1820 Commits

This Branch

This Branch

All Branches

Author	SHA1	Message	Date
enricobuehler	ddc28de32a	fix(gamestream): tell the client when the HOST ends the session ... On-glass on .173: a launched game exited, the host tore the session down correctly — and the client froze on its last frame instead of returning to its app list. The host's half was right; it just never said anything. On the native plane ending a session closes the QUIC connection with APP_EXITED, which every client renders as "game ended". The compat plane has no such channel: `end_session` stops the media threads, and stopping a UDP sender is silence, not a signal. Moonlight holds the ENet control stream for the whole session, so with no word from us it sat there until its own timeout eventually fired. The control loop already reacted to the CLIENT disconnecting; it had no path for the reverse. It now watches for its session being cleared out from under it and disconnects the peer, which is what Moonlight reads as "the session is over" — landing the player back in the app list, the same place quitting puts them. This is the signal every host-side end needed, not just a game exiting: the management stop and a `/cancel` racing another connection had the same silence.	2026-07-26 19:44:55 +02:00
enricobuehlerandClaude Opus 5	1839d7566b	feat(packaging/winget): install the Windows host with winget, from our own source ... `winget install unom.PunktfunkHost` after adding the source, and `winget upgrade` from then on. Windows had no update path at all before this — no self-update, no package manager — so keeping a host current meant noticing a release and re-running an installer by hand. Ships the manifest trio in winget-pkgs' own format (so submitting upstream later is a copy, not a rewrite) plus a release-time generator that substitutes only version, URL, hash and release-notes link. Everything reviewable — the silent switches, the agreements, the installation notes — stays in the checked-in templates rather than buried in a generator. Silent installs deliberately take the SAME task defaults the wizard shows: a per-channel default is a support trap. The disclosures the wizard puts on screen travel as manifest Agreements instead, shown before install and requiring acceptance — VB-Audio's bundling grant wants the user to see VB-CABLE's origin and donationware status, and a silent install shows them nothing otherwise. The console password can only be pointed at, never printed: it is generated per install and the notes are fixed at publish time. Self-hosted rather than the community repo, which gates on Defender/SmartScreen validation the self-signed installer cert would not clear today. The source is three endpoints — /information, /manifestSearch, /packageManifests/{id}. The reference implementation's other twenty are its admin API for mutating a CosmosDB; a catalogue generated at release time has nothing to mutate. It runs on unom-1 as a stock bun image with the two .mjs files bind-mounted, the same shape as the flatpak server, so there is no image to build or pull. The catalogue is derived from the RELEASES rather than local files: winget resolves --version and upgrade against the version list, so a source that only knew the newest release could neither pin an older one nor show an upgrade path from it. That also leaves no state to drift — re-running the build reproduces it exactly. NormalizedPackageNameAndPublisher is declared unsupported on purpose. winget derives it client-side with its own normalization, and a near-miss silently mis-correlates an installed host; ProductCode is exact and Inno gives us one. 28 checks drive the handler directly, and CI gates on them before anything reaches the box — a wrong response shape does not fail loudly, it just makes winget report "no package found". Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 19:09:13 +02:00
enricobuehlerandClaude Opus 5	3d89301336	fix(installer/windows): a silent install can't hang on a prompt, and an upgrade keeps your choices ... Three faults that only bite when nobody is watching the wizard — which is how a package manager always runs us, and how a re-run by hand half-behaves today. The conflicting-host warning used a plain MsgBox. That ignores /SUPPRESSMSGBOXES and displays even under /VERYSILENT, so an unattended install onto a box already running Sunshine/Apollo would sit on a modal dialog nobody can see or click. SuppressibleMsgBox with an IDNO default aborts instead, which is the honest answer for a combination the message itself calls unsupported. The GameStream and Public-firewall choices were re-applied on every run. A silent run has no wizard, so every task falls back to its script default and the installer confidently re-asserted it: a user who enabled GameStream had it turned off again, and one who opened the firewall on Public networks had it quietly revoked (that task is default-unchecked, so the reset ran the other way). Both params are now fresh-install-only, keyed off whether host.env already exists. `service install` already read an absent --gamestream as "keep host.env as-is"; --allow-public-network becomes tri-state the same way, resolving an absent flag from the marker the previous install recorded. It is strict on the value: a typo'd =of must not fall through to a marker that may say true, because that turns a mistyped opt-OUT into leaving Public open. Also brands the installer "Punktfunk Host" in the strings a user actually sees — Add/Remove Programs, the Start Menu group, the wizard task text. Paths are untouched; renaming those would relocate installs and orphan existing config. Verified on the windows-amd64 runner: ISCC compiles the script (with WithWeb defined, so the password page parses too), and cargo check + clippy -D warnings pass for punktfunk-host with nvenc,amf-qsv,qsv. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 19:09:13 +02:00
enricobuehlerandClaude Opus 5	1ee06defa6	feat(library): descriptive metadata on every entry — platform, year, genres, and friends ... Emulation-and-beyond libraries need more than a title and a poster. Every library shape (GameEntry, CustomEntry, CustomInput, ProviderEntryInput) now carries a shared, flattened GameMeta: platform, description, developer, publisher, release_year, genres, tags, region, players. All fields are optional and flat on the wire, so existing library.json files, provider plugins, and clients keep working unchanged. - Installed-store scanners (Steam, Lutris, Heroic, Epic, GOG, Xbox) stamp platform=PC; custom/provider entries carry whatever was authored. - GET /library grows a ?platform= filter (case-insensitive) beside ?provider=. - Console: the add/edit form gets a Details section (round-tripping every field, since update replaces the entry), the poster tile a platform badge (non-PC only — the store badge already implies PC) and the year. - plugin-kit: ProviderEntry accepts the same fields (new GameMeta schema, spread flat); SDK + OpenAPI spec regenerated. - pf-client-core decodes platform for future client badges. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 19:00:30 +02:00
enricobuehler	d8b7a86366	ci(plugin-kit): repair the file: dependency bun 1.3 links to itself ... The publish workflow has been failing at Typecheck with "cannot find module '@punktfunk/host'" — every import of the SDK, in a job whose previous step built that very SDK successfully. bun 1.3 installs a `file:` dependency by copying its DIRECTORIES but symlinking each top-level FILE to itself: `node_modules/@punktfunk/host/package.json -> package.json`, a dangling self-reference. So `dist/index.d.ts` arrives intact while the manifest that points at it does not, and resolution dies at the first step it takes. The types were never missing; nothing could find the front door. Replacing that tree with a real copy is the whole fix, and the step no-ops itself the moment bun links `file:` deps correctly again. Not a regression from anything in the kit — it has been broken since bun updated, and went unnoticed because nothing has been published since 0.1.4. Verified from a clean node_modules: install → repair → typecheck → 20 tests → build, all green. plugin-kit-v0.2.0	2026-07-26 18:58:49 +02:00
enricobuehler	f0e71e928a	fix(gamestream): the resolved launch command is what the backend nests ... A Windows/macOS build compiled the resolved-launch parameter out entirely and warned about it, because the two uses left were both Linux-gated. Both platforms want the same value: off Linux `set_launch_command` is a backend no-op and a library title resolves to no command at all (the interactive-session spawner launches it by id), so the cfg split was carrying an older distinction that no longer exists. Also covers the Windows half of `process_name` with a test — the pure name match plus a live scan for this test process, since `find`'s early return would otherwise skip the signal entirely and the test would pass vacuously. Gates: .133 check + clippy --all-targets clean with no warnings, 5/5 procscan (incl. live); .21 299 tests + fmt.	2026-07-26 18:28:13 +02:00
enricobuehler	ee722c2160	docs: what happens when a game ends, and when a session does ... The two behaviors are opt-in-shaped in different directions — one on by default, one off — so the page says which is which, what `always` costs, and that the keep-alive above it is a different clock governing a different thing. The precedence rule (a display kept forever stays up regardless) is spelled out rather than left to be discovered. The dedicated-game-session blurb claimed game-exit-ends-the-session as its own feature; that is now true everywhere, so it moves to where it belongs. Automation gains the `game.running` / `game.exited` rows and a short section on reacting to a game rather than a stream — the two are usually the same moment but not always, and anyone who has been polling the host to find out when a game finished can stop. docs-site: build + tsc clean.	2026-07-26 18:28:13 +02:00
enricobuehler	6a2c127ce9	feat(mgmt/web): show the running game, and who decides when it ends ... The host knows what it launched and, after a disconnect, that it is counting down to closing it. None of that was visible: the console showed a stream with no game, and a game on its way to being closed was something you found out about afterwards. The Dashboard gains a running-game card above the session card — box art matched against the catalog it already fetches, so no new endpoint and no request on the 2 s status poll. "End now" means the two different things the row's state implies: a live game ends by stopping its session (what then happens to the game follows the policy — stopping a session is not licence to close a game), while one already waiting out its reconnect window has no session left to stop and is ended directly. The settings card sits next to the display keep-alive policy because that is the same question one step out: keep-alive decides how long a *display* outlives a disconnect, this decides whether the *game* does. The copy says plainly what `always` costs, that a drop is not someone pressing Stop, and that a display kept forever is unaffected either way — the precedence rule that would otherwise surprise someone. Where a build enforces nothing (macOS, no launch path) the controls are shown disabled rather than hidden: "does nothing here" is information. Also fixes the story fixtures, which had gone stale against the `games[]` field Phase 1 added, and adds a story for the state the card exists for — a game whose client walked away. web: build + tsc clean, biome-formatted; en/de messages complete.	2026-07-26 18:28:13 +02:00
enricobuehler	110eabf281	feat(library/providers): let a provider say how to recognize its games ... A plugin's titles launch through the provider's own client, which hands off and exits — so the host had nothing left to watch, and both lifetime behaviors went quiet for exactly the entries a provider contributes. A `ProviderEntry` (and a manual custom entry) may now carry an optional `detect` hint: install dir, exe, or process name. It is deliberately a subset of what the host tracks internally. A Steam appid or a launcher's environment marker are things the host discovers for itself and would be meaningless — or dangerous — to take on someone's word; where a title is installed is something only the provider knows. The host's own findings win where both exist, so a stale export can never redirect the matcher, and a blank field is treated as absent rather than as "match everything" — an empty install dir would otherwise prefix-match every process on the box, and this feature can end processes. `process_name` is the weakest of the three and the only one typed by hand, so it is matched case-insensitively against the image's file name and nothing else: `retroarch` finds RetroArch, not a helper whose name merely starts the same way, and not a script that happens to live in a `retroarch/` directory. The never-adopt-a-pre-existing-process rule still bounds it. Also: the tray summary gains the running-game row (with the closing-in countdown for a game whose client is gone — visible at the machine without opening the console), the SDK mirrors the `game.*` events, and its generated client catches up with the endpoints Phase 1 added. Gates on .21: check + clippy --all-targets clean, 299 tests, fmt CI-parity, openapi regenerated (GameEntry still carries no `detect` outbound); SDK tsc + 54 tests green.	2026-07-26 18:28:13 +02:00
enricobuehler	98121ccd53	feat(session/gamestream): the compat plane learns a game's lifetime too ... Moonlight sessions had neither direction of the session⇄game binding: a game exiting left the player looking at a desktop, and a session ending never touched the game. The machinery for both landed with the native plane; this wires it up. The compat plane had no way to say "this is over" — RTSP carries no close code — so `/cancel`, the management stop and a game exiting all looked exactly like a client vanishing. They now set a session quit flag, which is what lets the virtual display skip its keep-alive linger for a real stop and what the end-game-on-session-end policy reads to tell a decision from a network blip. A drop still lingers, and still gets its reconnect window. Moonlight can't resume a session, so a client coming back for a game it left behind does it by launching the title again — that reclaims the waiting game before anything starts, rather than letting the old window close on the copy the player is now holding. Both planes now resolve a launch through one lookup (`resolve_launch`), so a GameStream title arrives with the same identity and the same detect signals a native one does; `resolve_session_launch`/`launch_command` were the older, identity-less half of that and are gone. A Moonlight game also has no live-session entry to hang off, so the status surface gains a slot for it — without it the Dashboard would show a stream with no game while one was running. Gates on .21: check + clippy --all-targets clean, 296 tests (293 + 3), fmt CI-parity.	2026-07-26 18:28:13 +02:00
enricobuehlerandClaude Fable 5	ed4e3d3ab6	feat(session): end-game/end-session lifetime on Windows ... Phase 2. The lease, the settings, the events and the status surface were already platform-neutral; what Windows lacked was a way to see processes and a way to ask one to close. `procscan` splits per-OS behind one contract. The Windows matcher is a Toolhelp snapshot plus each process's full image path (`QueryFullProcessImageNameW`), with creation times from `GetProcessTimes` enforcing the two rules the module exists for: never adopt a process that predates the launch, never trust a bare pid. Both matter more here than on Linux — the host is SYSTEM, so it can see and signal everything, and Windows recycles pids briskly. Path matching is case-insensitive and separator-aware, and normalizes the `\\?\` prefix `canonicalize` adds, without which a store-derived path would never compare equal to a live image path. There is no reaper argv and no readable environment on Windows, so a spec carrying only a Steam appid or an env marker matches nothing rather than falling through to an empty predicate. Steam's appid instead feeds a **veto**: its per-app `Running` flag in the user's hive can't be trusted to say a game IS running (Steam sets it around updates too, and leaves it stale after a crash), but it is exactly right for refusing to declare one gone. If the matcher can't see a game whose exe sits outside its manifest's install dir, ending the session would be a false positive the player feels immediately; honoring the veto only ever leaves a stream up. Termination asks before it insists, which on Windows needs a detail that fails silently if missed: `EnumWindows` only sees the calling thread's desktop, and the host's is session 0, which holds none of the user's windows. So the terminating thread binds to the input desktop first (the pattern `pf-inject`'s `sendinput.rs` uses), posts `WM_CLOSE` to the game's visible top-level windows, waits, and only then calls `TerminateProcess` on freshly re-verified pids. Without the desktop bind the polite pass finds nothing and every game dies unsaved. Job Objects, planned as WP2.3, are deferred with the reasoning in the design doc: every Windows launch either hands off through a launcher or the shell — where a job would wrap a shim that exits immediately — or is a direct exe whose path we already know, and wiring one in would touch the `CreateProcessAsUserW` token path the UAC and secure-desktop work depends on. Also: the watcher's steady-state poll now re-verifies known pids and only re-scans the whole table once they all vanish (which still catches a game that re-execs). On Windows a full scan is an `OpenProcess` per process on the box, so this is the difference between a negligible and a noticeable poll. Gates: Linux .21 check/clippy/fmt/293 tests green; Windows .133 check + clippy --all-targets clean, and the 4 procscan tests pass — including the live one that finds the test process in the real process table and rejects a wrong creation time. On-glass Windows (Steam/Epic/GOG/Xbox titles, the unsaved-progress check) still owed; it needs a box with games on it. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>	2026-07-26 18:28:13 +02:00
enricobuehlerandClaude Fable 5	10902f57d0	test(session/gamelease): prove it against real processes, not just fixtures ... The fixture tests establish that the parsing is right; they cannot establish that it describes this kernel. Two tests close that gap. procscan now scans the real /proc for a process it just started, which is the only version that would catch wrong `stat` field ordering, a broken uid check, or a mis-read uptime clock. Writing it found something worth keeping: a wrapper script that `exec`s a binary outside the install dir leaves no trace of that directory in either the image path or the command line, so the install-dir recipe cannot see it. A real game's binary lives under its install dir, so the fixture now models that instead — and the note is in the test for whoever wonders. The gamelease test drives a real child from Running through a host-requested end to Exited, and asserts the session-ending action does NOT fire for it — the difference between "the player quit" (end the session) and "we closed it" (do not). It needs ~12s to outlive the shim window and the exit confirmation, so it is #[ignore]d; run it with `-- --ignored gamelease`. Verified on .21. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>	2026-07-26 18:28:13 +02:00
enricobuehlerandClaude Fable 5	a17aa61c5a	fix(session/gamelease): four things the first pass got wrong ... Found reviewing the riskiest paths of the previous commit. - An install dir of `/games/x` matched a process running out of `/games/xyz`. The image-path check was already component-wise and fine; the command-line check compared raw bytes, so it needed the separator. Two library folders where one name prefixes the other is not a contrived case. - Ending a nested game force-releases kept displays, which is not per-display — so with another client streaming it could retire a display that was never ours. Skip it while anyone else has a live session: the failure mode becomes a game that keeps running, which is the default everywhere else anyway. - A lease nothing polls (no detect signals, or a platform without a matcher yet) sat at "launching" forever in the console. Report it as running: the host did just launch it, and with no watcher it is claiming nothing about the exit. - A game ended after its session was already gone reported no `game.exited` at all — its watcher had stopped with the session, so nothing was left to emit it. That is every grace expiry and every `POST /game/end`, i.e. exactly the ones an operator most wants to see. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>	2026-07-26 18:28:13 +02:00
enricobuehlerandClaude Fable 5	64c0ff96bc	feat(session): bind a session's life to its game's, in both directions (Linux) ... Two of the most-asked-for behaviors, and until now the host could only manage a sliver of one of them: end the streaming session when the launched game exits, and — when the operator asks — end the game when the session ends. Ending the session on game exit already worked, but only for a Steam title under a bare-spawn gamescope. Everywhere else the host spawned a launch and forgot it: the pid was logged and dropped, so on KWin, Mutter, wlroots or gamescope-attach a finished game left the client staring at a desktop. Ending a game was not possible at all — there was no primitive for it. The missing piece was never plumbing, it was knowledge: a launch says what to run, not what the game looks like once a launcher has handed off. So each store now also reports how to recognize its game (DetectSpec, previous commit's subject matter, folded in here), procscan turns that into live pids on Linux, and gamelease turns pids into a lifetime — with four kinds covering how the game got started: nested gamescope owns it; its display teardown ends the game child the host spawned it, in its own process group matched a launcher owns it; recognized by its store's signals untracked nothing identifies it — both behaviors stay off, and the host says so once rather than guessing A child that exits successfully within 5s was a launcher handing off, not the game: the lease re-resolves to matched (or to untracked) instead of reporting an exit. That single rule is what keeps steam://, epic:// and playnite:// launches from ending a session the moment they start. Ending a game is destructive, so three rules bound it. It is opt-in (game_on_session_end defaults to keep). A drop is not a decision — `always` waits out a reconnect window (5 min) that a returning client cancels by reclaiming its own game, matched on fingerprint and title. And only ever this session's game: a pid is adopted only if it started after the launch, so a copy the player already had open is never touched, and every pid is re-verified against its start time immediately before being signalled, so a recycled pid never is. Termination asks first (SIGTERM to the group, 10s) and only then insists. Also here, because they are the same decision seen from other angles: - A management stop is now a deliberate stop. `DELETE /session` sets quit before stop, so it behaves like a client pressing Stop — the display skips its keep-alive linger instead of lingering for a session nobody is coming back to. - `/status` reports what is running (games[]), including a game whose session has gone and which is waiting out its window, so the console can show it and offer `POST /game/end`. - New game.running / game.exited events, filterable like every other kind, which is the whole polling loop of the community plugin this replaces. - session_status::register takes a named struct: eleven positional arguments, half of them same-typed Arc<Atomic…>, is a transposition waiting to happen. Gates on Linux (.21): check, clippy --all-targets, fmt, and 291 tests green. Windows (Job Objects, Toolhelp matching) and the GameStream plane are phases 2 and 3; both are inert here, as is macOS, which has no launch path at all. Design + plan: punktfunk-planning/design/session-game-lifetime{,-implementation-plan}.md Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>	2026-07-26 18:28:13 +02:00
enricobuehlerandClaude Opus 5	0d9d78398c	fix(drivers/windows): name each virtual pad for what it is — one shared description read as "the setting did nothing" ... `pf_dualsense.inx` gave all four hardware ids a single %DeviceDesc%, so Device Manager labelled an emulated DualShock 4, DualSense Edge and Steam Deck pad "punktfunk Virtual DualSense". The HID layer was always per-type — device_type picks the PID (09CC for DS4), the report descriptor and the product string — but the one place a user goes to check said DualSense for every choice, which reads exactly like the controller-type setting being ignored. Split into four model lines over the same install section, one description each. No binding, service or descriptor change; stampinf's 9.9.MMdd.HHmm DriverVer increments on every build, so pnputil takes the update. InfVerif on the WDK runner: INF is VALID. Also correct the Slot.pref comment from the previous commit: emulating a DualShock 4 gives up adaptive triggers by construction. HidOutput::Trigger is emitted only by dualsense_proto, and a DS4 has no trigger-effect reports — the host never generates any to send. Rumble and the lightbar remain. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 16:56:44 +02:00
enricobuehlerandClaude Opus 5	6edd700910	fix(client/gamepad): honor an explicit controller type — the per-pad arrival was re-declaring the physical pad ... The "Controller type" setting reached the Hello and stopped there. Every pad then opened with a GamepadArrival carrying its DETECTED kind, and the host builds each virtual device from that arrival — the session default is only the fallback for a pad that never declares one. So "emulate my DualSense as a DualShock 4" put a DualSense on the host the instant the controller connected, and no amount of reconnecting helped. Apple and the native clients both; Android already applied the setting per pad. The setting now rides the declaration too: explicit wins for every slot, Automatic keeps per-pad detection so a mixed session stays honest. The physical kind still drives the LOCAL feedback paths — a DualSense emulated as a DualShock 4 keeps its lightbar and adaptive triggers, and a Deck keeps its rumble keep-alive. Regressed in 97c67b26 / 76be4c3e (multi-controller support); ships in 0.19.x. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 16:56:44 +02:00
enricobuehlerandClaude Fable 5	8362d57001	fix(vdisplay/windows): exclusive topology gets re-asserted — a verified isolate is not durable ... The CCD isolate verifies "sole active desktop" at commit time and is never checked again. On a hybrid Intel+NVIDIA box the internal panel is re-activated moments later (the isolated topology is deliberately not saved to the CCD database so teardown can restore the user's layout, and the post-isolate resize/HDR churn — or the panel's own driver, or display-poller software — re-resolves back to the stored layout). Proven on-glass: seconds after the verify, CCD showed the panel ACTIVE beside the virtual display while the host still believed it was exclusive — cursor and windows can land off-stream, and the lock screen can land on the physical panel. A group-scoped watchdog now re-queries every PUNKTFUNK_EXCLUSIVE_REASSERT_MS (default 2 s, 0 disables) while an EXCLUSIVE isolate is live and re-issues the isolate when a non-managed display crept back, logging who-is-fighting escalation (3×WARN → 1×ERROR → DEBUG). Gated on a new GroupState.ccd_exclusive flag so a Primary group's deliberately-lit panels are never "fixed". Cycles take the state lock via try_lock with a sliced sleep, so teardown's stop+join under the state lock cannot deadlock and is bounded by ~250 ms. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>	2026-07-26 16:53:19 +02:00
enricobuehlerandClaude Fable 5	6c2c8b6eab	feat(mgmt/web): surface the host.env encoder pin so a conflicting GPU choice isn't invisible ... listGpus gains encoder_pin (PUNKTFUNK_ENCODER when it actually pins something — unset/auto stay null), and the console's GPU card shows it: a muted note when the pin is compatible, an amber warning naming the pinned vendor and the next session's GPU when it contradicts the selection (the host overrides such a pin at session open — without this field the selection just looked broken). Docs updated to say the adapter wins. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>	2026-07-26 16:53:19 +02:00
enricobuehlerandClaude Fable 5	751d1de506	fix(encode/windows): a stale PUNKTFUNK_ENCODER pin no longer wedges a conflicting GPU selection ... On a hybrid box, picking a GPU in the web console whose vendor contradicts a host.env PUNKTFUNK_ENCODER pin produced an unrecoverable session: the pin won backend selection while the adapter followed the console, the wrong-vendor encoder failed deterministically at submit, the reset ladder burned its 5 in-place rebuilds on it, and the client reconnected into the identical wall forever (~10 s per cycle, no visible reason). Three legs: - windows_resolved_backend() now reconciles: a hardware pin whose vendor contradicts the selected GPU is overridden by the adapter-derived backend (capture + encode share one adapter, so honoring the pin can only fail); open_video warns loudly when a pin loses. The reconciliation is a pure, unit-tested table (resolve_windows_backend). - the QSV wrong-adapter bind is typed TerminalEncoderError, and the stream loop's reset ladder ends the session immediately on it instead of feeding a deterministic config error 5 futile rebuilds. - the un-pinned path was already correct (verified on-glass: NVIDIA preference + no pin = stable AV1 10-bit HDR on NVENC), so the override simply takes the conflict case onto that path. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>	2026-07-26 16:53:19 +02:00
enricobuehlerandClaude Opus 5	36e566052f	fix(apple/cursor): a pointer whose bitmap has not landed must not vanish ... `resetCursorRects` wore the host shape only when `hostCursors[st.serial]` hit, and fell through to `invisibleCursor` on a miss. But a miss is the ROUTINE case, not a degenerate one, and it is not a reason to hide the pointer: * State (0xD0) is a per-frame datagram and announces the new serial the moment the host QUEUES the bitmap on the reliable control stream, so on every single shape change the client knows a serial before it holds the pixels. * The shape ring drops the NEWEST under burst (CURSOR_SHAPE_QUEUE = 8) and the host never re-sends it — it only sends on a serial CHANGE — so a dropped serial stays un-backed until the pointer next changes shape. Crossing a toolbar flips arrow/I-beam/hand/resize several times a second, and each flip mints a fresh serial and a fresh bitmap, so bursts are ordinary. Either way the pointer BLINKED OUT rather than lagging, and in the dropped case it stayed gone for as long as the pointer held that shape — reported on glass as "the I-beam never appears over text fields, every other cursor is fine". Hold the last worn shape across the gap: only `st.visible == false` (the host says the pointer is hidden) may hide it now, never a missing bitmap. Worst case is a briefly stale pointer. This also makes two comments that already claimed this behaviour true — the shape-rejected warning's "keeping the previous cursor" and CURSOR_SHAPE_QUEUE's healing claim, both of which were fiction. Host side is exonerated: on .221 the poller sees the I-beam handle (0x10005, CURSOR_SHOWING set), and the monochrome AND-over-XOR conversion renders a correct glyph — black beam, white outline — so the bitmap that goes on the wire is good. swift build PunktfunkKit + cargo check punktfunk-core green; on-glass owed. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 14:13:08 +02:00
enricobuehlerandClaude Opus 5	6a9b3b0f28	fix(windows/cursor): a re-rendered pointer keeps its handle — re-probe the extent ... The GDI shape poller rasterises only when the HCURSOR VALUE changes. But Windows rebuilds the system cursors at a new size whenever the display scale under the pointer changes, and it does that behind a handle that never moves — the shared arrow is 0x10003 for the whole session. So the cache latched whatever size the pointer happened to have when the poller started and never let go. That window is not rare, it is the norm: a fresh virtual display is created at Windows' RECOMMENDED scale and only picks up the client's saved PerMonitorSettings override a beat later, while the poller starts within a second of the monitor appearing. Sample inside it and the session forwarded — and composited — a 96 px pointer over a 100 % desktop for its entire life, which reads on the client as a pointer 3x too large while every other thing on the streamed desktop is correctly sized. Scaling on the client cannot undo it: the bitmap is proportional to the video, it is just proportional to the WRONG scale. Re-read the bitmap's extent on a slow cadence whenever the handle is unchanged — dimensions only, no pixel copy, 4 Hz — and drop the cache when it moved, so the next tick re-rasterises and publishes a new serial. Observing the extent itself rather than a DPI proxy also covers the accessibility pointer-size slider and any other cause of a same-handle re-render. Windows-side clippy -D warnings green via scripts/wincheck.sh; on-glass owed. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 13:22:53 +02:00
enricobuehlerandClaude Opus 5	1421e235f2	fix(host): hdr-p010-selftest silently ignored its size argument ... `args` starts AT the subcommand (main builds it with `env::args().skip(1)`), so a subcommand's own optional arguments begin at index 1 — cf. `args.get(1)` in the `service` arm. This arm read `skip(2)`, swallowing the first optional argument. The documented invocation `hdr-p010-selftest 1920x1080 nvidia` therefore picked up the vendor (it is in the second slot) and ran at the 64x64 DEFAULT. A size-only `hdr-p010-selftest 1920x1080` parsed nothing whatsoever and still printed PASS. Nothing warned, because an unrecognised token is the only thing that errors and no token was ever inspected. The size is the entire point of the flag: the arm's own comment says heights like 1080 are not 16-aligned and exercise a different driver path. So the self-test has only ever validated the one geometry that exercises the least, and the sweep's W7 gate — 1920x1080 and 5120x2880 on the RTX box — could not actually run as written. Found by running that gate: both sizes printed "HDR P010 self-test (64x64 ...)" with byte-identical plane layouts (row_pitch=128, expected_total=12288 — those are 64x64's). Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 12:12:20 +02:00
enricobuehlerandClaude Opus 5	28bbaa5250	tools: check and lint Windows-only Rust from a Linux dev box (scripts/wincheck.sh) ... Linux CI never compiles `#[cfg(target_os = "windows")]` code, so a Windows-side edit was unverifiable until the Windows CI job ran or someone tested on a real box. The pf-capture sweep's Phases 3–6 used an ad-hoc version of this harness in a scratch directory; committing it means the next Windows change does not have to rediscover the recipe. The obvious in-tree command fails, and not because of the Windows code: cargo check --target x86_64-pc-windows-msvc -p pf-capture dies in build scripts that compile C for the target — audiopus_sys (cmake wants a VS generator), then ring (needs an MSVC C compiler plus the Windows SDK headers). Both enter through punktfunk-core's `quic` feature. So the script generates a workspace whose members are Cargo.toml copies with `src` SYMLINKED at the real crate directories — nothing to keep in sync, no copies to go stale — plus a ~30-line stub punktfunk-core carrying only `Mode` and `quic::HdrMeta`, which is provably all the Windows capture stack uses. rustls, ring and opus never enter the graph. Every path dep that is not a generated member points at the real crate. Covers pf-frame, pf-win-display and pf-capture with --all-targets, so the Windows `#[cfg(test)]` modules are type-checked too. ~10 s cold, ~1 s warm, into target/wincheck (gitignored). Verified non-vacuous: a deliberate type error planted in a windows-only file (idd_push/stall.rs) is caught and fails the run. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 12:07:22 +02:00
enricobuehlerandClaude Opus 5	0d69ebf60b	ci(windows): run pf-capture's tests instead of only type-checking them ... Phase 0 of the sweep added `clippy -p pf-capture --all-targets` to this workflow, which type-checks all 19 Windows `#[test]`s and executes none of them. They were decorative. The workflow's own comment explains why pf-encode's tests cannot run here — nvenc link-imports NvEncodeAPICreateInstance, which resolves only against the driver's import lib. That reasoning does not extend to pf-capture: it has no encoder dependency at all (`cargo tree -p pf-capture` lists no nvidia/ffmpeg/libvpl/pyrowave), so its test binary links against nothing the runner lacks. Confirmed empirically rather than reasoned: `cargo test --release -p pf-capture` was run on a Windows dev box against a workspace checkout — it linked, executed and passed, building in ~51 s off an existing release target dir. --release keeps it in the one C:\t\release tree, so it does not trip the C1069 disk exhaustion a second debug dep tree causes. 18 of the 19 run here (StallWatch + ring-generation masking, the cursor shape→wire truth table, and `f32_to_f16`'s rounding-carry/saturation edges); all are pure — no Win32, no device, no desktop. The 19th, `hdr_p010_selftest_intel_1080_live`, is `#[ignore]`d because it needs a real Intel adapter. This is Windows-only code no other job can execute, so it was the cheapest coverage still on the table. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 12:07:22 +02:00
enricobuehlerandClaude Opus 5	1423b63333	fix(gamestream): the HDR SDR-downgrade latch gated the capture offer but not the negotiation (X6) ... `pf_capture::hdr_capture_failed()` had exactly one consumer: `open_portal_monitor`, which uses it to drop the HDR offer (`want_hdr && !hdr_capture_failed()`). The RTSP negotiation consulted only `gnome_hdr_monitor_active()`. So once the latch was set — an HDR negotiation timed out, the monitor left HDR mode between probe and negotiation, NVIDIA EGL not listing LINEAR for XR30, a pre-50 Mutter — the host kept telling the client HDR while capturing and encoding SDR: `cfg.hdr` flowed to `open_portal_monitor`, which then silently opened the SDR offer. The client renders an SDR stream as PQ. Washed out, wrong gamut, and no error anywhere. The latch is sticky until host restart, so every reconnect repeated it. Consulted at RTSP honor time rather than folded into `host_hdr_capable()` for the same reason as the colour-mode probe next to it: that fn is the STATIC serverinfo capability (it decides whether to advertise SCM_HEVC_MAIN10 at all), and this is a live per-session fact. The native plane needs no equivalent — `capturer_supports_hdr()` is hard-false on Linux, and the latch is a fact about the portal capturer, which that plane never uses. Noted in handshake.rs so it isn't re-derived. Listed in the sweep's confirmed-defect register as cross-cutting/medium, then absent from every phase of its implementation plan. Verification is on-glass (force the latch, reconnect a client that requests HDR, confirm the Welcome/SDP reports SDR). Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 12:07:04 +02:00
enricobuehlerandClaude Opus 5	59e33ad11d	Merge origin/main into the pf-capture sweep ... Upstream landed `fc335b39` (negotiate the cursor around what the encoder can blend) on the same files this sweep restructured. Merged rather than rebased: the sweep MOVED ~2,000 lines out of `linux/mod.rs` into `pipewire.rs`/`portal.rs`/`pw_*.rs`, so a rebase would have re-fought the same conflict in up to nine commits; merging resolves it once with both sides in view. The repo already carries merge commits (`land/sweep-all`). Two conflicts, both resolved by keeping BOTH changes: * `linux/mod.rs` — `PortalCapturer::open` gains upstream's `want_metadata_cursor` AND keeps the sweep's `PortalSession` teardown, so the portal threads now take `(setup_tx, quit_rx, want_metadata_cursor)`. The second conflict was upstream editing inside the region Phase 5.1/5.3 moved out; the split wins and upstream's change is ported to where the code now lives: `choose_cursor_mode`'s new `want_metadata` ladder (4 arms — prefer Embedded when the encoder can't blend, and warn-then-take Metadata when Embedded isn't advertised) is now in `portal.rs`, taken verbatim. * `gamestream/stream.rs` — the pooled-capturer slot keeps upstream's third reuse key (`metadata_cursor`, beside HDR-ness) AND the sweep's `result.is_ok() && capturer.is_alive()` re-pool gate. Both guard the same slot against different failures: theirs against reusing a session negotiated for the wrong cursor mode, the sweep's (L2) against reusing a dead one. Everything else auto-merged, including the `want_metadata_cursor` thread through `host/capture.rs`'s facade and `native/stream.rs`'s `session_plan::cursor_blend_for`. Verified after the merge: workspace `cargo test` green, `clippy --workspace --all-targets` clean on Linux AND on x86_64-pc-windows-msvc, `cargo fmt --check --all` clean, pf-capture 38/38. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 11:44:22 +02:00
enricobuehlerandClaude Opus 5	14914bc72f	test(pf-capture): the suite the splits enable — 38 Linux + 19 Windows tests (Phase 6) ... The plan's priority order was "every producer- or OS-controlled parser, blend, geometry guard or negotiation decision", none of which had a test before this sweep (X3). Phase 5's splits are what made most of them reachable without a compositor or a driver. **6.1 `bitmap_extent`** — extracted from `update_cursor_meta`, which is the guard whose own SAFETY proof says a missing bound "SIGSEGVs inside the PipeWire `.process` callback (a segfault `catch_unwind` cannot catch)". Every input except the region size is producer-written. 5 tests: a bitmap that fits (with and without header/pixel offsets); the last-row rule (`stride·(bh−1) + row`, so a bitmap ending flush against the region is accepted rather than rejected by padding that is never read — and one byte short is rejected); anything past the region; degenerate geometry; and four distinct overflow vectors including the near-`i32::MAX` stride the SAFETY comment calls out. One assertion I first wrote was wrong, not the code — with a single row `stride·0 == 0`, so even a `usize::MAX`-wide row is arithmetically in range; the test now exercises the ≥2-row case that really overflows and says why the other is fine (the caller has already capped `bw` at 1024). **6.2 the composite blits** — 8 tests over `composite_cursor` / `composite_cursor_rgb10`: every packed `PixelFormat` arm lands the colour in its OWN channels (so a byte-order slip cannot pass); clipping off all four edges plus six fully-outside positions; zero-alpha, `visible: false` and no-bitmap-yet all drawing nothing; the integer alpha blend; NV12/Yuv444 declined rather than mis-blitted; and for the 10-bit path a bit-exact round trip of an untouched pixel (including the two alpha bits the repack must preserve), the R-at-bit-20 vs R-at-bit-0 distinction between `X2Rgb10` and `X2Bgr10`, and the same clipping. Plus `decode_bitmap_pixel`'s four byte orders. **6.4 the pod builders** — 4 tests. The `dataType` bitmask is pinned per Buffers pod, because each bit is load-bearing in a different direction: the mappable pod MUST include DmaBuf (or gamescope's modifier-bearing format pod wins and the buffer intersection is empty — a link silently stuck in "negotiating"), the SHM-only pod MUST exclude it (or Mutter hands dmabufs and the race-free download path is not), and the dmabuf pod MUST exclude the mappable types (or an HDR session can be handed a MemFd buffer, which Mutter paints 8-bit ARGB32 regardless of the negotiated 10-bit format). Also: every pod parses back through `Pod::from_bytes`; the HDR pods carry MANDATORY PQ + BT.2020 + LINEAR-modifier; and only the NV12 offer pins the colour matrix/range. The `dataType` reader parses the SPA property layout literally (`key, flags, size, type, value`) — a "find the first plausible-looking int" heuristic read the `size` word, which is also 4, and reported the wrong mask. **6.5 `FrameToken` generation masking (W14)** — 2 tests, with `IDD_GENERATION` parked two below the 24-bit boundary so the WRAP is what gets exercised: every minted generation is non-zero, fits the token's field, and survives the pack/unpack round trip `try_consume` performs; and the cleared- `latest` 0 sentinel never matches a live generation. **6.6 the cursor conversion (Windows)** — `convert`'s pixel logic extracted from its GDI plumbing into `mono_planes_to_rgba` / `apply_and_mask_alpha` / `alpha_is_empty`, which is what makes it testable at all (the caller needs a live `HCURSOR` and a screen DC). 6 tests: the four-state AND/XOR truth table in one row; transparency surviving without an invert neighbour; the invert outline covering all eight neighbours, overwriting only TRANSPARENT ones, and clipping at the edges; the alpha-less colour cursor taking alpha from the AND mask; and a short mask not panicking. **6.3 / 6.7** landed with the fixes they guard (`negotiation_plan` in 2.3, `f32_to_f16` in 3.5). 38 Linux tests, up from 6 at the start of the sweep — ci.yml already runs them. 19 `#[test]`s on the Windows side; Phase 0.1's `--all-targets` lint type-checks them all, but note it does not RUN them (the workflow lints only), so the Windows ones execute on a Windows box. clippy --all-targets clean on both targets. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 11:31:34 +02:00
enricobuehlerandClaude Opus 5	306f4a514d	refactor(pf-capture): structural splits + collapse the restated signal set (Phase 5) ... Refactors LAST, after every defect fix, so no behaviour change hides in a move diff — and so the newly-exposed review surface was already clean when it landed (the WP7 discipline). 28f8fc71's recorded trap (an inline `use super::X` inside a moved body silently changing meaning) was audited first: every `use super::` in the moved code is either at module level — where `super` still means `linux` — or inside a `mod tests` that moved with its parent. 5.1 `mod pipewire` → `linux/pipewire.rs`. It was 1,900 of that file's lines. `mod.rs` is a directory module, so a plain `mod pipewire;` resolves with no `#[path]`. 5.2 out of that, `linux/pw_pods.rs` (the 7 param-pod builders + `serialize_pod` + the PQ constant and its test) and `linux/pw_cursor.rs` (`CursorState`, `decode_bitmap_pixel`, `update_cursor_meta`, `dst_offsets`, both `composite_cursor*`). The pods are the crate's WIRE surface — a missing property is not a compile error but a link that stalls in `negotiating`; the cursor half is producer-driven and bounds-critical. Both are now pure enough to unit-test without a compositor, which is what Phase 6 needs. 5.3 `linux/portal.rs`: the ScreenCast/RemoteDesktop handshake, the cursor-mode choice and GNOME's BT.2100 probe — the async/tokio/zbus control plane, separated from the realtime half. Zero re-exports changed: `mod.rs` re-exports `gnome_hdr_monitor_active` at its old path. 5.4 `idd_push/open.rs` (the whole one-time construction path + `SharedObjectSa` + `AttachTexFail`) and `idd_push/compose_kick.rs`. Read `open.rs` when a session will not START and the parent when one stops flowing; the steady state stays with the parent by decision. Also moved the ~65-line stall REPORTING block out of `try_consume` (the hot loop) into `StallWatch::report`, taking its two correlation counters with it — they were capturer fields nothing else touched. 5.5 `windows/dxgi/selftest.rs`: `p010_reference`, both self-tests, `hdr_p010_convert_bars_on_luid`, `f32_to_f16` and the two test modules — the validation path, none of which runs in a session. The two `pub` entry points are re-exported, so `main.rs`'s subcommand and pf-encode's live e2e keep their paths. (An explicit `#[path]`, like `idd_push`'s children: this file is itself reached through a `#[path]`, so a bare `mod` would resolve to `windows/selftest.rs`.) 5.6 `CaptureSignals`. The seven shared flags/slots were created in `spawn_pipewire`, `_cb`-cloned one by one, passed as seven of `pipewire_thread`'s parameters, and then re-declared as fields on `PwHandles`, `PortalCapturer` AND `UserData` — the same list written four times, each with its own drifting copy of the doc comment. One `#[derive(Clone)]` struct instead (a refcount bump per field), which also makes it structurally obvious that producer and consumer share ONE set. And `CaptureOpts` for the four trailing `bool`s: four adjacent same-typed positional arguments are a silent-transposition footgun — swap `want_444` and `want_hdr` and it compiles, negotiates the wrong pod family, and surfaces as a black screen ten seconds later. 5.7 trait shape, targeted: `set_active(&self)` → `&mut self` (it took `&self` only because the flag happened to be an `Arc<AtomicBool>` — an implementation detail leaking into the contract); `capture_target_id` ⇒ `resize_output`'s pairing rule stated on both methods and in a cluster note; one-line cluster headers over the 13 methods. NOT done from 5.7: taking the gamescope targets as an `open_*` option instead of the trait method. It would put a Linux-only parameter on the cross-platform `open_virtual_output` and on the host's `capture_virtual_output` facade — a `#[cfg]`-shaped wart in two shared signatures to delete one already-`cfg`'d defaulted method whose lifecycle rule 2.5c had already made harmless. Wrong trade under 5.7's own "no churn for no defect fixed" principle. File sizes: `linux/mod.rs` 2,778 → 770 (target ≤900 ✓), `windows/dxgi.rs` 1,374 → 954, and `windows/idd_push.rs` 2,694 → 1,922. The last two miss the plan's ≤850/≤1,300 targets, which were computed against the ORIGINAL line counts — Phases 1–4 added several hundred lines of SAFETY proofs and defect rationale to exactly these files before the split. The moves themselves are the ones the plan specifies; closing the rest would mean inventing new splits it does not call for. Zero behaviour delta. pf-capture 20/20; workspace clippy --all-targets clean on Linux and on windows-msvc. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 11:25:39 +02:00
enricobuehlerandClaude Opus 5	530909a154	perf(pf-capture): take three per-frame costs off the Windows HDR hot path (Phase 4) ... **4.1 (W10) — the HDR convert allocated two views and mapped a constant buffer per frame, inside the ring slot's keyed-mutex hold.** So the driver's publisher sat blocked on that slot while the host created `CreateRenderTargetView`s and Mapped/Unmapped a 16-byte buffer whose contents never change. Both are lifetime-of-mode facts, not per-frame ones: - the two P010 plane RTVs now belong to the out-ring SLOT, built once in `ensure_out_ring` (`out_ring` becomes `Vec<OutSlot>`). A driver that rejects planar RTVs — the one hard requirement of this path — therefore fails at ring build, where such a failure belongs, instead of on the first frame. - `inv_src` becomes an IMMUTABLE constant buffer filled in `HdrP010Converter::new(device, w, h)`. The converter is already dropped by `recreate_ring` on every mode change, so it cannot go stale. This also deletes the `Map`'s silently-ignored failure: it could leave the chroma pass sampling at a garbage texel size with no error anywhere. **4.2 (W15) — `sdr_white_level_scale` ran inside the keyed-mutex hold too.** It is a CCD query that contends the display-config lock — precisely the contention `DescriptorPoller`'s doc says must stay off the frame path — and `prepare_blend_scratch` called it while holding the slot. Moved to `refresh_sdr_white_scale`, called at open and after each ring recreate (where the scratch is invalidated anyway). "Only on scratch rebuilds" is not the same as harmless when the thing being blocked is the producer. **4.3 (W11) — `VideoProcessorSetStreamAutoProcessingMode(vp, 0, false)`.** The comment claimed "no interpolation/auto-processing" while only setting the frame format; auto-processing's documented default is ENABLED, so vendor denoise/edge-enhancement was free to run inside every SDR `VideoProcessorBlt` — altering the pixels we encode and spending video-engine time on the desktop-capture hot path. Now actually disabled, with each call's purpose stated separately. **4.4 (the Linux CPU-path map cache) is deliberately NOT done.** The plan gates it on measurement ("measure first, and only if the CPU path still matters") and this box cannot measure a live capture session; it also needs a frame-buffer return path through `FramePayload::Cpu`, which is a larger change than the rest of this phase. Left for Phase 7's measured pass. Compile-verified for windows-msvc locally; pf-capture 20/20 on Linux; workspace clippy --all-targets clean on both targets. The GPU-capture check and the measured `cap_us` delta are owed to Phase 7.3. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 11:11:46 +02:00
enricobuehlerandClaude Opus 5	bc45287dc6	fix(pf-capture): Windows IDD-push defects — HDR pin, cursor, recreate, handles, f16 (Phase 3) ... Compile-verified for x86_64-pc-windows-msvc locally (a scratch workspace symlinking the real sources against a stub punktfunk-core — the `quic` feature's ring/opus C builds are what blocks an in-tree cross-check). Behaviour is owed the on-glass validation in sweep Phase 7.3. **3.1 (W1/F3) — the HDR pin asserted a flip it never verified.** `poll_display_hdr` discarded `set_advanced_color`'s `bool` and then wrote `now.hdr = self.client_10bit` — the DESIRED state in place of the observed one. On a display that cannot be flipped (the state this file already logs as "Downgrade point D" at open) that broke in both directions: wanting HDR, the fabricated `true` differed from `current`, so two poller samples drove `recreate_ring(true, …)` and rebuilt the ring FP16 while the driver composed 8-bit BGRA — every publish dropped by the driver's format guard, `recovering_since` expiring, `try_consume` bailing: a permanent 3-second reconnect loop. Wanting SDR, the fabricated `false` MATCHED `current`, so no recreate ever fired and the ring stayed BGRA against an FP16 composition — the same dropped-publish outcome, silently. Now it re-reads `advanced_color_enabled` and follows what the display actually composes, with a one-shot error naming want/observed/returned. A not-yet-settled read costs one debounce cycle, never a wrong ring, which is why this does not block the frame path on a settle poll the way `open_on` does. Downgrade point D's error now carries the same pair, so `Some(false)` (display says no) and `None` (the CCD read failed) are distinguishable. **3.2 (W2, W3, W6) — cursor correctness.** - W2: the poller's desktop rect was captured once at open and used forever, for BOTH the desktop→frame offset and the `in_rect` visibility test — while both mid-session mode-change paths (`resize_output`, `poll_display_hdr` → `recreate_ring`) keep the same poller. After an in-place resize the pointer was clipped to the old rect and offset by a stale origin. It is now a SEED: the poll thread re-queries on its existing 250 ms reattach cadence, keeping the last good value on `None` (a transient CCD failure must not park the rect at zero and report every position invisible), which keeps the CCD call off the encode thread as `DescriptorPoller` demands. - W3: `composite_forced` tested `cursor_sender.is_none()`, but §8.6's rationale is "no cursor CHANNEL" — and the delivery just above it is explicitly allowed to fail non-fatally, which is precisely the state needing the rescue. It was the one state that skipped it: a negotiated channel that failed to create or deliver left a cursor-excluded target with NO pointer at all. Now `cursor_shared.is_none()`, evaluated after that binding. - W6: `cursor()` degraded poller→shm correctly, but the BLEND path — the only consumer that matters in the composite model, since the Windows encode loop never attaches `frame.cursor` — read the poller directly with no `alive()` check and no fallback, so the documented fallback and the spawn-failure warning were both untrue for exactly those sessions (a dead poller meant pointer-less frames, not a degraded pointer). One `live_cursor()` now serves all three consumers and LATCHES the source, because the two keep independent serial namespaces and interleaving them poisons the client's shape cache. **3.3 (W4, W5, W14) — recreate hardening.** - W4: `recreate_ring` committed `display_hdr`/`width`/`height` BEFORE the fallible `create_ring_slots` (VRAM pressure at a large new mode — exactly when resizes happen), leaving a failed recreate emitting frames stamped with the new geometry against the old ring, the old generation and an unchanged header. Slots are built first; nothing after the commit point fails. - W5: a recreate never cleared the driver's status words, and `wait_for_attach` — the only classifier of TEX_FAIL/BIND_FAIL and the only source of the LUID rebind — runs at open ONLY. A stale `OPENED` therefore made a failed re-attach look healthy while the recover-or-drop bail reported nothing. Now cleared before the Release generation store (plus `status_logged`), and the 3 s bail prints the live `(driver_status, detail, render_luid)` the way `next_frame`'s 20 s bail already did. The four-field read is one `driver_diag()` helper instead of four copies of the same unsafe block. - W14: `IDD_GENERATION` is a full `u32` but the publish token carries 24 bits and `unpack` masks what it reads, so past 2²⁴ recreates `tok.generation != self.generation` would be permanently true — every frame rejected. Masked at the single mint point, and 0 skipped (it is also the cleared-`latest` sentinel). **3.4 (W8, W9, W12) — handle hygiene.** `shared_object_sa`'s security descriptor is a `LocalAlloc` nobody freed: leaked twice per open and once per ring recreate. It is now an RAII `SharedObjectSa` whose `Drop` `LocalFree`s it and whose `as_ptr()` only lends a borrow — which also makes the "descriptor must outlive the attributes" rule structural instead of a comment. The PyroWave fence's shared NT handle, created per capturer and never closed, becomes an `OwnedHandle` (the encoder holds its own duplicate, so closing ours is safe). `cursor_blend`'s `cbuf_scale` is cached only on a successful `Map` — caching unconditionally wedged the HDR/SDR cursor scale for the session after one transient failure. **3.5 (W7) — `f32_to_f16` swallowed the rounding carry.** `sign | half_exp | (half_mant + round)` ORs a mantissa carry into bit 10, so for every ODD biased exponent (bit 10 already set) the carry vanished and the result came back ~2× low: `1.9998779 → 1.0`, `0.49996948 → 0.25`. Only values one ULP below a power of two are affected — precisely what a gradient test pattern is full of — so this made `hdr-p010-selftest` FAIL a correct shader. Composed additively, with 4 tests (18 asserted bit patterns, a round-trip property over the self-test's scRGB values, saturation) — all verified numerically against a standalone reference on this box, including a scan confirming old-vs-new diverges ONLY on the carry cases. Phase 0.1's `--all-targets` lint is what lets these compile in CI at all. pf-capture 20/20 on Linux; workspace clippy --all-targets clean on Linux and windows-msvc. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 11:06:26 +02:00
enricobuehlerandClaude Opus 5	9a7b3a46d0	fix(pf-capture): gamescope cursor — auth without setenv, frame-space coords, live targets (2.5) ... **2.5a (L4) — stop `set_var`ing `XAUTHORITY` from a live multithreaded host.** The old connect swapped the process-global var around each `RustConnection::connect` under a mutex. That lock serialises this source against itself and nothing else: `getenv` takes no lock, so every OTHER thread's read raced it — and the concurrency is by construction, since `attach_gamescope_cursor` runs while the PipeWire thread is starting up (libspa plugin load, EGL/CUDA init). The primary path now parses the MIT-MAGIC-COOKIE-1 entry out of the given file itself and hands it to `DefaultStream::connect` + `RustConnection::connect_to_stream_with_auth_info` — the same two steps `RustConnection::connect` performs internally, minus its env-derived auth lookup — so nothing in this process touches the environment. The env swap survives only as a fallback for a file we cannot parse, and a wrong cookie pick cannot do damage: the server rejects it and the fallback (which does libxcb's full family/address match) takes over. Sharing `pf_vdisplay`'s process-wide env lock was not the fix — wrong layer, and it would still not fix `getenv`. **2.5b (L6) — publish the pointer in FRAME coordinates.** `QueryPointer` answers in the nested root's space, but `CursorOverlay::x/y`'s contract is frame pixels, and gamescope's `-w/-h` (nested root) and `-W/-H` (output + PipeWire node) are independent knobs — at `-W 1280 -H 720 -w 640 -h 360` they differ by 2×, so the pointer drew at a fraction of its real position. The root size comes free off the setup reply we already parse; the negotiated frame size arrives from the PipeWire thread's `param_changed` through a new `Arc<AtomicU64>` (`0` = not negotiated ⇒ pass through, as before). Scaling is computed in `i64` — a 5K coordinate times a 5K width overflows `i32`. Position only: the bitmap stays at root scale, warned once so a mismatched session is visible. **2.5c (L7) — the targets are a PROVIDER, not a snapshot.** The list was discovered once, before the game launched. gamescope creates the game's Xwayland at launch and advertises only the FIRST in any child's environ (verified on this box: `--xwayland-count 2` makes `:2` and `:3`, only `:2` is advertised), so the game's display was invisible — and when the connected Big Picture display then reported "gamescope is not drawing the pointer here", the source blanked the cursor for the whole game session, which is the exact regression the module doc says it fixed. The worker now re-runs the provider every 2 s: it adopts new Xwaylands, reconnects dead ones, and `spawn` no longer returns `None` on an empty list — a stream that starts before the game converges instead of staying cursorless. The provider is a host-facade closure, same one-way-edge shape as `FrameChannelSender`. **2.5d (L8) — bound the teardown join.** `Drop` joined the worker unbounded while the worker blocks in `RustConnection` replies with NO read timeout, so a peer that stops answering but keeps its socket open hung capturer teardown — on the session path. Now: a completion channel, `recv_timeout(250 ms)`, then detach with a warning (the thread only touches its own X connections and an `Arc`'d slot). 7 new tests, all host-runnable: the cookie reader against a hostile `.Xauthority` (wrong protocol, wrong display, wildcard entry, truncation mid-length-prefix, an over-long length prefix, a missing file), display-string parsing, and the root→frame mapping incl. the 5K overflow. pf-capture 20/20; workspace clippy --all-targets clean on Linux and windows-msvc. Phase 7.2 owes the on-glass nested-gamescope validation. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 10:56:05 +02:00
enricobuehlerandClaude Fable 5	3f4ad08869	test(encode/ffmpeg_win): extract the decision logic and pin it with unit tests ... The QSV open-failure fallback (1,400 lines, 23 unsafe, 0 tests) follows the vaapi.rs treatment: the device-free decisions now live in named functions with their contracts pinned — the per-vendor zero-copy default matrix (AMF on-glass-validated on, QSV opt-in), the PUNKTFUNK_FFWIN_POLL_MS clamp-before-µs-conversion (the 27.7-hour-spin class), the readback routing table with its mid-stream depth-change guard, the swscale source map, the QSV display-remoting latency contract (async_depth=1/low_power=1/ look_ahead=0/forced_idr=1/scenario) and the AMF no-B-frames contract, and the per-vendor zero-copy pool bind flags. A probe smoke rides along #[ignore]d for the runner. No FFI plumbing chased; no behavior changes. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>	2026-07-26 10:50:33 +02:00
enricobuehlerandClaude Fable 5	fc335b39e9	fix(host/encode): negotiate the cursor around what the encoder can blend ... EncoderCaps::blends_cursor's contract said the HOST must fall back to capturer-side compositing when a cursor-as-metadata session lands on an encoder that can't composite — but that host half was never built: open_video warned and the session streamed WITHOUT a pointer (confirmed on the VAAPI dmabuf and libav-NVENC CUDA paths; latent on vulkan RGB-direct/native-NV12). The negotiation is now caps-aware, ahead of capture, on both planes: * pf-encode grows cursor_blend_capable() — the pre-open dispatch mirror (sibling of linux_native_nv12_ok) answering whether the resolved backend composites frame.cursor; its pure core is test-pinned arm by arm. * Native plane: handshake::cursor_forward grants the cursor channel only where the resolved backend can blend (the capture-mouse flip makes the host draw the pointer on demand); denied sessions keep the pre-channel path — the compositor EMBEDS the pointer, never cursorless, never doubled. The Welcome's HOST_CAP_CURSOR bit is computed once and read back at both session-wiring sites instead of recomputed. SessionPlan::output_format additionally keeps every cursor-blend session off producer-native NV12 (the arm with no CSC to fold a cursor into), and vulkan RGB-direct now yields to a cursor-blend session even when pinned (EFC cannot composite; the open logs the override). Windows plans cursor_blend=false via the new shared cursor_blend_for() rule — the IDD capturer composites the pointer itself, and asking the encoder anyway fired the blends-cursor warn spuriously on every cursor-channel session. * GameStream plane: the hardcoded cursor_blend=true is gone. The portal source asks for cursor-as-metadata only when the resolved backend blends, otherwise negotiates an Embedded pointer (choose_cursor_mode's new ladder); the capturer pool now also keys on that mode. The virtual-output source passes false — its capture embeds the pointer where it can. The per-arm warns in vulkan_video (RGB-direct, native-NV12) are now structurally unreachable and removed. open_video's post-open check stays as the single backstop for what planning cannot see: a Vulkan-open falling back to VAAPI mid-session, and the gamescope residual (no embedded mode exists there, so a never-blending backend — H.264-on-AMD VAAPI, software — still streams cursorless; fixing that needs a compositing stage, deliberately not built in this pass). Zero-copy is preserved throughout — every fallback is a capture-negotiation change, never a readback. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>	2026-07-26 10:50:33 +02:00
enricobuehlerandClaude Opus 5	759fdf767c	fix(pf-capture): buffer-geometry guards + a genuine drop-oldest hand-off (sweep 2.4, 2.6) ... **2.4 — buffer geometry correctness on the Linux CPU path.** L5: the self-mmap de-pad mapped the fd from offset 0 and then indexed by `chunk.offset()` ALONE, ignoring `spa_data.mapoffset` — where that spa_data's region actually begins inside the fd. For a pooled-memfd producer (every buffer a slice of one fd) it therefore read the WRONG buffer, and the `needed > avail` guard could not catch it: `avail` came from the whole-fd mapping, so it was ample for any single buffer's span. PipeWire's own MAP_BUFFERS slice (the fallback) was always correct because it already starts at `mapoffset` — only the hand-rolled "fix" path was wrong, so the two paths now compute their base separately (checked add; both halves are producer-controlled). L15: when `fstat` failed, the mmap length was invented as `offset + needed` — producer-controlled geometry. That defeats the entire point of the fstat (the "buffer smaller than the frame span" guard then compares against the number the producer just supplied) and can map, and read, past the end of the object — a SIGBUS, not an `Err`. Without a real length we now decline to self-map and let PipeWire's own slice serve, which is bounded by construction. L12: `bytes_per_pixel()`'s catch-all answers 4 for NV12, so a producer-native NV12 buffer computed `row = 4w` against a stride of ~w and ALWAYS tripped the stride guard — reporting "chunk stride < row", i.e. blaming the producer for a host limitation. NV12 cannot be de-padded on this path at all (plane 1 is not even in `datas[0]`'s span), so reject it with a message that says which side is at fault and under what condition the NV12 offer is valid. L14: `spa_meta_bitmap` was read field-by-field through an aligned `*const` at a PRODUCER-controlled offset, with a SAFETY comment asserting an alignment the code never established. One `read_unaligned` of the `Copy` POD instead. **2.6 — hand-off semantics (L9, L10).** The frame channel was `sync_channel(8)` + `try_send`, i.e. drop-NEWEST, while the trait documented drop-oldest. Neither remedy the plan offered works: a `SyncSender` cannot pop, so "drain one and retry" is not expressible, and shrinking the bound makes staleness WORSE, not better (with bound N the queue holds the first N frames of a stall and everything after is discarded, so a larger N actually yields a fresher frame — the defect is the discarding, not the depth). Replaced with a one-deep OVERWRITING mailbox plus a depth-1 wakeup channel: publishing overwrites, so a stalled consumer loses the intermediate frames and is always handed the freshest one. `next_frame` used to return the OLDEST of eight stale frames. Falls out of it: - `wait_arrival` now honours its documented "must NOT consume" contract by peeking the slot. The `pending` field existed only to stash a frame the un-peekable channel forced it to consume; it is gone. - a queued `CapturedFrame` can own a dup'd dmabuf fd or a CUDA buffer, so the old 8-deep backlog could pin eight compositor buffers. One-deep by construction now. - L10: `set_active(false)` flushes the mailbox, so a reused capturer no longer opens the next stream with the previous session's last frame (wrong content, and a `pts_ns` from the old clock). Two lines, where flushing a queue + a `pending` stash was fiddlier. - the producer-liveness signal moves to the wakeup channel's `Disconnected`, and a final frame left in the slot as the thread exits is still served before the error. pf-capture 13/13; workspace clippy --all-targets clean on Linux and windows-msvc. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 10:48:18 +02:00
enricobuehlerandClaude Fable 5	f495b201e1	fix(encode): delete the write-only EncoderCaps::supports_hdr_metadata ... A caps field nothing reads is a contract nobody honors — and this one shipped write-only: its single reader anywhere in the workspace was a hardware-gated assertion inside pf-encode's own AMF smoke test. Both planes send the static HDR grade out-of-band unconditionally (the native 0xCE datagram per keyframe, the GameStream 0x010e control message), every first-party client reads exclusively that path, and none parse in-band SEI — so the host decision the field was reserved for (suppress out-of-band when the encoder embeds) can never validly exist. The field's doc contract had also rotted in two directions: it claimed set_hdr_meta no-ops when false (native AMF and QSV consume it regardless) and that only Windows direct-NVENC attaches in-band metadata (AMF and QSV do too). The in-band SEI/OBU emission itself is untouched — it stays a bonus for stock decoders, documented at the emit sites; the trait docs now describe the real routing. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>	2026-07-26 10:48:07 +02:00
enricobuehlerandClaude Opus 5	17eb8100a3	fix(pf-capture): portal teardown, observable death, one negotiation resolver (sweep 2.1–2.3) ... **2.1 (L1) — the portal thread leaked a live screencast session.** It parked on `future::pending()` and `Drop` stopped only the PipeWire thread, so every dropped portal capturer permanently leaked a thread + a 2-worker tokio runtime + a zbus connection — and because dropping that connection is what ends the compositor's cast (ashpd's `Session` has no `Drop`), the COMPOSITOR-side session leaked too. GameStream drops the pooled capturer on every HDR↔SDR reconnect flip and opens a fresh one, so the sessions accumulated: exactly the "second conflicting screencast" the pooling exists to prevent. Both portal threads now park on a `oneshot::Receiver` and `PortalSession::drop` fires it, then waits — BOUNDED (750 ms, then detach with a warning) — for a completion signal the thread sends after its runtime has been dropped. Bounded because teardown runs on the session path and the thread may be inside a D-Bus round-trip: an unbounded `join()` there would hang the host rather than leak. Every early-return path in `open` drops the session too, so a failed PipeWire spawn no longer strands a fresh cast. **2.2 (L2) — a dead capturer was pooled forever.** All three of the portal backend's terminal states are sticky and observable only by consuming a frame, and `Capturer` exposed no health predicate, so GameStream re-pooled unconditionally — after `result` was already an `Err`. One zerocopy-worker death or one compositor restart therefore wedged GameStream portal video at 10 s per reconnect attempt, permanently (that path has no rebuild closure). Adds `Capturer::is_alive()` (default `true`), implemented for `PortalCapturer` as `!broken && streaming && !thread.is_finished()` — the third term catches a dead PipeWire thread, which is otherwise indistinguishable from an idle desktop. A static desktop stays `Streaming`, so an idle-but-healthy capture is never reported dead. The host re-pools only on `result.is_ok() && capturer.is_alive()` and logs which of the two retired it. Both halves ship together: either alone leaves the wedge reachable. **2.3 (L3/F1) — a "VAAPI" downgrade latch was really a global zero-copy kill switch.** `spawn_pipewire` hand-mirrored the thread's `vaapi_passthrough` decision and the copy omitted `raw_dmabuf_import_disabled`, so once that latch fired the capturer still believed it had made the passthrough offer while the thread had already fallen back — and its timeout branch then latched a downgrade for an offer nobody made. That downgrade was `note_vaapi_dmabuf_failed`, which fed `pf_zerocopy::enabled()`, so ONE failed negotiation dropped every later session on the host — the NVENC EGL→CUDA path included — to CPU capture until restart. And since the raw-passthrough offer is also the PyroWave one on ANY vendor, a single PyroWave negotiation timeout was enough to do it on an NVIDIA box. Two fixes, both structural: - `negotiation_plan(NegotiationInputs) -> NegotiationPlan` is now the ONE resolver, consumed by the thread AND by `spawn_pipewire` — the mirror is deleted, not re-synced (WP7.6's shape), so the drift class is unrepresentable. It is pure (every env/latch read is an input), which is what makes it testable; the runtime-dependent half stays a method (`want_dmabuf` needs the modifier list the importer's construction actually yielded). The redundant `importer.is_none()` term goes: `build_importer` already excludes the passthrough, and that term is what made the decision look impure enough to "need" a mirror. `PUNKTFUNK_FORCE_SHM` was ALSO read in both places; now once. - the capture-side downgrade becomes `pf_zerocopy::note_raw_dmabuf_negotiation_failed`, which latches `RAW_DMABUF_DISABLED` — gating only the raw-passthrough decision. `enabled()` is now the env var alone, and `VAAPI_DMABUF_FAILED` is deleted. Also (L13, needed by 2.3's plan inputs): `PUNKTFUNK_PIPEWIRE_NV12` honoured only the exact string `"0"`, so `"0 "` or `"false"` read as force-ON — the bug class of ed525c4c. pf-host-config's explicit-off grammar is now exported as `env_on()` and the four in-crate copies (`PUNKTFUNK_ ZEROCOPY`/`_10BIT`/`_444`/`_CHACHA20`) plus this new caller share it. 7 new tests pin the four invariants that were prose-only comments, plus the latch property that made the mirror wrong. pf-capture 13/13; workspace clippy --all-targets clean on Linux and on x86_64-pc-windows-msvc. On-glass validation of 2.1/2.2/2.3 is owed (sweep Phase 7.1). Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 10:41:53 +02:00
enricobuehlerandClaude Fable 5	232b6d6be2	test(encode/vaapi): extract the open-time decision logic and pin it with unit tests ... The fallback backend under Vulkan Video — and the only AMD/Intel H.264 and 10-bit/HDR encoder — had 1,300 lines, 26 unsafe blocks, and zero tests. The device-free decisions now live in named functions with their contracts pinned: the entrypoint ladder + LP_MODE latch round-trip (the cross-GPU session-killer and the 8-bit-pins-10-bit under-advertisement are both key'd tests now), the PUNKTFUNK_VAAPI_LOW_POWER / _ASYNC_DEPTH grammars, the VUI ↔ scale_vaapi colour agreement (the Mesa-BT.601 hue-shift pin), the honest-downgrade depth table, the HEVC-Main10-only explicit profile, and the 10-bit probe gate. Probe + CPU-path encode round-trip ride along as #[ignore]d hardware smokes in the house style. No FFI plumbing was chased; no behavior changes. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>	2026-07-26 10:29:31 +02:00
enricobuehlerandClaude Opus 5	54a37aeb46	docs(pf-capture): truth pass over comments, docs and log strings (sweep Phase 1) ... Every item here is prose that actively misdirects a maintainer or an operator. Landing it before the defect fixes means those diffs get reviewed against accurate comments. 1.1 (X5) — split two merged doc blocks that documented the item ABOVE them: `hdr_meta`'s whole contract was glued onto `fn cursor()` in lib.rs (leaving `hdr_meta` undocumented), and `hdr_p010_selftest_at`'s was glued onto `hdr_p010_convert_bars_on_luid` in dxgi.rs. rustdoc for the trait's central HDR contract was simply wrong. 1.2 (L9 doc half) — the `Capturer` trait doc advertised a "bounded drop-oldest channel". The Linux portal's `sync_channel(8)` + `try_send` is drop-NEWEST: under a consumer stall the arriving frame is discarded and the encoder gets the stalest queued one. Say so. (Phase 2.6 changes the behaviour; this commit only stops the doc from lying about today's code.) 1.3 — delete the DONT_FIXATE claim from the dmabuf offer comment. `build_dmabuf_format` emits a plain MANDATORY `ChoiceEnum::Enum` and `param_changed` re-emits nothing, so there is no two-step DMA-BUF handshake here; libspa 0.9's `ChoiceFlags` cannot even express DONT_FIXATE. Recorded as a real follow-up instead of an implemented thing. 1.4 — `mainloop.run()` no longer "blocks until process exit": the quit channel attached above it stops the loop from `PortalCapturer::drop`. 1.5 (L11) — the 6-arm negotiation log ladder told a fully zero-copy PyroWave session "VAAPI encode with the CPU capture path — zero-copy was disabled", one line after logging that it had advertised the PyroWave device's dmabuf modifiers: the passthrough arm was gated on `pyrowave_modifiers.is_empty()`, so the pyrowave case fell through to the CPU warning. Ungate it; the CPU-path arm is now reachable only when no dmabuf is advertised at all, which is what it claims. Its parenthetical also gains the third real cause (the session asked for CPU frames). Control-flow-neutral for every other combination. 1.6 (W13) — `kick_dwm_compose` claimed a "sub-millisecond" round trip while its cursor-on-a-sibling-display branch sleeps 35 ms on the CALLER's thread. State the cost on the function and at both call sites, including which one is on the live frame path. 1.7 (W16) — DDA-era residue, DDA having been removed: - retarget the win32u hook's rationale + all four log strings: its job is no longer "keep DDA on one adapter" but "keep the virtual display on the adapter SET_RENDER_ADAPTER pinned" (a DXGI reparent surfaces as the driver's TEX_FAIL render-adapter mismatch). The hook stays installed. - the per-monitor-v2 DPI awareness rationale likewise: not DuplicateOutput1's E_ACCESSDENIED any more, but keeping cursor/window coordinates in the PHYSICAL pixels the host's CCD geometry (`source_desktop_rect`, `desktop_bounds`) is already in. - `HYBRID_HOOK_HITS` was write-only. Surface it as `hybrid_hook_hits()` on the IDD-push open line, which is the first point where DXGI has actually been exercised — the patch-readback check proves the bytes landed, only this proves DXGI reaches the export. - delete `hdr_p010_selftest()`: an unreachable 64×64 wrapper (main.rs calls `hdr_p010_selftest_at`); its description moves onto the function that survives. - `VideoConverter`'s docs promised a P010/BT.2020 output and a live scRGB input path. It pins `YCBCR_STUDIO_G22_LEFT_P709` unconditionally and its only caller always passes `scrgb_input: false`. 1.8 (X7) — the native handshake's 4:4:4 gate comment stated the OPPOSITE of what `pf_capture::capturer_supports_444` returns on Windows ("delivers subsampled NV12/P010 today, so it returns false there" — it forwards `resolved_backend_ingests_rgb_444()` and the IDD ring passes BGRA through for an SDR 4:4:4 session). No `.rs` behaviour delta: comments, doc comments and log strings, plus 1.5's control-flow-neutral ladder and 1.7's counter accessor. Linux tests 6/6; clippy --all-targets clean on both targets. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 10:27:06 +02:00
enricobuehlerandClaude Opus 5	cc848479c4	test(pf-encode): re-point the cpu_img size-change smoke at the CSC guard's contract ... vulkan_cpu_img_survives_a_source_size_change drove MISMATCHED source sizes through the then-lenient CSC arm as its vehicle for the staging cache hazard (format-only-keyed cpu_img → OOB copy while submit said Ok). e3354b6d's guard — correctly the equality check every sibling arm always had, against the MODE, not the coded extent, so the padded render-vs-coded tolerance in the direct arms is untouched — makes that scenario unrepresentable through submit and broke the test on main (.25 layers baseline read 13/13 instead of 14/12). Replaced by vulkan_csc_refuses_a_mismatched_source: refusal pinned in BOTH directions, plus the property that actually needs proving — a refused submit does not WEDGE the session (the bail lands after step 1's frame-type bookkeeping; the next well-sized frame must still encode, and an AU must come out). Verified on the 780M under validation layers: 8/8 vulkan tests, full-suite baseline restored to 14/12. The WP4.2 size-keyed staging stays as belt-and-braces; the hazard it fixed is now structurally unreachable through submit. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 10:20:37 +02:00
enricobuehlerandClaude Opus 5	6a2a153c0b	chore(pf-capture): make the crate verifiable — Windows CI lint + two denies (sweep Phase 0) ... Gate for the rest of design/pf-capture-sweep.md: today half this crate is compiled by no CI job at all, so the Windows-side findings can't even be type-checked. 0.1 (X1) — windows-host.yml lints `-p pf-capture --all-targets`. The host lint above it builds pf-capture only as a DEPENDENCY, so its `#[cfg(test)]` modules were never compiled anywhere: the 5 StallWatch tests, the DXGI HDR self-tests and the cursor-conversion tables were dead code in CI. The workflow's own comment already diagnosed this blind spot and fixed it for pf-encode; pf-capture never got the same treatment. No features on this crate, so no feature juggling and no extra dep tree. 0.2 (X2) — `#![deny(unsafe_op_in_unsafe_fn)]` beside the existing `deny(clippy::undocumented_unsafe_blocks)`. The crate declares "every unsafe block carries a SAFETY proof; enforce it" in ten file headers, but in edition 2021 that lint has nothing to fire on inside an `unsafe fn` — so the hardest FFI in the crate was exempt from its own program: the ring/slot construction, the fence signal, the blend scratch, and every D3D converter ctor/convert. 119 operations across 16 functions now carry a proof. `shared_object_sa` loses its `unsafe` marker instead (it states no precondition — only its RESULT is a raw pointer, which is the caller's business). 0.3 (X4) — drop the crate-wide `#![allow(dead_code)]`. Verified: zero warnings on either target without it, so it was hiding nothing the lint can see (W16's dead items are `pub`/written-once, so they need Phase 1.7's deletion instead). No behaviour change: only lint attributes, `unsafe { }` scoping and comments. Linux `cargo test -p pf-capture` 6/6, clippy --all-targets clean on both targets (Windows verified by cross-checking x86_64-pc-windows-msvc locally). Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 10:18:48 +02:00
enricobuehlerandClaude Opus 5	bf9386ecb2	test(pf-encode): pin the typed-EINVAL classifier's chain-survival contract (Phase 8) ... Rides on 25765c53 (the concurrent session landed the same typed-errno fix first — this branch's twin commit was dropped in the rebase; these are its surviving residuals): - Two tests pinning what the ladder's step-down rests on: the typed ffmpeg::Error survives with_context layers as a downcastable source (an eager format! anywhere between open_with and the ladder would silently break it), and the phrase WITHOUT the type no longer classifies. - The CudaHw::new eager-format guard note: those bail!s must NOT be converted to typed errors — a hwdevice/hwframes EINVAL is a config error no bitrate can fix, and enrolling it would burn ~10 doomed opens before surfacing the real failure. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 10:13:49 +02:00
enricobuehlerandClaude Opus 5	bf9fb3fb22	fix(pf-encode): enable VK_EXT_queue_family_foreign for the dmabuf acquires (Phase 8) ... Both Linux Vulkan encode backends named QUEUE_FAMILY_FOREIGN_EXT as the acquire barriers' src family without ever enabling the extension — spec-invalid on every device, tolerated by RADV. The audit filed vulkan_video's three sites; pyrowave's fresh-import acquire had the identical defect on its own device (critic catch). Enable when advertised (a fresh open-time enumerate — the rgb probe's is a probe-local and skipped entirely on native-NV12, so there was nothing to reuse; pf-presenter/dmabuf.rs is the in-repo precedent that already enables this extension). Not advertised → the core-1.1 QUEUE_FAMILY_EXTERNAL conservative substitute, chosen once at open and warn-logged (no fleet hardware takes that arm; such devices were never valid targets before). All four sites are acquire-only (src=FOREIGN, EXCLUSIVE images, oldLayout=UNDEFINED) — the swap is index-only. On-glass: 780M under validation layers — vulkan smokes + pyrowave smokes green, FOREIGN advertised and enabled, no fallback engaged. Shared ext_advertised helper in vk_util (cfg = the union of both consumers) with a unit test. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 10:08:44 +02:00
enricobuehlerandClaude Opus 5	1cefd37603	fix(pf-encode): arbitrate NVENC split-encode vs sub-frame readback (Phase 8) ... Verified against nvEncodeAPI.h's own splitEncodeMode doc (user-prompted — the audit's 'exclusive for HEVC' one-liner deserved checking): - H.264: split 'is not applicable' — hard-DISABLE the mode so the written config, CeilingKey, the split diagnostic log and the rejection-retry stay truthful (the retry used to re-open a byte-identical session after an H.264 'split rejection'). The libav path's operator arm gains the codec gate its auto arm always had. - HEVC: split 'not supported if … subframe mode' — when WE force split (TWO/THREE/AUTO_FORCED, the 4K120 throughput lever), sub-frame yields with a logged escape (PUNKTFUNK_SPLIT_ENCODE=0 chooses sub-frame). ⚠ Keyed on FORCED modes only, never != DISABLE: AUTO(0) is the resolver's fallthrough for every sub-950Mpix session, and the wider key would have disarmed the Phase-3 chunked-poll feature fleet-wide (critic catch). Under AUTO the driver arbitrates — the shipped state. - AV1: untouched — per-tile sub-frame + split are legal together. The arbitration is a pure nvenc_core fn called by each backend BEFORE the ladder, the ceiling key and the chunked-poll latch — all three see the post-arbitration truth. A drop inside build_init_params would have left poll_chunk busy-polling its whole budget every AU (numSlices stays 0 without reportSliceOffsets; both loop exits dead — critic catch). Linux latches subframe_forced beside subframe_on at query_caps (no env re-reads after open); Windows records the arbitrated state so reconfigure presents exactly the params the open had (also closes the pre-existing mid-session env-flip hazard there). Truth-table tests in nvenc_core; PUNKTFUNK_NVENC_SUBFRAME documented (it never was); PUNKTFUNK_SPLIT_ENCODE row updated. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 10:08:44 +02:00
enricobuehlerandClaude Fable 5	25765c53ec	fix(encode): classify libav-NVENC open failures by errno, not English strerror text ... The bitrate-probe ladder stepped down on format!("{e:#}").contains( "Invalid argument") — an English substring over the WHOLE context chain, which also fired on any other wrapped EINVAL (e.g. a CUDA-context errno) and gated a ~10-step ladder on strerror wording. The root ffmpeg::Error survives the anyhow chain; downcast and match Error::Other{errno:EINVAL} instead. Same fix for the intra-refresh ENOSYS probe in the open path. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>	2026-07-26 01:05:24 +02:00
enricobuehlerandClaude Fable 5	36589c39da	fix(gamestream): stall-recovery ladder — a wedged encoder no longer ends the stream ... The backends deliberately turn a wedged GPU into a bounded error so the caller can reset in place, but only the native plane ever did: on the GameStream path every submit/poll error propagated straight out of the video loop, costing Moonlight clients a full disconnect/reconnect. Port the native ladder (bounded resets + backoff + silent-wedge watchdog, same field-lesson backoff curve). Also keep the drop-recovery keyframe armed until it actually EMITS: it was consumed on read, so the coalesce gate could swallow it for good — leaving duplicate wire indices in the encoder's reference table for a later RFI to anchor on, exactly the stale-anchor case rfi.rs exists to prevent, and it opened only under congestion + loss, which is when RFI fires. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>	2026-07-26 01:05:23 +02:00
enricobuehlerandClaude Fable 5	e3354b6d5d	fix(encode/vulkan): guard the CSC source dimensions, and bound reset()'s wait ... The CSC path was the only backend arm that took frame.width/height on trust: the shader samples with clamped 1:1 texelFetch, so a mismatched frame silently streamed a cropped/edge-padded picture where every sibling errors into the encoder-rebuild path. The import cache now also carries the extent it imported at (a (st_dev, st_ino) hit alone doesn't prove the allocation still matches) and is dropped on reset(). reset() opened with an untimed device_wait_idle on the one thread whose every other wait is capped at ENCODE_FENCE_TIMEOUT_NS for exactly this reason — reset() runs BECAUSE the GPU looks wedged. Both vulkan-video and pyrowave now bound the wait and report "no in-place rebuild" on timeout instead of parking recovery on the suspect device; Drop keeps the unbounded wait (teardown must stay memory-safe against a wedged device). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>	2026-07-26 01:05:10 +02:00
enricobuehlerandClaude Opus 5	28f8fc71c4	refactor(pf-encode): split vulkan_video's construction tail into vk_build.rs (WP7.5) ... The ~820-line tail of free builders — make_frame/make_frame_csc/ make_frame_common, make_video_image, probe_rgb_direct, the H.265/AV1 parameter-set writers and the AV1 bit-writer — moves to a #[path] child module, the amf_sys.rs shape: the child sees the parent's private items (Frame and friends), so the split costs zero visibility churn. Six parent-called items went pub(super); five stay child-private (dead_code is per-item and each is used within the child). vulkan_video.rs drops 5,292 → 4,489 lines and the construction unsafe gets its own review surface; steady-state encode logic stays in the parent. ⚠ Trap recorded for future child-module splits: inline `use super::X` statements INSIDE moved fn bodies silently change meaning (super shifts one level) — vk_av1_encode/vk_valve_rgb imports needed crate:: paths. Proven on-glass, not just compiled: all 8 vulkan GPU smokes green under the validation layers on the 780M post-split (H.265 + AV1, RGB-direct, CSC, CPU paths — every moved constructor exercised). nvenc_cuda.rs and qsv.rs are DECLINED the same treatment, with evidence in the handoff doc: no equivalent self-contained seam — their candidate regions are ~150-line loader/accessor clusters interleaved with the encoders' own state types, and a thin-forwarder impl split is exactly the churn a no-defect phase penalizes. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 00:08:03 +02:00
enricobuehlerandClaude Opus 5	bc5ec5105c	refactor(pf-encode): one Linux backend resolver, consumed by dispatch AND mirrors (WP7.6) ... The Linux backend decision existed as open_video_backend's string match plus five partial hand-copies (the zero-copy-plane gate, the pyro advertisement gate, the software advertisement pin, the vulkan pref ceiling, resolved_backend_is_gpu). Windows solved this long ago — windows_resolved_backend() is consumed by its dispatch AND its mirrors, with labels still stamped at the open sites. Linux now has the twin: resolve_linux_backend (pure, lazy auto probe) + linux_resolved_backend (config wrapper, unknown→auto exactly as every mirror's old `_` arm). ⚠ This deliberately DEVIATES from the audit's shadow-assertion prescription, on both critics' findings: the shadow had no execution venue (open_video had ZERO test call sites; shipped hosts are --release) — unfalsifiable ceremony — and the file's own Windows half proves dispatch-consumes-resolver is safe: the mgmt record is protected by the label-from-the-open-site convention, not by resolver avoidance. One consumed table beats two tables plus an inert cross-check. Preservation riders from the critique, all applied: - pref_ceiling KEEPS its cfg!(vulkan-encode) branch (the resolver is feature-blind; dropping it re-creates advertise-then-die-at-open). - linux_zero_copy_is_vaapi's Vulkan|Software arm preserves the old `_` fallthrough EXACTLY — the vulkan-on-NVIDIA capture-plane mismatch and the software twin are FILED in the design doc, not fixed here. - resolved_backend_is_gpu splits as linux + not(any(windows, linux)) — never a target list, so no exotic target loses the fn. - The auto probe is lazy (impl FnOnce) — explicit prefs stay zero-probe (/serverinfo polls through these mirrors), pinned by a panicking closure in the resolver test. New coverage: the full alias table pinned, and a GPU-free dispatch test through the real open_video_backend (software arm, vendored openh264) — its first test call site anywhere. Config-latch seam guarded loudly. Cross-crate mirrors recorded out of scope in the design doc: session_plan::resolve_encoder, gamestream/serverinfo.rs base_codec_mode_support, capture.rs's "pyrowave" string. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 00:08:03 +02:00
enricobuehlerandClaude Opus 5	cb69cd3b8c	refactor(pf-encode): move the AMF C-ABI mirror to its own file (WP7.4) ... The ~410-line hand-mirrored AMF vtable ABI was already isolated in an inline `mod sys` — the remaining value of the audit's 'best split in the crate' is the FILE boundary: amf.rs drops 3379 → 2965 lines and the pure unsafe-FFI surface (25 unsafe fns, zero policy) is reviewable in isolation, which is the crate's stated review goal. Pure move: `#[path = "amf_sys.rs"] mod sys;` keeps the module name and every `sys::` call site byte-identical; the module was self-contained (only `use std::ffi::c_void`, no super:: references). The banner became the file's module docs; contents de-indented one level — rustfmt-clean on the first check, which is what proves the move byte-exact. Gated on .173: all five clippy combos + the amf-qsv,qsv test leg (38 passed — the live AMF matrix on real VCN silicon among them). Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 00:08:03 +02:00
enricobuehlerandClaude Opus 5	9491cc8759	refactor(pf-encode): extract the range-family RFI recovery policy (WP7.2) ... The two direct-NVENC backends carried hand-copied twins of the same loss-recovery decision: range validity, covering-range dedup, DPB window, clamp — ~30 duplicated lines each. The decision now lives once as nvenc_core::plan_range_recovery (the range half of WP7.2; the slot half is enc/rfi.rs), pure and unit-tested; each backend keeps its session gate, its unsafe per-timestamp driver loop, and its state stores. The step order is load-bearing and now pinned by tests: the covering dedup runs with the UNCLAMPED last and BEFORE the DPB window (a covered re-ask never touches the driver even when the range has since aged out of the DPB), the boundary at next_ts - RFI_DPB is inclusive, and the Invalidate carries the CLAMPED last — which is also what the caller records in last_rfi_range, exactly as the inline code stored it. A driver failure mid-loop still returns false with NO range recorded and no anchor armed. Decline deliberately clears nothing (neither twin touched pending_anchor on decline — same shape as Vulkan's non-clear, opposite of AMF/QSV; do not harmonize). The exact-cover → Covered test records EXISTING behavior including that a covered range survives a forced IDR with zero driver calls — a recorded fact, not an endorsement. RFI_DPB's import leaves both twins: its only per-backend use was the arithmetic that moved. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>	2026-07-26 00:08:03 +02:00