* 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
* 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
* feat: every HUD reading has its own switch and a row order
* feat: readings get a reorder dialog of their own
* refactor: drag alone moves a reading
* fix: the readings list is one column with a switch per row
* fix: one scrolling readings list, titled HUD info
* refactor: derive the readings order from the defaults instead of a second list
* fix: the readings dialog loses its doubled title and follows the saved order
* refactor: no title in the readings dialog, its actions sit beside the list
* refactor: readings dialog actions take the free side on landscape, the foot on portrait
* fix: the readings list follows the size it draws and drags with a live gap
* fix: the readings dialog fills the window and keeps scrolling under a held drag
* fix: reset restores every size's order, and the new HUD strings are translated
* refactor: one table for each size's rows and one for its text sizes
* refactor: the readings list swaps rows as the drag passes them, and the list animates the rest
* refactor: give the readings rows test tags for the local UI test
* refactor: HUD readings follow one configurable row table
* refactor: the Strip HUD draws its rows from the same table
* refactor: each size orders its rows by its own table, and the frame graph is a row
* fix: the readings dialog keeps the row order a drag leaves
* fix: each size keeps its own row order, and reset clears only the order
* refactor: the graph's size lives in one place, and the section markers are gone
* fix: a size change and the first save of a legacy config both keep the order
* chore: retrigger the build
* feat: a reading can be grabbed with Y and shuffled with the d-pad
* feat: controller hints float beside the dialog, and a moved row stays in view
* feat: Y grabs on a hold and drops on release, and a touch no longer flips the dialog
* fix: keep the key hints off the rows on narrow windows, and let Enter toggle a row
* fix(battlenet): a Battle.net download that stops moving ends as retryable, and a killed update comes back as an update
A job reply without a state or byte counts no longer counts as a started
download, and an Agent job that reports no change for ten minutes is
stopped with its Agent and fails with a Retry instead of waiting forever.
Updates remember their game folder in the saved queue, so one restored
after the app was closed runs agent-update on that folder instead of a
fresh install.
* fix(battlenet): a stalled job stops only the Agent it started
A game or Battle.net client running in the prefix keeps running when a
stalled Agent job is ended, and the progress test keeps the job alive
across polls with a controlled clock.
* fix(battlenet): only real progress holds off the stall timeout
The download rate and time estimate can change while no bytes move, so
the stall timer now resets only on byte counts, progress, state, phase,
pause or error.
* test(battlenet): bound the stalled-rate fake child's polls
* feat(storage): name, select and remove every prefix and store game
* feat(storage): bulk remove picks prefixes, shader caches or both
* fix(storage): count and clear a game's shader cache in every library
* fix(storage): guard shared prefixes, batches and active downloads
* fix(storage): count each shader cache once across games
* fix(storage): hold store uninstalls against downloads and keep every shortcut's options
* fix(stores): keep download cleanup and uninstall from overlapping
* fix(stores): wait for an installer's file discard before uninstalling
* fix(stores): a reinstall waits for the cancelled download's file cleanup
* fix(storage): fast first scan, readable 4:3 layout and every shader cache counted
* fix(storage): keep store sizes current and track discards from failing jobs
* fix(stores): uninstall waits for a cancelled update's cleanup
* test(stores): a reinstall waits for a failed job's discard
* feat(agent): DeeDee, a pet that floats over any app and opens the agent's chat
* feat(agent): DeeDee docks and opens its chat like a chat head; throws pick their side by velocity, for the pet and the HUD
* fix(hud): the strip tucks into its fps tab with a shorter pull, and a throw toward an edge counts
* fix(hud): the HUD shows in desktop sessions again
* feat(agent): DeeDee rests tucked past the screen edge like a chat head
* fix(agent): DeeDee opens its own chat in every window, themed, and a fling across no longer tucks the HUD away
* fix(agent): the chat grows from wherever DeeDee is as it hops to its corner
* fix(agent): DeeDee hides with the PC keyboard, falls back if it cannot float, can be opened by accessibility, and settles when a drag is cancelled
* fix(agent): B and Back close DeeDee's in-window chat before the window acts on them
* fix(agent): Back closes only the chat open in that window
* feat(apps): add an Appearance panel to the Apps page
A Palette button in the Apps header opens a panel for icon size, spacing and
hiding app names.
* feat(apps): support icon packs and per-app icons
The Appearance panel gains an Icon pack choice. Packs are found through the
ADW and Nova theme intents and must ship an appfilter.xml in res/xml, res/raw
or assets. Work profile apps keep their badge.
Apps a pack does not cover are drawn on its iconback, iconmask, iconupon and
scale tags when it has them, and keep their system icon otherwise. For packs
with an iconback, App color background paints that plate in each app's main
color instead of the pack's own.
With a pack selected, an app's long-press menu offers Change icon, which
opens a searchable picker listing icons that match the app first. Reset icon
removes one choice, and Reset in the panel removes them all.
Calendar apps use the pack's icon for today's date.
* feat(apps): translate the Appearance panel and icon strings
Adds the 19 new strings to all 32 languages, reusing each language's existing
wording where it has one. Icon size and spacing labels are kept short to fit
the panel's tabs.
* fix(apps): finish loading custom icons before marking them loaded
Two app list refreshes can start together on resume. The second one could
see the overrides as loaded before the first had read them, and build the
list without custom icons.
* feat(home): choose the page the Home button opens
As the Home app, DroidDeck always came back to whichever page was last
open, and a cold start landed on Steam. A new choice under Setup >
Launcher, shown while "Use as Home screen" is on, lets Home open the
Android apps page instead. The default stays Steam, which keeps the
current behaviour.
* fix(home): a Home request for the apps page closes a page open over the rail
* feat(stores): EA and Ubisoft stores browse their catalogs with prices, deals and search
* feat(stores): Get for free adds EA's free games to the account in the app
* fix(stores): an owned EA game matches its store card by its full product name
* feat(agent): an Agent tab in the launcher that talks to Claude Code outside any session
* feat(agent): the agent in the session drawer and as a bubble over the game
* fix(agent): connect loopback adb, show unread replies, and fit the drawer page to the screen
* feat(agent): first-run setup in the Agent tab, a harness choice with Codex, and no permission prompts
* fix(agent): install helpers off the main thread and offer sign-in only after the install
* fix(agent): connect adb only for commands that use it and run setup checks one at a time
* fix(agent): show why a setup check failed
* fix(agent): each harness has its own setup state and conversation
* fix(agent): start the adb server before connecting and retry the connection check
* fix(agent): switching harness clears the previous harness's setup results
* feat(agent): a header, a composer with send and stop, and Markdown replies in the agent chat
* fix(agent): leave room under the chat composer
* feat(agent): continue the conversation after a Lab build restarts the app
* feat(agent): show and remove what the agent stores in the Linux userland
* fix(agent): the build tools take about 5 GB
* feat(agent): DroidDeck opens on the Agent tab after restarting on a Lab build
* feat(agent): conversation history, model and plan mode, live replies, full tool input and output, processes, attachments, queued messages and a Stop that ends the turn
* fix(agent): check the Lab build less often while nothing is happening
* fix(agent): an old Lab build's progress is not shown as a failure, and a restart counts as success
* feat(agent): rebuild a kept Lab build from Updates, and tidy the chat's tool rows, costs, processes and times
* feat(agent): a setup step that prepares fast Lab builds for the installed version, kept alive while it runs
* fix(agent): the notification names a prepare and shows its percent without repeating the title
* fix(agent): run the agent's Lab builds in the app's own process and continue the conversation that asked after the restart
* fix(agent): setup's prepare runs the build itself instead of handing it to the app
* feat(agent): a turn that ends without a reply says so and offers Retry, the resume is retried once, and each turn's raw output is kept
* fix(agent): keep the previous turn's raw output too
* fix(agent): a resumed Claude Code turn no longer ends at the result for a background task the last run left behind
* fix(agent): install Codex's code-mode host with it, without which Codex runs no commands
* feat(agent): setup and storage collapse into a summary card like Setup's, and Codex tool rows show the command without its shell wrapper
* feat(agent): drop the agent bubble over sessions for now; the drawer keeps the agent
* fix(agent): the Agent tab shows once the Linux runtime is downloaded, even while its proot pack is still being fetched
* fix(agent): before the install brings adb, the connection step checks the app's own pairing instead of failing with adb not found
* fix(agent): proot's binding warnings are not shown as a setup step's reason
* fix(agent): the session drawer's Agent page fills the height under the drawer's header instead of a guess from the screen height
* fix(agent): write the browser and GitHub helpers before every setup step, not only the install
* fix(agent): find Wireless debugging's current port when the saved one no longer answers, so sign-ins open the browser after it restarts
* fix(agent): only take a Wireless debugging port that answers, since mDNS can still advertise an old one
* fix(agent): find a loopback Wireless debugging port by scanning when mDNS advertises none
* fix(agent): keep a Codex turn going through reconnect notices and read gh login's status reliably
* fix(agent): time out hung setup checks and keep their results and actions with their harness
* fix(agent): keep the draft across settings and show the Agent tab once the runtime installs
* fix(agent): keep the backslash when quoting apostrophes in a URL the browser helper opens
* fix(agent): refresh apt's package lists before installing, since a fresh Linux runtime ships without them
* fix(agent): sign-in links open through the app itself, so they work without Wireless debugging
* fix(agent): the connection step passes only when adb really reaches the device, not on adb's own error
* fix(agent): say Wireless debugging isn't reachable instead of adb's error
* fix(agent): keep prompts, tool ids, attachments and retries with their own turn and chat
* fix(agent): the service stops only once foreground, and Rebuild waits for a ready agent
* fix(agent): the Codex install says what it is downloading, so a long download doesn't look stuck
* fix(agent): pass the prompt to both CLIs the way they were tested on devices
* fix(agent): the browser helper uses the running app's files directory, falls back to adb when the app cannot open a link, and the link is not logged
* feat(lab): load an on-device Lab build over the installed app
* feat(lab): build Lab builds on the device and report the APK commit to agents
* fix(lab): dd-agent connects loopback adb before starting Claude Code
* feat(lab): tell the app's own agent to continue after a Lab build restart
* fix(lab): size the build heap from the device's memory
* feat(lab): keep earlier Lab builds and list them in Updates
* feat(lab): notify when DroidDeck comes back on a Lab build or turns one off
* feat(lab): estimate the time left and compile the installed commit ahead of the first build
* fix(lab): report a failed command as a failed build and update progress every five seconds
* feat(lab): a kept build for another version can be rebuilt
* feat(lab): record which installed commit the build caches are warm for
* fix(lab): the app's agent hands its builds to the app, so they outlive the agent's turn
* perf(lab): BuildConfig describes the installed base, so a Lab commit does not change it
* perf(lab): compile with the Kotlin daemon, since in-process Kotlin compiles always redo the whole app
* fix(lab): a build whose last task was up to date no longer counts as failed
* feat(lab): dd-lab-build --explain logs why tasks ran and Kotlin's build report
* fix(lab): keep a waiting build when another is staged or Lab is turned off, and never restart during a session
* fix(lab): keep the installed build's label and exact commit, and refuse AppLanguage changes
* fix(lab): setup verifies its downloads and waits for adb before preparing
* fix(agent): quote apostrophes in opened URLs and keep the sign-in pid
* feat: experimental native userland that runs the Linux side without proot
* fix(native): keep the visible name of link2symlink binaries
* fix(ci): run the native hook build through bash
* fix(native): address review findings in the hook and the ctl fallback
* fix(ctl): use the broadcast fallback only when content is missing
* fix(native): survive Android's seccomp filter in ld.so, libc and posix_spawn children
* docs(native): how to test under Android's seccomp filter
* fix(session): hide an unlistable /sys/bus/pci/devices so libpci does not abort Chromium's GPU init
* fix(native): nested syscall() redirects no longer loop with libblsession's openat2
* fix(native): controllers reach Steam and the install-script evaluator no longer blocks game moves
* fix(native): write the hook's NUL character literals as escapes
* fix(native): proot fast path turns off and greys out while running without proot
* fix(native): D-Bus logins work and Xwayland still admits root once peers read as root
* fix(native): scripts on session mounts outside the rootfs run with the rootfs shell
* fix(native): games on the SD card library start in their own folder
* fix(native): Wine drive links proot made point at their real places
* fix(native): at-calls, uid lookups and relative programs follow the session's mounts
* fix(native): close-from in spawned programs covers what is open at the spawn
* fix(native): service names resolve and the DNS fallback answers what was asked
* build(native): the hook is built with its Go support and tracked by the preload recipe
* fix(native): drive links into identity mounts stay, and relink and seccomp patch failures are logged
* fix(native): seccomp-sim's filter holds every allowed call
* fix(ctl): the broadcast fallback only stands in for a missing content command
* fix(native): pre-2.33 glibc stat calls present owners as root too
* fix(native): res_init and res_ninit call glibc, so Steam's store resolves names
* fix(ci): a preload lock older than a new output rebuilds that group, and generated headers are not hashed
* fix(native): a second seccomp patch run leaves other system calls alone
* fix(native): the terminal and bottom desktop keep proot's environment in a native session
* fix(native): any number of session mounts, and openat2 relative to a mounted folder
* fix(agent): a native command's stdin is written beside its timeout
* fix(ctl): native rejects --env without = and timeouts outside 0-600 s
* fix(native): the resolver fallback has the network's IPv6 servers, and lookups keep the fallback's validation errors
* fix(native): a spawn whose dup2 source close-from closed fails with EBADF
* fix(native): __getcwd_chk checks its buffer, and seccomp-sim -a takes only call numbers
* feat(native): guests see the session hostname from uname and gethostname
* feat(native): the terminal and bottom desktop run without proot
* fix(components): an origin APK without a newer preload output still proves the rest
* fix(native): seccomp-sim -a rejects empty fields
* fix(native): gethostbyname no longer answers failed lookups with Ubuntu and Debian addresses
* feat(native): store installers, root helpers, Comet and Xodus run without proot
* fix(native): the hook finds libraries, preloads and itself through a sandbox's mounts, and ldconfig -C makes its cache
* feat(native): Flatpak sandboxes run without proot, and apply_extra runs without a runtime
* fix(native): remove, glob and nftw take guest paths, and links in a mount are followed as the guest sees them
* fix(native): Flatpak sandboxes get the session's mounts, their libraries' RUNPATHs, and no rootfs-only settings
* fix(native): gethostbyname asks the network's DNS servers, nftw of / gives single slashes, and the library path fits mapped directories
* fix(native): a native process starts in its working directory's real place
* fix(native): the library path given to ld.so holds whole directories only
* fix(native): Wine drive links keep their guest paths, so proot finds the library after a native session
* fix(native): droiddeckctl native prepares the rootfs with the setting off
* fix(native): a SysV segment stays while any process has it attached
* fix(native): DNS lookups ask the network's own servers before public ones
* fix(native): a program's own seccomp traps reach its own SIGSYS handler
* feat(native): run native sessions under a device's own app seccomp filter for testing
* fix(native): SysV attaches and collection take a lock, a failed mark fails the attach, and collection closes its directory
* fix(native): shmat fails when it cannot take the segment lock
* fix(native): the SysV segment lock is retried when a signal interrupts it
* fix(native): process tools see the session's processes as root's
* fix(native): native processes are named after their program, not ld.so
* fix(native): a session's helpers keep the mode the session started in
* fix(native): a keyed SysV segment stays until IPC_RMID
* fix(native): a rootfs link to an absolute guest path is followed in the guest, so Wine finds its z: drive
* fix(native): a sandbox whose loader can't be patched runs under proot
* fix(native): creating a private SysV segment also collects stale ones
* fix(native): a session stopped during startup forgets its mode
* fix(xbox): a native backend is authorized by its process ancestry
* fix(native): Wine's drive links are followed in the guest from any prefix
* fix(native): only the startup that owns the running session publishes its mode
* fix(native): a SysV segment stays until IPC_RMID even after its maker ends
* fix(native): a sandbox mount nested in another mount gets its own place
* fix(native): a sandbox whose mount places can't be made runs under proot
* fix(dual-screen): drop the Dashboard notifications page and its notification listener
* test(session): freeze the clock in the memory-pressure policy harness
* fix(session): close the bottom panel from the docked menu's Close again
* fix(input): keep imported screen keys on screen when the gamepad is hidden
Screen keys used to be shown on their own, whatever the gamepad did. Since
they were merged into the gamepad layout they disappeared with a controller
connected, in desktop sessions, with on-screen controls set to Never, in
Steam + QAM mode and in Steam's touch mode.
Elements now carry an always-shown flag. The screen key migration sets it on
imported keys (and once on keys merged by earlier builds), and the on-screen
controls keep flagged elements drawn while the rest of the pad is hidden,
in buttons-only mode and with the quick toggle, without duplicating them in
the unfolded strip.
* fix(dual-screen): give the profile controls panel the bottom bar with swap
* fix(dual-screen): keep the layout editor on the bottom screen
The bottom panel was recreated above an open input editor or input
settings window whenever it closed and reopened (session stop/start, or
the bottom screen briefly reported as taken), hiding the editor. The panel
now waits while one of our input windows holds its display and reopens
when the last one closes.
An app arrival recorded before one of our windows covered the bottom
screen no longer marks the screen taken, the editor picks its display only
when a display is added or removed, and it skips a bottom screen another
app holds.
* fix(input): label keyed D-pads and sticks with their keys
Screen key D-pads and sticks used to show each direction's key (W, A, S, D)
when labels were on. The unified renderer drew no direction labels and only
offered Show label for buttons and gamepad sticks, so imported labels did
nothing.
D-pads and sticks bound to directional keys now draw each bound key's name
at its direction through the shared key-label helper, and Show label is
offered for every D-pad and stick.
* fix(input): remember the layout editor's snap choice
* refactor(input): drop controller pref setters and drawer field nothing uses
* fix(input): restore the Deck tabs and right trackpad in the unfolded strip
With Deck controls, the unfolded strip used to offer the Gamepad, Grips,
Left, Right and Both tabs and put a trackpad in the right stick slot. The
unified strip only showed the gamepad with two sticks.
For Deck pads the strip bar now carries the shared Deck panel tabs and
title, the other tabs show the same stock Deck panel layouts as the second
screen, and the docked gamepad takes its stick or trackpad per side from the
Deck gamepad layout. Other pads keep the plain two-stick strip.
* fix(input): carry the Deck gamepad stick or trackpad choice into its layout
The Deck controls Gamepad tab used to let each side switch between a stick
and a trackpad. That choice is now part of the editable Deck gamepad
layout, but the saved choice was never moved over and its setter was left
without callers.
The library upgrade now applies a saved non-default choice to the Deck
gamepad layout once, only while that layout is still the stock one, and the
unused setter is removed.
* chore(i18n): drop the Deck surface switch string nothing shows
* fix(input): offer Show label only where a label can be drawn
* fix(input): Show label follows the unsaved action and diagonal-only sticks
The element properties now get the editor's draft action, so changing a
stick or D-pad to keys offers Show label before saving. A stick bound only
on its diagonals counts as labelled too.
* fix(input): only sticks count their diagonal keys for Show label
Diagonal key names are drawn on sticks only, so a D-pad bound just on its
diagonals no longer offers a label it can't show.
* fix(input): restore on-screen touch behaviour: slide between buttons, mouse pass-through, keyboard on release, overlay haptics, stick reach and hit areas
* fix(dual-screen): restore second-screen bars, swapped Terminal and Deck panel tabs
* fix(input): fit on-screen layouts to the real view size, cutout, Steam + QAM mode and unfolded strip
* fix(input): unfolded strip keeps press haptics
* fix(input): restore layout import/export, element editor rows and layout tools
Bring back .icp layout import and export on the saved layouts page, the
per-element image, opacity and label controls, a uniform size slider and
reset positions in the layout editor, and non-destructive Steam, QAM and
keyboard button toggles stored on the layout style. Remove strings left
unused by the input profile unification.
* fix(input): keep hit slack to gamepad-shaped buttons and fit touch tests to saved layout geometry
* fix(input): button size regenerates the gamepad and scales spacing
A generated gamepad now stores its button size and is regenerated at that
size, so spacing follows the radius and the layout keeps regenerating on
rotation. Edited layouts scale positions from their anchor edge or centre
so gaps grow and shrink with the controls.
* fix(input): button size rescales moved gamepads around the generator
Gamepad layouts that no longer match the generated positions exactly were
scaled element by element from screen thirds, which split the face buttons
and D-pad across anchors. Gamepad controls now take the generator's
positions at the new size plus each control's own offset from the
generated position, so groups keep their shape and spacing.
* fix(components): DXVK Auto on Adreno 6xx keeps DroidDeck-Proton on DXVK-gplasync 2.7.1
* fix(components): DXVK Auto says so when the pinned DXVK could not be installed
* fix(components): a DXVK the player picked is shown as their choice, not a failed pinned install
* feat: move UBWC compatibility from Components to Settings
* fix: update driver menu tests for the moved UBWC restart check
* fix: load saved UBWC mode at startup and block Settings restart during runtime operations
* fix(fex): show which preset Use Default resolves to and rename FEX defaults to Stock FEX
* test(components): the FEX preset chip names its resolved preset instead of Use Default
* fix(fex): the game settings menu keeps Use Default beside the inherited preset
* fix(fex): inherited preset chip follows the current Default preset
## Problem and intent
Desktop sessions need configurable mouse, keyboard and application
actions from a controller. On dual-screen devices, a lower-screen
trackpad may already handle the pointer, leaving the physical controls
available for other assignments. This PR gives physical controllers and
local touch controls a shared profile/action model instead of separate
configuration paths.
This follows the centralization direction discussed in
[#1202](https://github.com/Droid-Deck/DroidDeck/pull/1202#issuecomment-6299647180)
and addresses the controller-input portion of #860 and #1356. Those
discussions are context, **not prior maintainer approval** of this
implementation. I would appreciate architectural feedback and guidance
on splitting or coordinating this work if that would make review easier.
## Profiles and configuration
- Controller stays within the existing Setup tabs. Game and Desktop each
have a default-profile selector, followed by one profile list and device
settings.
- Named profiles reference three independently reusable parts: Physical
controller, Main screen and Second screen. Layout geometry and action
bindings are stored separately. Profiles can share physical controls
while assigning different touch actions.
- The session menu selects a complete profile, or expands to mix the
three parts. A mixed selection can be saved as a new profile; Game and
Desktop retain separate selections.
- Both screens use the same full-size editor, with add, move, resize,
snapping, undo and shared properties. The main-screen editor has
movable, collapsible tools and is transparent above a running session.
When editing the secondary display, its entire area is the canvas and
the main display hosts tools/properties.
- The old fixed gamepad, fixed Deck controls and separate Screen keys
editor are replaced with editable presets and the shared renderer.
Existing positions, bindings, shapes, symbols, imported images and
toggle behavior migrate; legacy Screen keys become additional profiles
without replacing the user's selected profile. Migration retains a
backup before removing old preferences.
- Saved layouts have thumbnails and screen targeting. Actions include
compatible native output/remapping, keys/combinations, mouse buttons,
pointer movement, vertical/horizontal scrolling, directional keys,
existing menu/keyboard commands, and disabled. Buttons can use symbols,
shapes and toggle behavior.
## Architecture and latency
```mermaid
flowchart LR
Library[Profiles + reusable layouts and bindings] --> Router[InputRouter]
Physical[Physical controller adapters] --> Router
Touch[Main + secondary InputSurfaceView] --> Sources[Owned pad / Deck sources]
Sources --> Router
BLE[Steam Controller BLE adapter] --> Router
Router --> Native[Existing controller publication]
Native --> Steam[Steam Input / guest controller]
Router --> Held[Shared key and mouse ownership]
Held --> Guest[Existing guest input outputs]
Router --> UI[Existing menu / keyboard hosts]
```
Bindings resolve when the profile changes, outside event dispatch.
Native forwarding has no additional frame/timer queue; identity routes
reuse existing state/reports. Only synthesized continuous stick
pointer/scroll motion uses a frame callback. One source releasing a
key/button does not release another source's hold. Menu, profile,
suspension and removal transitions release old outputs and gate held
controls until neutral.
Steam continues to own game-specific mappings and action sets. Steam
Mobile Touch and VR retain their specialized paths. Raw device adapters
remain responsible for transport/calibration and must join routing
before publishing consumed controls; this does not claim that every
adapter has been unified.
Historical Thor measurements on the native Game path found
kernel-event-to-guest-observer upper-bound medians of **3.061 ms
baseline versus 2.999–3.403 ms candidate**, with a subsequent
installed-APK check at **3.039 ms / 300 matched edges**. These are
relative software-path measurements, not button-to-photon latency. Some
tails increased, simultaneous BLE contention remains unmeasured, and
**this UI revision has not had a new latency measurement**. The previous
cursor fix changed visual update intervals from roughly 33.3 ms to 16.7
ms without changing sensitivity or adding smoothing.
[Full design, migration and
boundaries](https://github.com/Delete228/DroidDeck/blob/codex/unified-controller-input/docs/development/unified-input-proposal.md).
## Validation
Updated through upstream 542bbf89c; current PR head 08526f828. The
latest CodeRabbit review generated no actionable comments.
- **1,017 tests passed in 133 suites after the latest upstream merge**,
with no failures, errors or skips: all input/UI tests, global reset,
component updates, PiP policy, runtime auto-update and Proton menu
auto-selection. The pre-merge review-fix run passed **967
input/UI/global-reset tests**. The subsequent second-screen refresh fix
passed **35 focused surface/presentation tests**, including three new
state-preservation regressions.
- Full local run before the final review follow-ups: **3,598 tests,
3,573 passed, 25 failed**. The 25 are unchanged upstream macOS
limitations: log-export tests requiring SecureDirectoryStream (24) and a
case-sensitive WoW add-on test (1). The global-reset regression also
passes. Earlier disk-exhausted attempts were discarded; the completed
run bounded temporary resources with one worker and JVM recycling,
without excluding tests or changing repository test configuration.
- APK build, runtime/session assets, native dependencies, alignment and
v1/v2/v3 signatures passed. Installed over the separate Input Test app
on **AYN Thor** with its environment retained. Local packaging omits
Rust store engines; the [full Build APK workflow for 08526f828
passed](https://github.com/Droid-Deck/DroidDeck/actions/runs/42295403480),
including Android/JVM, Python/native, Rust store-engine and Xbox checks.
- Review follow-ups keep migration IDs stable, preserve unreadable
supported-version libraries before recovery, retain forward-version
protection, release keyboard owners on focus loss, handle unavailable
presentation displays, and allow controller confirmation/cancellation
after key capture. Deck router cleanup checks ownership; BLE teardown
synchronously blocks late routing/publication, and shared Triton slot
ownership protects a replacement controller from stale instances. Legacy
Select-to-Steam migrates to Desktop without replacing saved bindings;
mouse backend transitions are serialized with publication. Profile
refresh now updates custom controls in place, preserving
terminal/dashboard state, the field keyboard and WoW chat. No physical
BLE-controller test or new latency measurement was performed.
- The latest 08526f828 APK is installed on Thor. Profile preferences
stayed byte-for-byte identical and the retained Steam environment
reached READY at 1280×720 / 60 Hz. The earlier capture-dialog fix was
also exercised on Thor. APK SHA-256:
`ed6ef888648468b4ebeb94e8f1ae74a195bb1f0a3f27e4fb3646d5462a4beb6b`.
- Upstream integration preserves the new WoW chat panel,
platform-pointer and PiP fixes. The fast-double-tap stick-click rule and
its six regressions were adapted to the shared renderer; WoW field
selection also preserves Ctrl held by a profile.
- Latest Thor UI pass: existing Setup navigation, profile/layout
previews, action persistence, centered element creation, main-screen
editing over Steam, secondary canvas with primary tools/properties, and
X returning from the docked menu to controls.
- Across the documented device sessions, the user tested native physical
buttons/sticks/triggers, Deck pads/grips, desktop
pointer/clicks/scrolling/arrows, Android keyboard with physical input,
Ctrl+F and Ctrl+A/X/V/S, simultaneous physical/touch holds, menu release
and Home/resume. These historical checks are distinguished from the
latest UI-only pass.
[Revision-specific results, APK hashes and
limitations](https://github.com/Delete228/DroidDeck/blob/codex/unified-controller-input/docs/development/unified-input-validation.md).
## Coordination and remaining work
- #1202 (jmarti326): shortcut capture/device identity and the
shared-assignment discussion informed this work.
- #1305 (kingokksa): shared PC-key definitions were adapted from that
PR; this does not absorb its entire feature set.
- #1414, #581, #1226 and #1150 remain relevant for external-device
input, calibration, keyboard routing and DualSense integration. The
DualSense coordination discussion in this PR describes adapter
boundaries and duplicate-stream suppression.
Import/export, automatic per-app selection, macros, multi-button
activation chords and automatic text-field keyboard detection remain out
of scope. A single binding emitting a key combination is implemented.
The configuration UI is now implemented and exercised on Thor. I am
happy to continue UX/accessibility/localization refinements here before
merge, or in follow-up PRs, according to maintainer feedback. The main
review questions remain the shared model, source ownership/lifecycle
contract, migration policy and integration with other input work.
## UI walkthrough on AYN Thor
Actual captures from the installed Input Test build. In paired images,
the main display is above the second display. The test application name
and local build badge are not production branding changes.
<details>
<summary>Controller settings: separate defaults for Desktop and Game,
followed by the profile library.</summary>
<img width="960" alt="Controller settings: separate defaults for Desktop
and Game, followed by the profile library."
src="https://github.com/user-attachments/assets/a4c156d6-9ded-43df-b0f8-e5f34028806d"
/>
</details>
<details>
<summary>Physical controller: bindings grouped by control type, with
each current action visible.</summary>
<img width="960" alt="Physical controller: bindings grouped by control
type, with each current action visible."
src="https://github.com/user-attachments/assets/a8a840e3-19ac-473f-9f55-b06ae9c637b5"
/>
</details>
<details>
<summary>Reusable layouts: bounded previews and separate screen
targets.</summary>
<img width="960" alt="Reusable layouts: bounded previews and separate
screen targets."
src="https://github.com/user-attachments/assets/cf0cf7d1-42f3-46cd-b50b-f01e1fa8ed41"
/>
</details>
<details>
<summary>Main-screen editor over the live Steam UI, with the movable
tools expanded. The background is Steam artwork, not a running
game.</summary>
<img width="960" alt="Main-screen editor over the live Steam UI, with
the movable tools expanded. The background is Steam artwork, not a
running game."
src="https://github.com/user-attachments/assets/0781f7c8-fd10-47e5-8a23-817f945a6289"
/>
</details>
<details>
<summary>Dual-screen editor: tools and selection on the main display,
full-size placement canvas on the second display.</summary>
<img width="720" alt="Dual-screen editor: tools and selection on the
main display, full-size placement canvas on the second display."
src="https://github.com/user-attachments/assets/f13d4ef6-be6d-4d95-9d42-e047e0a63b8a"
/>
</details>
<details>
<summary>The selected lower-screen trackpad stays visible while its
properties open on the main display.</summary>
<img width="720" alt="The selected lower-screen trackpad stays visible
while its properties open on the main display."
src="https://github.com/user-attachments/assets/26baf597-f037-419e-9601-e48f086c0d4f"
/>
</details>
<details>
<summary>Session menu: expand Profile to select the physical,
main-screen and second-screen parts independently.</summary>
<img width="720" alt="Session menu: expand Profile to select the
physical, main-screen and second-screen parts independently."
src="https://github.com/user-attachments/assets/27ac4d9c-68b3-4257-8353-23104e2343de"
/>
</details>
* feat(components): Auto updates for FEX, DXVK and VKD3D-Proton
Each component gets an Auto or Manual mode on the Default Components page.
Auto installs the newest DroidDeck-Components nightly into every Proton that
ships the component, matching each Proton's ARM64EC or x64 flavour, keeps the
current and replaced builds, and removes older builds it downloaded.
A game that ran on the previous build and exits within 30 seconds twice on the
new one rolls the line back; the failed build is skipped until a newer one is
published. Existing manual choices keep that component on Manual, and DXVK
stays on Manual while a GPU rule pins its version.
* fix(components): Auto notes per line, Original label and first-update rollback
Boot history on the Original is recorded before the first update so that
update can roll back, rollback notes stay with their own component, the
Original shows by name, and new Protons get the current build even when no
newer one is published.
* fix(components): Auto owns only its own downloads and follows in-place Proton updates
Auto deletes only builds it downloaded, re-swaps a Proton whose version
changed under an existing choice, reports a missing nightly listing as a
failed check, writes its state atomically so the page can read it without
waiting on a download, and words the rollback and download messages to
match what it does.
* fix(components): Auto keeps a rollback and reads updated Protons' own files
After a rollback only a build newer than every rejected one is taken, so an
older build still listed in the nightly cannot replace the restored one.
Flavour detection treats a choice saved for an older Proton version as not
applied, so a Proton Steam updated in place is read from its own files and
re-swapped. A failed state rename falls back to a direct write and is logged.
* fix(components): Keep the last good Auto state when a save fails
* fix(components): Serialise Auto state reads and writes
* Install the Linux runtime by itself on first launch
The runtime auto-update skipped devices with no runtime, so FEX, Proton,
DXVK and VKD3D waited for the first Steam launch to install it. Claim a
first install when the device never had a runtime, on a usable network,
outside battery saver and with a supported GPU, and check Proton as
soon as the runtime is ready. A runtime the user removed stays removed.
* fix(components): Check components as soon as Proton is installed
The Auto check stamped its hourly timer even when no Proton was
installed yet, so a fresh device waited an hour after Proton arrived.
Skip the stamp until a Proton exists and run the check right after a
Proton install finishes.
* fix(components): Apply stored builds on every check and retry unfinished rollbacks
A held or failed download no longer skips applying the stored build or
stamps the hourly timer. Stored builds are applied on every check, so a
new Proton is reconciled inside the hour, and a check requested while
one runs is queued. A rollback that could not restore or swap a Proton
is retried and reported only once it finishes. Removing the runtime
marks it as seen so it is not installed again, and the failed-save test
skips where the process can still write read-only files.
* fix(components): Keep rollback restores pending until applied and save state on failure
* fix(components): Hold newer builds until a pending rollback is applied
* fix(components): Drop pending restores for removed Protons
* fix(components): Never auto-install a non-ARM64X build over a Proton's own ARM64X DXVK or VKD3D
* fix(components): Read probe archives outside the lock and block auto updates while the original build is unknown
* fix(components): Require ARM64X when any Proton on the line needs it, and before each swap