- Fix A (USER32 crash): WM_CLOSE no longer destroys the native VST
window from the main thread (plugin editor children live on the
channel worker thread). WM_CLOSE -> hide + closeGUI on worker;
WM_DESTROY erases registry. LOAD_INSTRUMENT close uses closeGUI +
SW_HIDE, waits on hasAttachedView() instead of registry erase.
- Fix B: per-process SHM name SonicForge_DAW_IPC_{pid} so stale
bridges from old builds can never attach to a new app's ring buffer.
- Fix C (stall): opening editor for a plugin whose other channel still
has a live instance of the same DLL (Nexus) deadlocked the audio
loop (reload+createView on worker while audio thread processed the
other instance). Silence same-path channels via setReloading during
reload/attach, restore afterwards; audio resumes.
- Rebuild bridge + update install/ portable package.
- LOAD job posts WM_CLOSE to window created on main thread (DestroyWindow from worker crashed 0xc000041d, muting all tracks)
- handleOpenGui + attach jobs skip when Vst3Instrument::hasAttachedView() (state_ && guiAttached_)
- attach failure posts WM_CLOSE; post_close_gui erases entry only when it points at the same hwnd
- cap openVstGuiRetry at 5 to stop frontend retry spam
Root cause: Steinberg EventList defaults to maxSize=50 events. The STOP
all-notes-off sweep queues 128 note-offs per channel (16ch x 128 pitch),
so addEvent beyond 50 fails silently and the plugin never receives the
remaining note-offs — voices keep sustaining and the transport loops
forever until the bridge process is killed.
Fix: raise eventList.setMaxSize(4096) in loadPlugin. Also assign unique
noteIds per overlapping note-on (stack per pitch) so note-off releases
every voice at that pitch, and log addEvent result + channel in debug
output for verification.
- native bridge: dedicated watchdog thread (parent death / PID-reuse via GetProcessTimes) -> TerminateProcess self; transportStopped flag drops note-on after STOP + CC64 sustain release + allNotesOff on stop
- tauri: set_bridge_child overwrites managed BridgeProcess state (app.manage is no-op when state exists) so Destroyed handler kills restarted bridge
- ui: stopAllPlayback syncs isPlayingRef/subTabsRef + second transport stop at 150ms; onStatusChange resets __bridgeLoadedChannels at callback start (auto-restart never emits bridge-down); openVstGuiRetry waits for async C++ load before OPEN_GUI; loadVstToBridge uses synth_engine.plugin_path / listPlugins fallback and persists plugin_path on load; shrink track-header Synth/GUI buttons
- preloadTrackInstruments: also load VST3 tracks into bridge (no GUI) on project open
- onStatusChange(connected): reload all VST tracks after bridge restart/down recovery
- GUI button (track strip + piano roll): ensure-load then open — no white window when bridge restarted and instrument gone
- __bridgeLoadedChannels map: skip reload if already loaded (avoids plugin state reset); reset on bridge down
- piano roll: wrap Synth+GUI buttons in shrink-0 row so GUI stays on same line; compact track-strip GUI button
- Close other channels' editor windows before attaching a new one (Option B)
- needsReload()/setReloading() guard: skip processAudioBlock during
terminate+reload teardown to avoid UAF/hang when reopening GUI
- Wait for editor registry cleanup (5s timeout) before creating window
admin
3dtours
locpham