Fix: a failed launch was reported as success, and three defects non-ASCII found

[0xLeif]

Four defects from two sources — a real 22-minute drive and a translation round where every rune session was conducted in the language being translated rather than in English.

Stacked on #66, which is stacked on #65. Base is leif/screen-cell-model.

A launch that never happened reported success

before   start -- <not on PATH>   {"status":"ok","state":"exited","exit_code":127}
after    start -- <not on PATH>   {"status":"error","error":"…exited 127…not on PATH…"}
control  start -- true            {"status":"ok"}   ← unchanged

A caller checking status — the field whose entire job is to say whether the call worked — saw success. It was documented as "check state instead", which is the wrong shape of answer: an envelope shouldn't need a footnote to be read correctly.

Only 127 fails, deliberately. start -- true exits 0 immediately and is a successful launch of a program with nothing to do; treating any prompt exit as failure would break every short-lived child. And the session record is kept — start failing loudly is the fix, but deleting the transcript that shows why would trade one quiet failure for another.

An error that was confidently wrong

A session started in one directory and read from another got No such session, and that error told you to run rune session list — which is scoped to your project and returns nothing, actively confirming the wrong conclusion. rune knew where it was the whole time.

No session "s3" in this project (otherdir-676fbd13), but it exists in "rune-e66ba9aa". A project is the working directory, so cd there, or run rune session list --all-projects.

This has now caught three readers, two of whom had read the guide's warning about directory scoping first. That's the point at which a documented gotcha stops being a documentation problem.

Two defects that only non-ASCII input could find

The previous translation round sent English prompts, so it only ever exercised the output path. Conducting the session in-language exercised the input path, and found:

• CharacterWidth::ZERO omitted every Indic script — Latin, Greek, Cyrillic, Hebrew, Arabic and Thai were covered; Devanagari, Bengali, Tamil and the rest were not. हिन्दी was charged six columns for six codepoints. This is my own table, written hours earlier in Add: a cell model, so a wide glyph occupies the two columns it is drawn in #66.
• resync fed a character index to byteslice. On 日本語テキスト\e[1mAFTER the ESC is at character 7 and byte 21; it cut at 7, returning "\xAA\x9Eテキスト\e[1mAFTER" — both splitting a character and failing to drop the remainder it exists to drop.

I narrowed the first one against its report. The finding said "one column per codepoint", which overstates it: U+093F is a spacing mark and legitimately takes a column in wcwidth and xterm, so zeroing every Indic mark would be as wrong in the other direction. The fix is the Mn/Me subset — हिन्दी is now five columns, matching xterm, not the three a shaping engine draws. This follows wcwidth and doesn't try to settle shaping.

One claim that did not reproduce, and is not fixed

The same report said a send over ~1024 bytes jams every later send while rune answers settled: true, state: running. Measured:

  600 bytes   no jam
 1100 bytes   no jam          ← not a ~1024 threshold
 4096 bytes   no jam
20000 bytes   follow-up refused: status=error, "previous input is still being
              delivered to the child"; recovers by itself in ~10s

That's what ROADMAP.md already records. The reported shape looks like a probe reading clean_output without checking status — the same mistake this session made once already.

Also lands

The six in-language guide translations (docs/i18n/getting_started.*) and the round's report.

Test Plan

• 604 examples, 0 failures; rubocop clean; specsync check 32/32 files, 7414/7414 LOC
• Controls: launch_failure reverted fails 1 of 3; resync byte index 2; Indic ranges 2; project lookup 2
• One intermittent failure was seen once and did not recur in three further full runs; not identified, recorded rather than dismissed

🤖 Generated with Claude Code

https://claude.ai/code/session_018rf59AtQmJcodUJ6vXDZNY