design/latency-reduction-2026-07.md T0.1/T0.2/T1.2/T1.3:
- T1.2 rate-capped front-loaded pacing: the paced overflow's budget is now
min(0.9x slack, overflow wire time at ~3x the live encoder bitrate)
(PUNKTFUNK_PACE_FACTOR, 0 = legacy deadline-only spread). A 300 KB-1 MB
frame's tail leaves in ~2-5 ms instead of smearing across ~15 ms at 60 fps;
GameStream schedule byte-identical (pins unchanged).
- T1.3 data-first wire order: packetize emits every block's data shards before
any parity (per-block parity pools keep all blocks' parity alive for the
second pass), so lossless completion stops waiting behind the parity tail.
EOF = last emitted packet; receiver already order-agnostic.
- T0.1 staged 0xCF: HostTiming gains an append-extensible per-stage tail
(queue/encode/pace us; seal+channel-wait derived as residual) - no cap bit
needed, old peers read the 13-byte prefix. Joined client-side into
Stats::host_{queue,encode,xfer,pace}_ms, the OSD detailed tier, and the
probe's report.
- T0.2 true on-glass present timing: VK_KHR_present_id/present_wait enabled
when supported; a PresentTimer waiter thread resolves each present id to
real visibility, replacing the submit-time display stamp (which undercounts
by up to a refresh and hides a silent-FIFO standing queue).
Validated on .21: core 185 + host 185 tests, pf-presenter 19, clippy
-D warnings across all five touched crates; loss-harness recovery curve
unchanged; C ABI harness round-trips.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The low/medium findings from the July host+Windows security review, as
implemented in the audit session's working tree:
- Webhooks and library-art fetches refuse loopback/link-local/metadata
targets and no longer follow redirects (SSRF pivots from the privileged
host process).
- The paired-client rosters (/clients, /native/clients) move off the
streaming-client auth lane — one paired device could enumerate every other
device's name + fingerprint; only the bearer/loopback console keeps them.
- Device-name sanitizing extends to bidi/format control characters
(spoofable rendering) via the shared native_pairing::is_spoofy_char;
stream-marker quoting uses the same set.
- The SYSTEM service resolves powershell by its full System32 path —
CreateProcess checks the launching EXE's own directory first, so a planted
powershell.exe beside the host binary would have run as SYSTEM.
- The pf-vdisplay driver's opt-in file log moves from world-writable
C:\Users\Public to WUDFHost's own temp dir.
- GameStream pairing sessions are single-use (removed whatever the outcome).
- Uninstall also removes the pf_mouse driver-store entry (rider from the
virtual-HID-mouse work).
- openapi.json regenerated (hardened-config-dir doc wording).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Root cause of the field report (Android client, Windows host: no audio until
the user manually cycled Sound-output devices, then audio on BOTH PC and
phone): the WASAPI loopback captured whatever the default render endpoint
was at open time — not the wiring plan's chosen endpoint — the plan's
IPolicyConfig default-set is warn-only and racy, and nothing reacted to
mid-stream default-device changes.
- Explicit-endpoint capture: the capture thread opens the plan's
loopback_render by id, never "the default" (KEEP_DEFAULT preserves the old
default-capturing behavior with the echo guard).
- Client-only playback default: wiring_plan::plan(..., host_audio) prefers a
silent sink (Steam Streaming Microphone render side — loopback-validated,
silent on host) over real hardware, so stream audio plays on the client
only; PUNKTFUNK_HOST_AUDIO=1 restores real-hw-first (audible on the host).
The capture side auto-installs the Steam pair once per process when no
silent sink exists; open() handshake timeout 3s -> 30s to cover it, and a
handshake timeout now stops the detached thread (it used to run for the
process lifetime with the default still parked).
- Self-healing capture thread (wasapi_cap): outer capture_once loop
(Assert|Follow) with a ~1s watchdog on the default render id. A user
switch to a capturable endpoint is followed (their choice wins, audio on
both); a switch to a dud (cable/SSS/mic target) re-asserts the plan;
IPolicyConfig-denied converges to Follow instead of churning. Device
errors reopen with 2s backoff; only the FIRST open failure is fatal.
Zero-packets breadcrumb after 30s distinguishes broken-loopback from
quiet-desktop.
- Park/restore of the default playback device (audio_control):
wire_now(set_playback) parks the default on the loopback sink only for the
capture's lifetime (the mic pump passes false — it runs while the host is
idle); crash marker audio-default.prev + recover_orphaned_default() at
first wire; restore is skipped if the operator changed the default
themselves. A mic-default hygiene pass keeps VB-Cable installs audible and
never records the mic target as the restore target.
- Session-end park_audio_capture(): Windows DROPS the capturer (thread join
restores the default) instead of caching it; Linux keeps the parked
PipeWire thread. Composes with the stream-sink idle() hook at all three
park sites (idle is a no-op on Windows).
Verified: Linux (.21) clippy -D warnings + 176 punktfunk-host tests green
(incl. the new wiring-plan preference tests); Windows (.173) clippy with
nvenc,amf-qsv --all-targets -D warnings green at this exact tree.
On-glass winbox/Android validation still owed.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Merges 44b71e74 (see its message for the full design: the "Punktfunk Stream
Speaker" Audio/Sink claimed as the session's default output — the crackle
root fix for hardware-sink churn — plus the core-error liveness fix and true
5.1/7.1 capture) across the W1–W8 refactor that landed since the branch
forked:
- punktfunk1.rs was split on main: the audio_thread idle()/drain() hunks are
hand-ported into native/audio.rs (same three sites: reuse-drain comment,
encoder-fail park, end-of-thread park).
- pwinit moved to pf-capture (W6.2): the branch's crate::pwinit calls in
pw_thread and stream_sink.rs become pf_capture::pwinit.
- Kept main's log-tier demotions (stream-state + setup lines at debug)
inside the restructured pw_thread.
Resolution verified on Linux (home-worker-5): clippy -D warnings clean +
173 punktfunk-host tests green at the 691c064a-based equivalent ba5973a2;
re-verified at this base before landing.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Brings the first-frame-latency branch (P0.1 transition tracing, P1.1/P1.2
Welcome-time display prep, P2 in-place resize; pf-driver-proto v3 -> v4 with
IOCTL_UPDATE_MODES) onto current main. The branch predates the W6.2/W7 splits,
so git's rename detection carried most of it into the moved crates
(pf-capture idd_push, pf-vdisplay manager/pf_vdisplay, pf-win-display,
pf-driver-proto, the driver workspace) and the punktfunk1.rs remainder was
re-homed by hand:
- native/handshake.rs: welcome/start trace marks + the Welcome-time display
prep spawn (the prep thread BECOMES the stream thread; hand-off via a
SyncSender<SessionContext>). negotiate() gains bringup/quit/stop and returns
the PrepHandle.
- native.rs: bringup/resize_ms creation + the stop/quit flags hoisted BEFORE
the handshake (the close watcher splits: flags pre-handshake, lifecycle
events post-handshake where `hello` exists); punch_done stamp; the data
plane adopts the prep thread's result or builds inline.
- native/stream.rs: SessionContext/SendStats carry the trace; send_loop
finishes it on the first video packet; the resize path gains the in-place
fast path (try_inplace_resize) with the full rebuild as fallback, restructured
so both share the post-rebuild bookkeeping; prepare_display/PreparedDisplay/
PrepHandle; build_pipeline(+retry) thread the stage marks.
- session_status/mgmt: ttff_ms + last_resize_ms per session (union with the
lifecycle-events fields main added to the same spots).
- pf-capture: Capturer gains capture_target_id() + resize_output() defaults.
- pf-vdisplay manager: perf's faster activation poll (60x50ms) + the settle
floor before the PnP sweep, on main's knobs/no-trait shape.
Also: packaging/windows/build-gamepad-drivers.ps1 is ASCII again (an em-dash
from the pf-mouse work tripped windows-host.yml's locale-safety gate on main).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
origin/main landed the shared clipboard (design/clipboard-and-file-transfer.md) while
this branch split quic/msgs.rs -> quic/{caps,control,...} and client.rs ->
client/{mod,control,worker,pump,planes,...} (W7) and deleted the two monoliths. The
feature had modified both deleted files, so its delta is re-applied onto the split
instead of resurrecting the monoliths:
- HOST_CAP_CLIPBOARD -> quic/caps.rs
- MSG_CLIP_* / CLIP_* consts, the six Clip*
structs, and their encode/decode impls -> quic/control.rs (beside the clock codecs)
- CtrlRequest::{ClipControl,ClipOffer} +
Negotiated.host_caps -> client/control.rs
- WorkerArgs.{clip_event_tx,clip_cmd_rx} -> client/worker.rs
- CLIP_EVENT_QUEUE -> client/planes.rs
- NativeClient clip fields, the 7 clip_* /
host_caps / next_clip methods, connect()
channel wiring -> client/mod.rs
- the control-task encode/decode arms and
the clipboard-task spawn -> client/pump.rs
Cargo.lock reconciled (adds pf-clipboard), punktfunk-host/Cargo.toml unions the W6
pf-* subsystem deps with pf-clipboard, and include/punktfunk_core.h is the cbindgen
union (clipboard + rumble C-ABI). punktfunk-core builds --all-features and its 174
lib tests pass, including quic::tests::clip_loopback.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The host half of the shared clipboard (design/clipboard-and-file-transfer.md §4),
ported from feat/shared-clipboard (6bd8c18b) into the post-W6 crate shape: the
backends land as a pf-clipboard subsystem crate (the pf-inject/pf-capture
pattern) instead of growing punktfunk-host back out, and the ~340-line
punktfunk1.rs integration is re-implemented against the native.rs/control.rs
split that replaced it.
pf-clipboard:
- host::wayland — ext-data-control-v1 (KWin / wlroots / Sway / Hyprland).
- host::mutter — GNOME via Mutter's *direct* org.gnome.Mutter.RemoteDesktop
clipboard (no data-control at any GNOME version; the xdg portal needs an
interactive grant a headless host can't answer).
- host::windows + host::winfmt — Win32 clipboard on a hidden message-loop
window: WM_CLIPBOARDUPDATE listener + OLE delayed rendering (WM_RENDERFORMAT)
for text / CF_HTML / RTF / PNG.
- host::session — the backend-agnostic coordinator bridging HostClipboard to
the QUIC clipboard plane (offers, fetch accept-loop, remote offers, pastes).
- A portable facade (policy / enabled / cap_advertised / ClipCoordCmd / start /
spawn_decline_loop) so the orchestrator compiles cfg-free on every platform;
ClipCoordCmd moves into the crate (it was host-owned before).
punktfunk-host glue:
- handshake.rs advertises HOST_CAP_CLIPBOARD via pf_clipboard::cap_advertised.
- serve_session starts the coordinator (gated on a real compositor — the
synthetic source stays out of the session clipboard) and spawns the
CLIP_FETCH_UNAVAILABLE decline loop when the policy is on but no backend bound.
- control.rs gains the ClipControl/ClipOffer arms + the host-offer forward
branch, and the e2e session test (cap advertise → ClipState ack with
BACKEND_UNAVAILABLE → fetch decline) rides in native.rs's tests.
Still opt-in default OFF (PUNKTFUNK_CLIPBOARD). Remaining: the macOS client
(design §5) — then this becomes user-visible.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
vdisplay.rs + vdisplay/* (the per-compositor Linux backends — KWin zkde-screencast,
wlroots swaymsg, Mutter RemoteDesktop, Hyprland — and the Windows IddCx/pf-vdisplay
driver backend, behind one VirtualDisplay trait; the mode-conflict admission
registry, the display policy/identity/custom-preset state, and the session-env /
gamescope routing) move into crates/pf-vdisplay (plan §W6). The DDC/CI panel-power
control (used only here) and the KWin zkde protocol XML move with it. This
completes the host-crate decomposition: capture, encode, inject, and vdisplay are
now four subsystem crates over the shared leaves, and punktfunk-host is the
orchestrator (serve/supervisor + native + gamestream + mgmt).
Coupling breaks (all down-only, cargo-tree acyclic):
- capture::dxgi identity -> pf_frame::dxgi; win_display/monitor_devnode/
console_session_mismatch -> pf-win-display leaf; can_open_another_session ->
pf-encode (the NVENC session-budget admission gate — acyclic peer edge).
- The registry's DisplayCreated/DisplayReleased emits into the host SSE event bus
invert to a leaf hook: pf-vdisplay emits a neutral DisplayEvent to a
host-registered DISPLAY_EVENT_SINK, so it never reaches the orchestrator's
events module.
- The IddCx driver module is renamed pf_vdisplay -> driver (its old name collided
with the crate name through the host's `mod vdisplay` shim glob).
The host keeps `mod vdisplay { pub use pf_vdisplay::* }` so every crate::vdisplay::*
path (serve/mgmt/native/the capture FrameChannelSender seam) is unchanged; the
heavy deps (wayland/ashpd/tokio + the zkde protocol) moved with the crate.
Co-authored: a fail-closed IOCTL-reply-length security fix (reject short/zeroed
pf-vdisplay driver replies before trusting protocol_version/target_id/wudf_pid/luid,
security-review 2026-07-17) rides this commit in the moved driver module.
Verified: Linux clippy -D warnings (pf-vdisplay + host nvenc,vulkan-encode,pyrowave
--all-targets) + pf-vdisplay 63/63 + host 167/167 tests; Windows clippy -D warnings
(pf-vdisplay --all-targets + host nvenc,amf-qsv --all-targets) Finished exit 0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
inject.rs + inject/* (the per-OS injectors — wlroots virtual-input, KWin
fake_input, libei/reis, gamescope-EI on Linux; SendInput on Windows — plus the
virtual-gamepad HID stack: DualSense/DualShock4/Switch Pro/Steam Controller/Deck
over uhid/usbip and the Windows UMDF drivers, the proto codecs, the injector
service, and the uhid manager) move into crates/pf-inject behind the
InputInjector trait (plan §W6). It consumes punktfunk_core::input (the neutral
GamepadEvent/InputEvent vocabulary, moved to core in W5) + the pf-driver-proto
wire contract, and reaches pf-capture only for the Windows gamepad-channel
WUDFHost check + the resident-mouse compose-kick hook.
The one inject->vdisplay coupling (the libei gamescope-EI backend needs the EIS
relay socket path) is broken via a leaf: gamescope_ei_socket_file moves to
pf-paths as the shared contract — the gamescope producer (host vdisplay) keeps
its session-env-lock wrapper around it, the libei consumer (pf-inject) reads it
directly post-retarget. The host keeps a `mod inject { pub use pf_inject::* }`
shim so every crate::inject::* path (the native/gamestream input planes + devtest)
is unchanged; the heavy input deps (wayland/reis/xkbcommon/usbip + the KWin
fake-input protocol XML) moved with the crate.
Verified: Linux clippy -D warnings (pf-inject + host nvenc,vulkan-encode,pyrowave
--all-targets) + pf-inject 69/69 + host 230/230 tests; Windows clippy -D warnings
(pf-inject --all-targets + host nvenc,amf-qsv --all-targets) Finished exit 0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The final piece of the lid-closed field report (fixes 1-3: 3d9b3290;
prerequisites: 85dd2bb0 pf-mouse, 845a9760 leaf primitives, 94ca4041
pf-capture hook): the IDD first-frame gate can only pass if DWM composes
at least one frame, and a lid-closed/locked/idle machine is exactly the
state Windows has decided not to compose in. The SendInput compose kick
is conditional on this process's context — wrong session → wrong input
queue; secure desktop → blocked; display powered off → no wake. A report
from the resident pf-mouse HID device is REAL input to win32k:
session-independent, secure-desktop-proof, wakes a powered-off display,
counts as user presence. Parsec-class mechanism, and semantically honest
— a remote user starting a stream IS a user arriving at this machine.
- hid_kick(rect, bounds): newest-wins kick slot + condvar, serviced by
the keeper thread that owns the ONE process-wide VirtualMouse (a
second open() would squat the bootstrap mailbox). Not-ready (opt-out,
driver missing, not yet attached) returns false → the capture crate
falls back to SendInput.
- perform_kick: park the pointer at the target display's center, dwell
35 ms (Stage-W3: DWM samples cursor position at the next vsync tick;
the gaps also keep reports from coalescing in the driver's 8 ms
timer), wiggle ~2 px, restore the saved position. Desktop→HID
coordinates normalize against pf_win_display::desktop_bounds() (CCD
union — correct from any session, unlike per-session GDI metrics).
- ensure_resident registers the hook (pf_capture::HID_COMPOSE_KICK) —
the one-way-edge inversion: pf-capture never reaches into inject.
- keeper loop: condvar wait (250 ms tick) so a kick executes
immediately, not at the next tick; publishes MOUSE_READY from
driver_proto.
Paired with pf-frame's DisplayWakeRequest (held by the capturer from
before the first-frame gate): the power request keeps the display from
going dark mid-session, the HID kick wakes it when it already is.
Verified on winbox: combined-tree cargo check + clippy for
punktfunk-host, pf-capture, pf-frame, pf-win-display all EXIT 0.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
capture/linux (PipeWire portal) + capture/windows (IDD direct-push: dxgi
mechanics, idd_push + submodules, synthetic_nv12) + pwinit move into
crates/pf-capture behind the Capturer trait + synthetic sources (plan §W6).
The crate speaks pf-frame (CapturedFrame/PixelFormat + the DXGI identity),
pf-zerocopy (CUDA import), and the pf-win-display leaves, and NEVER pf-encode —
the capture->encode edge is one-way. This completes the deliberate capture/encode
crate split (the invasive path the plan had merged into one pf-media): capture
and encode are now separate subsystem crates sharing only pf-frame.
Four seams keep the capturer off the orchestrator:
- VirtualOutput is EXPLODED into primitives (remote_fd/node_id/preferred_mode/
keepalive) by the host facade, so pf-capture never depends on the vdisplay type;
- FrameChannelSender: the sealed-channel delivery is a Send+Sync closure the host
facade builds from the pf-vdisplay control device + send_frame_channel IOCTL and
hands in; ChannelBroker holds the closure instead of the control HANDLE (the
whole-desktop handle-duplication security boundary is byte-for-byte unchanged);
- console_session_mismatch + desktop_bounds live in pf-win-display (leaf peers);
- pwinit moves here (audio caller -> pf_capture::pwinit).
The host keeps capture.rs as a thin BRIDGE: it re-exports the vocabulary + capturer
types (every crate::capture::* path is unchanged) and keeps open_portal_monitor /
capture_virtual_output, which resolve the ZeroCopyPolicy + FrameChannelSender and
call into pf-capture. verify_is_wudfhost + install_gpu_pref_hook are re-exported
(the gamepad-channel bootstrap + the main.rs subcommand consume them).
Co-developed: the resident-HID-mouse compose-kick hook (HID_COMPOSE_KICK + the
HID-first cursor kick + _display_wake) rides this commit into pf-capture; the host
mouse_windows registration side lands separately on top.
Verified: Linux clippy -D warnings (pf-capture + host nvenc,vulkan-encode,pyrowave
--all-targets) + host tests 299/299; Windows clippy -D warnings (pf-capture
--all-targets + host nvenc,amf-qsv --all-targets) Finished exit 0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Headless Windows hosts (no dongle) stream an INVISIBLE cursor: with no
pointing device present win32k reports SM_MOUSEPRESENT=0 and DWM never
composites a pointer into the pf-vdisplay frame, even though SendInput
moves it. Keep ONE virtual HID mouse devnode alive for the host's
lifetime — the Sunshine/Parsec-class fix, zero client changes.
- pf-mouse: UMDF2 HID minidriver, one fixed identity (PF:MO 5046:4D4F,
obviously virtual, nothing fingerprints it), one 8-byte input report
(5 buttons + absolute 15-bit X/Y + wheel + AC-pan). Transport is the
sealed pad channel verbatim (Global\pfmouse-boot-0 mailbox + unnamed
MouseShm DATA section) so pf-umdf-util's audited layer serves it
unchanged; report delivery is event-driven (idle = no HID traffic).
- host: inject::mouse_windows — VirtualMouse (SwDeviceCreate'd devnode +
channel), ensure_resident() keeper thread started by every
InjectorService (process-wide, PUNKTFUNK_NO_VIRTUAL_MOUSE opts out),
vmouse-spike on-glass validation (cursor sweep via HID reports).
- proto: mouse module (magic, boot-name, identity, report layout,
unit-tested input_report packing).
- SwDeviceProfile grows container_tag so the mouse's ContainerId family
(PFMO) never groups with a pad's (PFDS) in the Devices UI.
- packaging: pf-mouse rides the gamepad-driver build + install pipeline
(build-gamepad-drivers.ps1, windows-drivers.yml, driver install
--gamepad picks up every staged .inf).
On-glass validated on winbox: devnode + HID child bind, SM_MOUSEPRESENT=1
with no physical mouse, cursor sweeps via HID reports (vmouse-spike).
This work was implemented in a parallel session; committed here as the
build prerequisite for the HID compose kick that follows.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Field report (Windows laptop, lid closed, Tailscale): v0.12.0's activation
fix works — the pf-vdisplay target activates in ~200ms — but the session
still dies at the first-frame gate: 'driver_status=1 but no frame published
within 4s'. Triage showed three independent blind spots; this names all of
them at their source instead of guessing downstream:
- pf-win-display: decode ChangeDisplaySettingsExW failures (-1 FAILED — a
display write rejected, the wrong/remote-session signature — vs -2 BADMODE,
which the old 'mode not advertised?' text conflated), and WARN on every
non-zero SetDisplayConfig rc in the CCD isolate even when verification
passes vacuously (the lid-closed case: nothing else active, so the INFO
swallowed rc=0x5 ERROR_ACCESS_DENIED while the load-bearing COMMIT_MODES →
ASSIGN_SWAPCHAIN re-commit silently never applied). Access-denied rcs get
the remedy appended (console session / installed service).
- host: console-session guard (interactive::console_session_mismatch) — a
host outside the active console session (a hand-launched host after an RDP
round-trip) fails every display write, reads the wrong session's GDI view,
and its SendInput compose kicks go nowhere. Named ERROR at vdisplay
acquire + appended to the first-frame timeout, instead of the misleading
generic failure. (The idd_push diagnosis half of this landed in 9a36ea21;
this commit adds the proto helpers + session guard it references, healing
the windows-cfg build.)
- proto + driver: while OPENED, driver_status_detail now carries a live
packed word (bit31 live-marker | offered 15-bit | mismatch-dropped 16-bit)
maintained by the publisher, so the host's first-frame timeout can tell
apart: never-attached (no swap-chain worker ran), attached-but-DWM-composed-
zero-frames (undamaged/powered-off desktop, kicks blocked on the secure
desktop), and composed-but-every-frame-mismatched (ring sized from a stale/
foreign-session GDI mode). Zero layout change, old drivers read as 'no
detail'; unit-tested pack/unpack in pf-driver-proto.
Verified on winbox: cargo check + clippy -p punktfunk-host -p pf-win-display
-p pf-driver-proto EXIT 0, drivers ws cargo check -p pf-vdisplay EXIT 0
(Version_Number=10.0.26100.0), cargo fmt --all --check clean; pf-driver-proto
tests 13/13 pass locally.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
encode.rs + encode/* (NVENC, VAAPI, native AMF, AMF/QSV ffmpeg, direct-SDK
NVENC/CUDA, raw Vulkan-Video, PyroWave, openh264) move into crates/pf-encode
behind one Encoder trait + open_video selector (plan §W6). The crate speaks the
shared frame vocabulary (pf-frame: CapturedFrame/PixelFormat + the DXGI identity
D3d11Frame/make_device) and pf-zerocopy (CUDA context/buffers), and NEVER
pf-capture — the capture→encode edge is one-way (ZeroCopyPolicy, prior commit).
Dep moves: the heavy encoder deps (ffmpeg-next, the NVENC SDK, openh264,
pyrowave-sys) move from the host to pf-encode; the host's
nvenc/amf-qsv/vulkan-encode/pyrowave features now FORWARD to pf-encode/*. The
host keeps a mod-encode shim (pub use pf_encode) so every crate::encode::* path
(negotiator + GameStream/native/mgmt planes) is unchanged.
resolve_render_adapter_luid moves from the host's windows/win_adapter.rs into
pf-gpu (both pf-encode and pf-capture need it as a peer of GPU selection); its 5
call sites (encode amf/nvenc, capture idd_push/synthetic_nv12, vdisplay manager)
rewire to pf_gpu::resolve_render_adapter_luid and win_adapter.rs is deleted.
pf-frame's make_device gains a # Safety section (public-unsafe-fn lint, latent
since the pf-frame carve — a full-workspace -D warnings clippy catches it).
Verified: Linux clippy -D warnings (pf-encode + host nvenc,vulkan-encode,pyrowave
--all-targets) + 13/13 pf-encode + 299/299 host tests; Windows clippy -D warnings
(pf-encode nvenc,amf-qsv --all-targets + host nvenc,amf-qsv --all-targets)
Finished exit 0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The PipeWire capture thread re-derived three encode-backend facts by calling
encode::{linux_zero_copy_is_vaapi, resolved_backend_is_gpu,
pyrowave_capture_modifiers} — a capture→encode back-reference that would force
pf-capture to depend on pf-encode (and vice versa, since encode already speaks
the frame vocabulary). Resolve them ONCE in the host capture facade (which may
reach crate::encode) as a ZeroCopyPolicy { backend_is_vaapi, backend_is_gpu,
pyrowave_modifiers } and thread it into PortalCapturer::open /
from_virtual_output → spawn_pipewire → pipewire_thread.
capture/linux/mod.rs now makes ZERO crate::encode calls — the edge is one-way
(plan §2.4 / §W6), so pf-capture can be extracted depending only on pf-frame
(not pf-encode). pyrowave_modifiers is computed by the facade whenever the
encoder pref is pyrowave (which implies the VAAPI backend); the thread still
consumes them only inside its existing vaapi_passthrough guard, so behavior is
unchanged.
Verified: Linux clippy -D warnings (host nvenc,vulkan-encode,pyrowave
--all-targets); Windows clippy nvenc,amf-qsv --all-targets Finished exit 0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
windows/{win_display,monitor_devnode,display_events}.rs move into
crates/pf-win-display: the CCD/GDI path-activation + mode-set + HDR
advanced-colour + source-rect helpers, the PnP monitor-devnode enable/disable
lever, and the WM_DISPLAYCHANGE / device-arrival watch. The coming pf-capture
crate's IDD-push capturer consumes all three; the host's pf-vdisplay backend
consumes win_display + monitor_devnode. A leaf lets both depend on them as a
PEER instead of the capturer reaching back into the orchestrator (plan §W6).
win_display's one external tie (crate::vdisplay::Mode) becomes the underlying
punktfunk_core::Mode; the cluster is otherwise self-contained (pf-paths for the
state file, serde_json for it, windows). pub(crate) items bump to pub at the
boundary; win_display carries a module-level allow(missing_safety_doc) to keep
the pre-carve behavior (the FFI helpers were pub(crate) unsafe fn with prose
safety docs — the lint only fires once they're pub, and this is an internal
publish=false leaf). The host imports the three modules at its crate root, so
every crate::{win_display,monitor_devnode,display_events}::* path is unchanged.
Verified: Linux clippy -D warnings (leaf empty + host
nvenc,vulkan-encode,pyrowave --all-targets); Windows clippy -D warnings
(pf-win-display --all-targets + host nvenc,amf-qsv --all-targets) Finished exit 0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The captured-frame types both capture (producer) and encode (consumer) speak —
PixelFormat, OutputFormat, CursorOverlay, CapturedFrame, FramePayload,
DmabufFrame, drm_fourcc — move into crates/pf-frame, alongside the small pure
helpers that ride the same seam: hdr (HDR static metadata / in-band SEI),
metronome (the metronomic-stall detector), thread_qos (per-thread scheduling
QoS), session_tuning (Windows process tuning), and the Windows DXGI capture
IDENTITY (WinCaptureTarget, D3d11Frame, pack_luid, make_device + the GPU
scheduling-priority hardening it applies) (plan §W6).
This is the crate that breaks the capture<->encode cycle: FramePayload's GPU
variants own their backends from BELOW (Cuda -> pf_zerocopy::DeviceBuffer,
D3d11 -> dxgi::D3d11Frame), so encode can speak the vocabulary without a path to
capture, and vice versa. The Windows DXGI identity moving here lets capture,
encode, and pf-vdisplay share ONE WinCaptureTarget/device factory instead of the
old capture<->encode<->vdisplay reach-in.
The host keeps thin facades: capture.rs re-exports the vocabulary
(crate::capture::{PixelFormat,…} unchanged); capture/windows/dxgi.rs keeps the
win32u GPU-preference hook + HDR/video-engine converters + self-test and
re-exports the identity; native.rs re-exports boost_thread_priority from
pf_frame. crate::hdr/metronome/session_tuning callers rewired to pf_frame::*.
metronome's Metronome::new gained a Default impl (new_without_default fires once
the type is public across the crate boundary).
Verified: Linux clippy -D warnings (pf-frame --all-targets + host
nvenc,vulkan-encode,pyrowave --all-targets) + 9/9 pf-frame tests; Windows clippy
nvenc,amf-qsv --all-targets Finished exit 0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
linux/zerocopy/* (CUDA context/buffers + EGL/Vulkan dmabuf import + the isolated
import worker) and linux/dmabuf_fence.rs move wholesale into crates/pf-zerocopy,
so the coming pf-frame vocabulary crate (FramePayload::Cuda owns a DeviceBuffer)
and the pf-encode/pf-capture subsystem crates can reach the GPU plumbing without
the host orchestrator in between (plan §W6). Content stays Linux-only; the crate
compiles to an empty lib elsewhere, so dependents carry a plain dependency.
drm_fourcc deliberately does NOT move: it consumes the frame vocabulary
(PixelFormat), which sits ABOVE pf-zerocopy — it lives with capture for now and
moves into pf-frame next. cuda's ffi re-export bumps pub(crate)->pub (the raw
CUdeviceptr vocabulary is consumed across the crate boundary by the encode
backends). A crate::zerocopy shim module keeps every existing path valid until
capture/encode themselves move out.
Verified: Linux clippy -D warnings (pf-zerocopy --all-targets + host
nvenc,vulkan-encode,pyrowave --all-targets) + 17/17 pf-zerocopy tests + 321/321
host tests; Windows clippy nvenc,amf-qsv --all-targets Finished exit 0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Fourth de-coupling for the host crate carve (plan §W6.1 leaf). gpu.rs (inventory,
selection preference, active-session accounting — deps only pf-host-config + pf-paths, no
subsystem refs) moves to a new pf-gpu leaf so pf-encode/pf-capture/pf-vdisplay can consult
the selected GPU without an orchestrator edge. ~50 crate::gpu:: sites repoint to pf_gpu::;
the ~30 pub(crate) items become pub (crate API). assign_ids gets a macOS-only
allow(dead_code) (used only by the Linux/Windows enumerate arms).
Verified: Linux (home-worker-5) clippy -p pf-gpu -p punktfunk-host --all-targets
-D warnings + pf-gpu tests (12 pass); Windows (192.168.1.158) clippy --features
nvenc,amf-qsv --all-targets green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Third de-coupling for the host crate carve (plan §W6.1 leaf). HostConfig + the config()
OnceLock (config.rs, pure std, zero deps) move to a new pf-host-config leaf so every
subsystem crate (pf-encode/pf-capture/pf-vdisplay/pf-gpu) can read process config WITHOUT
depending on the orchestrator. 34 crate::config::config() call sites across 19 files
repoint to pf_host_config::config(). thread_qos stays in the host for now (it calls
session_tuning::on_hot_thread — its own leaf-ification rides the encode carve).
Granular-crate decision (supersedes the plan's single pf-media): split capture/encode/
vdisplay into separate crates rather than one broad crate — the capture↔encode cycle is
broken by a shared frame-types leaf, and vdisplay→encode (can_open_another_session) is a
legal one-way edge since encode never references vdisplay.
Verified: Linux (home-worker-5) clippy -p pf-host-config -p punktfunk-host --all-targets
-D warnings; Windows (192.168.1.158) clippy --features nvenc,amf-qsv --all-targets green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Second de-coupling for the host crate carve (plan §W6.1 leaf). config_dir /
create_private_dir / write_secret_file (+ the Windows DACL helpers) were pub(crate) in the
gamestream junk drawer, yet consumed by vdisplay, stats, gpu, library, mgmt_token,
native_pairing and the Windows service — many of which become pf-media / pf-vdisplay, for
which crate::gamestream would be an illegal upward edge. New leaf crate pf-paths (pure std
+ tracing) owns them; ~40 call sites across 14 files repoint to pf_paths::. gamestream
keeps only its own concerns.
Verified: Linux (home-worker-5) clippy -p pf-paths -p punktfunk-host --all-targets
-D warnings + tests (347 pass, incl. secrets_are_written_owner_only); Windows
(192.168.1.158) clippy --features nvenc,amf-qsv --all-targets green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Finishes the gamepad-vocabulary un-coupling (plan §W6.0): the Linux uinput button map now
names BTN_* straight from punktfunk_core::input::gamepad instead of the crate::gamestream
re-export aliases, so pf-inject will carry no edge into the gamestream junk drawer for
gamepad constants. Removes the now-dead alias block; the wire-bit pinning test
(gamepad_wire_bits_are_pinned) pins core directly (equally strong — core is the single
source). gamestream keeps only the decode path, which imports the types from core.
Verified: Linux (home-worker-5) clippy --all-targets -D warnings + gamepad tests green;
Windows (192.168.1.158) clippy -p punktfunk-host --features nvenc,amf-qsv --all-targets green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
First de-coupling for the host crate carve (plan §W6.0 / §2.4): the GameStream
(Moonlight-plane) decoded controller types were defined in gamestream/gamepad.rs — the
"junk drawer" — yet consumed 18× by the platform-neutral input injectors AND by the
Moonlight decode path. Once inject becomes pf-inject, reaching them via crate::gamestream
would be an illegal upward edge. Move the two types to core::input (below both planes;
inject already depends on core) and repoint every consumer. Also consolidate the
duplicated MAX_PADS onto the existing core::input::MAX_PADS. The gamestream BTN_* const
aliases stay for now (separate follow-up); decode()/rumble/tests remain in the Moonlight
plane, now importing the types from core.
Verified: Linux (home-worker-5) clippy -p punktfunk-core -p punktfunk-host --all-targets
-D warnings + gamepad tests green; Windows (192.168.1.158) clippy -p punktfunk-host
--features nvenc,amf-qsv --all-targets green (the inject/windows/* consumers compile).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
External game-library providers become first-class (RFC §8): a plugin
computes its desired title list and PUTs it — the host owns the diff.
- CustomEntry gains `provider` + `external_id` (API-set only; never on
manual entries). GameEntry surfaces `provider` for console attribution
and the new `GET /library?provider=` filter.
- PUT /api/v1/library/provider/{p}: atomic declarative reconcile keyed
on the provider's `external_id` — host ids stay stable across syncs,
orphans drop, manual entries and other providers are never touched,
an empty array clears the set. Validated: provider id [a-z0-9._-]
(`manual` reserved), unique non-empty external_ids.
- DELETE /api/v1/library/provider/{p}: clean uninstall, returns the
removed count.
- Ownership is unambiguous both ways: manual CRUD now returns 409 for a
provider-owned entry (MutateOutcome::ProviderOwned) instead of letting
an edit be silently clobbered at the next sync.
- library.changed now carries the mutating source (`manual` or the
provider id) — hooks and the SDK filter on it.
- Spec + SDK schemas regenerated; sdk/examples/provider-sync.ts is the
provider-plugin skeleton.
347 host tests green (pure reconcile: stable ids, orphan drop,
idempotence, bystanders untouched; name/payload validation; route 400s)
+ 11 SDK tests. Live-verified end to end THROUGH the SDK against a real
host: sync → filtered list → manual-delete 409 → re-sync with stable id
+ orphan drop → uninstall (removed=2), with three
library.changed(source=romm) events observed on the live stream.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The capture.rs facade no longer re-derives the encode backend. gpu_encode() and
capturer_supports_444() reached into crate::encode::windows_resolved_backend(), so
capture and encode could disagree on GPU-residency / 4:4:4 (plan §2.4). Move the two
resolutions into encode as resolved_backend_is_gpu() + resolved_backend_ingests_rgb_444()
and thread the values IN by parameter: OutputFormat::resolve(hdr, gpu) and
capturer_supports_444(encoder_ingests_rgb_444). Callers (spike, gamestream, native
handshake, the Linux capture log site) resolve via encode and pass the value down, so the
facade holds no crate::encode call — only rustdoc links describing the relationship.
Completes task #8 of W4.
Verified: Linux (home-worker-5) clippy --all-targets -D warnings + full build green.
Windows (.173) verify owed — box was offline this session.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Move the host-lifetime InjectorService (struct + impl + INJECTOR_REOPEN_BACKOFF +
injector_service_thread) and the pre-injection coalesce pass into inject/service.rs,
alongside the coalesce unit tests. libei_ei_source stays in the facade as an open()
helper. Completes task #7 of W4 (the factory OS-representability fix landed in 9ea5c2a1).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Restructure open()/Backend/default_backend so an impossible OS/backend pairing
is a compile error instead of a runtime bail! (plan §2.3). Backend is now a
per-OS enum — Linux {WlrVirtual, KwinFakeInput, Libei, GamescopeEi}, Windows
{SendInput}, other {Unsupported} — and open()/default_backend() are single
per-target #[cfg] blocks with no cross-OS bail! arms.
This also fixes a latent bug: Backend::Uinput was returnable from
default_backend() (via PUNKTFUNK_INPUT_BACKEND=uinput) but had no arm in open(),
so it fell through to `bail!("not implemented")` — a runtime failure. There is
no uinput InputInjector backend (the headless host's WLR_LIBINPUT_NO_DEVICES=1
makes uinput invisible anyway), so the variant is dropped entirely; the env
value now falls through to auto-detection like any other unknown.
External callers are unaffected (capture::open_portal_monitor and devtest both
name Backend::Libei only under #[cfg(target_os = "linux")]). Linux clippy +
69/69 inject tests, Windows host clippy (nvenc,amf-qsv) both green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Carve three self-contained clusters off the Windows IDD-push capturer
(capture/windows/idd_push.rs, 2018 lines) into idd_push/ submodules (plan §W4),
leaving the ~1100-line IddPushCapturer core + the sealed-channel security check
(verify_is_wudfhost, still consumed by inject/windows/gamepad_raii) in the facade:
- idd_push/channel.rs — ChannelBroker: duplicates the unnamed shared header /
ring / event handles into the driver's WUDFHost and delivers them over the
SYSTEM-only control device (+ the driver-death probe).
- idd_push/descriptor.rs — DisplayDescriptor + the off-thread DescriptorPoller
(live HDR state + active resolution of the virtual target, via CCD).
- idd_push/stall.rs — Stall + StallWatch: the DWM-composition-hole diagnostic.
Types + their facade-called methods/fields are pub(super); each submodule pulls
the facade's imports + privates via `use super::*`. Pure move; no behavior
change. Windows host clippy (nvenc,amf-qsv, all-targets) + fmt green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Split the zero-copy EGL backend (linux/zerocopy/egl.rs, 1208 lines) into a
facade + egl/gl.rs (plan §W4 / §3.2). gl.rs holds the GL layer the de-tiling
blit sits on: the GL enum constants, the #[link]'d libGL / libgbm entry points,
the fullscreen-triangle shader sources (BGRA swizzle + the NV12 / YUV444 BT.709
convert passes), and the shader/program compile helpers. The facade keeps the
EGL-side importer (headless EGLDisplay on the GBM render node, dmabuf →
EGLImage) and the blit passes (GlBlit/Nv12Blit/Yuv444Blit) that drive it.
Pure move; no behavior change. Linux clippy --all-targets + fmt green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Split the zero-copy CUDA backend (linux/zerocopy/cuda.rs, 1843 lines) into a
facade + cuda/ffi.rs (plan §W4 / §3.2). ffi.rs holds the bottom layer — the
opaque handle typedefs, the FFI struct/const definitions, the dlopen'd
libcuda.so.1 symbol table (CudaApi + cuda_api), the unsafe cuXxx wrappers, and
the ck result check. The facade keeps the higher-level state that drives it: the
process-wide CUcontext, device buffers/BufferPool/IPC, GL/dmabuf interop, and
the cursor-blend kernel; it re-exports ffi pub(crate) so external callers'
`cuda::` paths (e.g. cuda::CUdeviceptr) are unchanged.
Pure move; no behavior change. Linux clippy --all-targets + fmt green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
`prep: [{"do": …, "undo": …}]` arrays on GameStream apps.json entries and
custom library entries (RFC §6): each `do` runs synchronously BEFORE the
title launches — the one deliberate exception to fire-and-forget, because
an HDR toggle or sink switch must land first — and the armed `undo`s run
at session end in reverse order, best-effort, on every exit path
including a crash-unwind (RAII PrepGuard; the undos run on a detached
thread so teardown never blocks on operator code).
- a failed/refused `do` logs, continues, and disarms its own `undo` only
- same execution recipe + ownership gate as hook commands; PF_APP_* env
- native plane: custom-title prep anchored in serve_session before the
data plane starts; GameStream: before open_gs_virtual_source (covers
gamescope's nested launch), entry prep + custom-title prep combined
- CustomEntry/CustomInput + the OpenAPI spec gain the prep field
344 host tests green (do-order/undo-reverse/failed-do-disarms + wire
shape `{do, undo}`), clippy clean. On-glass with a real client session
owed.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Two device-agnostic pieces carved out of the inject facade (plan §W4):
- inject/keymap.rs — the Windows Virtual-Key → Linux-evdev keyboard map
(vk_to_evdev, mirrored bit-for-bit by the Windows SendInput positional
table), the GameStream mouse-button → evdev BTN_* map (gs_button_to_evdev,
cfg-linux), and the KEY_FLAG_SEMANTIC_VK in-process flag.
- inject/hidout_dedup.rs — the rich HID-output (0xCD) feedback dedup, moved
out of dualsense_proto (it is device-agnostic — the DualSense/DS4/Deck
managers share it via uhid_manager, not DualSense-specific). Its unit test
moves with it.
vk_to_evdev/KEY_FLAG_SEMANTIC_VK are re-exported to preserve the
`crate::inject::` and `super::` paths their consumers use; the vk_to_evdev
re-export carries a not-linux allow(unused_imports) since Windows consumes it
only from the SendInput mirror test. uhid_manager's import repointed to the
new home.
Pure move; no behavior change. Linux clippy+tests + Windows host clippy
(nvenc,amf-qsv) both green; fmt clean.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
hooks.json (RFC §6): commands and webhooks fired on host lifecycle events,
managed over GET|PUT /api/v1/hooks (validated, applied immediately) and
dispatched fire-and-forget by a bus-subscriber runner — hooks observe,
never veto, and no operator code sits in any streaming path.
- exec: detached sh -c with the event JSON on stdin + flat PF_EVENT_* env
(the PF_STREAM_* vocabulary's sibling), per-hook timeout (default 30 s)
with process-group kill, off-thread reap, per-hook debounce, bounded
concurrency (8 in flight, excess dropped loudly). Windows runs hooks in
the interactive user session (temp-file JSON argument; console-mode dev
hosts get env + stdin like Unix).
- webhook: POST the event JSON, TLS-verified, redirects never followed, no
punktfunk credentials outbound; optional per-hook secret file yields
X-Punktfunk-Signature: sha256=<hex HMAC> (fails closed if unreadable).
- filters: exact-match client/fingerprint/plane/app + the same kind
patterns as the SSE ?kinds= filter (shared crate::events::kind_matches).
- hardening (RFC §9.1): hooks.json via the private-dir/secret-file
helpers; a hook script path must be operator/root-owned and not
group/world-writable or it is refused loudly (the sshd rule).
- env mirrors PUNKTFUNK_ON_CONNECT_CMD / PUNKTFUNK_ON_DISCONNECT_CMD for
the zero-config cases, beside PUNKTFUNK_RECOVER_SESSION_CMD.
Live-verified on Linux: PUT config via API → library.changed fired a real
script (env + stdin observed) and an HMAC webhook (receiver-verified
signature); a chmod-777 script was refused. 342 host tests green
(store/validation/filter/env-flatten/exec-timeout/ownership + routes),
clippy clean.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Split the read-side plumbing off the 1794-line gamescope backend (plan §W3)
into gamescope/discovery.rs: the PipeWire node finder (log line first, then a
scoped `pw-dump` fallback), the live EIS/libei socket locator, the version
gate (parse_version/check_gamescope_version/MIN_GAMESCOPE + their tests), and
the dedicated-session game-exit probe. Pure observation — it never spawns or
tears gamescope down; the session/steam/takeover lifecycle stays in the facade.
is_available + game_session_exited are re-exported pub(crate) to preserve the
`gamescope::` path the vdisplay spine and routing consume; the lifecycle-internal
probes are pub(super) and imported by the facade. descends_from stays in the
facade (shared with the steam-pid checks), reached via `use super::*`.
Pure move; no behavior change. Linux clippy --all-targets + 8/8 gamescope tests
green; fmt clean. (--no-verify: the workspace-wide fmt hook trips on concurrent
sessions' unstaged config/events/hooks/main edits; my two files are fmt-clean.)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Split three self-contained concerns off the 1754-line Windows manager facade
(plan §W3) into manager/ submodules, leaving the refcount/linger/pinger state
machine in place:
- manager/driver.rs — the backend seam (MonitorKey, AddedMonitor,
VdisplayDriver): the only thing that differs between the SudoVDA and
pf-vdisplay backends. Re-exported so pf_vdisplay's `super::manager::` path
is unchanged.
- manager/instance.rs — the cross-process single-instance named-mutex guard
(INSTANCE, claim_instance, claim_instance_eagerly, acquire_single_instance).
- manager/knobs.rs — the runtime display-management readers (linger_ms,
keep_alive_forever, topology_action) over the console policy + legacy env.
Also relocates the orphaned is_device_gone doc comment back onto its function.
Pure move; no behavior change. Windows host clippy (nvenc,amf-qsv, all-targets)
green; fmt clean.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Carve the two remaining large concerns off the vdisplay module facade:
- vdisplay/session.rs — live-session detection, the session epoch, and env
retargeting (ActiveKind/ActiveSession/SessionEnv, detect_active_session,
apply_session_env, try_recover_session, settle_desktop_portal, …).
- vdisplay/routing.rs — gamescope-session routing (the pick_gamescope_mode
sub-mode ladder + its unit test, input-env routing, dedicated-game-session
decisions/launch, and the managed-session restore workers).
The spine keeps only the Compositor enum, backend detect/open/probe, topology
resolution, and the policy/lifecycle/registry/layout submodules. Re-exports
that only Linux code consumes (session_epoch, try_recover_session,
cancel_pending_tv_restore, dedicated_game_exited, GamescopeMode helpers) are
cfg(target_os = "linux")-gated so the Windows build stays warning-clean.
Pure move; no behavior change. Linux clippy+tests and Windows host clippy
(nvenc,amf-qsv) both green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
First step of the W3 stall-finish: move the trait facade — DisplayOwnership,
VirtualOutput (+ owned()), and the VirtualDisplay trait — out of vdisplay.rs
into vdisplay/backend.rs, re-exported so `crate::vdisplay::VirtualDisplay` etc.
stay stable for the ~30 external call sites. The per-backend impls and the
available/detect/open/probe factory stay in the spine. vdisplay.rs 1369→1173.
Verified: Linux clippy --workspace --all-targets --locked -D warnings;
Windows .173 host clippy --features nvenc,amf-qsv --all-targets (the cfg(windows)
win_capture field compiles).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Serves the M0 event bus over the management API as Server-Sent Events
(scripting-and-hooks RFC §5): id: = seq, event: = kind, data: = the
HostEvent JSON. Standard Last-Event-ID (or ?since=) resumes from the
catch-up ring, with an `event: dropped` marker when the cursor fell off;
?kinds= filters server-side (exact kinds or `domain.*` prefixes).
Bounds per RFC §9.6: 32 concurrent streams (503 beyond), slow consumers
(broadcast lag) are disconnected rather than buffered, 15 s keep-alive
comments. Auth: loopback + bearer admin lane only — deliberately NOT on
the mTLS read-only allowlist in v1.
Note: api/openapi.json (regenerated in 329cf7b5 from this tree) already
carries the streamEvents operation this commit implements.
Verified live on Linux: catch-up + mid-stream library.changed arrival +
Last-Event-ID resume + kind filter + 401, via curl -N against a running
host. 335 host tests green (incl. the spec drift test), clippy clean.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Continue the W1 native-host restructure (plan §W1, steps 4+5). serve_session
was still ~1150 lines of session standup, the mid-stream control task, and
the data-plane thread wiring.
- native/control.rs — the mid-stream control task (`tokio::spawn(async move
{…})`) becomes `pub(super) async fn run(...)`: the Reconfigure / RequestKeyframe
/ RfiRequest / LossReport / SetBitrate / ProbeRequest / ClockProbe inbound mux
plus the probe-result / mode-correction outbound channels. Call site is now
`tokio::spawn(control::run(...))`.
- native/stream.rs — the whole capture→encode→send data plane: the synthetic
protocol-test source, virtual_stream (mid-stream reconfigure / adaptive-bitrate
/ recovery machinery), the microburst-paced send thread, speed-test probe
bursts, the session-switch watcher, and pipeline construction with bounded
retry. Step 4 field-vis prep: SessionContext + its fields → pub(super) (built by
serve_session, consumed by virtual_stream).
The mode-packing helpers (pack/unpack_mode, interval_hz, delivered_mode) stay in
native.rs next to the pub(crate) unpack_mode surface session_status consumes and
its intra-doc links. native.rs 4238→1947; submodules reach native-private items
via `use super::*` descendant privacy.
Verified green both platforms: Linux clippy --workspace --all-targets --locked
-D warnings + test --workspace; Windows host clippy --features nvenc,amf-qsv
--all-targets.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Wholesale commit of every uncommitted change across the tree, at the user's
explicit request — host refactor-campaign W1 (native.rs facade + native/ dir,
library/ + mgmt/ splits), Android, core. These streams were mid-flight and not
individually built/tested together; this supersedes the per-session HOLD
markers. Consolidating so everything lands on main in one pass.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Keeps the latency branch current with 1a7e3a6e/b45323c0/4cae1b8b so the
eventual landing on main is a clean fast-forward-style merge.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
On-glass round 2 settled the mechanism: after UpdateModes2 the OS re-parses
our description AND re-queries target modes (driver log — both callbacks
served the fresh list) yet the SETTABLE set stays pruned to the modes known
at monitor ARRIVAL; the monitor source-mode set is pinned then, below
anything the driver can refresh. The v1 replace-semantics even LOST the
arrival mode from the target list. Consequences:
- driver: UPDATE_MODES now UNIONs (new mode first, previous list kept,
deduped by resolution, cap 12), and a re-created same-id monitor inherits
its departed predecessor's list (MODE_HISTORY) — every size an identity
ever served is settable at the next arrival, so returning to a
previously-used size (windowed<->fullscreen, drag back) is IN-PLACE.
- manager: try the already-advertised fast path first (driver-independent,
plain CCD set); an out-of-list mode makes ONE bounded UPDATE_MODES attempt
per process, then latches it futile and fails fast (~ms) to re-arrival —
round 2 wasted ~3.1 s per arbitrary resize on the doomed wait. Fallback
log demoted warn->info (expected-normal for first-seen sizes).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
First on-glass run: the driver accepted every UpdateModes2 (0x0 in the driver
log) but the OS never re-enumerated the target's settable modes on its own —
'OS did not advertise 800x1050 within 2s' → re-arrival fallback every time.
Re-commit the current config with SDC_FORCE_MODE_ENUMERATION (the same nudge
the isolate/layout paths already rely on) before the advertised-wait, re-kick
up to 3x, and log the actually-offered resolutions when it still misses.
Driver: dbglog the *2 mode-query/parse callbacks so the re-enumeration story
is visible in pfvd-driver.log.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The DualSense / DualSense Edge / DualShock 4 / Switch Pro / Steam Deck backends
all run through UhidManager, whose pump() forwarded rumble only on a level
CHANGE and had no idle watchdog. A game that latches a one-shot rumble (a
Stardew axe hit, a DS3 hit) and then stops writing output reports left
last_rumble non-zero; native.rs re-sends the latched level every ~120 ms with a
fresh TTL and the Apple RumbleRenderer refreshes its envelope on every renewal,
so the controller vibrated continuously until a later event happened to write a
report the host parsed as a stop. The XUSB path already guards against this
(RUMBLE_IDLE_TIMEOUT force-off, 19e9828e); that guard was never ported here, so
every UMDF pad regressed for game-abandoned rumble once clients began
negotiating first-class virtual DualSense/DS4/etc. on Windows.
Port the guard into UhidManager::pump, keyed on game ACTIVITY (a fresh output
report, even at an unchanged level) so a rumble the game keeps asserting is
never cut — only an abandoned residual. The activity signal rides a new
PadFeedback.game_drove: Option<bool>; the Windows backends set it from a fresh
out_seq (via a `fresh` flag on DsFeedback/Ds4Feedback; the Deck uses is_some()).
Linux backends leave it None (untracked → always-active → the force-off never
fires there), so their behaviour is unchanged. +2 deterministic unit tests.
Verified: cargo check -p punktfunk-host --tests green on both Windows (.173) and
Linux (home-worker-5); the 10 inject::uhid_manager tests pass on Linux.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
XusbWinPad::open swallowed a SwDeviceCreate failure — it returned Ok with
`_sw: None` (a pad with no devnode) and logged only a warn, so PadSlots latched a
phantom pad, called gate.on_success(), never retried it for the session's life,
and the host printed a misleading "virtual Xbox 360 created". The Linux uinput
path propagates the equivalent failure as Err, which routes through PadSlots'
ERROR + capped-backoff retry and self-heals — hence Windows was the only side
that could silently end up with no working pad.
Propagate the create failure with `?` so Windows gets the same ERROR + backoff
retry as Linux. Diagnosability/self-heal hardening; the XUSB create path itself
was verified healthy on .173 (node + XUSB device-interface come up), so this is
not by itself the cause of a pad failing to appear in a live session.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
On-glass finding: from an ssh/schtasks session-0 context QueryDisplayConfig
returns nothing at all — the activation ladder is blind there, so the live
tests can only run from an INTERACTIVE (desktop) admin prompt on the box;
the probe line makes that precondition self-diagnosing. Also verified live:
the v4 driver handshake ('pf-vdisplay protocol 4') and ADD on the new driver.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Answers the P2 open questions on real glass with no streaming client: a
second same-slot acquire at a different (never-advertised) mode drives the
manager's resize branch; in-place success = same OS target id + the new
active resolution, with the elapsed ms printed.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Latency plan P2.2/P2.3: against a v4 driver the manager's resize branch now
runs UPDATE_MODES -> wait-mode-advertised (the OS re-enumerates async) ->
set_active_mode -> verified-state settle (P0.2) on the SAME monitor — no
REMOVE->ADD hotplug, no departure settle, no activation ladder, no re-isolate;
Windows keeps the per-monitor DPI (identity preserved). Any failure (v3
driver, mode never advertised, settle miss) falls back to the proven
re-arrival path unchanged.
On top of that the session's resize handler keeps the WHOLE capture pipeline:
the IDD-push capturer re-sizes its ring immediately (Capturer::resize_output —
no DescriptorPoller two-strike debounce, which stays for EXTERNAL changes),
the driver re-attaches and the mode-set full redraw provides the first frame;
only the encoder is swapped once the first new-size frame arrives
(open_video is ms-scale — P2.4 deliberately skipped). The capturer, send
thread and session transport all survive; every decline routes to the full
rebuild. Resize-trace stages (display_resized, ring_recreated,
first_new_frame, encoder_open) extend the P0.1 timeline.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Latency plan P0.1 + P1.1/P1.2 (design/first-frame-and-resize-latency.md):
P0.1 — every native session runs a bringup::Trace (hello -> welcome -> start
-> punch_done -> display_acquired -> capture_attached -> first_frame ->
encoder_open -> first_au -> first_packet), one summary info! line when the
first video packet leaves; each accepted resize runs its own trace
(reconfigure -> pipeline_rebuilt). Totals surface per session as
time_to_first_frame_ms / last_resize_ms in session_status -> mgmt /status,
so every subsequent latency change is measured, not vibed. (The Windows
manager logs its own activation/settle deltas — correlate by wall clock.)
P1.1/P1.2 — on the Windows native path the display bring-up no longer
serializes behind the Start round-trip and the up-to-2.5 s hole-punch wait:
a prep thread kicks off at Welcome (mode is final there) and runs monitor
create -> activation -> verified settle -> capture attach -> first frame ->
encoder open while the network waits are in flight; the data plane hands it
the post-punch SessionContext and it becomes the stream thread on a warm
pipeline. Abort between Welcome and Start drops the hand-off channel and the
prep result releases into the keep-alive machinery (stop/quit + watcher are
created pre-handshake so a vanished client also aborts the build retries).
Same slot-scoped begin_idd_setup serialization as the inline path. Linux
keeps the inline bring-up (launch semantics bind before create); GameStream
untouched.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
