Beagle SCM

SUBS-033: status lists a newly-added (staged-gitlink) sub's files as unk, not one mount row

A submodule that is declared, mounted, and gitlink-staged (be put <sub>) but not yet committed has all its internal files listed individually as unk by be status, instead of being filtered behind one staged-gitlink (new <sub>) row. Real case: the journal's be sub (?/beagle-ext) shows 29 be/* files as unk. html/beagle don't, because their gitlinks are already committed. Method: Issues. Relates to SUBS-013 (the be put staging side).

Input

Context

• In beagle-journal: be is in .gitmodules, mounted (be/.be anchor), gitlink staged via be put be (put be#c828169707… in .be/wtlog), NOT yet posted. be status → 29 be/* lines as unk + ? …, 29 unk; recursion status:be separately shows ? 29 ok. So the files are double-counted.
• Root cause: sniff/CLASS.c class_merge_step records a submodule prefix (which drops the sub's internal wt rows via class_under_submodule) ONLY when a COMMITTED baseline gitlink row exists: base != NULL && ok64Lit(base->verb, 0) == RON_s (CLASS.c:188). A staged-only gitlink put (base == NULL, a put <path>#<40hex> row) is not recognized → be/* surface as CLASS_WT_ONLY.
• html/beagle are filtered because their 160000 gitlinks are in the committed baseline tree; be's gitlink is only staged.

Goals

• A mounted sub whose gitlink is staged (put <path>#<40hex>) but uncommitted is treated as a submodule boundary: its internal files are filtered and it surfaces as one staged-gitlink row. Matches git add <submodule>.
• be status in beagle-journal shows new be and zero be/* unk.

Constraints

• Minimal fix in sniff/CLASS.c; reuse class_remember_submodule. Repro-first (CLAUDE.md §17), as a case in a registered table/dir test under beagle/test/.
• Don't regress committed-gitlink subs nor regular (fragment-less) staged file puts. Don't touch test/CMakeLists.txt / test/put/* (live WIP there).

WIP

Design decisions

• At CLASS.c:188 also fire the gitlink branch when put != NULL && u8csLen(put->uri.fragment) == 40 (a staged gitlink bump). Discriminator: a gitlink put row carries a 40-hex fragment; a regular file put has none.
• The sub's own path sorts before its children (be < be/…), so the prefix is recorded before any descendant row arrives — ordering already holds.

TODOs

• Repro: parent + mounted sub, be put <sub>, assert be status lists no <sub>/* unk — test/all/03-staged-sub-unk (fails pre-fix, passes post).
• Fix CLASS.c gitlink detection to include the staged-put (40-hex) case.
• Repro green + full ctest green (398/399; lone fail be-post-relpath is a pre-existing ssh/PATH env issue, unrelated to sniff/).

Blockers and bummers

• Concurrent journal post diverged the beagle lines at ab4519af (trunk vs the ?/beagle/.beagle-journal dot-branch the journal pins); reconciled by weave.

Outcome

• 707bbbbe — SUBS-033 fix landed on beagle trunk ? (CLASS.c staged-gitlink detection + test/all/03-staged-sub-unk). Verified: journal be/ unk → 0.
• 0b60bf6a — weave-merge of 707bbbbe into the journal's beagle line (?/beagle/.beagle-journal, over 121a98ae); clean, content-conflict=0.
• Journal beagle gitlink bump to 0b60bf6a STAGED (put beagle#0b60bf6a); awaiting the human's journal post.
• CLOSED (audit 2026-07-07): fix 707bbbbe on beagle trunk, all TODOs green; the journal-post residual is bookkeeping outside this ticket.