Beagle SCM

STATUS-018: a staged path's live con never renders — the put branch preempts the con check

classify tests the /todo/ULOG/ULOG-004 con set only in its NO-STAGED-INTENT arm, so a con row on a path that also carries a put row is unreachable: the row renders ...V (staged change) and counts.con stays 0, while post — which reads wtl.conflicts() directly — refuses the same path. Status and post disagree on one ulog. A conflicting get RE-STAGES an already-staged path, so this masks EVERY conflict on staged work, which is the normal shape of a post that had to merge.

Input

Context

• Live evidence, //TODO-005 after a half-failed post: log.conflicts() → ["views/todo/todo.js"], classify buckets that path put, counts.con = 0, be status prints ...V, be post throws conflict in tracked file views/todo/todo.js.
• Root cause: shared/classify.js:456 if (onDisk && conSet.has(path)) sits AFTER the staged-intent branch (shared/classify.js:443, push({bucket: b ? "put" : "new"}); continue;).
• The trigger is the get itself: a merge-get re-writes a put <path> row for a path that was already staged, THEN the con <path> row — so the mask is automatic.
• be/test/status/conflict only covers the UNSTAGED path (no put row lands, ...! renders), which is why this passed CI.
• /todo/POST/POST-035's gate and /todo/DIS/DIS-080's "resolution == posted" are both intact — post is the correct side here.

Goals

A live con row outranks staged intent, so the two readers of one ulog cannot disagree.

• A path in wtl.conflicts() buckets con and paints the wt char ! whether or not it carries a put row; counts.con tallies it.
• The staged fact is not lost: the row stays staged (the quad's own column), only the wt char changes.
• Ack semantics UNCHANGED: a put/delete row LATER than the con row still resolves it (verified: edit + put + post lands with no --force).
• status stays row-driven — no byte scan, no marker parsing (STATUS-017).

Constraints

• ONE accessor for both sides: classify must consume wtl.conflicts(), never re-derive a con set, so post's gate and status cannot drift again.
• Regold any golden asserting ...V on a staged-and-conflicted path.
• No change to conflicts() itself — the accessor is right; only its consumer in classify is wrong.

WIP

Design decisions

• Shape of the fix: hoist the conSet test above the staged-intent branch, keeping the onDisk guard (a con row for a path no longer on disk is not a wt conflict).

TODOs

• repro test: stage the path, then a conflicting get; assert ...! + 1 con and that post's refusal and status agree
• hoist the con test above the staged branch in shared/classify.js
• keep staged true on the con row; check quad.js renders both facts
• regold whatever asserted ...V for this shape
• extend be/test/status/conflict with the staged variant (its unstaged case stays)

Blockers and bummers

• /todo/DIFF/DIFF-020 needs this: the status row is where a conflict is NOTICED, and its ! char is what routes the row's nav to diff (views/status/status.js:326).

Outcome