Beagle SCM

LITE-024 DONE: file.c:123:45 refs — lex the line:col suffix, land on the line

gritzko 2026-08-14: compiler/grep-style refs abc/TCP.c:12:24 (path, line, optional col) nav to the LINE, not just the file. DOG-033 already fuses (dir/)*base.ext into ONE F token and LITE-015 makes clicking it open the file through the door + FSEG resolver; this ticket extends the lexed shape with the :line(:col)? suffix and lands the opened view on that line:col via the LITE-023 cursor. lite/ pager consumer only; the lexer rule extension is ///jab/dog ground, same split as DOG-033.

Input

Context

The DOG-033 rule (FREE.c.rl + MKDT.c.rl, on_key) stops at the extension — abc/TCP.c:12:24 lexes as the F path + :12:24 crumbs, so a click opens the file at the TOP. The LITE-015 click path: pager hands the F token bytes to the door verbatim (openTarget in lite/main.js), stat then resolve.resolveAt FSEG fallback, chooser on several hits.

• Lexers are ragel machines in dog/tok/ — NO manual parsing in C, no regex anywhere; the suffix extends the ONE fused rule in both grammars, regenerated twins.
• LITE-023 gives the pager a cursor to land on; without it landed, degrade to scroll-to-line — the tickets stay independently landable.
• The : bar rides the same door (LITE-015), so typed file.c:123:45 navs identically for free once the door understands the suffix.

Goals

Lexer suffix + door split + pager landing, verified end to end in lite.

• Rule: the DOG-033 path shape gains an optional (":" dgt+ (":" dgt+)?) tail, still ONE F token, both FREE.c.rl and MKDT.c.rl. Bare times/ratios (12:30) still never match — the path+ext prefix is required; ABC-123, 1.5, a/b behavior unchanged.
• Door (lite/main.js): shed the trailing :line(:col)? off the F bytes BEFORE stat/FSEG resolve — the ONE split point both entrances (click, : bar) share; return the landing (line, col) alongside the hunks.
• Pager: apply the landing — scroll the line into view, cursor on line:col (LITE-023); no LITE-023 → scroll only. The pager still does no path math (LITE-015 law).
• Chooser leg: an ambiguous suffixed ref keeps the suffix through the chooser — a row click lands on the line too.
• Miss stays the quiet plain-words bar message; a suffix on a non-resolving path changes nothing.

Constraints

• Repo split: dog/tok rule + fixtures in a ///jab/dog worktree (jab/ itself is READ-ONLY); lite door/pager/tests in a ///lite worktree. No VC ops in either.
• Ragel only in C; the door's suffix shed in JS touches only the one token's bytes (a trailing-:digits split, not URI machinery).
• Tokenizer fixtures pin the fused boundaries: a/b.c:12, a/b.c:12:34, b.c: (no digits — suffix NOT fused), 12:30 alone (never a ref).
• Verify through the real UI path: a pty probe clicks a file.c:NN:NN ref in a viewed file and asserts the pushed view lands on the line (status line / cursor row), plus the : bar leg.

WIP

Design decisions

• The door owns the split and the landing rides its return value — keeps the pager arg-blind and gives the bar the feature for free; splitting in the pager would duplicate the shed at every entrance.

TODOs

• FREE.c.rl + MKDT.c.rl tail — ALREADY LANDED as DOG-034 (pathanchor, both tiers + TOK01 fixtures); nothing to write
• door: suffix shed + landing in the return, chooser keeps the suffix
• pager: scroll-to-line (degrade path; cursor landing is one line in _land() post-merge)
• pty test: click lands on the line; : bar leg
• gritzko: merge order — wt base #2dc61618 is behind the 14:00 1d0b187b LITE-023 landing, SAME three files (main.js, view/bro.js, view/pager.js); then the one-line cursor landing

Blockers and bummers

• ///jab/dog is STALE: head f8fc9d17 (Aug 12) lacks DOG-033/034; jab get ///jab/dog/ refetches the same, jab get ? goes BACKWARDS to 17Jul. The DOG-034 rule lives only in ///quickjab's dog sub (39158d0c). Runtime matters for the tests: installed ~/bin/jab fuses NO path tokens — run refline with LITEJAB=<quickjab build>.
• Worker did NOT get over its dirty wt (snapshot at ~/tmp/LITE-024-snapshot/); the merge onto 1d0b187b awaits gritzko's order.

Outcome

Worker wts under work/LITE-024/ (lite base ///lite/#2dc61618, quickjab head 60459dfd/dog 39158d0c for the runtime build), 5 files jab put in the lite wt, NO commit. Orchestrator re-ran refline in the wt: 23 bad under stock jab (pre-DOG-033 tokenizer, expected), PASS [lite/refline] 32 checks, 0 bad under the DOG-034 quickjab build.

• Lexer half: zero new code — DOG-034 already carries (":" ron64{1,10}){1,2} on both DOG-033 tiers; every ticket fixture green under the quickjab build (a/b.c:12, a/b.c:12:34 fuse to ONE F; b.c:, 12:30, localhost:8080, example.com:443 never; ABC-123/1.5 unchanged); ctest -R TOK 2/2.
• main.js +34/−4: digitTail()/splitRef() — the ONE split point in openTarget before stat/FSEG; landing returns as hunks.land = {line, col}; openPartial hands the tail to the chooser.
• view/bro.js +4/−2: chooser rows keep the tail in their hidden U targets, so a chosen row still lands.
• view/pager.js +24: _openPush applies hunks.land via _land() — line → byte → display row → v.scroll, stores v.land; scroll-only degrade as specced, no path math in the pager; miss message unchanged.
• test/refline/{pty.js,run.sh} new (248+76): 32 checks, real openpty + SGR presses — tail shed, land shapes (:12:4, :12, none, lone colon), chooser tails hidden at paint, click lands scroll===12 + bar #L12, : bar leg identical, chooser-row click lands line 5, miss quiet; plus a REAL-LEXER leg (committed src/see.c, the tokenizer's own F span abc/FSW.c:12:4, clicked → line 12). RED-first: 20 of 28 bad against the landed door/pager.
• Full lite suite vs pristine-clone baseline, same runtime: IDENTICAL fails (blob 12/13, cat 16/35, list 17/78, tree 25/47 — pre-existing), all else PASS + new refline.

Suggested commit (lite wt, post-merge): LITE-024: refs land on the line, door sheds :line:col