Beagle SCM

BEE-074: bee follows search and symbol anchors — the matcher port, def templates, the hook passes by

With QJAB-015 the lexer fuses file:'text', file:"snippet", file:/re/ and file:name, and bee lands every one of them on the top of the file. Resolve each to a byte span in the current blob, first match wins. The symbol anchor is sugar: per-file-type spot templates with the name bound. The three matchers exist in pure JS in be and port as they are.

Input

Context

• door.js:517 splitRef peels anchors by lastIndexOf(":"): digitTail (door.js:496), permaTail (door.js:507).
• index/perma.js:324:Ab follow lands :line:hashlet; isHashlet (:49) and mintHashlet (:75) already demand a non-digit.
• index/lindex.js:382:PS mints a LINK row per F token, keyed on the target's segments; the anchor plays no part in the key.
• The matchers: beagle-ext/views/spot/match.js:462 make(mode, body, ext): grep, regex, and the flat-token spot with brackets.
• SYM rows (BEE-063) and the trigram lane (BEE-071) narrow a loose target before any blob is opened.
• door.js:98 rootArg already turns a //name/rel arg into a path through mnt.byName, refusing an unknown name in words.

Goals

• splitRef takes its parts from tok.link (QJAB-015); digitTail and permaTail go.
• index/match.js: the port of be's three matchers, make(kind, body, ext).run(bytes, toks) -> [{lo, hi}].
• index/defs.js: ext -> [kind, template] rules in order; .c/.h share one: fn, struct, enum, union, typedef, macro, proto.
• file:name binds the name into s rule by rule: the first rule to hit wins, then its first match; the kind labels the hunk.
• bee see and the web board land on the match span; a miss says so and lands on the file top, never on a guess.
• bee mint and the commit hook leave search and symbol anchors byte-identical.
• A scope link (.c:/re/, abc/:'x', abc/.c:"f(x)") is a query: every match, as hunks in the search view, never one landing.
• Scope candidates come off the lanes first: trigrams for grep, regex literals and snippets, SYM rows for .c:name.
• //repo/… resolves through mnt.byName as rootArg does; an unknown repo is a miss in words; LINK rows key it (BEE-002).
• Rendered pages link them too (BEE-070): the href carries the anchor escaped, the page scrolls to the span.
• Cross-check test: every N-tagged token of a fixture corpus resolves back to itself through the template table.
• test/anchor/: each anchor kind, a colon in a body, a miss, mint passing by; the repro comes first (work 3).
• test/scope/: ext, dir and dir+ext scopes confine the scan; .c:name finds a def; //repo/ crosses repos; an unknown repo.

Constraints

• No regex and no hand peeling on ref text; the parts come from the lexer (work, NOT TO DO 2).
• DEF (dog/tok/DEF.c:85) keeps indexing; templates only resolve a GIVEN name, which is why they can stay naive.
• Existing suites green, no test edits; comment budgets per comment; README.mkd and INDEXES.mkd updated.

WIP

Design decisions

• Templates over N tags for resolution: a bound name needs no keyword guards, and a balanced A beats DEF's [^(){};]*.
• Definition before prototype in the .c list, so X.h:name lands on the declaration and X.c:name on the body.
• The def table is data, one row per rule: ['fn', 's(A){B}'], ['struct', 'struct s{B}'], ['proto', 's(A);'].
• A scope link and the grep/spot verbs are one code path; the link is only another spelling of the same query.

TODOs

• repro: test/anchor/ red on all four kinds
• splitRef over tok.link
• index/match.js, the port
• index/defs.js: .c/.h, .js/.ts, .go, .py, .rs, .rl
• see, the board and the rendered page land on the span
• mint and the hook pass them by
• scope queries over the lanes, test/scope/
• //repo/ links through mnt.byName
• the N-tag cross-check

Blockers and bummers

• ANSWERED: #define is ONE H token and the name a separate S beside it, so the macro rule is #define s.
• test/perma/pty.js:79:ix now fails: it pins a/b.c:k4 as NO anchor, and the grammar reads :k4 as a symbol.
• That is DOG-050's own ruled change (its bummer 1); no test was edited here, so the suite is left red for gritzko.
• No N tag ever reaches JS: DEFMark (dog/tok/DEF.h:29) is not bound into quickjab's tok.parse.
• So the cross-check runs over a WRITTEN corpus per ext instead, in test/scope/scope.js; a QJAB binding would free it.
• be's match.js:262:vT restores b.pos in the gap branch, so any two-space needle hangs; the port advances the scan.
• A one-letter symbol (union U) IS a placeholder in the template alphabet; a span must carry the name, or it is refused.
• An empty { } body is not expressible by {B}: trait Four { } is on the commented allow-list of the cross-check.
• Two rules per callable shape, s(A){B} and s(){B}, since a placeholder binds at least one token.
• bee see fuses argv words while tok.link refuses the join, so :'a b' survives the CLI without hand parsing.
• A SCOPE mints no LINK row: .c names no file, and a row on it would make every .c a suspect of every scope link.
• uri._parse("x.c:name") reads x.c as a SCHEME, so an explicit [t](x.c:name) href goes out verbatim.
• Pre-existing: [t](x.c:12) does the same, so BEE-070 rests on the autolink leg; not touched here.
• test/coldfan leg 5 failed once and passed three reruns: its own comment calls that race pre-existing.
• The .rl lexer tags whitespace S, not W; the matcher reads an all-blank token as a gap whatever its tag.

Outcome

Working tree only, nothing committed. door.js splits every ref through tok.link and resolves a search or symbol anchor to a byte span in the blob; index/match.js is the ported matcher, index/defs.js the def templates; view/grep.js grew scope() over the verb's own run, view/see.js fuses a spaced body and shows a query's hunks; index/lindex.js skips a scope. test/anchor/ (12 + 31) and test/scope/ (7 + 14) are new and green. 77 of 77 suites passed before, 78 of 79 after: the one red is test/perma, on the :k4 expectation DOG-050 already ruled amended.