Beagle SCM

JS-047: io.setMtime over a new abc/FILE utimensat leaf

The one native primitive the write verbs need: restamp a worktree file's mtime so a later be reads it clean (native SNIFFAtStampPath, sniff/AT.c:761, calls utimensat raw, no abc wrapper). Add a FILE wrapper in libabc, bind io.setMtime(path, ron60). Pure libabc — no keeper/dog. Blocks the write verbs. Part of JS-046. Method: Issues.

Input

Context

• JABC has io.stat/lstat/chmod/readlink/symlink (js/io.cpp, JS-042) but NO mtime setter; abc/FILE.h has no utimens wrapper either (grep confirms none).
• Without it every JS put/post leaves files looking dirty → the next be re-stages the whole tree (the failure put_visit_tracked guards, PUT.c:64).

Goals

• abc/FILE.h/.c: a FILESetMtime(path8b, ron60) wrapper over utimensat(AT_FDCWD, …, AT_SYMLINK_NOFOLLOW), error-propagating (ok64).
• js/io.cpp: io.setMtime(path, ron60BigInt) leaf over it; update API.md + INDEX.md (the io.cpp module section).

Constraints

• libabc only. ron60→timespec like the helper at sniff/AT.c:688. ASAN + ctest green. NOFOLLOW so a symlink stamps the link, not its target (matches native).

WIP

Design decisions

• ron60 spans 2000–2099 in ms (per the codec); split into sec + nsec for the timespec. Reuse the AT.c conversion math rather than reinventing it.

TODOs

• Repro-first ctest (extend JABCstat or add JABCmtime): setMtime(path, r) then io.lstat(path).mtime round-trips r exactly. (test/stat.js:71-78.)
• Implement FILESetMtime + io.setMtime; update API.md + INDEX.md. (ea9de45b leaf + 4fd1d639 binding/docs.)

Blockers and bummers

• Blocks JS-049/JS-050/JS-051 (every stamping path). File-disjoint from JS-048 → runs in parallel.

Outcome

• LANDED ea9de45b (abc FILESetMtime utimensat leaf, now dog/abc/FILE.{h,c}) + 4fd1d639 (io.setMtime binding + docs; the jab-fork lineage carries it as cf47a81). Round-trip test lives in the jab runtime repo test/stat.js (JABCstat). Unblocked JS-049/JS-050/JS-051 — all landed 2026-06-22.