Commit Graph

11 Commits

Author SHA1 Message Date
679b243974 Audio render-hook M3: wire DLL + host hooked mode + loopback fallback
End-to-end plumbing of the render-hook audio path.

Hook (coop_hook.dll):
- CMake: build audio_hook.cpp, link ole32/mmdevapi, NTDDI_WIN10_CO.
- dllmain: CoInitializeEx(MTA) on the worker thread; install the audio hooks
  even before the ring exists (so streams are counted); open the host's
  coop_audio_<pid> ring when it appears and attach it (enabling capture+silence);
  remove_audio_hooks on clean detach.

Host (coop_host.exe):
- AudioMirror now creates the shared audio ring (owns capture_enabled) and tries
  the Hooked path first: waits ~1s for the hook to publish a format, then
  re-renders the game's frames from the ring with AUTOCONVERTPCM (no echo, since
  the hook silences the game locally).
- Automatic fallback: if the ring can't be created, no format arrives in time,
  or the render client won't initialize, it disables capture (so the game stays
  audible) and reverts to the existing process-loopback path (echo, no regress).
- Exposes Source (Hooked/Loopback/None) for the upcoming panel indicator.

Loopback render loop kept intact as run_loopback. Manual end-to-end (M5) and the
panel UI (M4) are next.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-19 12:02:29 +02:00
4e3814a072 Audio render-hook M2: render hook + in-process self-test (concept proven)
Implements the WASAPI render-hook (hook/src/audio_hook.{hpp,cpp}) and an
in-process self-test that proves COM vtable discovery and GetBuffer/ReleaseBuffer
interception with no game and no second Steam account.

- audio_hook.cpp: anchors on IMMDevice::Activate (idx 3) off our own default
  endpoint (shared vtable), then hooks IAudioClient::Initialize (3) /
  GetService (14) and IAudioRenderClient::GetBuffer (3) / ReleaseBuffer (4) off
  live game pointers. Copies primary-stream frames into the audio ring and
  releases with AUDCLNT_BUFFERFLAGS_SILENT (+ memset belt-and-suspenders), only
  while the host-owned capture_enabled flag is set. Stream counting runs always;
  on a ring overrun it keeps playing locally rather than going silent.
- ipc_client.hpp: publish_audio_stream / note_audio_frames /
  set_audio_streams_seen write the render-stream debug fields into HookStatus.
- tests/audio_hook_test.cpp: installs the hooks, renders a tone through WASAPI
  in-process, and asserts exactly one stream, frames pushed to the ring, the
  ring carries the non-silent tone, and the primary was silenced. PASS:
  streams_seen=1, frames_captured=32640.
- plan doc: correct GetService vtable index 13 -> 14 (SetEventHandle is 13).

coop_hook DLL wiring + host consumer/fallback come next (M3).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-19 11:57:28 +02:00
8e93312925 Audio render-hook M1: shared audio ring + protocol diag fields
First milestone of the injection render-hook audio path (see
docs/audio-render-hook-plan.md) that fixes the local audio echo without a
virtual device.

- common/include/coop/audio_ring.hpp: new lock-free SPSC shared-memory ring
  for PCM, separate from the input/status SharedBlock. Free-running 64-bit
  positions (release/acquire), format handshake, host-owned capture_enabled
  gate, drop-whole-packet overrun policy.
- common/include/coop/protocol.hpp: add AudioStreamInfo + audio_streams_seen /
  audio_streams[] to the always-present HookStatus for the render-stream-count
  debug view; bump kProtocolVersion 3->4 (new members appended).
- tests/audio_ring_test.cpp: in-process unit test (push/pop integrity,
  wrap-around, format handshake, overrun/drop). No hook or audio device.
- docs/audio-render-hook-plan.md: the green-lit design this implements.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-19 11:49:18 +02:00
8059f0e488 docs: update README for Phases 0-2 done, add roadmap + lessons
- Reflect audio mirror (Phase 2) and video mirror (Phase 1b) as implemented;
  drop stale "(current)"/"still ahead" framing and the broken docs/ link.
- Add a Roadmap section for the planned work: toggle/generalize the debug UI,
  fix the local audio echo, Present-hook video path, Steam Input, x86 support.
- Add Lessons learned (RPT focus requirement, no exclusive fullscreen, the
  ActivateAudioInterfaceAsync agile-handler gotcha, WGC occlusion, audio echo).
- Document clangd/compile_commands setup and the CTest suite.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-19 10:27:57 +02:00
663d86e6ec Phase 2: audio mirror via WASAPI process loopback
Mirror the real game's audio so Steam Remote Play Together (which streams the
host's own audio session) carries it to the guest. The host captures the game
by PID via WASAPI process loopback and re-renders it on the default endpoint;
the game still plays locally too (accepted "double audio" for now).

- ProcessLoopbackCapture: process-loopback capture client, frame-sink + stats.
  The completion handler must be agile (IAgileObject) or
  ActivateAudioInterfaceAsync rejects every call with E_ILLEGAL_METHOD_CALL.
- AudioMirror: wraps capture with an event-driven render client and a primed
  ring buffer; AudioPanel drives it from the injected game's window/PID.
- coop_tone: standalone WASAPI sine-wave process used as a known audio source.
- audio_loopback_test (CTest): captures coop_tone by PID and asserts non-silent
  audio arrives, so the path is verifiable without a second Steam account.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-19 10:21:27 +02:00
ba545ba64a Tooling: generate compile_commands.json for clangd
The Visual Studio CMake generator does not emit compile_commands.json, so
clangd had no include paths or flags and reported hundreds of false errors.
Add a parallel Ninja config (build-clangd/) that produces the database:

- CMAKE_EXPORT_COMPILE_COMMANDS ON (honored by Ninja, ignored by VS)
- gen-compile-commands.bat configures build-clangd/ under vcvars64
- .clangd points clangd's CompilationDatabase at build-clangd/
- ignore /build-clangd

clangd's clang-cl driver auto-resolves the MSVC/Windows SDK system includes,
so the cl.exe-based database is sufficient.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-19 10:21:10 +02:00
e38df6cf88 Phase 1b: video mirror via Windows Graphics Capture
Mirror the injected game's window into the host's borderless window so Steam
RPT streams a live copy of the game. No injection needed for video.

- capture/window_capture: WGC capture of a target HWND. Wraps our D3D11 device
  as an IDirect3DDevice, creates a free-threaded frame pool + capture session,
  hides the cursor and (best-effort) the capture border. Frames arrive on a WGC
  thread and are handed to the render thread, which copies the newest into a
  shader-resource texture and draws it -- keeping all D3D11 context use on one
  thread. Handles window resize via frame-pool Recreate.
- capture/frame_renderer: fullscreen-triangle shader that blits the captured
  texture letterboxed (aspect-preserved) into the window.
- capture_panel: "Mirror game window" toggle + lifecycle; target HWND comes
  from the hook's reported game window.
- main: winrt apartment init; mirrored frame drawn as background, ImGui on top.
- CMake: link windowsapp + d3dcompiler.

Verified: builds clean; host starts (apartment init + shader compile succeed).
Visual capture quality/latency to be judged on a real game.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-19 01:03:59 +02:00
a0a12d69fe Phase 1a: per-slot poll viz + input-path diagnostics
Surfaced by a game (Life is Strange: Before the Storm) that ignores
controller input when it lacks true OS focus even though it still polls
XInput. To find the focus-gated detection path, instrument the hook.

Protocol v3 status back-channel now reports:
- per-slot XInputGetState and XInputGetCapabilities counters (replacing the
  single aggregate), so the overlay shows exactly which slots the game polls
  and how fast;
- focus-API call counts (GetForegroundWindow/GetActiveWindow/GetFocus) to see
  whether the game consults the APIs we spoof;
- input-path diagnostics: whether the process registered Raw Input for a
  gamepad usage and whether it set RIDEV_INPUTSINK (background delivery), and
  whether a DirectInput dll is loaded.

Host overlay gains a per-slot poll table and an "Input path" section. The DLL
refreshes input diagnostics each worker tick via GetRegisteredRawInputDevices.
hook_selftest updated for per-slot counters; passes.

This is diagnostic-only: once a real run shows which path LiS uses, the
targeted focus fix (e.g. forcing RIDEV_INPUTSINK or DI background coop) follows.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-19 00:38:07 +02:00
df4325d21b Phase 1a: focus spoofing + hook observability
Two problems surfaced in testing: (1) no way to tell whether the injected
hook was actually the input source, and (2) the final design needs the tool
window focused for Steam RPT capture, which would pause/silence games that
react to focus loss. Both are addressed here.

- Focus spoofing (hook/focus_spoof): find the game's main window, subclass it
  to rewrite/swallow WM_ACTIVATE/ACTIVATEAPP/NCACTIVATE/KILLFOCUS, and inline-
  hook GetForegroundWindow/GetActiveWindow/GetFocus to always report the game
  as active. The game keeps running and polling while unfocused.
- Status back-channel (protocol v2): the DLL reports attached/focus-spoof
  flags, game pid/hwnd, a heartbeat, and a cumulative XInputGetState counter.
  The host overlay turns the counter into a live poll rate, so "is the hook
  working" is directly observable.
- Synthetic test-input toggle in the host: forwards a known automated pattern
  (stick circle + periodic A) to prove forwarding independent of the physical
  pad.
- hook_selftest extended to assert the status channel; passes.

Documented the windowed/borderless requirement and the new observable test
flow in the README.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-19 00:13:06 +02:00
e370c8dcc5 Phase 1a: input forwarding via DLL injection + XInput hook
The host can now inject coop_hook.dll into a running game and forward
controller state to it over shared memory, so the game reads the host's
(eventually the guest's) input and nothing else.

- hook/: coop_hook.dll. DllMain spawns a worker that opens the shared-memory
  channel (named by the game's pid) and installs SafetyHook inline hooks on
  XInputGetState/GetStateEx/GetCapabilities/SetState. Detours synthesize state
  from shared memory; unmanaged slots report disconnected, hiding physical pads.
- host/: process picker (Toolhelp32), CreateRemoteThread(LoadLibraryW) injector
  with an IsWow64Process2 bitness guard, IPC server publishing pads each frame,
  and an ImGui Injection panel wiring it together.
- tests/: hook_selftest exercises the IPC seqlock + hook detours in-process
  (no game/controller needed); passes.

Build: SafetyHook wired in (COOP_BUILD_HOOK=ON), Zydis via FetchContent.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-18 23:21:20 +02:00
cf058aecfa Phase 0: donor-launch spike foundation
Scaffold CoopAllTheThings: Remote Play Together for any XInput game via a
mirror app under a donor appid (real game keeps its own appid, so DRM,
achievements, and playtime stay intact).

- Build: CMake skeleton, ImGui + SafetyHook submodules (no vcpkg)
- common/: host<->hook IPC contract (seqlock pad state, shared-memory RAII)
- host/: borderless D3D11 window + ImGui overlay listing visible XInput pads,
  behind an InputSource interface (Steam Input slots in later)
- README documents the Phase 0 donor-launch validation procedure, anti-cheat
  limitation, and XInput/bitness constraints

Phase 0 validates the riskiest assumption (Steam RPT streams an arbitrary
window under a donor appid and routes guest input to it) before capture and
injection are built.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-18 22:48:31 +02:00