Beagle SCM

DOG-033 DONE: tokenizer path refs — ticket OR (dir/)*base.ext fuses to one F

Ruling (gritzko 2026-08-14): file references in code comments and markdown get the ticket-code treatment — reuse the issue-key detection in place, the rule becomes ticket OR (dir/)*base.ext. One fused F token per reference, minted by the SHARED tokenizer, so no consumer ever scans text for paths itself and the pager's click can never drift from the lexer.

Input

Context

Ticket codes are clickable because FREE.c.rl:122 (uc ucnum* "-" dgt+) and MKDT.c.rl:135 fuse the code into one F nav-target token; a resolver then maps the bytes to todo/TOPIC/TOPIC-123. A file reference like diff/color.js in a C comment lexes as word/punct crumbs (diff / color . js) — nothing to click, nothing to hand a resolver.

• FREE is the comment-interior lexer every language tokenizer re-lexes comments with; MKDT covers markdown/StrictMark — the two rules cover "comments, Markdown and StrictMark", same as ticket codes today.
• LITE-015 consumes the token: click hands the bytes to the door, which resolves them through the LITE-011 FSEG descent.

Goals

The issue-key rule becomes issue-key OR partial-path; a path reference paints and clicks exactly like a ticket code, in both grammars.

• Rule shape: (seg "/")* seg "." ext — a dotted extension is REQUIRED (a/b is not a path ref), fused to ONE F token via the existing on_key, longest-match over plain word/number.
• Both FREE.c.rl and MKDT.c.rl, regenerated .rl.c twins; tok.parse parity between the two (be's shared/ticket.js precedent: one rule, two grammars, no drift).
• Existing F consumers stay QUIET on a non-key F: be's shared/ticket.js probes todo/ and misses (a quiet null already), lite's door stat-misses — no guards to add.

Constraints

• Ragel only — never manual parsing in C, no regexes anywhere.
• MKDT ordering: code/link/emph spans keep outranking the new rule exactly as they outrank issue keys (a path inside backticks stays in its span).
• No new tag: F is THE nav-target tag — same paint (THEME slot 56), same anchor semantics; the tok32 vocabulary (dog/tok/TOK.h) does not change.
• Segment charset from the existing word classes (no leading-dot segment, so ../x and dotfiles stay crumbs); tests: diff/color.js, bare TCP.c, a/b (no ext, NOT fused), ABC-123 (unchanged), 1.5 (still a number).

WIP

Design decisions

TODOs

• [ ] FREE.c.rl: the fused path rule via on_key, next to the issue-key rule
• [ ] MKDT.c.rl: same rule, ordered under code/link/emph
• [ ] regen the .rl.c twins, tokenizer fixtures for path / non-path lines

Blockers and bummers

Outcome