mirror of
https://github.com/ddeverill/SteamlessController.git
synced 2026-10-21 22:22:16 -05:00
v1.19
84
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Veri7ion
|
1eb2a59d6d |
Fix tray icon restoration after Explorer restart
Fixes the tray icon not being restored when Windows Explorer is restarted. TrayApp already registers for the TaskbarCreated shell notification and calls AddTrayIcon() when it is received. However, the window used to receive this notification was created as a message-only window using HWND_MESSAGE. Message-only windows do not receive the broadcast shell notification sent when Explorer recreates the taskbar, so AddTrayIcon() was never called and the application was left running without a tray icon. This changes the tray window to a hidden top-level window, allowing it to receive TaskbarCreated while remaining invisible. Result After restarting Windows Explorer, the Steamless Controller tray icon is automatically registered again without needing to restart the application.v1.19 |
||
|
Dylan Deverill
|
fde18e600e | Update README.md | ||
|
Dylan Deverill
|
6db91983ed | Update README.md | ||
|
Dylan DeverillandClaude Opus 5
|
9a323d9152 |
Bump the installer to 1.18
Both copies of the number move together: resource.h feeds the VERSIONINFO resource and the startup line in the event log, InnoInstallerScript.iss feeds Programs and Features. Two components rather than three, matching the release tags and the previous bumps. 1.17.3 was a hotfix number; the patch field goes back to 0. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>v1.18 |
||
|
Dylan DeverillandClaude Opus 5
|
9d656c2809 |
Stop unrelated balloons offering a driver download
ShowAlertBalloon defaulted its action to BalloonAction::ViGEmDownload, so every caller that did not think about the question sent the user to the ViGEmBus releases page when the notification was clicked. Five did: a Bluetooth connection dropping, or a USB controller being unplugged another process holding the device a device cycle that never took the devnode down Steam refusing to hand the controller back a devnode that needs re-enabling in Device Manager None of those are ViGEmBus problems and none of them are improved by installing it. The advice was actively misleading for the last one, where the fix is Device Manager, and for the disconnect balloons, which are routine. Doing nothing is the default now, so forgetting to consider it is inert and opening something has to be asked for. The two balloons that genuinely act already say so: ShowViGEmBalloon sets the action directly, and only when the manager established the bus is actually missing rather than present and unusable; the disabled-devnode balloon passes EnableDisabledDevice. The member's initial value moves to None for the same reason, and None leads the enum so the harmless case is what a default-constructed value lands on. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
745ad65b0c |
Narrow the release cycle to the interface that had the controller
The acquire path has cycled a single interface since the receiver work; the release path still took down all four. Only one slot is ever occupied, so the other three were being disabled and re-enabled for nothing — and each one handed Steam another arrival to react to, when the only arrival that matters is on the interface it will actually read the controller from. The cost was hidden until the reopen veto was fixed, because the vetoes dominated: a release cycle ran 15.47s, of which about 13s was two vetoed interfaces. Measured end to end by launching a Steam game in auto mode 2, which is the only thing that exercises this path: 10:33:30.394 AUTO: steam state 2, want=0 10:33:30.394 RELEASE: closing all device handles 10:33:30.459 RELEASE: cycling device so Steam sees an arrival 10:33:31.512 CYCLE: done after 1063 ms 10:33:31.518 CONNECT mi_02, and the rest after it 1.06s against 15.47s, one interface cycled rather than four, no Kernel-PnP 225 events, and the reopens land after the cycle instead of 26ms into it. The handoff itself is confirmed rather than assumed: quitting the game had our next claim refused with "another process holds a write handle", so Steam had genuinely adopted the controller in the interval. We then took it back on one narrowed cycle, 2.2s from the game exiting to game mode being live. Unconditional here, unlike the acquire path, which narrows only its first attempt so that a wrong guess cannot keep costing it across retries. A release has no second attempt to protect, and an empty path already deletes the request file and falls back to cycling everything. LastLivePath is read before ReleaseDevices rather than after. The value survives either way — it is only ever assigned, never cleared — but taking it while the slots are still up has the two paths asking the same question at the same point. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
f30595c0f8 |
Stop reopening the controller during our own device cycle
WM_DEVICECHANGE called ControllerManager::OnDeviceChange unconditionally, and OnDeviceChange opens a read/write handle on every interface it finds. An open handle vetoes a device removal no matter who owns it, so the app spent every cycle fighting handles it had reopened milliseconds after asking for the cycle. A release on this machine, with the log timestamps: 23:22:31.671 RELEASE: closing all device handles 23:22:31.701 RELEASE: cycling device so Steam sees an arrival 23:22:31.727 CONNECT mi_02 26ms after asking for the cycle 23:22:31.737 CONNECT mi_03 23:22:31.749 CONNECT mi_05 23:22:31.762 CONNECT mi_04 23:22:37.107 DISCONNECT mi_02 the helper finally wins it 23:22:47.210 cycle done 15.47s Kernel-PnP 225 names SteamlessController.exe as what stopped the removal, at 23:22:37 and again at 23:22:43. The helper's log prices it: an unobstructed interface cycles in about 1.2s, a vetoed one in about 6.5s. Three vetoes are on record for one evening, two in a single release. The second hole is the same root cause and hurts acquisition. The test for "the cycle has delivered" was ControllerManager::IsConnected, which is true whenever any Steam Controller interface enumerates. A receiver publishes four and a narrowed cycle takes down one, so the surviving three held that true for the whole cycle: the guard came down on the first arrival of anything HID-shaped, including the device being cycled, before it had even gone. The comment above it claimed to identify "our controller specifically, not merely something HID-shaped". On a puck it never could. Arrival is now judged by the interfaces the cycle actually took down going away and coming back, which is the only honest signal available. Tracked through SteamController::EnumerateAll, because HidDevice::Enumerate opens each candidate with a zero access mask and closes it before moving on — unlike OpenSlot it leaves nothing behind for PnP to trip over, so watching a cycle cannot veto it. While a cycle is in flight no slot is opened at all. RestartControllerDevices records what the cycle owes before starting it, and consumes the narrowing request rather than reading it, so one caller's narrowing cannot leak into another caller's cycle. EndCycle is the single exit, with a 20s watchdog behind it. Every normal ending is an event that might not arrive, and a flag left raised means the device is never opened again and the tray reports no controller until restart — the previous code cleared the flag from five places and had no backstop at all. It skips the resync for the caller that is about to release the device anyway, where reopening would be the very veto this prevents. Measured after, on the same hardware: the acquire cycle runs 1.13s against a first interface that used to take 6.5s, no CONNECT appears between the release and the cycle completing, and no Kernel-PnP 225 events are logged. Both faults predate the acquisition work on this branch; origin/main is byte-identical here. Narrowing the cycle to one interface made the second one bite harder, since only one of four interfaces now ever leaves. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
408413a041 |
Reconcile Steam's state on a timer, so a lost edge is not permanent
SteamWatcher announces a state change exactly once, through a single PostMessage. Nothing repeats it. If that message is lost, dropped, or arrives while nothing is pumping, the app keeps a stale view of Steam for the rest of the session — a game launches and the controller is never handed back. This is not hypothetical. An instance on this machine sat at steam=1 while a game was running, and the release only happened once something else moved the state again. The cause of the lost edge is still unknown; this bounds its cost either way, which is the part worth having before the cause is found. The timer compares the watcher's current state against the last one actually acted on and applies the difference. Any single lost edge now costs seconds rather than the rest of the session. Logged loudly when it fires, because it should never fire. If it happens in the field the log says so, rather than the reporter having to notice the controller behaving oddly and describe it. m_lastAppliedSteamState is recorded before ApplySteamState's Manual check, so leaving Manual mode does not find a stale value and report a change that never actually went missing. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
f10a18347c |
Take the controller back the instant the cycle returns it
Acquiring against a running Steam was losing a race we were not really running. The old sequence waited for WM_DEVICECHANGE, enumerated every HID interface on the machine, opened the device shared, closed it, and only then reopened it exclusively. Steam had to be anywhere in that. Measured first, because every timing figure in #79 so far had come from reading log line order and that was wrong twice. A losing attempt cost 656ms, of which about 545ms was the retry burst sleeping, 46.6ms was eleven full enumerations, and 0.2ms was the claim itself. A winning attempt took 31ms start to finish. The in-process work was never the problem; the time was going somewhere the log could not see. The claim now happens before the probe. A live slot answers a state probe in a report or two but an empty one costs the whole timeout, so walking the list probe-first left the occupied slot unclaimed until every empty slot ahead of it had timed out — up to 75ms on a four-slot receiver, against the ~180ms Steam takes to reopen the device here and under 73ms on the machine in #79. A sweep claims every slot up front and the per-slot path uses that result; slots that turn out to be empty hand theirs straight back. The pounce thread stops waiting to be told. WM_DEVICECHANGE is one of the last events in an arrival, not the first: PnP registers the interface, dispatches to every listener, and ours queues behind whatever else is pending. Since we are the ones who asked for the cycle, a thread armed before it spins CreateFileW on the paths we already know and takes the device exclusively on the first open. It only keeps a handle after seeing the path disappear — grabbing earlier would hold the very handle that vetoes our own cycle, and PnP does not care that the handle is ours. Adopted slots skip the claim sweep, because reopening would give the device back for exactly as long as it takes to ask for it again. Adoption happens before the in-flight guard. A caught handle is proof the cycle delivered, and adopting after the guard meant sitting on a device we had already won for the whole CYCLE_MIN_GAP_MS: measured at 4030ms between the grab and the adopt, now 19ms. The burst no longer runs on a blocked claim. A held handle does not free itself, so retrying it ten times at 50ms intervals only delayed the cycle that could actually free it. It still runs for a slot that has not started streaming, which is what it is good for. The cycle is narrowed to the occupied interface. A receiver publishes one per slot and only one is ever occupied, so cycling all of them multiplied the downtime and gave Steam an arrival per interface to react to. The app names the target in cycle.request and the helper consumes it on read, so a stale request cannot narrow a later cycle. Only on the first attempt: if narrowing did not work the assumption behind it may be wrong, and later attempts cycle everything. Measured after: a losing attempt is 15ms rather than 656ms, one enumeration rather than eleven, and the whole blocked-claim-to-game-mode path is 1.20s, nearly all of it the devnode being down. Two things make this observable rather than another round of inference. The acquire logs where its milliseconds went, won or lost. And when we do settle for a shared handle, the helper is asked who is holding the device, so the answer lands in a log the reporter already sends instead of needing a command run by hand — the helper's own scan only runs when PnP vetoes a cycle, and the case worth explaining is the opposite one, where every cycle succeeds and the controller is taken back anyway. The scan's counters also stop lying. "queried" counts handles whose names were resolved, which is far fewer than those examined and is not a coverage figure — candidates are skipped when their process already has a known holder, when OpenProcess is refused, when DuplicateHandle fails, or when the object is not device-like. Read as coverage it says 7% when the scan in fact examined everything, which is how a complete answer got mistaken for a partial one. It now reports examined against total, and says that a deadline means a wedged worker rather than a short scan. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
5e220d4724 |
Cycle the controller through the helper even when already elevated
RestartControllerDevices took a shortcut on an elevated run: rather than firing the helper, it cycled every interface itself, in a loop, on the UI thread. Each interface disables, waits a second for the removal to propagate, and enables, so a four-interface receiver spent about five seconds inside that loop with no message pump running. Nothing was pumping, so every WM_DEVICECHANGE the cycle produced sat in the queue until it finished. The first interface was back — and taken by Steam — seconds before we drained those messages and claimed. A reporter's log in #79 shows it exactly: RELEASE at 15:49:26.049, then five seconds of no log lines at all, then four CONNECTs inside 22ms as the queue emptied, then the claim refused. Eleven of that session's fourteen cycles ran this path and lost; the three that went through the helper are the ones that behaved. It also explains the app appearing to hang mid-cycle, reported as the tray menu opening as a white rectangle: the UI thread was inside the cycle. The helper is a separate process, so the cycle runs while this one keeps pumping and the arrival drives the claim as soon as it lands. Measured on the same elevated configuration: helper starts 56ms after the request, whole cycle-to-claim 1.27s, message loop live throughout. Dropping the shortcut also removes a fork that had already drifted — the in-process path wrote nothing to cycle.log, so an elevated run left no record of a cycle having happened at all, which is why the helper's holder scan never appeared in that reporter's logs. The three EnsureCycleTaskRegistered call sites no longer skip registration when elevated. They assumed an elevated run never needed the task; it does now. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
c3f58de856 |
Stamp the build into the binary and into the first line of its log
A release carried no version anywhere a user or a maintainer could see it. The exe had no VERSIONINFO, so its properties showed nothing, and events.log opened with a bare "=== SteamlessController started ===". Twice while chasing #79 a reporter's log arrived and there was no way to tell which of two builds had written it, which cost a round trip each time. resource.h now holds the version once, and both consumers read it from there: app.rc builds a VERSIONINFO resource so Explorer and Programs and Features can report it, and the startup line names it. Keeping them on one definition means a binary and its log can never disagree about what they are. InnoInstallerScript.iss still carries its own copy of the number. Inno can read it back out of the built exe instead, which would leave one place to edit, but that makes the installer refuse to compile before the app is built and is a change to how releases are cut rather than to the app. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
234d016ac8 |
Settle for a shared handle when cycling cannot win the controller
Game mode refused a shared handle outright whenever steam.exe was running, on the reasoning that the holder was probably Steam and that refusing would escalate to a device cycle that took the handle back. On machines where that cycle never wins the reopen race, it took nothing back and the app simply never worked: #79 counted 731 blocked claims in a single day against zero successes, and both of that reporter's successes across the whole log came through the shared path with Steam closed, never through an exclusive claim. No exclusive claim is obtainable against a live write handle, however polite its holder. Windows share arbitration runs in both directions: FILE_SHARE_READ asks the kernel to deny write access to everyone else, and it must refuse while such a handle exists. Measured on a puck receiver with Steam running, even GENERIC_READ with FILE_SHARE_READ returns ERROR_SHARING_VIOLATION, so it is our share mode being rejected rather than our access. "Steam only holds it shared" and "we can take it exclusively" were never compatible statements. Shared is now the last resort rather than a refusal. The claim still prefers exclusivity and a blocked claim still escalates to a device cycle exactly as before, so machines that win that race keep winning it. Only once cycling is over — the budget spent, or a cycle shown to have invalidated no handles — does the acquire path retry with sharing allowed, immediately before it would otherwise report failure. Two callers that rebuild an already-running game mode pass it too: a slot arriving late should join rather than be dropped, and a platform change should not cost the user their controller. Sharing is not free, which is why it is last. Steam keeps driving the same controller, and its stock desktop layout binds both the left stick and the D-pad to arrow keys. Measured with a dual-channel sampler: 16 D-pad presses produced 16 matching global UP/DOWN_ARROW keystrokes 1-2ms later, press and release both mirrored. Those are real SendInput keystrokes, so they reach processes Steam cannot inject into — the doubling was first reported in a Game Pass title. The tray reports the state rather than hiding it: a distinct icon and a "(sharing with Steam)" tooltip, so a user chasing doubled input knows where to look. Slots are probed quickly before they are probed properly. A receiver publishes an interface per slot and only one has a controller in it, but every empty slot cost the full 250ms state-report timeout before the live one was even reached — up to 750ms, spent before the first claim attempt. Steam reclaims the device about 180ms after it re-arrives, so that ordering handed it the race whenever re-enumeration put the live slot last. A 25ms pass now runs first, since a live slot streams continuously and answers within a few reports. Measured across a device cycle with the live slot arriving last: arrival to exclusive claim fell from 1697ms to 99ms. A quick probe that comes up short is not evidence of an empty slot, so it neither records the slot as silent nor logs it; otherwise a controller slow to resume streaming would be latched out of the full pass that follows. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
6a513b0fb7 |
Bump the installer to 1.17
Covers the new control mode and the profile management that came with it. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
588b2cc9bc |
Let a profile for a game that is gone still be reached
A profile whose game is no longer installed could not be seen, edited or deleted. SendInitState walks the installed list and looks profiles up against it, so an entry with no installed game was never sent to the page — while staying live in m_gameProfiles, still matched by MatchProfile, and under the new mode still turning the controller on. Invisible and working at once, and the case where deleting a profile is most likely to be what someone wants. What kept it unreachable is that the page names games by index into the installed list, and an orphan has no index. Raw ids cannot cross that channel: they hold non-ASCII and backslashes the parser does not unescape. So the indices are extended instead — one appended entry per orphaned profile, past the real games, with m_installedCount marking where they start. Apply and delete then work through the paths that already existed, unchanged, and SendInitState's profile loop picks the orphans up on its own because it iterates the same list. Their names come from displayName, which was stored for exactly this: naming a profile when the installed list cannot. Profiles older than that field fall back to the raw id. Nothing is pruned automatically. Not being in the installed list is not the same as being uninstalled — a deleted Start Menu shortcut, a Steam library on a drive that is currently disconnected, or a game whose shortcut was never created all look identical from here, and quietly deleting a profile in any of those cases would destroy real work to tidy a cosmetic problem. They are labelled and left, and if the game returns they simply stop being orphans. An enumeration that finds nothing at all is treated as knowing nothing rather than as everything being uninstalled. A COM failure reading shell:AppsFolder is far likelier than an empty PC, and flagging every profile as missing on that evidence would invite someone to delete all of them. Deleting an orphan drops its picker entry along with the profile. The entry only existed because the profile referred to it, and left behind it stops being pinned — after which the only section it could appear in is the one labelled INSTALLED GAMES, which it never was. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
2d93ceba28 |
Let a game profile follow the default, and be deleted
Both of these follow from the mode added in the previous commit. Profiles used to be rare — you made one when you wanted a game to play differently — and it turned them routine, because under "off unless a game profile is running" a profile is also how you say a game should have the controller at all. Two things that did not matter when profiles were rare matter once there are a dozen. The first is that every profile was a full copy of the controls, forked from the default at the moment it was created and never tracking it again. A profile made only to turn the controller on had no business carrying controls of its own, and improving the default did nothing for any of them. useDefaultMappings makes such a profile a follower: ActiveProfile resolves it to the default rather than copying anything, so editing the default reaches every game following it without a single stored profile being touched. Games never saved before start out following; games already saved keep what they were saved as, and the stored controls stay written either way, so unticking gives a game back exactly the controls it had. False is the stored default, because every profile written before this had controls of its own and reading those back as followers would have discarded them. The second is that there was no way to remove a profile. Creating one is a couple of clicks and undoing it was impossible, which is a poor trade in any case and a worse one now that an unwanted profile also turns the controller on by itself. The delete path reuses what was already there: the registry writer already pruned subkeys past the end of a shrunken map, and ActiveProfile already resolves a missing id to the default, so the runtime consequence of deleting the live game's profile is a fallback that already existed. Like applying, it makes no control decision while the window is open — a row may be listening for a button press, and dropping the controller under it would end that mid-capture. Closing the window re-evaluates, which is what actually lets go. The page keeps one modal rather than gaining a second. Escape, the backdrop and the keydown handler all assume exactly one thing can sit above everything else, and a second would have had to be taught to each of them. Applying now pushes unconditionally instead of comparing the edited id against the live one. That comparison cannot see the new case: edit the default while a following game is live and the ids do not match, but the live controls have just changed. Only ActiveProfile knows about following, so it decides. The game index the page sends moved into GameIndexFrom, shared by apply and delete so one place decides what a valid index is. It also tightened apply, which used to let a malformed index fall through to the same branch as an absent one and be read as the default profile. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
c243576ec2 |
Add an auto mode that runs only for games with a profile
The existing auto modes all answer "is Steam busy": Manual defers to the tray toggle, and the two Steam modes yield whenever Steam, or a Steam game, is running. None of them can express the thing people keep asking for, which is a controller that stays out of the way at the desktop and comes to life in a game. OffUnlessProfile answers a different question, and only that one: is a game the user has made a profile for in front? That keeps the outcome readable off the mode name. An earlier attempt at this — an enable/disable setting on the default profile, layered over whichever auto mode happened to be selected — did not: the result depended on the mode, on the Steam rule and on the profile together, and could not be predicted from any one of them. It is also the one mode that does not consult Steam. A Steam game launched from the Steam client always publishes RunningAppID, so both existing modes yield for it and a per-game profile could never apply — per-game profiles were dead for most of what a Steam Controller owner actually plays. Having made a profile for a game is the user saying they want us driving it rather than Steam Input, and the acquire path already knows how to take the device from a live Steam. With nothing matched the mode wants nothing, so it never contends with Steam except over games it was told about. Making that work meant inverting who drives whom. OnForegroundChanged used to return early unless the controller was already ours, on the grounds that a yielded controller has no bindings to switch; it is now the path that decides whether we hold the device at all, so it has to run precisely when we do not. ActivateProfile split into SelectProfile, which records what matched, PushActiveProfile, which applies it, and EvaluateControl, which acquires or releases — the control decision reads the selection, and leaving them joined had ReleaseControl and profile switching calling each other. Releases wait five seconds. Every release and re-acquire round trip unplugs and replugs the virtual pad, which a running game sees as the controller being pulled out, and alt-tabbing to a browser mid-game is ordinary. Steam turning up is exempt, since the whole point of yielding is that Steam is about to open the device. For the same reason PushActiveProfile only runs when control is wanted: pushing the default profile on the way out of a game rebuilt the pad under it within a frame, five seconds before the grace period had any say. ShowAlertBalloon now logs what it raises. Notifications were the one thing the app did that left no trace, so a report of seeing one twice could not be answered — the log showed a profile loading once and said nothing about how many times the toast was drawn. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
fde32bc650 |
Document what the installer needs that a clone does not have
Compiling resources\InnoInstallerScript.iss on a fresh checkout fails, and the error names a file with no obvious way to obtain it. Two prerequisites were undocumented. The ViGEmBus installer the script packages is not in the repository. It is a 6 MB vendored binary and *.exe is gitignored — by the blanket "loose binaries committed by mistake" rule rather than by anything written for it, which is why nobody noticed. Left out rather than committed: the version is pinned in a single #define, and a stale 6 MB binary in history is harder to walk back than a line of prose. The script also packages SteamlessDeviceCycle and ViGEmBusProbe, and the build section above only ever told anyone to build SteamlessController. Both are now in the build commands and in the target table, which listed neither. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
598d1bf210 |
Add a touch keyboard binding that needs no elevation to click
A paddle bound to Ctrl+Win+O opens osk.exe, which carries uiAccess="true" in its manifest and so runs at High integrity. UIPI then discards every trackpad click aimed at it: the keyboard appears and not one of its keys can be pressed. Elevating this app fixes that and costs a UAC prompt on every launch, plus the cursor-freeze bug that keeps the tray unelevated. The touch keyboard is a different window, and that is what makes this work. Measured on Windows 11: TabTip.exe runs High but owns no visible window, while TextInputHost.exe — which owns "Windows Input Experience", the surface the user actually clicks — runs at Medium, the same level as this app. So the clicks land with nothing elevated anywhere. TabTip is asked to show the keyboard through ITipInvocation::Toggle. Neither that interface nor CLSID_UIHostNoLaunch is in the SDK, so both are declared from their registered IIDs. The class has no LocalServer32 — COM will not start the server — so a machine where TabTip is not already running gets it started and polled for. The toggle runs on a worker thread. It is a cross-process call that can block, and every caller is a controller read loop running at report rate. Closes #71. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>v1.17 v1.16 |
||
|
Dylan DeverillandClaude Opus 5
|
29f7e1e50e |
Remove the vendored ViGEmClient copy that is not built
third_party carried two copies of ViGEmClient. CMake compiles b66d02d; the
1.16.18.0 directory is referenced by nothing outside itself — not the build,
not a script, not a document.
It has already cost one misdiagnosis. #66 was filed reporting that a log line
was untrustworthy, on the grounds that vigem_target_add could not return
VIGEM_ERROR_TARGET_NOT_PLUGGED_IN — reading the copy that is not built. In the
copy that is, it can and does:
|
||
|
Dylan DeverillandClaude Opus 5
|
9a7f90d7f9 |
Keep a log line from ending at its first non-ASCII character
Both logs format wide strings with %ls, which fprintf converts through the C locale. The default locale encodes nothing above ASCII, and it does not fail loudly when it meets something: it abandons the rest of the line. Measured with the same call shape EventLog uses — narrow format string, %ls argument, FILE* stream: default : [before utf8ctype: [before — after éü] The first line is the whole output. No closing bracket, no newline, no error. This surfaced on a bus report ending in an em dash, which logged its opening clause and stopped mid-sentence, and was patched at that one call site. The flaw is not confined there. Every value in these logs worth reading is one that can carry non-ASCII: a game's name in PROFILE, the foreground window and executable path in INJECT, the process holding the controller and the veto name in ACQUIRE, and the same holder lists again in the helper's cycle.log. Device paths are ASCII by construction and were the only reason this went unnoticed. Setting LC_CTYPE at both entry points fixes every site at once, including sites not yet written, which converting the fifteen existing call sites would not. LC_CTYPE alone, so number and date formatting keep the C locale's behaviour, and at the top of wWinMain so it lands before the first log line and before any thread exists to race it. The local converter added with the bus report goes away with it — one mechanism rather than two, and no call site left doing it differently for reasons nobody will remember. No console fallout: both binaries are WIN32 subsystem with nowhere to print, and the console probes are separate executables that do not set the locale. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
c7866ecfa7 |
Check that ViGEmBus actually installed instead of assuming it did
With another program's rebranded ViGEm already on the machine, setup's driver install silently does nothing and reports success. The user finds out when they click Enable Steamless Mode and no controller reaches the game. Reported in #68, where the only ViGEm driver package left afterwards was the Oculus fork's; running the ViGEmBus installer by hand then offered Modify/Repair/ Remove rather than a fresh install, so the bundle had registered itself without a working driver ever landing. Everything downstream of that — the connect failures, the retry storm, and on 1.13 the device cycling that stranded a reporter's HID collections — starts here. The preceding commits make that state survivable and diagnosable; this one keeps it from being reached. Setup now asks the same question the app asks at runtime: is a device publishing the ViGEm bus interface present? Every cheaper check was measured and found to lie. The bundle exits 0 when it detects a related product and declines to act, so the exit code says nothing. The registry records the product whether or not a driver landed. And the driver's version is no guide either, since 1.22.0 ships a binary still stamped 1.21.442.0 — which cost an afternoon before it was pinned down. ViGEmBusProbe carries the check, and is installed rather than left a build artefact: its other job is being the thing to ask a user to run when a controller never reaches a game. Exit code 0 means at least one usable bus, 1 means none. Verified both directions against a real bus and a disabled one. A failed check does not fail the install. The app is fine, and it now picks the driver up on its own once it appears, so the message says what to install and that no reinstall is needed afterwards. Silent installs log the finding instead of stopping on a dialog nobody is present to close. It informs rather than repairs. Running msiexec /fa or the bundle's own repair would be acting on shared-identity MSIs on the one kind of machine known to have other products' drivers registered against them, which is not a risk worth taking to save a download. The installer code itself is unverified: there is no Inno Setup on the machine this was written on, so the Pascal has not been compiled and no install has been run. It also assumes ssPostInstall follows the non-postinstall [Run] entries. That assumption fails loudly if wrong — every healthy install would warn — which is why the check sits there rather than behind a wrapper built to guarantee the ordering. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>1.15 |
||
|
Dylan DeverillandClaude Opus 5
|
af1c815cf5 |
Stop treating a missing ViGEm bus as a contested controller
A virtual pad that could not be created was reported as GameModeOutcome::Blocked — the same answer the manager gives when another process holds the controller. The acquire path answers Blocked by restarting the controller's devnode to invalidate somebody's handle, so a ViGEm problem drove device cycles at hardware with nothing wrong with it. That is what left a reporter's puck collections stuck at problem code 22 in #65. VirtualPadUnavailable is now its own outcome, and the acquire path returns on it ahead of all the contention machinery: no retry counter, no cycling, no holder scan. Measured against a disabled bus with Steam closed, so that contention could not account for the result: zero device cycles across the whole episode, where the previous build spent three. Blocked still outranks it in the aggregate, because contention is the one of the two that a cycle can actually resolve. The failing slot releases its exclusive handle. It claimed the device and switched lizard mode off before discovering it had nowhere to send input, then held that claim for nothing — locking Steam out, and making the app itself the handle that vetoes its own device cycle, which is the veto #65's logs recorded. The driver notice fires once. The acquire burst re-entered EnableGameMode eleven times in half a second and every pass re-armed the balloon, which is the toast storm reported in #68 — Windows was still replaying queued toasts minutes after the app had been closed. The burst is called off as soon as a missing pad is the answer, the latch is cleared only by a pad that actually came up, and this path retries every 30 seconds rather than every four. Retrying no longer disturbs the controller. Learning that the bus is still absent by attempting an enable costs an exclusive claim and a lizard mode toggle every time, so once the absence is established it is re-established by enumerating, which touches no device at all. The tray menu re-enumerates before it is drawn. Releasing clears the slot list and only a device event refilled it, so after a manual disable the toggle grayed itself out as "(no controller detected)" and stayed that way until the app was restarted — #68's third item, and why a restart was the only recovery. The bus report reaches the log as UTF-8. EventLog's %ls goes through fprintf, which converts through the C locale and abandons the line at the first character it cannot encode: the report logged as "no virtual gamepad bus is present" and stopped dead at the em dash that followed. Converted at this call site only — the remaining %ls arguments carry device paths and process names and have the same flaw wherever those are not ASCII. Recovery no longer needs a restart. With the retry running, re-enabling the bus brought game mode up 30 seconds later with no relaunch and no clicking. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
255f4f9cad |
Ask the system which virtual gamepad buses exist instead of guessing
vigem_connect returns VIGEM_ERROR_BUS_NOT_FOUND both when no bus driver is installed and when one is present and did not answer, and the client never records which product's bus it tried. The app inferred "driver missing" from that error, so a driverMissing=1 that happened to be right was indistinguishable from one that was wrong — which is most of why #65 took as long as it did to pin down. ViGEmBusInfo enumerates the interface GUID directly. The failure line is now followed by what the machine actually has, and driverMissing is set from an empty enumeration rather than inferred from an error code with several causes. The GUID comes from the vendored client's own header rather than a copy, so updating third_party cannot leave a stale value behind here. That matters because the header instructs forks to change it, and several products ship one: measured on the #68 reporter's machine, which carries both the Oculus and Virtual Desktop forks, neither publishes this GUID and only the Nefarius bus appears. An empty result therefore does mean no usable bus, which is what makes it safe to drive the "go install the driver" advice from. It also retires the theory that the client had been connecting to a fork. A disabled bus reads the same as an absent one — the devnode stays present and reports a problem, but its interface is withdrawn entirely — so that case belongs in the empty-list wording rather than in a state of its own. ViGEmBusProbe prints the same inventory, since "vigem_connect failed" is not a question a reporter's log can answer on its own. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
c4efa63328 |
Split the remap page's script across a third string literal
The per-game profile work grew the script half of the page to 26 366 bytes, and MSVC truncates any single string literal past ~16 380 with C2026, so main did not compile at all in a release build: RemapWindow.cpp(647,1): error C2026: string too big, trailing characters truncated The 65 535 figure the comment cited is the limit on the concatenated result, not on each piece, which is what let the literal grow past the cap without anyone noticing. Corrected that, and split the script at the per-game picker so both halves sit near 13 K. The whole document still joins to roughly 40 K, well inside the concatenation limit. Nothing about the page itself changes — a raw literal boundary is not content, and the newline it introduces falls between two functions. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
459483dee0 |
Bump installer version to 1.15
Covers the per-game control profile work since 1.14. Paddle bindings, both trackpads and the virtual pad's platform are configured in one window and can be saved per game, replacing the tray checkboxes that applied to everything at once. The saved profile follows whichever game is in front, and key bindings can now carry Ctrl, Alt, Shift and Win. Existing settings are read forward rather than reset. Values written by earlier builds decode unchanged — the modifier field occupies bits that were always zero — and the retired trackpad toggles are converted once into the per-pad equivalent they described: the pad that drove the mouse becomes a pointer whose click is the left button it was always hardwired to, and on PlayStation the other pad keeps feeding the DS4 touchpad, which is what it had been doing all along. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
a087c236b2 |
Add per-game control profiles
Everything about how the controller behaves — paddle bindings, both trackpads, and which kind of pad the game sees — is now configured in one window and saved per game, rather than spread across tray checkboxes that applied everywhere at once. Installed games come from the Start Menu and the packaged-app list. Steam does not shortcut its library with .lnk files, so its games are read from the .url shortcuts it does write; Xbox and Store titles register no shortcut file at all and come from the shell's app list instead. None of these sources can tell a game from any other program, so the picker searches rather than scrolls — a saved profile is pinned, everything else is reached by typing. Trackpads are configured per pad. The previous pair of toggles could only ever describe one pad doing one thing, because a single TrackpadMouse instance was switched between pads by a bool; there are two now. A pad can move the pointer, scroll, feed the DS4 touchpad, or do nothing, and its click is an ordinary binding dispatched down the same path as the paddles — it used to be hardwired to a left click that only fired while that pad drove the mouse. Key bindings can carry Ctrl, Alt, Shift and Win. The modifiers ride in the free top byte of the packed binding, so values written by earlier builds decode unchanged. Windows is chosen from a toggle rather than pressed: the Start menu opens before the page is told anything, which is what made that key unbindable before. Choosing modifiers rather than typing them also keeps combinations like Alt+Tab from being claimed by the system mid-capture. Ctrl+Alt+Delete is refused outright, being the Secure Attention Sequence that no ordinary process can send. Profiles follow the foreground application, and nothing polls to do it. An unelevated process cannot subscribe to process-start events — the WMI class that reports them requires administrator rights, as does an ETW kernel session — but SetWinEventHook reports foreground changes for the cost of one hook and no thread. Focus is also the better question than "is it running": a launcher that starts a game and stays resident answers yes long after the game has exited. Steam profiles are keyed by app id, which a foreground window cannot supply, so Steam's manifests are read to map that id to an install directory; a game's launcher, its anti-cheat service and the game itself all live under it and all resolve to the same profile. Auto Mode still decides whether the controller is ours at all, so a profile only loads while we hold the pad. A balloon names the profile when one loads, under the notification setting, because a profile that quietly fails to match is otherwise invisible. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
838cd7ae68 |
Widen the disable settle window to what a receiver actually needs
Measured on a four-slot receiver (PID 1304), the gap between ConfigFlags landing in the registry and the devnode admitting to problem 22 ran from 96 ms to 330 ms across its slots, on an idle machine: 22:27:30.533 MI_03=0/1 flag written, node still reports healthy 22:27:30.863 MI_03=22/1 node catches up, 330 ms later The previous window was 450 ms, which covers that measurement and very little else. Judging a disable inside the gap is the failure this window exists to prevent — it reads as "the disable was refused", which skips the re-enable and clears the pending record while the flag is already committed, leaving a devnode switched off with nothing tracking it. A slower or busier machine has no headroom there, so size it for the bad case: ~950 ms. It is only ever spent on the path about to conclude the disable failed, and a device that goes down promptly still pays none of it. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>v1.14 |
||
|
Dylan DeverillandClaude Opus 5
|
bda3a24756 |
Stop the app from cycling the controller to escape its own handle
Reported against 1.13 (#65): the puck's vendor collections sat at problem 22 after the app cycled them, and did so again within seconds of every boot because the acquire runs at every startup. The holder scan in the user's log names the blocker plainly: HOLDERS: all=[SteamlessController.exe (7696)] reportable=[none] FELL BACK TO RESTART (transport=dongle err=50 veto=open handle ...) The app releases its handles before asking for a cycle, but it reopens too soon. TryAcquireController opens the device at the top, every arrival the cycle produces re-enters it, and the existing spacing check sits below that open — it only stopped a second cycle stacking on the first, never the reopen. PnP does not care that the handle vetoing the teardown belongs to us. So the cycle is refused, and each refused cycle is a chance to leave the devnode switched off. Hold the claim off while a cycle is in flight. Raised where cycles are started so the recovery path is covered too, cleared by the arrival that ends the cycle so the reopen race against Steam is not slowed — measured at 61 ms from arrival to game mode, with ten re-entries suppressed in between. The helper also reports when every handle it found was one of ours, and the app then releases rather than cycling again; steam.exe is deliberately excluded, since it holds the device by design and yields. Two further defects the report exposed: A disable was judged by reading the devnode's status once, immediately. Read too early it says "nothing went down", which skips the re-enable entirely and clears the pending record, so a disable landing a moment later is left in place with nothing tracking it and a cycle verdict claiming the opposite — which is how a FELL BACK TO RESTART verdict and a disabled collection came to coexist. Poll for it to settle instead, only on the path about to conclude the disable was refused. Enabling in place does not always work. A node can sit at problem 22 with ConfigFlags clear, and for those DICS_ENABLE reports success and changes nothing however often it is asked, while rebuilding the node does work. Fall back to remove and re-enumerate, in the cycle and in the recovery path. Confirmed enabled after the rebuild, not merely removed: a persistent ConfigFlags disable is read again by the rebuilt node and disables it just the same, and treating removal as proof would report success over a dead device and discard the record that is the only way back. Verified by forcing the in-place enable off — the rebuild ran and correctly reported STILL DISABLED rather than a false success. Not covered here: vigem_connect failing against the Oculus fork of the ViGEm bus, which is what kept the app retrying long enough for this to bite. That is a separate issue. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
197a7b0f58 |
Bump installer version to 1.14
Covers the disabled-devnode work since 1.13. A device cycle switches the devnode off and back on about a second later, and the disable is global — it lands in the registry at once and outlives the process, a replug, and a reboot. Killing the app inside that second left the controller switched off with nothing running to put it back, recoverable only through Device Manager. The intent to disable is now recorded before the disable and cleared only once the devnode confirms it is back, so an unfinished cycle is undone on the next run. A devnode disabled by something else — Device Manager, regedit, pnputil, a tool that hides HID devices — is no longer indistinguishable from an empty USB port. It is named at startup and can be switched back on from the tray rather than silently, since it may well be deliberate. Also adds a heartbeat to the event log. Idle and stopped previously looked identical there, which is what made the fault behind this release hard to read after the fact. Still the only place carrying the app version, and nothing in the app compares version strings. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
9e289a2178 |
Log a heartbeat so an idle app and a stopped one stop looking alike
Nothing is written while the app sits idle with nothing changing — AUTO lines fire on Steam state changes, not on state — and nothing is written while it is wedged or dead either. A gap in the log therefore means either "healthy, with nothing to say" or "stopped working", and after the fact there is no way to tell which. That is not a hypothetical shortcoming. Diagnosing a controller that had gone missing turned on a 33-hour gap that read as a hang and was very likely an idle app with a game in the foreground, and nothing in the record could settle it. Beat every 15 minutes with the state that matters: connected, game mode, Steam state, auto mode, uptime. Idling for a month costs a fraction of the log's 512 KB, and the resolution bounds a stop to within a quarter hour, which only has to beat "somewhere in the last 33 hours". Driven by the window timer deliberately. A background thread would prove only that the process still exists; this proves the message loop is being pumped, which is what "not hung" means here — the acquire and release paths both run on it and both can sit inside SetupAPI for seconds. A beat that arrives late also says so. Lateness is measured with unbiased interrupt time, which does not advance while the machine is suspended, so a lid being shut does not read as a stall. Measured by suspending the UI thread for 20 s: one beat on resume reporting 23 s blocked, timers having coalesced rather than fired four times. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
25e4f86446 |
Put back a controller devnode left disabled by an unfinished cycle
A cycle disables the devnode and re-enables it about a second later. The disable is global, so it lands in the registry immediately and outlives the process that asked for it — and a replug, and a reboot. If the process does not survive that second, the controller is left switched off with nothing running to put it back, and recovering it needs Device Manager. The window is not theoretical. On the elevated path the cycle runs inside the long-lived tray app rather than the second-long helper, so being killed mid-cycle is an ordinary thing for a user to do. Measured by ending the helper the instant ConfigFlags went to 1: the HID collection stays disabled across replug and reboot, and nothing in the app notices. Record the intent to disable before performing it, committed to disk, and drop the record only once the devnode confirms it is back. What is left behind is an unfinished cycle, and the next run undoes it — in process when elevated, through the helper's scheduled task otherwise, since enabling a devnode needs administrator rights. A cycle that ends stranded deliberately leaves its record armed. The helper also reconciles before cycling: a switched-off controller enumerates nothing, so otherwise the cycle finds no interfaces and silently does nothing, which looks exactly like a controller that is not plugged in. The record names the devnode and its parent. The disable is aimed at the HID collection child and lands there, but recording both costs nothing and keeps recovery independent of that. A devnode can also be disabled by something else entirely — Device Manager, regedit, pnputil, a tool that hides HID devices. That is not undone silently, because it may well be deliberate. It is no longer invisible either: it publishes no interfaces, so from in here it is indistinguishable from an empty USB port, while the device still lights up and still chimes on arrival. Name it at startup and offer a tray item to switch it back on, which routes through the same recovery path rather than growing a second way to enable a devnode. Also log the cycle on the release path. It announced nothing at all, so a device left disabled by an interrupted release appeared to come from nowhere — the acquire path has always said what it was doing. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
a46b169f29 |
Bump installer version to 1.13
Covers the input-blocking work since 1.12: the app now detects when Windows is discarding what it injects — an elevated window in the foreground blocks it, and haptics keep working throughout, so the controller looks alive while the trackpad and buttons do nothing — names the application responsible in a tray balloon, and records enough in the event log to tell that apart from a cursor clip, a filtering hook, or a state report too short for the trackpad path. 1.12 was cut from a working-tree edit that was never committed, so it has no commit of its own and the version history steps 1.11 to 1.13. The released 1.12 build carried the same code as 1.11. Still the only place carrying the app version, and nothing in the app compares version strings. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>v1.13 |
||
|
Dylan DeverillandClaude Opus 5
|
09fe805119 |
Tell the user when Windows is discarding the input this app injects
Users reported the trackpad going dead after a takeover — haptics still
firing, cursor not moving — and recovering the moment focus moved to
another window or the window in front was closed. Haptics are HID writes
and the virtual pad is ViGEm, so both keep working regardless; SendInput
was the one stage of the pipeline that could fail and leave no trace.
Reproduced on this machine with an elevated Device Manager in front:
- 137 px of movement injected, cursor pinned at (910,781) for 5s
- foreground mmc.exe at High integrity, this process at Medium
- SendInput returned success for every one of those events
That last point decides the shape of the fix. UIPI is documented as
failing the call with ERROR_ACCESS_DENIED, but on Windows 11 26200 the
call reports success and the input is dropped afterwards, so no
return-value check can see it. The condition is only observable by asking,
before injecting, whether the foreground window outranks this process.
So that is what Send now does, on every event:
- one check covers cursor movement, trackpad clicks, paddle buttons and
keys, none of which can be judged from their own result. A blocked
paddle press previously produced no evidence anywhere
- the foreground lookup is cached against the window it describes;
injection happens at report rate, foreground changes do not
- the state changing is what gets logged, not the events: one line
naming the blocker, one when it stops
- the tray balloon says which application is blocking input and that
switching away from it restores the controller, rate-limited to one
per blocker per minute so alt-tabbing cannot spam it
- the check never gates the event. It observes and reports; if it is
ever wrong about a system, the input goes out exactly as before
A foreground process whose token cannot be read is deliberately not
named. From Low integrity, OpenProcess succeeds while OpenProcessToken
fails with ERROR_ACCESS_DENIED, and protected processes refuse it too —
so an unreadable level means unknown, not blocked. Guessing would put a
wrong application's name in front of the user; the snapshot records
"unreadable" instead.
Backing this up, for the causes that produce no error and no integrity
mismatch: injected movement is checked against the cursor actually
moving, which is what catches a cursor clip a game left behind or a hook
filtering injected events. It stays quiet whenever a blocker is already
known, so it cannot contradict the better explanation. A refused call is
still logged with a dropped-event count and a recovery line, as a cheap
catch-all for systems where the call does fail.
The read loop also logs the shape of the first state report it gets.
TrackpadMouse requires 30 bytes where the haptics need 28, so a short
report would produce working haptics and a dead cursor — the same symptom
from a completely different cause, and not one that can be told apart
from a user's description. This machine reports 0x42 at 54 bytes over
USB, which rules it out here but not on transports we have less evidence
for.
EventLog::Write is annotated _Printf_format_string_ so a mismatched
specifier is caught by /analyze. It is not enforced in the normal build,
but these lines are read long after the fact by someone trying to explain
a fault they cannot reproduce, and the paths that log least often are
exactly the ones no test run exercises.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
||
|
Dylan DeverillandClaude Opus 5
|
e1e252619a |
Bump installer version to 1.11
Covers the two fixes since 1.10: trackpad haptics were playing on the opposite pad, and a failed controller handoff now reports which process is actually holding the device rather than assuming Steam is. Still the only place carrying the app version, and nothing in the app compares version strings. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>v1.12 |
||
|
Dylan DeverillandClaude Opus 5
|
350161785d |
Report what is actually holding the controller instead of blaming Steam
When an exclusive claim failed, the app assumed Steam was refusing to let
go, burned three device cycles over ~26 seconds, and told the user to
close Steam. All of that could be wrong: any process with the device open
blocks the claim, and no amount of cycling evicts it.
Measured while a process held an open handle and never yielded:
- disabling the HID collection is vetoed (ERROR_GEN_FAILURE, ~6s)
- query-removing it is vetoed (PNP_VetoOutstandingOpen, ~5.4s)
- query-removing the USB parent is vetoed the same way, naming the
child as the blocker
Consent propagates up the tree, so no devnode teardown reachable from
user mode breaks such a handle. The cycle works against Steam only
because Steam cooperates: it yields on DBT_DEVICEQUERYREMOVE and reopens
on arrival, which is why the reopen is a race rather than a standoff.
Against an uncooperative holder there is no eviction path, only
detection — so detection is what this adds.
- CM_Query_And_Remove_SubTreeW reports why a teardown was refused, and
the verdict crosses to the tray app through a status file
- a cycle that broke no handles stops the retry loop immediately, since
repeating it cannot do more than the last one did
- the handle table is scanned to name the processes holding the device,
which is the only actionable output left when eviction is impossible
- messages say what was established, and suggest a replug: a physical
disconnect is a surprise removal that no veto can stop
The event log carries the unfiltered holder list and the scan's coverage,
not just the conclusion — an empty result can mean nothing was holding
the device, or that the scan never reached it, and that difference cannot
be recovered from a user's machine after the fact.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
v1.11
|
||
|
Dylan DeverillandClaude Opus 5
|
cb6b1f3cc6 |
Fix trackpad haptics playing on the opposite pad
The side byte on the haptic output reports was inverted: 1 is the right pad and 2 is the left, not the reverse. Both the click pulse and the movement tick went out through the same helper, so both landed on the wrong actuator. This survived since the haptics went in because it is imperceptible with one thumb down — an LRA firing in the opposite grip still reads as "the controller ticked" and gets attributed to the finger that is moving. It takes a second contact point to localise. The values are now named constants carrying the confirmed mapping so an inversion cannot slip back in silently. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
42f4a23850 |
Allow back paddles to be bound to middle and thumb mouse buttons
Kind::MouseButton was reserved for these in the binding refactor but had no emission arm and no way to pick one. Wire both up: middle click, and the two thumb buttons that games label Mouse 4 and Mouse 5. They were previously unbindable by any route. Middle and the thumb buttons can be captured by pressing them, since none is needed to operate the remap window. Left and right deliberately cannot — the user has to keep clicking Rebind and Cancel while a row listens — so those stay pickable from the catalog instead, as before. Capture suppresses the browser defaults these carry: autoscroll on middle, and history navigation on the thumb buttons. The window's HTML had to be split into two concatenated literals. It crossed MSVC's ~16KB limit for a single string literal (C2026) at 17KB, which fails the build outright rather than truncating quietly. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
7b2391dd95 |
Repeat held key bindings at the system typematic rate
Holding a paddle bound to a letter produced exactly one character. The emission loop only acts on edges, so a held paddle sent one key-down and nothing further until release. That was the correct consequence of the design rather than a fault in the send path: auto-repeat is not something Windows synthesises for a key it believes is down. A physical keyboard repeats because its own firmware keeps sending make codes while the key is held, and Windows forwards each one. Injected input has no firmware behind it, so the repeat has to be generated here. Re-send the press while the paddle is held, after the user's configured delay and at their configured rate — SPI_GETKEYBOARDDELAY and SPI_GETKEYBOARDSPEED — so a paddle behaves like the key it is bound to rather than at a rate invented here. Both are captured at press time, so the rate cannot shift mid-hold. The read loop already ticks at report rate, so this is a per-frame elapsed check with no new thread or timer. Keys only. Mice do not repeat when held, so mouse-button bindings keep their one-shot behaviour, and gamepad bindings are held state in the pad report, where a repeat would mean nothing. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
b5b7e28cf1 |
Allow back paddles to be bound to keyboard keys
Press-to-bind only listened to the controller, so the remap window could offer gamepad buttons and mouse clicks but nothing from the keyboard. Capture a key press while a row is listening and bind that instead. Capture happens in the remap page's keydown handler rather than through a keyboard hook. The page is already focused while listening, and a low-level hook is a heavy, AV-bait mechanism for a convenience feature. The page reports the physical KeyboardEvent.code, which KeyInput turns into a virtual-key code, so a binding made on one layout means the same physical key on another. Escape stays the cancel affordance and is deliberately not bindable. Keys are delivered with SendInput as scan codes rather than virtual keys, because software that reads scan codes directly sees nothing useful otherwise. Note that injected input does not reach every game — anti-cheat and some input paths ignore it — so this is realistically a desktop, emulator, and browser feature. Gamepad bindings remain the answer in game. Extended keys are flagged from an explicit set. MapVirtualKey is documented to report the 0xE0 prefix under MAPVK_VK_TO_VSC_EX but does not do so for the navigation cluster, which shares scan codes with the numpad — a test caught Up arrow being delivered as numpad 8. Labels come from GetKeyNameText against the active layout, so a non-US keyboard shows its own key names, and are JSON-escaped on the way to the page: the backslash key really is named "\". Paddle release tracking now records what the press actually sent instead of re-reading the binding on release, so rebinding a paddle mid-hold releases the input that was pressed rather than stranding it down. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
0839f79f16 |
Represent paddle bindings as a tagged value, not a flat enum
Keyboard and extended mouse bindings can't be enumerated the way gamepad
buttons can — a key binding is a value drawn from ~256 virtual-key codes,
not a member of a closed set. BackButtonConfig held four bare
BackButtonAction values, so there was nowhere to put one.
Introduce BackButtonBinding, a kind/code pair covering built-in actions,
keyboard VK codes, and extended mouse buttons. No behaviour change: only
Kind::Action is produced or consumed today. This is the groundwork for
capturing keys and mouse buttons in the remap window.
Both formats extend without a migration:
- Registry: packed as (kind << 16) | code. Every value written by earlier
builds is a bare BackButtonAction of 0-18, so its kind bits are already
zero and it decodes as Kind::Action. Action bindings still pack to their
bare value, so a downgrade also reads settings written by this build.
- WebView2: built-in actions keep their bare ids ("A", "leftMouse"), and
the open-ended kinds are prefixed ("key:32", "mouse:x1"). Flat strings
keep the hand-rolled JSON reader in RemapWindow usable, and the JS input
catalog needs no changes.
Unrecognised values — corrupt, or written by a future build that knows
kinds this one does not — degrade to None rather than being applied.
Verified the invariants with a standalone test over the full value space:
legacy decoding, pack and id round-trips for every kind, rejection of
malformed input, and the shipped defaults.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
||
|
Dylan DeverillandClaude Opus 5
|
ba215b2429 |
Fix custom back button mappings being ignored in mouse click mode
"Back Buttons as Mouse Click" hijacked L4 and R4 unconditionally: the virtual controller skipped their bindings entirely, so a paddle mapped to RT silently did nothing while the option was on. Only the upper paddles were affected, so the symptom read as "two of my four paddles are dead" rather than pointing at the toggle. The same gate also suppressed the Left/Right Click bindings the remap window already offered, meaning the feature disabled the very capability it existed to provide. The toggle was a strict subset of the remap window, which already offers left and right click per paddle, so retire it rather than teach it to cooperate: - Remove the backMouseEnabled gates, the tray item, and the duplicate L4/R4 handling in TrackpadMouse. Paddle mouse output now flows through the single edge-detection path in ReadLoop. - Default L4 and R4 to left click so a fresh install still clicks out of the box. - Add an explicit Off option to the picker. None was already a real "do nothing", but Reset now yields left click, so without this there would be no way to disable a paddle at all. - Migrate BackButtons=1 into real bindings, filling in only paddles with no binding of their own so an explicit mapping is never overwritten, then delete the value so it runs once. TrackpadMouse::Reset used to release a paddle held while game mode was torn down. Removing its L4/R4 handling would have stranded the mouse button down, so that cleanup moves to ReleasePaddleMouseButtons, which also covers right click and runs on the enable, disable, and destructor paths. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
0833d51868 |
Bump installer version to 1.10 (#54)
Covers the two fixes merged since 1.9: manual mode can now take the controller back from a running Steam (#53), and a failed ViGEm target add no longer orphans a pad that holds an XInput slot until reboot (#52). The installer script is still the only place carrying the app version, and nothing in the app compares version strings -- AppVersion is a display and registry value only, with no VersionInfoVersion field derived from it. So 1.10 sorting below 1.9 lexically has no effect beyond how it reads in Programs and Features. Co-authored-by: Claude Opus 5 <noreply@anthropic.com>V1.10 |
||
|
Dylan DeverillandClaude Opus 5
|
9ef3609da5 |
Fix orphaned ViGEm pads left behind by a failed target add (#52)
vigem_target_add() plugs the child device in with IOCTL_VIGEM_PLUGIN_TARGET, then waits for it to come up with IOCTL_VIGEM_WAIT_DEVICE_READY. When that wait fails, the cleanup path calls vigem_target_remove() -- but target->State is only set to VIGEM_TARGET_CONNECTED on the two success paths above it, so the removal trips its own "State != VIGEM_TARGET_CONNECTED" guard and returns without ever sending IOCTL_VIGEM_UNPLUG_TARGET. The device stays plugged into the bus with no owning process, holding an XInput slot until the machine reboots. Observed locally: a pad created at 19:57:25 outlived the app by five hours and several sessions, and every later session had to take the next serial. The rapid enable/disable churn during device cycling is what triggers the wait timeout in the first place. Mark the target connected before the cleanup so the unplug is actually sent, and pin the return to VIGEM_ERROR_TARGET_NOT_PLUGGED_IN -- a successful unplug would otherwise report VIGEM_ERROR_NONE and register a target in pTargetsList that is no longer plugged in. This is a local patch to the vendored b66d02d ViGEmClient snapshot; upstream is archived, so no future sync will bring it in. Also: - VirtualController records which ViGEm call failed and its error code; ControllerManager logs both. The old driverMissing line could not distinguish a missing bus from a failed target add, which is exactly the distinction needed to spot this leak. - ShowViGEmBalloon() now latches the way ShowElevationBalloon() already did. Every failed enable calls back into UpdateTrayIcon -- several times a second during an acquire-retry loop -- so an unavailable driver produced a wall of identical balloons. The latch clears once the driver is reachable again so a genuine mid-session disappearance still warns. UpdateTrayIcon no longer returns early on vigemMissing, so the icon and tooltip keep tracking the controller instead of freezing. - Reword that balloon: a disabled bus devnode reports the same VIGEM_ERROR_BUS_NOT_FOUND as an uninstalled one, and "is not installed" sends users looking in the wrong place. Co-authored-by: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Opus 5
|
94f1bf94dd |
Share one acquire/release path across manual and auto modes (#53)
Manual mode could not take the controller back from a running Steam. The tray toggle called ControllerManager::EnableGameMode() once and discarded the result, so a Blocked outcome -- Steam holding the exclusive write handle -- was simply dropped. The auto modes wrapped the same call in TryAcquireController(), which escalates a blocked claim to a device cycle via the elevated helper and retries on re-arrival. Only closing Steam freed the pad in manual mode. The two paths diverged because acquiring is asynchronous: a blocked claim fires a cycle through Task Scheduler and comes back seconds later as a WM_DEVICECHANGE, so the retry loop lives in timers and message handlers rather than on the stack. Each of those re-entry points has to re-answer "do we still want control?", and the code answered it by recomputing m_autoMode != Manual && WantControl(steamState) -- deriving intent from Steam. That works in the auto modes, where Steam's state is the intent. Manual mode had nothing to derive from: its intent lived in a menu click that was never recorded anywhere. WantControl() said as much, returning false for Manual with the comment "tray toggle decides" -- and the tray toggle's decision went unstored. Store it. m_wantControl replaces the derived guard at all three re-entry points (the IDT_ACQUIRE and IDT_WAKE_POLL timers, and the arrival branch of WM_DEVICECHANGE), leaving WantControl() with a single caller in ApplySteamState. The toggle sets it and calls the shared path, so manual mode gets the retry burst, device cycling, wake polling, and the failure balloon it never had. Release is unified the same way: ReleaseControl() lifts the yield branch out of ApplySteamState, substituting m_steamWatcher.GetState() for the SteamState parameter the manual caller has no access to. Manual toggle-off now also cycles the devnode, so a running Steam sees an arrival event and re-adopts the pad instead of leaving it unclaimed. Three follow-on details: - Manual toggle-on calls EnsureCycleTaskRegistered(), matching what IDM_MODE_GAME already does, so the one-time UAC prompt lands on a menu click rather than ambushing the first cycle. - SetAutoMode(Manual) adopts IsGameModeActive() rather than inheriting a stale m_wantControl from the auto mode being left. - The toggle's menu label tracks m_wantControl in manual mode, not the live handle state. Acquisition can spend several seconds cycling, and a second click during that window should cancel rather than start over. Both AUTO: log lines in the acquire path are now reachable from manual mode, so they are relabelled ACQUIRE: -- these are exactly the lines read when diagnosing a failed handoff, and the old prefix pointed away from it. Co-authored-by: Claude Opus 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Fable 5
|
f2da1ce31e |
Bump installer version to 1.9 (#50)
Matches the README, which advertises Bluetooth support as new in 1.9. The installer script is the only place carrying the app version. Co-authored-by: Claude Fable 5 <noreply@anthropic.com>v1.9 |
||
|
|
345bf15ca3 |
Fix puck activation after firmware updates (#44)
The receiver stopped working after a controller firmware update. Original diagnosis and fix by @danbosscher; extended here for Bluetooth, Steam handoff, and several defects found during QA on the hardware. The puck publishes a controller interface per slot plus a management endpoint, and on current firmware an empty slot rejects the lizard-mode command outright. - Enumerate only the controller collection (FF00:0001), excluding the puck's management endpoint at FF00:0002 - Wait for a live 0x42/0x45 state report before sending feature commands, and skip slots that never produce one - Prefer an exclusive claim but accept a shared handle when Windows keeps a compatible write handle open — only while Steam is absent, since a handle held with Steam running is probably Steam's, and driving the controller alongside it is worse than refusing. Refusing is what escalates to the device cycle that takes the handle back. - Add the Nereid receiver PID (0x1305) Defects found by QA and fixed here: - The battery payload is [chargeState, percent] on every transport, not just Bluetooth. A 92% battery was reported, and fed to the DS4 battery level, as 1%. - Game mode could not tell an idle receiver from a contested one, so a controller that was merely switched off triggered three device cycles, restarting the receiver's devnodes for nothing. - A receiver raises no device event when a slot becomes occupied, so waking a controller went unnoticed for as long as it was left. Poll for it instead. - Overlapping cycles undid successful claims: a cycle is asynchronous, and the arrivals it generated re-entered the acquire path and fired another on top of it, ripping the device back out a second after game mode came up. Cycles are now spaced. - Failure was announced while the last cycle was still landing — the alert fired one second before success. The verdict now waits out a grace period. Verified on the puck, Bluetooth, and wired: handoff to Steam on game launch, reclaim afterwards, and no duplicate input. Reclaiming takes around nine seconds on a four-slot receiver, dominated by SetupDiCallClassInstaller tearing down each interface in turn. Three attempts to shorten it (batching the cycle, a shorter settle, cycling only the contested interface) all regressed the reclaim and were reverted; it needs to be worked out against hardware rather than reasoned about. Co-Authored-By: Dan Bosscher <danbosscher@users.noreply.github.com> Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Fable 5
|
719f76a911 |
Scale the remap window for DPI and make it resizable (#47)
main.cpp declares PER_MONITOR_AWARE_V2, so Windows scales nothing for us, but the remap window was created with its 760x668 design size as raw physical pixels. WebView2 correctly renders its CSS at the monitor's scale inside that frame, so at 200% the content needed 1520x1336 in a 760x668 window — the reported "super small" window with cramped content. WS_THICKFRAME was also simply absent, so there was no sizing border at all. The existing WM_NCHITTEST handler already had a comment about letting Windows handle "the resize border", so that code was written expecting a resizable window that never was one. - Size from GetDpiForWindow, clamped to the monitor work area: 760x668 at 200% is taller than a 1080p screen - Centre on the monitor under the cursor rather than the primary one, which also picks up the right scaling on a mixed-DPI desktop - WS_THICKFRAME to resize. Its border is in the non-client area, outside the rect WebView2 occupies, so it stays grabbable even though the web content covers the whole client area - WM_GETMINMAXINFO for a scaled minimum, WM_DPICHANGED to follow the window between monitors at the physical size the user set Window class background changed from white to the page colour: resizing repaints the frame before WebView2 catches up, and white flashed against the dark UI once dragging borders became possible. Co-authored-by: Claude Fable 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Fable 5
|
7feb53f3ea |
Validate the cycle task points at the current helper
EnsureCycleTaskRegistered only checked that the scheduled task existed, never that it pointed at the current helper. A task left behind by an older or relocated install still runs and schtasks still reports success, but it executes a helper that may be gone or predate current device support — so the cycle does nothing, the controller is never reclaimed from Steam, and nothing anywhere reports a failure. Not theoretical: an uninstall left a task pointing into Program Files while the app ran from a build folder, and every cycle the app ordered was a silent no-op. From the app's side a successful schtasks /Run is indistinguishable from a working cycle, which made it expensive to find. Query the task as XML and compare its <Command> against our own helper path, re-registering on mismatch and logging that it happened. The XML declares encoding="UTF-16" while redirected output is single-byte, so it is read narrow; the path is compared against both codepage renderings rather than guessing which the console used. Also register at startup when the loaded mode is OffOnlyInGame, instead of waiting for a cycle to discover the problem. Without it a missing or stale task is only noticed during a cycle — which happens exactly when a game launches — so the UAC prompt ambushed the user mid-launch. A no-op when the task is present and correct, so installed users see no change. Installer users never hit any of this: setup runs --register under the elevation it already has. Verified by relocating the app: launching from the registered location is silent, launching from a copy logs the stale task and re-registers to the new path. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
Dylan DeverillandClaude Fable 5
|
50a8ae246d |
Add Bluetooth LE support for the Steam Controller
The controller now works wired, through the dongle, or paired over Bluetooth, with the full feature set: game mode, rumble, trackpad haptics, battery, PlayStation-mode gyro, and taking the controller back from a running Steam. Windows exposes the entire vendor protocol over HID-over-GATT under PID 0x1303 — feature reports, output reports, and the 0x45 state report (54 bytes, ~157 Hz) all work through the existing HidDevice stack, so support is a thin layer rather than a new transport backend. - Add SC2026_BT_PID to enumeration and a per-slot Transport tag, detected from the interface path (HOGP GUID / bthenum markers win over PID tokens) and logged on CONNECT/DISCONNECT - Swap the battery report 0x43 payload bytes on Bluetooth slots — the firmware sends [chargeState, percent], reversed from USB/dongle - Bluetooth-specific disconnect alert copy, and a notification when acquisition gives up rather than a silently dead pad Device cycling had to be reworked for reclaiming the controller from Steam. DIF_PROPERTYCHANGE reports success while leaving the device and Steam's handle untouched; CM_Query_And_Remove_SubTree is vetoed by open handles; pnputil /restart-device works once, then leaves the devnode reboot-pending and refuses thereafter. Disable+enable works repeatedly. Two traps found against live hardware. SetupDiCallClassInstaller returned ERROR_INVALID_DATA for a disable Windows had genuinely performed — the app logged the device removal a tenth of a second before the call reported failure — so return codes cannot gate the re-enable, or the controller is stranded disabled and needs Device Manager to recover. CM_Get_DevNode_Status is now the source of truth: whether the node went down decides if a real cycle happened, the re-enable always runs and retries until the node confirms it is enabled, and only a stranded device counts as failure. Some devices also reject a global scope change, so DICS_FLAG_CONFIGSPECIFIC is tried as well. SteamProbe becomes a transport discovery tool: detailed enumeration with transport detection, interface selection, non-fatal lizard-off with a raw-dump fallback, an output-report test, report-rate measurement, and --enum / --imu modes. The windowless cycle helper gains its own cycle.log, since it cannot share the tray app's events.log (held open _SH_DENYWR) and was previously impossible to diagnose. Verified on hardware: game mode, rumble, haptics, battery, IMU streaming, clean removal on power-off/sleep/range loss with automatic reconnect, yielding to Steam on game launch, and reclaiming afterwards (~1.2s). Also confirmed the firmware enforces one host transport at a time, so no cross-transport deduplication is needed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
Dylan Deverill
|
5c3f4aacf0 | Update README.md |