MARK-009: mark_refdef is a hand-rolled byte scanner — ragel-ize the refdef grammar

Now: OPEN
Sev: LOW

mark_refdef (MARK.c:76-101) parses the refdef line grammar with a manual scan loop (while (!u8csEmpty(s) && u8csAt(s,0) != ']') u8csUsed(s,1); …) — slice-typed but still hand parsing, against the module's own doctrine (mark/README.mkd:15 "never a hand-rolled byte scan"; abc "never do manual parsing, use ragel"). It also runs TWICE per line (collector pass 1, again per-line in mark_blocks MARK.c:646). From the 2026-07-07 mark/ diagnostic review; code confirmed in-tree. Method Issues.

Input

Context

The MARKE ragel lexer already classifies lines — the refdef fields (label, url, title) belong in the same machine or a sibling one.
Double scanning is a smell of the same missing single classification point.

Goals

Refdef grammar lives in ragel (MARKE or a dedicated machine); mark_refdef's manual loops go; one scan per line.

Constraints

Pure refactor — byte-identical HTML on the whole wiki render (old vs new binary diff empty); regen MARKE.rl.c from the .rl source.

WIP

Design decisions

(pending)

TODOs

Ragel refdef machine; retire the manual scan; wiki render byte-identical; suite green.

Blockers and bummers

None known.

Outcome

(pending)