIBRARY_PATH. Debian 13 matches the runtime's own base, so only
sonames the runtime lacks are added. Each package falls back to
snapshot.debian.org when the deb.debian.org pool has moved on.
- WebKit starts its helper processes from its built-in
/usr/lib/x86_64-linux-gnu/webkit2gtk-4.1 path, so each launch links that
folder in the runtime to the layer's copy.
- An installed AppImage launcher keeps starting through its AppRun; the
next update check moves it to the .deb.
Closure checked offline: every ELF in the .deb and the layer resolves with
the runtime's own ld-linux-x86-64.so.2 inside a sandbox holding only the
runtime, the layer and libGLESv2 (0 missing, 0 version or undefined
symbol errors with LD_BIND_NOW).
Based on zooluis's research in #2268.
Co-authored-by: zooluis <150430329+zooluis@users.noreply.github.com>
* fix(epic): provide desktop GLib data paths
* fix(epic): keep archive symlinks from redirecting the desktop layer and icon
---------
Co-authored-by: zooluis <150430329+zooluis@users.noreply.github.com>
* feat(updates): install any kept Preview build, link builds to their PR and commit, flag stale test builds
* fix(updates): address review of the Updates tab history and test-build watch
* fix(updates): notify on switch to a followed PR, keep higher-version rebuilds installable
* fix(ci): let full reruns replace every artifact, note only expired PR builds
* fix(ci): leave the rebuild note only when the PR build is too old to rerun
* fix(ci): poll unfinished PR builds together and queue version-bump runs
* fix(ci): also rebuild PR builds that were running when the versionCode changed
* fix(ci): rebuild fork and just-finished PR builds, recheck the PR head before a rerun
* fix(ci): fail the rebuild sweep when finding candidates fails
* fix(ci): time the bump from the rebuild run's creation and scan unfinished builds of any age
* fix(ci): check the latest build of every open PR instead of scanning run history
* fix(ci): count PR builds that finish between the version push and the rebuild run
* fix(ci): use the push time from the event as the rebuild boundary