08be8fea77
When a capture sidecar stopped/restarted, the bridge video thread got EPIPE
on the FIFO write, set g_port_stop[port]=1, and the port went dead — requiring
a full bridge restart to recover. Subsequent record attempts on that port would
hang in 'connecting' forever.
Fix: mirror the audio thread pattern — on EPIPE, close the FIFO and loop back
to open() blocking for the next reader. Hardware lock errors (SDK failures)
still stop the port via g_port_stop as before. Only reader-disconnect (EPIPE)
now recovers gracefully.
This was the cause of port 6 (Ghost) failure in the burn test.
🤖 Generated with Claude Code
|
||
|---|---|---|
| .. | ||
| capture | ||
| mam-api | ||
| node-agent | ||
| playout | ||
| premiere-plugin | ||
| premiere-plugin-uxp | ||
| web-ui | ||
| worker | ||