DIS-061 DONT: be/ JS post advances the tracked branch on EVERY commit — superseded by DIS-071 (the track-URI model); live defects refiled as DIS-076 (the split + ref mis-keying), DIS-072 (pins), DIS-074 (bare post)
Now:
DONT
trees (AUTHORITATIVE) separates the two motions, made UNIFORM by ruling 2026-07-13: a commit moves only the worktree (it goes "ahead"), fast-forwarding the tracked branch is a SEPARATE explicit post ?branch — no commit-time FF for any branch shape (supersedes the 2026-07-12 tip==base sub-ruling AND overrules the post's own example-6 line "A regular branch would advance on commit", trees.mkd:211). The be/ JS does the opposite, by prior ruling: postOne FF-advances cur's attached branch atomically with every commit (post.js:719-721, gated only on !att.detached), so a sub commit moves its pin-branch with no parent commit anywhere (submount.js:319-320, SUBS-040 I2; wiki Worktree:12 "post advances the tip"). The post overturns that ruling; this ticket carries the flip plus the adjacent bare-post/own-branch/normalization defects.
Input
Context
Commit = branch advance today: post.js:719-721 advanceRef(reader, reader.shard, branchKey, …) on every commit; branchKey defaults to cur's attached branch (post.js:601-602, 624).
Bare post has NO fast-forward arm: clean wt → POSTNONE: no changes since base (post.js:666-673); dirty-unnamed → POSTNOMSG (post.js:675-677). The advance arm needs an explicit Query slot (post.js:618-622).
post ?<cur's own branch> REFUSES (POSTNONE: cur's own branch — nothing to advance, post.js:263-264) — the exact inverse of the post's bare-post meaning.
Normalization bug: resolveTarget keeps the raw absolute query /jab/.beagle (post.js:239-249) while wtlog's curBranch is the stripped key .beagle (wtlog.js:39-57); the own-branch guard misses, resolveRef misses (store.js:420-442), advanceRef store.set()s a bogus ref literally keyed /jab/.beagle (post.js:281). post's query path never routes through branchlib.parse/key.
The overturned ruling lives in: SUBS-040:15,26,30 (I2 "the sub's recursive postOne advances the CURRENT branch ref … which IS the dot-branch") and Worktree:12 ("Branch is GET-owned … post advances the tip").
What already matches trees: post-order recursion, gitlink bump, recursive top commit (postSubs post.js:486-545); post "msg" commit path (post.js:97-158, 675-710); status ahead/behind divergence (status.js:463-475, 323-345).
Goals
Split commit from branch-advance: postOne moves only the wt cur row; the wt reads "ahead" in status until the branch catches up.
Pin-branch ref advances at the PARENT's gitlink-bump commit (postSubs), never at the child's own commit — the reflection model per the ruling below.
Route every post Query through branchlib.parse/key so ?/jab/.beagle resolves to the stored .beagle key; repro for the bogus-ref mint first.
Amend the overturned spec spots: Worktree:12, Submodules:28 ("child worktree commits update the local synthetic branch"), POST bare-row (POST.mkd:27 "reuse absorbed msg" narrows to under-an-active-PATCH only), SUBS-040 ruling note.
Constraints
RULED (gritzko, 2026-07-13, SUPERSEDES the 2026-07-12 chain-anchor reading): a pin-branch tip IS the gitlink in the IMMEDIATE parent's BASE commit, refreshed by the parent on ANY move of its base (post, get) — level-by-level reflection, exactly trees.mkd:87-93 ("commit the immediate parent … the problem may move one level up"). Pin refs are materialized rows the parent maintains for its immediate children; pin-branches are synthetic and EXEMPT from the uniform explicit-post ?branch rule below. The only way OUT of a pin-chain is retracking (get ?other, e.g. get ? = trunk) or detaching — no put/post needed, the branch already exists.
RULED (gritzko, 2026-07-13, recursion + scope): every op is SCOPED to the context repo (at ///jab/dog it affects dog + its sub abc, NEVER the parent jab); clean recursion iff the sub is BASED ON the pin. get: a sub based BEHIND the pin that FFs cleanly is FF'd; ahead/diverged → skip, stays "adv" (post/patch territory). Bare post: commit ALL changes to make the tree clean — sub dirty files bottom-up, then the resulting gitlinks; an off-pin sub's advance is absorbed too. Explicit post (put/deletes present): commit exactly the staged set; a sub with put files gets a new commit whose gitlink is committed one level up, within scope. post-to-remote and patch are STORE-driven, independent of tree state; patch merges on all levels from each sub's de-facto base + local edits. Views recur always (if applicable); put/delete recur only on an explicit path, never bare. status must show each sub's base vs pin ahead/behind (counts + post/miss commit lists) and the "adv" dir entry in the parent.
RULED (gritzko, 2026-07-13, UNIFORM — supersedes the 2026-07-12 tip==base sub-ruling): post 'msg' advances the WORKTREE only, for every branch shape; a plain branch advances ONLY by an explicit post ?branch (the POST.mkd:30 arm), never as a commit side-effect. RESOLVED 2026-07-13: pin-branches anchor at the immediate parent's WT base (bullet above), so trees example 6 needs no extra post ? for the pins — only the top's own plain branch lags until explicitly posted.
Large behavioral change: every be/test/ leg that assumes commit==advance will move — inventory first, amend only with explicit permission per /meta/work.
Repro-first per the norms; keep DIS-054/DIS-058/DIS-059 post.js changes reconciled at land time.
WIP
Design decisions
(pending the ruling above)
TODOs
obtain the pin-advance ruling — RULED 2026-07-12: chain-anchored, moves only when the commits reach the chain root (top/recursive commit)
obtain the plain-branch sub-ruling — re-RULED 2026-07-13 UNIFORM: no commit-time FF at all, advance = explicit post ?branch
obtain the root-anchor sub-ruling — RESOLVED 2026-07-13: neither root form; pin tip = immediate parent's base gitlink, refreshed on any parent move
repro: commit on ANY tracked branch → wt ahead, ref UNMOVED; post ?branch → FF'ed; immediate-parent commit → child pin ref MOVES to the new base's gitlink, parent now ahead of ITS OWN pin; post ?/jab/.beagle → no bogus key
repro, recursion legs: get FFs a behind sub to the pin, SKIPS a diverged one ("adv" stays); bare post absorbs an off-pin sub + dirty sub files bottom-up; explicit post bubbles a put sub's gitlink one level up; scope leg — run at the sub, the parent stays untouched
postOne: DROP advanceRef from the commit path entirely; branch advance lives ONLY in the explicit Query arm
recursive post: the parent refreshes its immediate children's pin refs on ANY base move (post, get) — materialized rows, no lazy chain derivation
branchlib normalization in resolveTarget/advanceBranch
test inventory: legs assuming commit==advance; request user for amendments
spec sweep, wiki half — DONE 2026-07-13: Worktree.mkd:12 (uniform), Submodules.mkd intro+Recursion (pin=parent-base gitlink, gate, FF-behind, post forms, store-driven push/patch, views, scope), POST.mkd:5,9-10,13-15 (uniform + sub commit forms)
propose the post edits (user-owned): trees.mkd:211 "A regular branch would advance on commit" is the one line the uniform ruling overrules (:87-93 stands as written under the parent-base anchoring)
verify get ? (bare query = trunk) retracks a mounted sub off its pin-branch — the trees unpin example; add the leg to the repro set
Blockers and bummers
(none — both rulings made 2026-07-12; the ticket is fully actionable)
STALE 2026-07-15: its authority is gone — blog/trees.mkd no longer exists (only img/trees.jpg survives), so every [trees] cite below dangles; worktree replaces it and RULED 2026-07-15 replaces the rulings frozen here.
OVERRULED: the pin-branch exemption (pin refs as materialized rows the immediate parent refreshes) — RULED 2026-07-15 law #4: a sub tracks the parent's PIN URI //WT/path/to/sub, synthetic pin-branches are killed outright (DIS-072).
Refiled, nothing lost: the commit-vs-advance split + the ?/abs/branch mis-keying → DIS-076; the pin half → DIS-072; the bare-post/own-branch half → DIS-074; the model → DIS-071.
