Beagle SCM

DIFF-020: a get/post-merge conflict renders as plain local edits — no EXPECTED axis, no yellow

/todo/DIFF/DIFF-016 landed the conflict render, but its whole axis is gated on PATCH rows: expectedCtx returns null when patchTheirs() is empty, and a merge-get writes no patch row. So a conflicted file diffs as ordinary green/red — the woven both-sides tokens are indistinguishable from a hand edit, which is exactly what /todo/DIS/DIS-080 made the diff view responsible for. The wtlog already carries the two pointers the EXPECTED reading needs; nothing new has to be written.

Input

Context

• views/diff/diff.js:392-398 — expectedCtx() starts with expected.theirsShas(log) and returns null on an empty list; shared/expected.js:22-25 — that list is wtlog.patchTheirs(), PATCH rows only.
• views/diff/diff.js:398 builds cons: log.conflicts() and NOTHING reads it — grep finds no consumer, though /todo/ULOG/ULOG-004 goal 2 named the diff yellow spans as one.
• Measured, get-merge fixture (f.txt woven to XY): diff --color paints 48;5;217 + 48;5;157, the LOCAL rm/in washes; no 153/117 blue-orange, no 229/227 yellow.
• The machinery itself is fine: be/test/diff/20-patch-conflict PASSes with cq 227 — a patch-row conflict does wash yellow.
• The pointers are in the rows: after a conflicted get, curTip IS theirs (measured 56cfcde0) and the barrier row before it names the pre-get base (fc8ff3f3) — the same backward scan conflicts() already runs stops on exactly that row.
• Ours BYTES are unrecoverable from the ulog (a put row is path-only, no blob sha), and the EXPECTED model never needs them — the ours axis is "wt != EXPECTED".

Goals

A live con path diffs in CONFLICT MODE, rendered identically to patch mode.

• ONE patch/conflict mode (ruling gritzko 2026-08-03): 4 provenance washes — ours in/rm green/red, theirs in/rm pale blue/orange — no per-origin render.
• A conflict renders structurally: in-touching-in interleaves green/blue at one anchor; in-touching-rm nests green interior to an orange run (/wiki/Dirty shapes).
• Yellow retires with the mode: DIFF-016's 229/227 pair goes, be/test/diff/20-patch-conflict (asserts cq 227) regolds, patch-mode output changes too.
• The theirs/base pointers come from the rows: conflicts() returns the barrier sha and the pre-barrier tip alongside each path (accessor change, NO new ulog rows).
• shared/expected.js expectedOf(k, path, base, shas, cache) is reused as-is with base = the pre-get tip and shas = [the get's sha].
• A path with no con row keeps today's rendering exactly — zero extra work when nothing is conflicted (DIFF-016's own no-op rule).

Constraints

• Weave-based only; no diff3, no marker parsing (DIFF-016 Constraints).
• --plain output stays byte-identical — provenance is a COLOUR distinction, asserted on --color SGR digits (DIFF-016's deliberate no-fallback ruling).
• Verify through the real UI path, pager pty, not TLV probes ([verify-through-real-ui-path]).
• The con row stays path-only unless gritzko rules otherwise — base+theirs are derivable from row order.

WIP

Design decisions

• RULING (gritzko 2026-08-03): one patch/conflict mode, 4 washes; a con path's FROM is the PRE-merge base off the rows — provenance needs it, uniform base-relativity yields for that one path.
• The banner mark off conflicts() stays: with yellow gone, the file-level conflict signal is status ! + the banner, so it is load-bearing, not a fallback.
• OPEN: a clean abutment (green at an orange hole's EDGE) renders close to a real collision (/wiki/Dirty: adjacency in text != overlap in the weave); a 5th collision wash only if gritzko rules it.

TODOs

• SIBLING accessor on the ULOG-004 scan (conflictOrigins() → [{path, base, theirs}]); conflicts() KEEPS returning bare paths — post.js:886 and classify string-test it, and a shape change means a lockstep edit in two parallel tickets
• feed expectedCtx from that when patchTheirs() is empty; keep the patch arm untouched
• consume ec.cons — the path-granular cross-check DIFF-016 specified and never wired
• banner mark for a conflicted path (survives either way, and is the whole fallback arm)
• tests: get-merge conflict shows the 4 washes, 20-patch-conflict regolds off yellow, --plain byte-identical, pty render

Blockers and bummers

Outcome