Beagle SCM

LOG-001: be log / jab log follow non-spine (merge) parents too, rendering non-spine commits in grey

be log (and the JS jab log) walk only the FIRST-PARENT spine, so commits merged in as a second (or later) parent never appear — even though they are fully reachable. Live: a merge 213c9003 whose 2nd parent is d6f7e138 (DIS-053) → be log | grep d6f7e138 is EMPTY; the commit is reachable (a parent of cur) but off the first-parent line be log prints. Goal: log also follows non-spine parents so merged-in commits show, rendered in GREY to distinguish them from the first-parent spine. Both the C (graf/LOG.c) and JS (be/views/log/log.js) versions, at parity. Method: Issues.

Input

Context

• C be log: graf/LOG.c GRAFLog walks the first-parent chain only (note the "Hard cap on the first-parent walk" ~:51); graflog_render_commit / graflog_pack tag each row with a tok32 tag (L/W/S) the theme colours.
• JS jab log: be/views/log/log.js mirrors GRAFLog (one row per commit, newest-first) — the same first-parent-only walk.
• Live repro: merge 213c9003 (parent1 04aca8e9, parent2 d6f7e138). be log shows the parent1 spine; d6f7e138 (parent2, DIS-053) is reachable but absent. be map: (graf/MAP.c GRAFMap) shows the full branch tree, but log itself never surfaces merged-in commits.

Goals

• be log and jab log follow non-spine parents (a merge's 2nd+ parents and their chains), so merged-in commits appear in the listing — reachable history is visible, not only the first-parent line.
• Non-spine commits render in GREY (a distinct tok32 tag → SGR) so the first-parent spine stays visually primary and merged-in commits read as secondary.
• C and JS output at parity (same rows, same grey marking, same order).

Constraints

• Respect the existing walk bound (the #N count / first-parent cap) so a merge-heavy/deep DAG doesn't explode; apply a sensible cap across the non-spine walk too. Topo- or time-order the combined listing.
• Grey via the tok32 tag → SGR theme (dog/THEME for C, be/view/theme.js for JS), NOT a hand-rolled SGR; the plain (non-colour) sink stays uncoloured. Keep the tag in parity across C and JS.
• Reuse the DAG navigation (DAGParents / DAGAncestors / DAGTopoSort in graf/); no new graph index. ABC style; ragel/typed only. Repro-first (CLAUDE §17).

WIP

Design decisions

• Listing shape (JS, chosen): INTERLEAVE non-spine commits into the spine listing in time order (newest-first by commit ts, the key fileHistory/aheadBehind use), no indent/marker — the grey tag alone marks them secondary. Combined #N cap.
• Grey tag (JS, chosen): reuse the existing Q tok32 slot ("unk/dir" = grey, FG16(90)) — already grey in BOTH dog/THEME (C binding paints it) AND be/view/theme.js, so NO new tag/theme edit was needed; whole non-spine row is Q-tagged.
• GREY TAG (C, chosen) = Q, ALIGNED to the landed JS choice (same tok32 byte). Q reused, no new slot/SGR change; dog/THEME doc-comment only. Non-spine row: EVERY span tagged Q; spine keeps per-column tags.
• Listing shape (C): ONE flat list of the tip's full closure, TOPO-ordered (DAGTopoSort) emitted reversed = newest-first (as file-history). Topo (not time) keeps tied-author-ts merges deterministic (be-log-projector #N cap relies on it). Spine = first-parent chain from the commit BODY. Bound: #N AND LOG_MAX_WALK over the closure; index-miss → linear fallback. Differs from JS time-interleave.

TODOs

• Repro-first (JS): hermetic merge tip M=[mainline C2, side C1]; jab log (via log.js branchHistory) OMITS C1 pre-fix (RED proven), shows it GREY after.
• C: graf/LOG.c GRAFLog/graflog_branch walks the full closure (DAGAncestors + DAGTopoSort, bounded); non-spine rows greyed via graflog_pack tag Q.
• JS: be/views/log/log.js follows non-spine parents (bounded), greys them via tag Q; NO be/view/theme.js edit needed (Q already grey there).
• Bound holds (LOG_MAX_WALK + #N over the closure); plain sink unchanged (both sides verified: 2nd-parent present + uncoloured in plain).
• Strict be-vs-jab byte-parity test: DEFERRED by ruling (see Blockers) — C topo-orders, JS time-interleaves; rows + grey tag Q match, order may differ on tied-ts merges. Revive as its own ticket if byte parity ever matters.

Blockers and bummers

• Order divergence: C topo-orders the closure, JS time-interleaves — same rows + same grey tag Q, but row order can differ on a tied-author-ts merge. Acceptable for now (colour, not order, marks merged-in); a strict byte-parity test is deferred.

Outcome

• LANDED (both sides, 26Jun): JS be bc040627 "LOG-001 jab log follows non-spine parents, greyed"; C dog 21ffe0c4 / beagle pin 7f9163dd "LOG-001 be log follows non-spine parents, greyed". Audit 2026-07-07: code verified in-tree (graf/LOG.c DAGAncestors+DAGTopoSort walk, LOG_TAG_GREY 'Q', graflog_branch_linear fallback, graf/test/log-merge.sh; log.js nonspine rows). Byte-parity test deferred (see TODOs). CLOSED.
• (pending-commit) JS side done. Fix: be/views/log/log.js — branchHistory (:170 rewrite, +non-spine BFS) returns [{sha,nonspine}]; appendRow (:340) greys a non-spine row with tag Q; handler (:411) normalises rows. Test: be/test/log/nonspine.js (+be/test/CMakeLists.txt reg) — RED pre-fix, GREEN post-fix; full be/ suite green (only the pre-existing bro-universal xfail-inversion remains). Grey tag used: Q. STOPPED before commit (human gates).
• (pending-commit) C side done. ROOT CAUSE: graf/LOG.c graflog_branch walked the first-parent spine only, so a merge's 2nd+ parent was reachable but never listed. FIX (graf/LOG.c): graflog_branch (:395) takes the FULL closure (DAGAncestors), topo-sorts (DAGTopoSort), emits newest-first; spine marked via body first-parent (graflog_first_parent :339); non-spine greyed with tag Q (LOG_TAG_GREY ~:221); index-miss → graflog_branch_linear. dog/THEME.c/.h doc-comment only. REPRO graf/test/log-merge.sh (CMake GRAFlogMerge): hermetic 2-parent merge — 2nd-parent surfaces + greys, spine primary, plain uncoloured. RED proven (linear path → "2nd-parent absent"); GREEN after. ctest 378/378. Grey tag Q. STOPPED before commit (human gates).