test(pty-tape): settle the prompt queue before Ctrl-T

[0thernet]

Fixes the CI flake in real PTY: TAPE gestures echo range commands at 80x24 (expected "verse"; PR #185, Check run 38088154653). Also fixes a dawg race that turned up while proving it.

1. Test: waited on signals that arrive before the queue drains

The test typed seven lines (bars 8, two adds, two sections, track drums, track bass), then waited for:

• the session to have 2 notes: true after line 3;
• the header to say · bass: true from launch, because bass is the default track.

The CI frame shows exactly this: rev 3, header bass, no section row. Ctrl-T opened TAPE before section verse ran. #173's settle() was never used here. The fix: t.settle() before Ctrl-T (all input acted on, nothing running), then the old checks become plain assertions, and Ctrl-T uses t.type(). The TAPE tiles test had the same shape and gets the same fix.

I audited the other PTY tests that send several lines then wait. Each waits on the last line's effect, which is sound with the FIFO queue from #173.

2. dawg: TAPE keys typed ahead reduced from stale state

A TAPE key reduces from the landed score and playhead, but its commands go into the prompt queue. With commands still queued (an agent turn, a slow write, a loaded box), [ [ [ computed every shrink from the same loop. One press was lost, and the loop ended at 2 bars instead of 1. Typing ahead in a real session does the same. Now, while the queue is busy, a command-typing TAPE key waits its turn as a deferred step (PromptQueue items may be () => string[]). It reduces from what the lines before it left, and its commands run next. When the queue is idle, keys reduce immediately as before.

Proof

• Deterministic adversary. A local-only 300 ms delay before each queued line, which widens the window the CI runner hit. origin/main: both TAPE-gesture tests fail with the CI error (Expected to contain: "verse"). Test fix only: those pass, but TAPE tiles fails (one-bar loop, the [ race). Final head: 6/6 pty-tape tests pass.
• Stress (8 PTY files, 37 tests, 20 iterations, 16 CPU hogs, host-run heavy): origin/main 0/20; final head 0/20. The 37-test loop does not reach this race on this machine. The deterministic adversary above is the reproduction. Local bun run check on the final head: 4027 pass, 0 fail.

No assertions were weakened and no retries were added.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated
dawg	Ignored	Preview	Oct 10, 2026 11:01pm UTC