Repository navigation
Description
Summary
When the WebView2 browser process crashes, Dot X.exe keeps running with no window. It does not detect the crash, does not recreate the webview, and does not exit. The plugin API also stops answering, so plugins see timeouts (channelTargets.getForChannel) or The resource id … is invalid until the user kills Dot X in Task Manager.
Environment
- Dot X 2.0.0-beta.3 (beta track), Windows 11 Pro 10.0.26200
- WebView2 runtime 157.0.4341.57
- Plugins: media-control, mqtt-plugin, dotx-streamdeck-companion, channel-targets-test
What happened (2026-11-18)
| Time (local) | Event |
|---|---|
| 11:12:24 | Windows: "Your system is low on virtual memory" (System log, event 26) |
| 11:12:25–27 | WebView2 browser process (ptype=browser, --webview-exe-name=Dot X.exe) aborts with 0x80000003 in msedge.dll+0x77d8b15. Crashpad dump written to %LOCALAPPDATA%\com.dot-x.dev\EBWebView\Crashpad\reports\ |
| after that | No msedgewebview2.exe processes left for com.dot-x.dev. Dot X.exe (and its 3 deno plugin hosts) keep running, with no window |
| 11:12 → 13:50 | Stream Deck plugin logs about 6,100 Dot X API request timed out: channelTargets.getForChannel. Earlier sessions also logged Failed to get channel targets: The resource id 4138292103 is invalid. |
| 13:50 | Killed and relaunched Dot X manually; everything works again |
The same thing happened on 2026-11-08: an identical ptype=browser, 0x80000003 crash on WebView2 156.0.4314.58 / Dot X 2.0.0-beta.20261030134907.1, about a minute after a low-memory warning. Windows named an external process (StreamDeck.exe at about 62 GB of commit) as the cause that time. So the trigger is outside Dot X. What Dot X does after the crash is the problem.
Expected
Dot X recovers when the webview dies, or at least fails visibly:
- Detect the crash: WebView2
ProcessFailed(COREWEBVIEW2_PROCESS_FAILED_KIND_BROWSER_PROCESS_EXITED, and render-process failures too). In Tauri/wry this means handling the process-failed event, or watching for the webview going away. - Recover: recreate the webview/window. If that isn't possible, log it and exit cleanly (or restart), so autostart/tray users aren't left with a process that looks alive but does nothing.
- Don't route the plugin API through the webview, or fail fast when it's gone. Today plugins only get 3 s timeouts, and the
resource id … is invaliderrors suggest that handlers depend on webview-owned Tauri resources. - Log something in Dot X's own logs when this happens. Right now the only trace is the Crashpad dump.
Repro idea
Start Dot X, then kill the msedgewebview2.exe browser process whose command line contains --webview-exe-name=Dot X.exe (the one without --type=). Dot X.exe stays alive with no window, and plugin API calls start timing out.
The crash dump is available on request.
Activity