[pdfinn]

Summary

sed_test's ShortScripts failed in Linux AMD64 Build & Test on #1054, which does not touch sed. The log has only --- FAIL: ShortScripts and the address line, with no assertion message. The cause is a race in the test.

1. runsed() spawned sed on a pipe and only then wrote abc\n into it.
2. Some scripts are refused before sed reads anything: an unclosed s/, a lone { or }, : with no label. For those, sed can exit and close the read end before step 1's write.
3. emu/port/devpipe.c's pipewrite then kills the writer, which is the test itself, with write on closed pipe.
4. run() catches that as a failure with no message.

Unclosed has the same exposure.

Changes

• tests/sed_test.b, runsed(): the input is written, and the write end dropped, before sed is spawned. The test still holds the read end at that point, and the pipe buffers the 4 bytes, so sed can't close the pipe under the write. A short write is reported as crash: cannot write input rather than passing silently.

Testing

Linux AMD64, headless build.

• Before the fix: 16/16 passes (-c1 and -c0, 8 each). It is timing-dependent, which is why it rarely shows.
• Before the fix, with a 50 ms sleep before the write: ShortScripts and Unclosed fail on every run, with the same bare FAIL as CI.
• After the fix: passes under -c1 and -c0. It still passes with a 50 ms sleep left in after the spawn.
• /tests/runner.dis -v: not run in full locally; this PR's CI runs it.

Checklist

• Code follows the existing style of the file(s) modified
• No .dis build artifacts committed from appl/ or tests/
• Documentation updated (if applicable): n/a
• No secrets, API keys, or credentials included

---

Generated by Claude Code