tui: skip audio init when no playback device on the node runtime #61673

Activity

1. opencodejr commented on Nov 4, 2026

   

   opencodejrBot

   on Nov 4, 2026

   Contributor

   More actions

   Thanks for the detailed write-up. I could reproduce the underlying problem.

   Setup: x86_64 Linux container with libasound.so.2 installed and no /dev/snd, @opencode/[email protected], attention.sound: true, and three prompts to a simulated model. Each "Session done" sound makes libasound write a new 35-line batch to the TUI's stderr: cannot find card '0' → Unknown PCM default → the dmix variants. That is 35, then 70, then 105 lines over three turns. It still happens the same way on the latest dev (8568952), where packages/tui/src/audio.ts neither checks for a playback device nor stops calling start() again after a failed start.

   Caveats:

   • I used the shipped Bun binary, not a Node build. Since fix(tui): keep native and child stderr off the rendered screen #59128 isn't merged, I couldn't isolate the Node-only "one batch per session" part on its own. Today the batch repeats on every attention event on either runtime.
   • The harness sends the TUI's stderr to a log file, so the attached recording shows the session but not the ALSA text on screen. The script below counts the lines instead.

   Related: #41763 (original report), #59128 (Bun-side capture and the no-re-probe change), and #51842 / #51854 (the device check you suggested, closed without merging).

   recording.mp4

   Reproduction script (drive.ts)

   // Repro for #61673 / #41763: on Linux with libasound.so.2 but no /dev/snd,
   // every attention sound ("Session done") makes libasound write a batch of
   // "ALSA lib ..." diagnostics to the TUI process's fd 2. In a real terminal fd 2
   // is the tty, so the batch paints over the rendered TUI. Drive redirects the
   // TUI's stderr to a log file, so this script counts the lines there.
   import { Effect, Stream } from "effect"
   import { readdirSync, readFileSync, statSync } from "node:fs"
   import { join } from "node:path"
   import { tmpdir } from "node:os"
   import { Llm, OpenCodeDriver } from "opencode-drive"
   
   const start = Date.now()
   function stderrLogs(dir: string, depth = 0): string[] {
     if (depth > 5) return []
     let out: string[] = []
     let entries: string[] = []
     try { entries = readdirSync(dir) } catch { return [] }
     for (const e of entries) {
       const p = join(dir, e)
       let s
       try { s = statSync(p) } catch { continue }
       if (s.isDirectory()) out = out.concat(stderrLogs(p, depth + 1))
       else if (e.endsWith(".stderr.log") && s.mtimeMs >= start) out.push(p)
     }
     return out
   }
   const report = (label: string) =>
     Effect.sync(() => {
       for (const f of stderrLogs(tmpdir())) {
         const text = readFileSync(f, "utf8")
         const alsa = text.split("\n").filter((l) => l.startsWith("ALSA lib")).length
         console.log(`[${label}] ${f}: ${alsa} "ALSA lib" lines`)
         if (alsa && label === "end") console.log(text.split("\n").filter((l) => l.startsWith("ALSA lib")).slice(0, 8).join("\n"))
       }
     })
   
   export default OpenCodeDriver.use(
     {
       ...(process.env.OC_DEV ? { opencode: { dev: process.env.OC_DEV } } : {}),
       tui: { recording: true },
       tuiConfig: { attention: { sound: true } },
     },
     ({ ui, llm }) =>
       Effect.gen(function* () {
         yield* ui.screenshot("before")
         yield* llm.serve((_request, index) => Stream.make(Llm.text(`Reply number ${index + 1}.`)))
         for (const n of [1, 2, 3]) {
           yield* ui.submit(`hello ${n}`)
           yield* Effect.sleep("6 seconds")
           yield* ui.screenshot(`after-turn-${n}`)
           yield* report(`after turn ${n}`)
         }
         yield* report("end")
       }),
   )

   Run it with opencode-drive: opencode-drive run ./drive.ts. It drives opencode from your PATH; set OC_DEV to the path of an OpenCode checkout to run that from source instead.
2. opencodejr

   added

   reproduced

   on Nov 4, 2026