Beagle SCM

POST-032: weave conflicts on pending edits must mark+record, never hard-err (GETCONF)

Input

Context

Landing WORK-001 (2026-07-18): posting the WORK-001 wt onto its tracked ///be/ hit ONE genuine weave conflict (views/bro/pager.js, the driveSpell(s, …) line) and the flow HARD-ERRED instead of treating the conflict as a normal outcome.

• Attempt 1 (post run in the be tree): the weave fenced pager.js and wrote a con row (26718LLuCr) — then ABORTED with no commit; the tip stayed at the prior rev with the fence stranded in the working tree.
• Attempt 2 (:post from the WORK-001 wt): the wt-local commit LANDED (post ?#c32ddd17 in WORK-001/.be), then the target-side get leg died with err: be get: GETCONF 1 file(s) merged with conflicts — resolve the ma…, re-fencing be's pager.js and writing a SECOND con row (26718LQaF2) — target tip unmoved.
• Self-hosting corollary observed: a fence woven into a LOADED .js of the be tree makes every next jab there die at parse (SyntaxError: Unexpected token '<<') — the marking itself is per spec; the hard-error duplication is not.

Goals

post (and its get leg) treat a pending-edit conflict as a NORMAL merge outcome, per the ruling below: complete the weave, leave the fences, write the con wtlog row, report conflict status — and never exit through the GETCONF hard-error path.

• One attempt = one con row; a repeated post over an unresolved conflict must not stack duplicate rows.
• Exit status/report says "merged with conflicts" as a STATE, not an error; the rest of the op (unconflicted files, log rows) completes.

Constraints

• RULING (gritzko, chat 2026-07-18): "Conflicts for pending edits are not any special. Must merge, mark conflict, make wtlog record. Erring out is not correct." Matches the bare-track post spec (weave uncommitted + FF base, conflict status); the dirty-refusal/hard-err gate contradicts it.
• get is likely affected the same way (GETCONF originates in the get path; bare get over dirty edits should mark+record too) — fix the shared leg, not a post-only wrapper.
• User-facing text in plain words; no bare C codes like GETCONF in the error line.

WIP

Design decisions

• SALVAGE (be reset 2026-07-18, the resolved hybrid lived only in be's working tree): the CORRECT pager.js merge of main's BE-039 shape with WORK-001's BRO-025 is — _runSpell(spell, ctxOverride) and _actSpell(spell, ctxOverride); in both, const ctx = ctxOverride || discover.navTree(this._context()); and driveSpell(s, ctx || undefined); view.call.context = ctx || ""; main's _fillAuth compose-by-path stays AS IS (wt URIs compose by path).
• SALVAGE cont.: the click site becomes const o = SPELL.parseOspell(target); then mutation-gate on o.spell's verb → _actSpell(o.spell, o.context || undefined) else _runSpell(o.spell, o.context || undefined).
• WARNING: WORK-001's own pager.js carries the REGRESSIVE side (drops navTree threading in _runSpell and _fillAuth's compose-by-path) — merge the hybrid INTO work/WORK-001 BEFORE the next landing attempt, else the same line re-conflicts or the regression lands silently.

TODOs

• red-first repro: post a wt onto a tracked tree with one conflicting pending edit → expect fences + ONE con row + conflict status, tip advanced per spec, no hard error
• same for bare get over a conflicting pending edit
• dedup: re-run over unresolved fences adds no second con row
• locate the GETCONF throw path; route it into the conflict-status report

Blockers and bummers

• Evidence trail: be/.be rows 26718LLuCr + 26718LQaF2 (duplicate con), WORK-001/.be post ?#c32ddd17 (the stranded local commit awaiting a working target-side leg).
• Post-reset state (2026-07-18): be's working tree reset — the fence, the duplicate con fallout, AND the resolved hybrid all gone from be; the FULL WORK-001 change survives in work/WORK-001 (8 staged files + test sub + local commit c32ddd17); GRAF-001 landed earlier as ade2448b and is unaffected.
• Related unlanded: the GET-050 draft (same-pin re-point no-op + wtlog-write-before-checkout atomicity) with its fix+tests staged in work/GET-REPOINT.

Outcome