MCPcopy Create free account
hub / github.com/colbymchenry/codegraph / getSegmentMatches

Method getSegmentMatches

src/index.ts:2713–2789  ·  view source on GitHub ↗

* Graph-derived prompt matching for the front-load hook's MEDIUM tier: * which indexed symbols do these prose words name? "state machine des * commandes" → `OrderStateMachine`, in any human language whose technical * nouns are Latin script — no keyword list involved. * * Precision com

(words: string[], limit: number = 6)

Source from the content-addressed store, hash-verified

2711 * returned symbol is guaranteed to exist right now.
2712 */
2713 getSegmentMatches(words: string[], limit: number = 6): SegmentMatch[] {
2714 if (words.length === 0) return [];
2715 // Variant → original word (plural folding), for coverage accounting.
2716 const variantToWord = new Map<string, string>();
2717 for (const word of words) {
2718 for (const variant of segmentLookupVariants(word)) {
2719 if (!variantToWord.has(variant)) variantToWord.set(variant, word);
2720 }
2721 }
2722 const variants = [...variantToWord.keys()];
2723
2724 // Tier A: co-occurrence. The SQL folds variants back to their original
2725 // word (#1146), so minWords=2 means two distinct PROMPT WORDS — a name
2726 // matching both `service` and `services` can't tie with (or crowd past
2727 // the LIMIT) a genuine two-word match. The JS re-check below recomputes
2728 // the fold from live segments as the honesty layer.
2729 const variantPairs = [...variantToWord.entries()].map(([segment, word]) => ({ segment, word }));
2730 const candidates: Array<{ name: string; matchedWords: Set<string> }> = [];
2731 for (const hit of this.queries.getSegmentCoOccurrence(variantPairs, 2, 24)) {
2732 const matched = this.wordsMatchingName(hit.name, variantToWord);
2733 if (matched.size >= 2) candidates.push({ name: hit.name, matchedWords: matched });
2734 }
2735
2736 // Tier B: single rare word. Only when co-occurrence found nothing — a
2737 // co-occurring name is categorically stronger evidence — and under
2738 // stricter rules, because one word is thin: the word must be ≥5 chars
2739 // (measured FPs: "this", "typo"); the segment must appear in AT LEAST TWO
2740 // names (a concept the codebase is about clusters across names —
2741 // CheckoutService/CheckoutController — while a prose coincidence is a
2742 // singleton: measured FP "deploy to PRODUCTION" → the one name
2743 // matchesNonProductionDir); and the candidate name must have ≥2 segments
2744 // (a bare common verb matching a bare function name — "write" → `write` —
2745 // is prose coincidence, not the user naming a symbol).
2746 if (candidates.length === 0) {
2747 const singleWordVariants = variants.filter((v) => variantToWord.get(v)!.length >= 5);
2748 const counts = this.queries.getSegmentNameCounts(singleWordVariants);
2749 const rare = [...counts.entries()]
2750 .filter(([, n]) => n >= 2 && n <= CodeGraph.SEGMENT_RARITY_CEILING)
2751 .sort((a, b) => a[1] - b[1])
2752 .slice(0, 2);
2753 for (const [variant] of rare) {
2754 const word = variantToWord.get(variant)!;
2755 for (const name of this.queries.getNamesForSegment(variant, 12)) {
2756 if (splitIdentifierSegments(name).length < 2) continue;
2757 candidates.push({ name, matchedWords: new Set([word]) });
2758 }
2759 }
2760 }
2761
2762 // Verify against nodes (the honesty gate) and pick a representative
2763 // definition per name. A name whose only nodes are file/import kind has
2764 // no real definition to point at — surfacing the import statement instead
2765 // reads as a matched symbol but isn't one (#1144) — so it's skipped, the
2766 // same way an orphaned vocab row is. (Import names no longer enter the
2767 // vocab at write time, but rows written before that exclusion persist
2768 // until the next full index.)
2769 const out: SegmentMatch[] = [];
2770 const seen = new Set<string>();

Callers 3

findRelevantContextMethod · 0.95
mainFunction · 0.80

Calls 13

wordsMatchingNameMethod · 0.95
segmentLookupVariantsFunction · 0.90
splitIdentifierSegmentsFunction · 0.90
hasMethod · 0.80
mapMethod · 0.80
pushMethod · 0.80
getSegmentNameCountsMethod · 0.80
getNamesForSegmentMethod · 0.80
getMethod · 0.65
getNodesByNameMethod · 0.65
setMethod · 0.45

Tested by

no test coverage detected