sensemaking 0.24.3 → 0.24.5

package/dist/esm/text/segment.js

@@ -6,7 +6,8 @@ import { stemmer } from 'stemmer';
 export const UNSPACED_SCRIPTS = '\\p{scx=Han}\\p{scx=Hiragana}\\p{scx=Katakana}\\p{scx=Thai}\\p{scx=Khmer}\\p{scx=Lao}\\p{scx=Myanmar}';
 // A run is script BASE characters with their combining marks attached; a bare mark after a
 // Latin letter (decomposed é) never starts one.
-const RUN = new RegExp(`((?:[${UNSPACED_SCRIPTS}]\\p{M}*)+)`, 'gu');
+const RUN_BODY = `(?:[${UNSPACED_SCRIPTS}]\\p{M}*)+`;
+const RUN = new RegExp(`(${RUN_BODY})`, 'gu');
 const HAS_RUN = new RegExp(`[${UNSPACED_SCRIPTS}]`, 'u');
 // Whether text holds a script that marks no word boundaries, the predicate every store's
 // sidecar population and query split turns on.
@@ -121,26 +122,150 @@ const QUALIFIER = /(^|[\s(])(-?)(title|summary|text)\s*:\s*$/;
 // Raw title/summary/text drop punctuation as unicode61's token separator, so `数数` and `数。数`
 // falsely match there; only the barriered `_seg` columns are safe from it, hence this fallback target.
 const SIDECAR_COLUMNS = '{title_seg summary_seg text_seg}:';
+const AUTHORED_COLUMNS = '{title summary text}:';
+const QUOTED_RUN_TOKEN = new RegExp(`"(${RUN_BODY})"`, 'gu');
+const FTS_WORD = /[\p{L}\p{N}_]/u;
+function isFtsWord(char) {
+    return char !== undefined && FTS_WORD.test(char);
+}
+function quoteEnd(text, start) {
+    for(let i = start + 1; i < text.length; i++){
+        if (text[i] !== '"') continue;
+        if (text[i + 1] === '"') {
+            i++;
+            continue;
+        }
+        return i + 1;
+    }
+    return text.length;
+}
+function nearEnd(text, open) {
+    let depth = 1;
+    let quoted = false;
+    for(let i = open + 1; i < text.length; i++){
+        const char = text[i];
+        if (char === '"') {
+            if (quoted && text[i + 1] === '"') {
+                i++;
+            } else {
+                quoted = !quoted;
+            }
+        } else if (!quoted && char === '(') {
+            depth++;
+        } else if (!quoted && char === ')' && --depth === 0) {
+            return i + 1;
+        }
+    }
+    return text.length;
+}
+// Lexical routing only: SQLite remains responsible for validating the expression. The scanner
+// recognizes NEAR groups and the one FTS string governed by ^, while quoted text shields both.
+function positionalSpans(terms) {
+    const spans = [];
+    for(let i = 0; i < terms.length; i++){
+        if (terms[i] === '"') {
+            i = quoteEnd(terms, i) - 1;
+            continue;
+        }
+        if (terms.startsWith('NEAR', i) && !isFtsWord(terms[i - 1])) {
+            var _terms_open;
+            let open = i + 4;
+            while(/\s/u.test((_terms_open = terms[open]) !== null && _terms_open !== void 0 ? _terms_open : ''))open++;
+            if (terms[open] === '(') {
+                const end = nearEnd(terms, open);
+                if (HAS_RUN.test(terms.slice(open + 1, end))) spans.push({
+                    start: i,
+                    end,
+                    qualified: QUALIFIER.test(terms.slice(0, i))
+                });
+                i = end - 1;
+                continue;
+            }
+        }
+        if (terms[i] === '^' && !isFtsWord(terms[i - 1])) {
+            var _terms_operand;
+            let operand = i + 1;
+            while(/\s/u.test((_terms_operand = terms[operand]) !== null && _terms_operand !== void 0 ? _terms_operand : ''))operand++;
+            const end = terms[operand] === '"' ? quoteEnd(terms, operand) : (()=>{
+                var _terms_cursor;
+                let cursor = operand;
+                while(/[\p{L}\p{N}_]/u.test((_terms_cursor = terms[cursor]) !== null && _terms_cursor !== void 0 ? _terms_cursor : ''))cursor++;
+                return cursor;
+            })();
+            if (end > operand && HAS_RUN.test(terms.slice(operand, end))) spans.push({
+                start: i,
+                end,
+                qualified: QUALIFIER.test(terms.slice(0, i))
+            });
+            i = Math.max(i, end - 1);
+        }
+    }
+    return spans;
+}
+function routePositionalSpans(terms) {
+    const spans = positionalSpans(terms);
+    if (spans.length === 0) return terms;
+    let out = '';
+    let cursor = 0;
+    for (const span of spans){
+        out += terms.slice(cursor, span.start);
+        if (!span.qualified) out += AUTHORED_COLUMNS;
+        out += terms.slice(span.start, span.end);
+        cursor = span.end;
+    }
+    return out + terms.slice(cursor);
+}
+function inPositionalSpan(start, spans) {
+    return spans.some((span)=>start >= span.start && start < span.end);
+}
 // A run's punctuation-free groups, each its own quoted phrase (bare token if one grapheme),
 // space-joined -- the same split points segmentField barriers, so query and index agree.
-function runQuery(run, columnPrefix) {
-    return splitOnPunctuation(run).map((g)=>`${columnPrefix}${g.length > 1 ? `"${g.join(' ')}"` : g[0]}`).join(' ');
+function runQuery(run, columnPrefix, quoteSingle = false) {
+    return splitOnPunctuation(run).map((g)=>`${columnPrefix}${g.length > 1 || quoteSingle ? `"${g.join(' ')}"` : g[0]}`).join(' ');
+}
+// Retarget complete quoted run tokens without interpreting the expression around them. Positional
+// spans have already been routed to authored columns, so sidecar barriers cannot change meaning.
+function quotedRunTokens(terms) {
+    const spans = positionalSpans(terms);
+    let out = '';
+    let cursor = 0;
+    for (const match of terms.matchAll(QUOTED_RUN_TOKEN)){
+        const start = match.index;
+        const run = match[1];
+        if (inPositionalSpan(start, spans) || PUNCTUATION.test(run)) continue;
+        out += terms.slice(cursor, start);
+        const qualifier = out.match(QUALIFIER);
+        if (qualifier) {
+            out = `${out.slice(0, qualifier.index)}${qualifier[1]}${runQuery(run, `${qualifier[2]}${qualifier[3]}_seg:`, true)}`;
+        } else {
+            out += runQuery(run, SIDECAR_COLUMNS, true);
+        }
+        cursor = start + match[0].length;
+    }
+    return cursor === 0 ? terms : out + terms.slice(cursor);
 }
 // Query side: each unspaced run becomes phrases of its graphemes, matching segmentField. A
 // qualifier maps to its `_seg` column; unqualified maps to all three (SIDECAR_COLUMNS).
 export function segmentMatch(terms) {
     if (!HAS_RUN.test(terms)) return terms;
+    terms = routePositionalSpans(terms);
+    terms = quotedRunTokens(terms);
+    const spans = positionalSpans(terms);
     let out = '';
     let quoted = false;
+    let offset = 0;
     const pieces = terms.split(RUN); // split keeps captured runs at odd indices
     for(let i = 0; i < pieces.length; i++){
         if (i % 2 === 0) {
             for (const ch of pieces[i])if (ch === '"') quoted = !quoted;
             out += pieces[i];
+            offset += pieces[i].length;
             continue;
         }
-        if (quoted) {
+        const protectedRun = inPositionalSpan(offset, spans);
+        if (quoted || protectedRun) {
             out += pieces[i]; // an author's phrase is matched as written
+            offset += pieces[i].length;
             continue;
         }
         const m = out.match(QUALIFIER);
@@ -149,6 +274,7 @@ export function segmentMatch(terms) {
         } else {
             out += runQuery(pieces[i], SIDECAR_COLUMNS);
         }
+        offset += pieces[i].length;
     }
     return out;
 }

package/dist/esm/text/segment.js.map

@@ -1 +1 @@
-{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/text/segment.ts"],"sourcesContent":["import { stemmer } from 'stemmer';\n\n// Word boundaries FTS5's tokenizers cannot find on their own. Contract: seg(query) must appear in\n// seg(document) for any substring query; word mode is context-dependent and breaks that, so grapheme mode (UAX #29) is used.\n\n// Scripts written without word spaces: a closed set of writing systems.\n// Script_Extensions, not Script: the katakana long-vowel mark ー is Script=Common.\nexport const UNSPACED_SCRIPTS = '\\\\p{scx=Han}\\\\p{scx=Hiragana}\\\\p{scx=Katakana}\\\\p{scx=Thai}\\\\p{scx=Khmer}\\\\p{scx=Lao}\\\\p{scx=Myanmar}';\n// A run is script BASE characters with their combining marks attached; a bare mark after a\n// Latin letter (decomposed é) never starts one.\nconst RUN = new RegExp(`((?:[${UNSPACED_SCRIPTS}]\\\\p{M}*)+)`, 'gu');\nconst HAS_RUN = new RegExp(`[${UNSPACED_SCRIPTS}]`, 'u');\n\n// Whether text holds a script that marks no word boundaries, the predicate every store's\n// sidecar population and query split turns on.\nexport function hasUnspacedRun(text: string): boolean {\n  return HAS_RUN.test(text);\n}\n\nexport interface SearchToken {\n  text: string;\n  start: number;\n  end: number;\n  unspaced: boolean;\n}\n\n// Search tokens share the native separator contract: letters, marks, and numbers form tokens;\n// underscore, apostrophe, and hyphen separate them. Unspaced-script runs stay whole so callers\n// can apply substring semantics, while adjacent Latin text still gets its own token.\nexport function searchTokens(text: string): SearchToken[] {\n  const tokens: SearchToken[] = [];\n  for (const match of text.matchAll(/[\\p{L}\\p{N}][\\p{L}\\p{M}\\p{N}]*/gu)) {\n    const start = match.index;\n    const end = start + match[0].length;\n    let cursor = start;\n    for (const run of text.slice(start, end).matchAll(RUN)) {\n      const runStart = start + run.index;\n      if (runStart > cursor) tokens.push({ text: text.slice(cursor, runStart), start: cursor, end: runStart, unspaced: false });\n      const runEnd = runStart + run[0].length;\n      tokens.push({ text: run[0], start: runStart, end: runEnd, unspaced: true });\n      cursor = runEnd;\n    }\n    if (cursor < end) tokens.push({ text: text.slice(cursor, end), start: cursor, end, unspaced: false });\n  }\n  return tokens;\n}\n\n// Phrase membership is checked after native ranking. A CJK token matches inside one authored\n// run; Latin tokens compare folded Porter stems. Each call scans one field once per candidate.\nexport function matchesSearchPhrase(text: string, phrase: SearchToken[]): boolean {\n  const haystack = searchTokens(text);\n  if (phrase.length === 0 || haystack.length < phrase.length) return false;\n  for (let start = 0; start <= haystack.length - phrase.length; start++) {\n    let matches = true;\n    for (let i = 0; i < phrase.length; i++) {\n      const query = phrase[i];\n      const authored = haystack[start + i];\n      if (query.unspaced) {\n        if (!authored.unspaced || !authored.text.includes(query.text)) matches = false;\n      } else if (authored.unspaced || stemmer(foldForSearch(authored.text)) !== stemmer(foldForSearch(query.text))) {\n        matches = false;\n      }\n      if (!matches) break;\n    }\n    if (matches) return true;\n  }\n  return false;\n}\n\n// Character spans of unspaced-script runs in text, the same boundary the index segments on --\n// the snippet marker uses this to keep substring semantics inside a run and word/stem matching outside it.\nexport function unspacedRuns(text: string): Array<{ start: number; end: number }> {\n  return Array.from(text.matchAll(RUN), (m) => ({ start: m.index, end: m.index + m[0].length }));\n}\n// Grapheme clusters, ECMA-402/UAX #29: base char plus its marks, ZWJ sequences, Hangul jamo.\n// Built on first use and kept: construction is 6.6 ms, and a tree with no unspaced-script run\n// never segments at all, so no command pays it at module load.\nlet graphemeSegmenter: Intl.Segmenter | undefined;\n// unicode61 drops Unicode punctuation as a separator. Some of it (。、「」) has Script_Extensions\n// into an unspaced script, so RUN keeps it -- a grapheme matching this becomes a split point.\nconst PUNCTUATION = /\\p{P}/u;\n// Token barrier between separate runs (or a punctuation split within one) so their graphemes\n// are never phrase-adjacent. U+A7F7: a letter (so FTS5 keeps it as a token) no one types.\nconst BARRIER = 'ꟷ';\n\nfunction graphemes(run: string): string[] {\n  graphemeSegmenter ??= new Intl.Segmenter(undefined, { granularity: 'grapheme' });\n  return Array.from(graphemeSegmenter.segment(run), (s) => s.segment);\n}\n\n// A run's graphemes, cut into punctuation-free groups at every punctuation grapheme (dropped,\n// matching unicode61). The one place punctuation is classified, so index and query agree.\nfunction splitOnPunctuation(run: string): string[][] {\n  const groups: string[][] = [[]];\n  for (const g of graphemes(run)) {\n    if (PUNCTUATION.test(g)) groups.push([]);\n    else groups[groups.length - 1].push(g);\n  }\n  return groups.filter((g) => g.length > 0);\n}\n\n// Index side: '' when the field has no unspaced-script run. Otherwise each run explodes into\n// its graphemes, barrier-delimited from its neighbors and from a punctuation split within itself.\nexport function segmentField(text: string): string {\n  if (!HAS_RUN.test(text)) return '';\n  const out = text.replace(RUN, (run) => {\n    const body = splitOnPunctuation(run)\n      .map((g) => g.join(' '))\n      .join(` ${BARRIER} `);\n    return ` ${BARRIER} ${body} ${BARRIER} `;\n  });\n  return out.replace(/\\s+/g, ' ').trim();\n}\n\n// A `title:`/`summary:`/`text:` qualifier directly before a run that is about to become a\n// quoted grapheme phrase, so the rewrite can retarget it at the matching `_seg` column.\nconst QUALIFIER = /(^|[\\s(])(-?)(title|summary|text)\\s*:\\s*$/;\n\n// Raw title/summary/text drop punctuation as unicode61's token separator, so `数数` and `数。数`\n// falsely match there; only the barriered `_seg` columns are safe from it, hence this fallback target.\nconst SIDECAR_COLUMNS = '{title_seg summary_seg text_seg}:';\n\n// A run's punctuation-free groups, each its own quoted phrase (bare token if one grapheme),\n// space-joined -- the same split points segmentField barriers, so query and index agree.\nfunction runQuery(run: string, columnPrefix: string): string {\n  return splitOnPunctuation(run)\n    .map((g) => `${columnPrefix}${g.length > 1 ? `\"${g.join(' ')}\"` : g[0]}`)\n    .join(' ');\n}\n\n// Query side: each unspaced run becomes phrases of its graphemes, matching segmentField. A\n// qualifier maps to its `_seg` column; unqualified maps to all three (SIDECAR_COLUMNS).\nexport function segmentMatch(terms: string): string {\n  if (!HAS_RUN.test(terms)) return terms;\n  let out = '';\n  let quoted = false;\n  const pieces = terms.split(RUN); // split keeps captured runs at odd indices\n  for (let i = 0; i < pieces.length; i++) {\n    if (i % 2 === 0) {\n      for (const ch of pieces[i]) if (ch === '\"') quoted = !quoted;\n      out += pieces[i];\n      continue;\n    }\n    if (quoted) {\n      out += pieces[i]; // an author's phrase is matched as written\n      continue;\n    }\n    const m = out.match(QUALIFIER);\n    if (m) {\n      out = `${out.slice(0, m.index)}${m[1]}${runQuery(pieces[i], `${m[2]}${m[3]}_seg:`)}`;\n    } else {\n      out += runQuery(pieces[i], SIDECAR_COLUMNS);\n    }\n  }\n  return out;\n}\n\n// FTS tokenizers compare letters without combining marks. Keep the authored string for output and\n// offsets, but use this key when JS needs to compare a matched word to its source segment.\nexport function foldForSearch(text: string): string {\n  return text.normalize('NFD').replace(/\\p{M}/gu, '').toLowerCase();\n}\n"],"names":["stemmer","UNSPACED_SCRIPTS","RUN","RegExp","HAS_RUN","hasUnspacedRun","text","test","searchTokens","tokens","match","matchAll","start","index","end","length","cursor","run","slice","runStart","push","unspaced","runEnd","matchesSearchPhrase","phrase","haystack","matches","i","query","authored","includes","foldForSearch","unspacedRuns","Array","from","m","graphemeSegmenter","PUNCTUATION","BARRIER","graphemes","Intl","Segmenter","undefined","granularity","segment","s","splitOnPunctuation","groups","g","filter","segmentField","out","replace","body","map","join","trim","QUALIFIER","SIDECAR_COLUMNS","runQuery","columnPrefix","segmentMatch","terms","quoted","pieces","split","ch","normalize","toLowerCase"],"mappings":"AAAA,SAASA,OAAO,QAAQ,UAAU;AAElC,kGAAkG;AAClG,6HAA6H;AAE7H,wEAAwE;AACxE,kFAAkF;AAClF,OAAO,MAAMC,mBAAmB,wGAAwG;AACxI,2FAA2F;AAC3F,gDAAgD;AAChD,MAAMC,MAAM,IAAIC,OAAO,CAAC,KAAK,EAAEF,iBAAiB,WAAW,CAAC,EAAE;AAC9D,MAAMG,UAAU,IAAID,OAAO,CAAC,CAAC,EAAEF,iBAAiB,CAAC,CAAC,EAAE;AAEpD,yFAAyF;AACzF,+CAA+C;AAC/C,OAAO,SAASI,eAAeC,IAAY;IACzC,OAAOF,QAAQG,IAAI,CAACD;AACtB;AASA,8FAA8F;AAC9F,+FAA+F;AAC/F,qFAAqF;AACrF,OAAO,SAASE,aAAaF,IAAY;IACvC,MAAMG,SAAwB,EAAE;IAChC,KAAK,MAAMC,SAASJ,KAAKK,QAAQ,CAAC,oCAAqC;QACrE,MAAMC,QAAQF,MAAMG,KAAK;QACzB,MAAMC,MAAMF,QAAQF,KAAK,CAAC,EAAE,CAACK,MAAM;QACnC,IAAIC,SAASJ;QACb,KAAK,MAAMK,OAAOX,KAAKY,KAAK,CAACN,OAAOE,KAAKH,QAAQ,CAACT,KAAM;YACtD,MAAMiB,WAAWP,QAAQK,IAAIJ,KAAK;YAClC,IAAIM,WAAWH,QAAQP,OAAOW,IAAI,CAAC;gBAAEd,MAAMA,KAAKY,KAAK,CAACF,QAAQG;gBAAWP,OAAOI;gBAAQF,KAAKK;gBAAUE,UAAU;YAAM;YACvH,MAAMC,SAASH,WAAWF,GAAG,CAAC,EAAE,CAACF,MAAM;YACvCN,OAAOW,IAAI,CAAC;gBAAEd,MAAMW,GAAG,CAAC,EAAE;gBAAEL,OAAOO;gBAAUL,KAAKQ;gBAAQD,UAAU;YAAK;YACzEL,SAASM;QACX;QACA,IAAIN,SAASF,KAAKL,OAAOW,IAAI,CAAC;YAAEd,MAAMA,KAAKY,KAAK,CAACF,QAAQF;YAAMF,OAAOI;YAAQF;YAAKO,UAAU;QAAM;IACrG;IACA,OAAOZ;AACT;AAEA,6FAA6F;AAC7F,+FAA+F;AAC/F,OAAO,SAASc,oBAAoBjB,IAAY,EAAEkB,MAAqB;IACrE,MAAMC,WAAWjB,aAAaF;IAC9B,IAAIkB,OAAOT,MAAM,KAAK,KAAKU,SAASV,MAAM,GAAGS,OAAOT,MAAM,EAAE,OAAO;IACnE,IAAK,IAAIH,QAAQ,GAAGA,SAASa,SAASV,MAAM,GAAGS,OAAOT,MAAM,EAAEH,QAAS;QACrE,IAAIc,UAAU;QACd,IAAK,IAAIC,IAAI,GAAGA,IAAIH,OAAOT,MAAM,EAAEY,IAAK;YACtC,MAAMC,QAAQJ,MAAM,CAACG,EAAE;YACvB,MAAME,WAAWJ,QAAQ,CAACb,QAAQe,EAAE;YACpC,IAAIC,MAAMP,QAAQ,EAAE;gBAClB,IAAI,CAACQ,SAASR,QAAQ,IAAI,CAACQ,SAASvB,IAAI,CAACwB,QAAQ,CAACF,MAAMtB,IAAI,GAAGoB,UAAU;YAC3E,OAAO,IAAIG,SAASR,QAAQ,IAAIrB,QAAQ+B,cAAcF,SAASvB,IAAI,OAAON,QAAQ+B,cAAcH,MAAMtB,IAAI,IAAI;gBAC5GoB,UAAU;YACZ;YACA,IAAI,CAACA,SAAS;QAChB;QACA,IAAIA,SAAS,OAAO;IACtB;IACA,OAAO;AACT;AAEA,8FAA8F;AAC9F,2GAA2G;AAC3G,OAAO,SAASM,aAAa1B,IAAY;IACvC,OAAO2B,MAAMC,IAAI,CAAC5B,KAAKK,QAAQ,CAACT,MAAM,CAACiC,IAAO,CAAA;YAAEvB,OAAOuB,EAAEtB,KAAK;YAAEC,KAAKqB,EAAEtB,KAAK,GAAGsB,CAAC,CAAC,EAAE,CAACpB,MAAM;QAAC,CAAA;AAC7F;AACA,6FAA6F;AAC7F,8FAA8F;AAC9F,+DAA+D;AAC/D,IAAIqB;AACJ,8FAA8F;AAC9F,8FAA8F;AAC9F,MAAMC,cAAc;AACpB,6FAA6F;AAC7F,0FAA0F;AAC1F,MAAMC,UAAU;AAEhB,SAASC,UAAUtB,GAAW;IAC5BmB,8BAAAA,+BAAAA,oBAAAA,oBAAsB,IAAII,KAAKC,SAAS,CAACC,WAAW;QAAEC,aAAa;IAAW;IAC9E,OAAOV,MAAMC,IAAI,CAACE,kBAAkBQ,OAAO,CAAC3B,MAAM,CAAC4B,IAAMA,EAAED,OAAO;AACpE;AAEA,8FAA8F;AAC9F,0FAA0F;AAC1F,SAASE,mBAAmB7B,GAAW;IACrC,MAAM8B,SAAqB;QAAC,EAAE;KAAC;IAC/B,KAAK,MAAMC,KAAKT,UAAUtB,KAAM;QAC9B,IAAIoB,YAAY9B,IAAI,CAACyC,IAAID,OAAO3B,IAAI,CAAC,EAAE;aAClC2B,MAAM,CAACA,OAAOhC,MAAM,GAAG,EAAE,CAACK,IAAI,CAAC4B;IACtC;IACA,OAAOD,OAAOE,MAAM,CAAC,CAACD,IAAMA,EAAEjC,MAAM,GAAG;AACzC;AAEA,6FAA6F;AAC7F,kGAAkG;AAClG,OAAO,SAASmC,aAAa5C,IAAY;IACvC,IAAI,CAACF,QAAQG,IAAI,CAACD,OAAO,OAAO;IAChC,MAAM6C,MAAM7C,KAAK8C,OAAO,CAAClD,KAAK,CAACe;QAC7B,MAAMoC,OAAOP,mBAAmB7B,KAC7BqC,GAAG,CAAC,CAACN,IAAMA,EAAEO,IAAI,CAAC,MAClBA,IAAI,CAAC,CAAC,CAAC,EAAEjB,QAAQ,CAAC,CAAC;QACtB,OAAO,CAAC,CAAC,EAAEA,QAAQ,CAAC,EAAEe,KAAK,CAAC,EAAEf,QAAQ,CAAC,CAAC;IAC1C;IACA,OAAOa,IAAIC,OAAO,CAAC,QAAQ,KAAKI,IAAI;AACtC;AAEA,0FAA0F;AAC1F,wFAAwF;AACxF,MAAMC,YAAY;AAElB,4FAA4F;AAC5F,uGAAuG;AACvG,MAAMC,kBAAkB;AAExB,4FAA4F;AAC5F,yFAAyF;AACzF,SAASC,SAAS1C,GAAW,EAAE2C,YAAoB;IACjD,OAAOd,mBAAmB7B,KACvBqC,GAAG,CAAC,CAACN,IAAM,GAAGY,eAAeZ,EAAEjC,MAAM,GAAG,IAAI,CAAC,CAAC,EAAEiC,EAAEO,IAAI,CAAC,KAAK,CAAC,CAAC,GAAGP,CAAC,CAAC,EAAE,EAAE,EACvEO,IAAI,CAAC;AACV;AAEA,2FAA2F;AAC3F,wFAAwF;AACxF,OAAO,SAASM,aAAaC,KAAa;IACxC,IAAI,CAAC1D,QAAQG,IAAI,CAACuD,QAAQ,OAAOA;IACjC,IAAIX,MAAM;IACV,IAAIY,SAAS;IACb,MAAMC,SAASF,MAAMG,KAAK,CAAC/D,MAAM,2CAA2C;IAC5E,IAAK,IAAIyB,IAAI,GAAGA,IAAIqC,OAAOjD,MAAM,EAAEY,IAAK;QACtC,IAAIA,IAAI,MAAM,GAAG;YACf,KAAK,MAAMuC,MAAMF,MAAM,CAACrC,EAAE,CAAE,IAAIuC,OAAO,KAAKH,SAAS,CAACA;YACtDZ,OAAOa,MAAM,CAACrC,EAAE;YAChB;QACF;QACA,IAAIoC,QAAQ;YACVZ,OAAOa,MAAM,CAACrC,EAAE,EAAE,2CAA2C;YAC7D;QACF;QACA,MAAMQ,IAAIgB,IAAIzC,KAAK,CAAC+C;QACpB,IAAItB,GAAG;YACLgB,MAAM,GAAGA,IAAIjC,KAAK,CAAC,GAAGiB,EAAEtB,KAAK,IAAIsB,CAAC,CAAC,EAAE,GAAGwB,SAASK,MAAM,CAACrC,EAAE,EAAE,GAAGQ,CAAC,CAAC,EAAE,GAAGA,CAAC,CAAC,EAAE,CAAC,KAAK,CAAC,GAAG;QACtF,OAAO;YACLgB,OAAOQ,SAASK,MAAM,CAACrC,EAAE,EAAE+B;QAC7B;IACF;IACA,OAAOP;AACT;AAEA,kGAAkG;AAClG,2FAA2F;AAC3F,OAAO,SAASpB,cAAczB,IAAY;IACxC,OAAOA,KAAK6D,SAAS,CAAC,OAAOf,OAAO,CAAC,WAAW,IAAIgB,WAAW;AACjE"}
+{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/text/segment.ts"],"sourcesContent":["import { stemmer } from 'stemmer';\n\n// Word boundaries FTS5's tokenizers cannot find on their own. Contract: seg(query) must appear in\n// seg(document) for any substring query; word mode is context-dependent and breaks that, so grapheme mode (UAX #29) is used.\n\n// Scripts written without word spaces: a closed set of writing systems.\n// Script_Extensions, not Script: the katakana long-vowel mark ー is Script=Common.\nexport const UNSPACED_SCRIPTS = '\\\\p{scx=Han}\\\\p{scx=Hiragana}\\\\p{scx=Katakana}\\\\p{scx=Thai}\\\\p{scx=Khmer}\\\\p{scx=Lao}\\\\p{scx=Myanmar}';\n// A run is script BASE characters with their combining marks attached; a bare mark after a\n// Latin letter (decomposed é) never starts one.\nconst RUN_BODY = `(?:[${UNSPACED_SCRIPTS}]\\\\p{M}*)+`;\nconst RUN = new RegExp(`(${RUN_BODY})`, 'gu');\nconst HAS_RUN = new RegExp(`[${UNSPACED_SCRIPTS}]`, 'u');\n\n// Whether text holds a script that marks no word boundaries, the predicate every store's\n// sidecar population and query split turns on.\nexport function hasUnspacedRun(text: string): boolean {\n  return HAS_RUN.test(text);\n}\n\nexport interface SearchToken {\n  text: string;\n  start: number;\n  end: number;\n  unspaced: boolean;\n}\n\n// Search tokens share the native separator contract: letters, marks, and numbers form tokens;\n// underscore, apostrophe, and hyphen separate them. Unspaced-script runs stay whole so callers\n// can apply substring semantics, while adjacent Latin text still gets its own token.\nexport function searchTokens(text: string): SearchToken[] {\n  const tokens: SearchToken[] = [];\n  for (const match of text.matchAll(/[\\p{L}\\p{N}][\\p{L}\\p{M}\\p{N}]*/gu)) {\n    const start = match.index;\n    const end = start + match[0].length;\n    let cursor = start;\n    for (const run of text.slice(start, end).matchAll(RUN)) {\n      const runStart = start + run.index;\n      if (runStart > cursor) tokens.push({ text: text.slice(cursor, runStart), start: cursor, end: runStart, unspaced: false });\n      const runEnd = runStart + run[0].length;\n      tokens.push({ text: run[0], start: runStart, end: runEnd, unspaced: true });\n      cursor = runEnd;\n    }\n    if (cursor < end) tokens.push({ text: text.slice(cursor, end), start: cursor, end, unspaced: false });\n  }\n  return tokens;\n}\n\n// Phrase membership is checked after native ranking. A CJK token matches inside one authored\n// run; Latin tokens compare folded Porter stems. Each call scans one field once per candidate.\nexport function matchesSearchPhrase(text: string, phrase: SearchToken[]): boolean {\n  const haystack = searchTokens(text);\n  if (phrase.length === 0 || haystack.length < phrase.length) return false;\n  for (let start = 0; start <= haystack.length - phrase.length; start++) {\n    let matches = true;\n    for (let i = 0; i < phrase.length; i++) {\n      const query = phrase[i];\n      const authored = haystack[start + i];\n      if (query.unspaced) {\n        if (!authored.unspaced || !authored.text.includes(query.text)) matches = false;\n      } else if (authored.unspaced || stemmer(foldForSearch(authored.text)) !== stemmer(foldForSearch(query.text))) {\n        matches = false;\n      }\n      if (!matches) break;\n    }\n    if (matches) return true;\n  }\n  return false;\n}\n\n// Character spans of unspaced-script runs in text, the same boundary the index segments on --\n// the snippet marker uses this to keep substring semantics inside a run and word/stem matching outside it.\nexport function unspacedRuns(text: string): Array<{ start: number; end: number }> {\n  return Array.from(text.matchAll(RUN), (m) => ({ start: m.index, end: m.index + m[0].length }));\n}\n// Grapheme clusters, ECMA-402/UAX #29: base char plus its marks, ZWJ sequences, Hangul jamo.\n// Built on first use and kept: construction is 6.6 ms, and a tree with no unspaced-script run\n// never segments at all, so no command pays it at module load.\nlet graphemeSegmenter: Intl.Segmenter | undefined;\n// unicode61 drops Unicode punctuation as a separator. Some of it (。、「」) has Script_Extensions\n// into an unspaced script, so RUN keeps it -- a grapheme matching this becomes a split point.\nconst PUNCTUATION = /\\p{P}/u;\n// Token barrier between separate runs (or a punctuation split within one) so their graphemes\n// are never phrase-adjacent. U+A7F7: a letter (so FTS5 keeps it as a token) no one types.\nconst BARRIER = 'ꟷ';\n\nfunction graphemes(run: string): string[] {\n  graphemeSegmenter ??= new Intl.Segmenter(undefined, { granularity: 'grapheme' });\n  return Array.from(graphemeSegmenter.segment(run), (s) => s.segment);\n}\n\n// A run's graphemes, cut into punctuation-free groups at every punctuation grapheme (dropped,\n// matching unicode61). The one place punctuation is classified, so index and query agree.\nfunction splitOnPunctuation(run: string): string[][] {\n  const groups: string[][] = [[]];\n  for (const g of graphemes(run)) {\n    if (PUNCTUATION.test(g)) groups.push([]);\n    else groups[groups.length - 1].push(g);\n  }\n  return groups.filter((g) => g.length > 0);\n}\n\n// Index side: '' when the field has no unspaced-script run. Otherwise each run explodes into\n// its graphemes, barrier-delimited from its neighbors and from a punctuation split within itself.\nexport function segmentField(text: string): string {\n  if (!HAS_RUN.test(text)) return '';\n  const out = text.replace(RUN, (run) => {\n    const body = splitOnPunctuation(run)\n      .map((g) => g.join(' '))\n      .join(` ${BARRIER} `);\n    return ` ${BARRIER} ${body} ${BARRIER} `;\n  });\n  return out.replace(/\\s+/g, ' ').trim();\n}\n\n// A `title:`/`summary:`/`text:` qualifier directly before a run that is about to become a\n// quoted grapheme phrase, so the rewrite can retarget it at the matching `_seg` column.\nconst QUALIFIER = /(^|[\\s(])(-?)(title|summary|text)\\s*:\\s*$/;\n\n// Raw title/summary/text drop punctuation as unicode61's token separator, so `数数` and `数。数`\n// falsely match there; only the barriered `_seg` columns are safe from it, hence this fallback target.\nconst SIDECAR_COLUMNS = '{title_seg summary_seg text_seg}:';\nconst AUTHORED_COLUMNS = '{title summary text}:';\nconst QUOTED_RUN_TOKEN = new RegExp(`\"(${RUN_BODY})\"`, 'gu');\n\ninterface PositionalSpan {\n  start: number;\n  end: number;\n  qualified: boolean;\n}\n\nconst FTS_WORD = /[\\p{L}\\p{N}_]/u;\n\nfunction isFtsWord(char: string | undefined): boolean {\n  return char !== undefined && FTS_WORD.test(char);\n}\n\nfunction quoteEnd(text: string, start: number): number {\n  for (let i = start + 1; i < text.length; i++) {\n    if (text[i] !== '\"') continue;\n    if (text[i + 1] === '\"') {\n      i++;\n      continue;\n    }\n    return i + 1;\n  }\n  return text.length;\n}\n\nfunction nearEnd(text: string, open: number): number {\n  let depth = 1;\n  let quoted = false;\n  for (let i = open + 1; i < text.length; i++) {\n    const char = text[i];\n    if (char === '\"') {\n      if (quoted && text[i + 1] === '\"') {\n        i++;\n      } else {\n        quoted = !quoted;\n      }\n    } else if (!quoted && char === '(') {\n      depth++;\n    } else if (!quoted && char === ')' && --depth === 0) {\n      return i + 1;\n    }\n  }\n  return text.length;\n}\n\n// Lexical routing only: SQLite remains responsible for validating the expression. The scanner\n// recognizes NEAR groups and the one FTS string governed by ^, while quoted text shields both.\nfunction positionalSpans(terms: string): PositionalSpan[] {\n  const spans: PositionalSpan[] = [];\n  for (let i = 0; i < terms.length; i++) {\n    if (terms[i] === '\"') {\n      i = quoteEnd(terms, i) - 1;\n      continue;\n    }\n    if (terms.startsWith('NEAR', i) && !isFtsWord(terms[i - 1])) {\n      let open = i + 4;\n      while (/\\s/u.test(terms[open] ?? '')) open++;\n      if (terms[open] === '(') {\n        const end = nearEnd(terms, open);\n        if (HAS_RUN.test(terms.slice(open + 1, end))) spans.push({ start: i, end, qualified: QUALIFIER.test(terms.slice(0, i)) });\n        i = end - 1;\n        continue;\n      }\n    }\n    if (terms[i] === '^' && !isFtsWord(terms[i - 1])) {\n      let operand = i + 1;\n      while (/\\s/u.test(terms[operand] ?? '')) operand++;\n      const end =\n        terms[operand] === '\"'\n          ? quoteEnd(terms, operand)\n          : (() => {\n              let cursor = operand;\n              while (/[\\p{L}\\p{N}_]/u.test(terms[cursor] ?? '')) cursor++;\n              return cursor;\n            })();\n      if (end > operand && HAS_RUN.test(terms.slice(operand, end))) spans.push({ start: i, end, qualified: QUALIFIER.test(terms.slice(0, i)) });\n      i = Math.max(i, end - 1);\n    }\n  }\n  return spans;\n}\n\nfunction routePositionalSpans(terms: string): string {\n  const spans = positionalSpans(terms);\n  if (spans.length === 0) return terms;\n  let out = '';\n  let cursor = 0;\n  for (const span of spans) {\n    out += terms.slice(cursor, span.start);\n    if (!span.qualified) out += AUTHORED_COLUMNS;\n    out += terms.slice(span.start, span.end);\n    cursor = span.end;\n  }\n  return out + terms.slice(cursor);\n}\n\nfunction inPositionalSpan(start: number, spans: PositionalSpan[]): boolean {\n  return spans.some((span) => start >= span.start && start < span.end);\n}\n\n// A run's punctuation-free groups, each its own quoted phrase (bare token if one grapheme),\n// space-joined -- the same split points segmentField barriers, so query and index agree.\nfunction runQuery(run: string, columnPrefix: string, quoteSingle = false): string {\n  return splitOnPunctuation(run)\n    .map((g) => `${columnPrefix}${g.length > 1 || quoteSingle ? `\"${g.join(' ')}\"` : g[0]}`)\n    .join(' ');\n}\n\n// Retarget complete quoted run tokens without interpreting the expression around them. Positional\n// spans have already been routed to authored columns, so sidecar barriers cannot change meaning.\nfunction quotedRunTokens(terms: string): string {\n  const spans = positionalSpans(terms);\n  let out = '';\n  let cursor = 0;\n  for (const match of terms.matchAll(QUOTED_RUN_TOKEN)) {\n    const start = match.index;\n    const run = match[1];\n    if (inPositionalSpan(start, spans) || PUNCTUATION.test(run)) continue;\n    out += terms.slice(cursor, start);\n    const qualifier = out.match(QUALIFIER);\n    if (qualifier) {\n      out = `${out.slice(0, qualifier.index)}${qualifier[1]}${runQuery(run, `${qualifier[2]}${qualifier[3]}_seg:`, true)}`;\n    } else {\n      out += runQuery(run, SIDECAR_COLUMNS, true);\n    }\n    cursor = start + match[0].length;\n  }\n  return cursor === 0 ? terms : out + terms.slice(cursor);\n}\n\n// Query side: each unspaced run becomes phrases of its graphemes, matching segmentField. A\n// qualifier maps to its `_seg` column; unqualified maps to all three (SIDECAR_COLUMNS).\nexport function segmentMatch(terms: string): string {\n  if (!HAS_RUN.test(terms)) return terms;\n  terms = routePositionalSpans(terms);\n  terms = quotedRunTokens(terms);\n  const spans = positionalSpans(terms);\n  let out = '';\n  let quoted = false;\n  let offset = 0;\n  const pieces = terms.split(RUN); // split keeps captured runs at odd indices\n  for (let i = 0; i < pieces.length; i++) {\n    if (i % 2 === 0) {\n      for (const ch of pieces[i]) if (ch === '\"') quoted = !quoted;\n      out += pieces[i];\n      offset += pieces[i].length;\n      continue;\n    }\n    const protectedRun = inPositionalSpan(offset, spans);\n    if (quoted || protectedRun) {\n      out += pieces[i]; // an author's phrase is matched as written\n      offset += pieces[i].length;\n      continue;\n    }\n    const m = out.match(QUALIFIER);\n    if (m) {\n      out = `${out.slice(0, m.index)}${m[1]}${runQuery(pieces[i], `${m[2]}${m[3]}_seg:`)}`;\n    } else {\n      out += runQuery(pieces[i], SIDECAR_COLUMNS);\n    }\n    offset += pieces[i].length;\n  }\n  return out;\n}\n\n// FTS tokenizers compare letters without combining marks. Keep the authored string for output and\n// offsets, but use this key when JS needs to compare a matched word to its source segment.\nexport function foldForSearch(text: string): string {\n  return text.normalize('NFD').replace(/\\p{M}/gu, '').toLowerCase();\n}\n"],"names":["stemmer","UNSPACED_SCRIPTS","RUN_BODY","RUN","RegExp","HAS_RUN","hasUnspacedRun","text","test","searchTokens","tokens","match","matchAll","start","index","end","length","cursor","run","slice","runStart","push","unspaced","runEnd","matchesSearchPhrase","phrase","haystack","matches","i","query","authored","includes","foldForSearch","unspacedRuns","Array","from","m","graphemeSegmenter","PUNCTUATION","BARRIER","graphemes","Intl","Segmenter","undefined","granularity","segment","s","splitOnPunctuation","groups","g","filter","segmentField","out","replace","body","map","join","trim","QUALIFIER","SIDECAR_COLUMNS","AUTHORED_COLUMNS","QUOTED_RUN_TOKEN","FTS_WORD","isFtsWord","char","quoteEnd","nearEnd","open","depth","quoted","positionalSpans","terms","spans","startsWith","qualified","operand","Math","max","routePositionalSpans","span","inPositionalSpan","some","runQuery","columnPrefix","quoteSingle","quotedRunTokens","qualifier","segmentMatch","offset","pieces","split","ch","protectedRun","normalize","toLowerCase"],"mappings":"AAAA,SAASA,OAAO,QAAQ,UAAU;AAElC,kGAAkG;AAClG,6HAA6H;AAE7H,wEAAwE;AACxE,kFAAkF;AAClF,OAAO,MAAMC,mBAAmB,wGAAwG;AACxI,2FAA2F;AAC3F,gDAAgD;AAChD,MAAMC,WAAW,CAAC,IAAI,EAAED,iBAAiB,UAAU,CAAC;AACpD,MAAME,MAAM,IAAIC,OAAO,CAAC,CAAC,EAAEF,SAAS,CAAC,CAAC,EAAE;AACxC,MAAMG,UAAU,IAAID,OAAO,CAAC,CAAC,EAAEH,iBAAiB,CAAC,CAAC,EAAE;AAEpD,yFAAyF;AACzF,+CAA+C;AAC/C,OAAO,SAASK,eAAeC,IAAY;IACzC,OAAOF,QAAQG,IAAI,CAACD;AACtB;AASA,8FAA8F;AAC9F,+FAA+F;AAC/F,qFAAqF;AACrF,OAAO,SAASE,aAAaF,IAAY;IACvC,MAAMG,SAAwB,EAAE;IAChC,KAAK,MAAMC,SAASJ,KAAKK,QAAQ,CAAC,oCAAqC;QACrE,MAAMC,QAAQF,MAAMG,KAAK;QACzB,MAAMC,MAAMF,QAAQF,KAAK,CAAC,EAAE,CAACK,MAAM;QACnC,IAAIC,SAASJ;QACb,KAAK,MAAMK,OAAOX,KAAKY,KAAK,CAACN,OAAOE,KAAKH,QAAQ,CAACT,KAAM;YACtD,MAAMiB,WAAWP,QAAQK,IAAIJ,KAAK;YAClC,IAAIM,WAAWH,QAAQP,OAAOW,IAAI,CAAC;gBAAEd,MAAMA,KAAKY,KAAK,CAACF,QAAQG;gBAAWP,OAAOI;gBAAQF,KAAKK;gBAAUE,UAAU;YAAM;YACvH,MAAMC,SAASH,WAAWF,GAAG,CAAC,EAAE,CAACF,MAAM;YACvCN,OAAOW,IAAI,CAAC;gBAAEd,MAAMW,GAAG,CAAC,EAAE;gBAAEL,OAAOO;gBAAUL,KAAKQ;gBAAQD,UAAU;YAAK;YACzEL,SAASM;QACX;QACA,IAAIN,SAASF,KAAKL,OAAOW,IAAI,CAAC;YAAEd,MAAMA,KAAKY,KAAK,CAACF,QAAQF;YAAMF,OAAOI;YAAQF;YAAKO,UAAU;QAAM;IACrG;IACA,OAAOZ;AACT;AAEA,6FAA6F;AAC7F,+FAA+F;AAC/F,OAAO,SAASc,oBAAoBjB,IAAY,EAAEkB,MAAqB;IACrE,MAAMC,WAAWjB,aAAaF;IAC9B,IAAIkB,OAAOT,MAAM,KAAK,KAAKU,SAASV,MAAM,GAAGS,OAAOT,MAAM,EAAE,OAAO;IACnE,IAAK,IAAIH,QAAQ,GAAGA,SAASa,SAASV,MAAM,GAAGS,OAAOT,MAAM,EAAEH,QAAS;QACrE,IAAIc,UAAU;QACd,IAAK,IAAIC,IAAI,GAAGA,IAAIH,OAAOT,MAAM,EAAEY,IAAK;YACtC,MAAMC,QAAQJ,MAAM,CAACG,EAAE;YACvB,MAAME,WAAWJ,QAAQ,CAACb,QAAQe,EAAE;YACpC,IAAIC,MAAMP,QAAQ,EAAE;gBAClB,IAAI,CAACQ,SAASR,QAAQ,IAAI,CAACQ,SAASvB,IAAI,CAACwB,QAAQ,CAACF,MAAMtB,IAAI,GAAGoB,UAAU;YAC3E,OAAO,IAAIG,SAASR,QAAQ,IAAItB,QAAQgC,cAAcF,SAASvB,IAAI,OAAOP,QAAQgC,cAAcH,MAAMtB,IAAI,IAAI;gBAC5GoB,UAAU;YACZ;YACA,IAAI,CAACA,SAAS;QAChB;QACA,IAAIA,SAAS,OAAO;IACtB;IACA,OAAO;AACT;AAEA,8FAA8F;AAC9F,2GAA2G;AAC3G,OAAO,SAASM,aAAa1B,IAAY;IACvC,OAAO2B,MAAMC,IAAI,CAAC5B,KAAKK,QAAQ,CAACT,MAAM,CAACiC,IAAO,CAAA;YAAEvB,OAAOuB,EAAEtB,KAAK;YAAEC,KAAKqB,EAAEtB,KAAK,GAAGsB,CAAC,CAAC,EAAE,CAACpB,MAAM;QAAC,CAAA;AAC7F;AACA,6FAA6F;AAC7F,8FAA8F;AAC9F,+DAA+D;AAC/D,IAAIqB;AACJ,8FAA8F;AAC9F,8FAA8F;AAC9F,MAAMC,cAAc;AACpB,6FAA6F;AAC7F,0FAA0F;AAC1F,MAAMC,UAAU;AAEhB,SAASC,UAAUtB,GAAW;IAC5BmB,8BAAAA,+BAAAA,oBAAAA,oBAAsB,IAAII,KAAKC,SAAS,CAACC,WAAW;QAAEC,aAAa;IAAW;IAC9E,OAAOV,MAAMC,IAAI,CAACE,kBAAkBQ,OAAO,CAAC3B,MAAM,CAAC4B,IAAMA,EAAED,OAAO;AACpE;AAEA,8FAA8F;AAC9F,0FAA0F;AAC1F,SAASE,mBAAmB7B,GAAW;IACrC,MAAM8B,SAAqB;QAAC,EAAE;KAAC;IAC/B,KAAK,MAAMC,KAAKT,UAAUtB,KAAM;QAC9B,IAAIoB,YAAY9B,IAAI,CAACyC,IAAID,OAAO3B,IAAI,CAAC,EAAE;aAClC2B,MAAM,CAACA,OAAOhC,MAAM,GAAG,EAAE,CAACK,IAAI,CAAC4B;IACtC;IACA,OAAOD,OAAOE,MAAM,CAAC,CAACD,IAAMA,EAAEjC,MAAM,GAAG;AACzC;AAEA,6FAA6F;AAC7F,kGAAkG;AAClG,OAAO,SAASmC,aAAa5C,IAAY;IACvC,IAAI,CAACF,QAAQG,IAAI,CAACD,OAAO,OAAO;IAChC,MAAM6C,MAAM7C,KAAK8C,OAAO,CAAClD,KAAK,CAACe;QAC7B,MAAMoC,OAAOP,mBAAmB7B,KAC7BqC,GAAG,CAAC,CAACN,IAAMA,EAAEO,IAAI,CAAC,MAClBA,IAAI,CAAC,CAAC,CAAC,EAAEjB,QAAQ,CAAC,CAAC;QACtB,OAAO,CAAC,CAAC,EAAEA,QAAQ,CAAC,EAAEe,KAAK,CAAC,EAAEf,QAAQ,CAAC,CAAC;IAC1C;IACA,OAAOa,IAAIC,OAAO,CAAC,QAAQ,KAAKI,IAAI;AACtC;AAEA,0FAA0F;AAC1F,wFAAwF;AACxF,MAAMC,YAAY;AAElB,4FAA4F;AAC5F,uGAAuG;AACvG,MAAMC,kBAAkB;AACxB,MAAMC,mBAAmB;AACzB,MAAMC,mBAAmB,IAAIzD,OAAO,CAAC,EAAE,EAAEF,SAAS,EAAE,CAAC,EAAE;AAQvD,MAAM4D,WAAW;AAEjB,SAASC,UAAUC,IAAwB;IACzC,OAAOA,SAASrB,aAAamB,SAAStD,IAAI,CAACwD;AAC7C;AAEA,SAASC,SAAS1D,IAAY,EAAEM,KAAa;IAC3C,IAAK,IAAIe,IAAIf,QAAQ,GAAGe,IAAIrB,KAAKS,MAAM,EAAEY,IAAK;QAC5C,IAAIrB,IAAI,CAACqB,EAAE,KAAK,KAAK;QACrB,IAAIrB,IAAI,CAACqB,IAAI,EAAE,KAAK,KAAK;YACvBA;YACA;QACF;QACA,OAAOA,IAAI;IACb;IACA,OAAOrB,KAAKS,MAAM;AACpB;AAEA,SAASkD,QAAQ3D,IAAY,EAAE4D,IAAY;IACzC,IAAIC,QAAQ;IACZ,IAAIC,SAAS;IACb,IAAK,IAAIzC,IAAIuC,OAAO,GAAGvC,IAAIrB,KAAKS,MAAM,EAAEY,IAAK;QAC3C,MAAMoC,OAAOzD,IAAI,CAACqB,EAAE;QACpB,IAAIoC,SAAS,KAAK;YAChB,IAAIK,UAAU9D,IAAI,CAACqB,IAAI,EAAE,KAAK,KAAK;gBACjCA;YACF,OAAO;gBACLyC,SAAS,CAACA;YACZ;QACF,OAAO,IAAI,CAACA,UAAUL,SAAS,KAAK;YAClCI;QACF,OAAO,IAAI,CAACC,UAAUL,SAAS,OAAO,EAAEI,UAAU,GAAG;YACnD,OAAOxC,IAAI;QACb;IACF;IACA,OAAOrB,KAAKS,MAAM;AACpB;AAEA,8FAA8F;AAC9F,+FAA+F;AAC/F,SAASsD,gBAAgBC,KAAa;IACpC,MAAMC,QAA0B,EAAE;IAClC,IAAK,IAAI5C,IAAI,GAAGA,IAAI2C,MAAMvD,MAAM,EAAEY,IAAK;QACrC,IAAI2C,KAAK,CAAC3C,EAAE,KAAK,KAAK;YACpBA,IAAIqC,SAASM,OAAO3C,KAAK;YACzB;QACF;QACA,IAAI2C,MAAME,UAAU,CAAC,QAAQ7C,MAAM,CAACmC,UAAUQ,KAAK,CAAC3C,IAAI,EAAE,GAAG;gBAEzC2C;YADlB,IAAIJ,OAAOvC,IAAI;YACf,MAAO,MAAMpB,IAAI,EAAC+D,cAAAA,KAAK,CAACJ,KAAK,cAAXI,yBAAAA,cAAe,IAAKJ;YACtC,IAAII,KAAK,CAACJ,KAAK,KAAK,KAAK;gBACvB,MAAMpD,MAAMmD,QAAQK,OAAOJ;gBAC3B,IAAI9D,QAAQG,IAAI,CAAC+D,MAAMpD,KAAK,CAACgD,OAAO,GAAGpD,OAAOyD,MAAMnD,IAAI,CAAC;oBAAER,OAAOe;oBAAGb;oBAAK2D,WAAWhB,UAAUlD,IAAI,CAAC+D,MAAMpD,KAAK,CAAC,GAAGS;gBAAI;gBACvHA,IAAIb,MAAM;gBACV;YACF;QACF;QACA,IAAIwD,KAAK,CAAC3C,EAAE,KAAK,OAAO,CAACmC,UAAUQ,KAAK,CAAC3C,IAAI,EAAE,GAAG;gBAE9B2C;YADlB,IAAII,UAAU/C,IAAI;YAClB,MAAO,MAAMpB,IAAI,EAAC+D,iBAAAA,KAAK,CAACI,QAAQ,cAAdJ,4BAAAA,iBAAkB,IAAKI;YACzC,MAAM5D,MACJwD,KAAK,CAACI,QAAQ,KAAK,MACfV,SAASM,OAAOI,WAChB,AAAC,CAAA;oBAE8BJ;gBAD7B,IAAItD,SAAS0D;gBACb,MAAO,iBAAiBnE,IAAI,EAAC+D,gBAAAA,KAAK,CAACtD,OAAO,cAAbsD,2BAAAA,gBAAiB,IAAKtD;gBACnD,OAAOA;YACT,CAAA;YACN,IAAIF,MAAM4D,WAAWtE,QAAQG,IAAI,CAAC+D,MAAMpD,KAAK,CAACwD,SAAS5D,OAAOyD,MAAMnD,IAAI,CAAC;gBAAER,OAAOe;gBAAGb;gBAAK2D,WAAWhB,UAAUlD,IAAI,CAAC+D,MAAMpD,KAAK,CAAC,GAAGS;YAAI;YACvIA,IAAIgD,KAAKC,GAAG,CAACjD,GAAGb,MAAM;QACxB;IACF;IACA,OAAOyD;AACT;AAEA,SAASM,qBAAqBP,KAAa;IACzC,MAAMC,QAAQF,gBAAgBC;IAC9B,IAAIC,MAAMxD,MAAM,KAAK,GAAG,OAAOuD;IAC/B,IAAInB,MAAM;IACV,IAAInC,SAAS;IACb,KAAK,MAAM8D,QAAQP,MAAO;QACxBpB,OAAOmB,MAAMpD,KAAK,CAACF,QAAQ8D,KAAKlE,KAAK;QACrC,IAAI,CAACkE,KAAKL,SAAS,EAAEtB,OAAOQ;QAC5BR,OAAOmB,MAAMpD,KAAK,CAAC4D,KAAKlE,KAAK,EAAEkE,KAAKhE,GAAG;QACvCE,SAAS8D,KAAKhE,GAAG;IACnB;IACA,OAAOqC,MAAMmB,MAAMpD,KAAK,CAACF;AAC3B;AAEA,SAAS+D,iBAAiBnE,KAAa,EAAE2D,KAAuB;IAC9D,OAAOA,MAAMS,IAAI,CAAC,CAACF,OAASlE,SAASkE,KAAKlE,KAAK,IAAIA,QAAQkE,KAAKhE,GAAG;AACrE;AAEA,4FAA4F;AAC5F,yFAAyF;AACzF,SAASmE,SAAShE,GAAW,EAAEiE,YAAoB,EAAEC,cAAc,KAAK;IACtE,OAAOrC,mBAAmB7B,KACvBqC,GAAG,CAAC,CAACN,IAAM,GAAGkC,eAAelC,EAAEjC,MAAM,GAAG,KAAKoE,cAAc,CAAC,CAAC,EAAEnC,EAAEO,IAAI,CAAC,KAAK,CAAC,CAAC,GAAGP,CAAC,CAAC,EAAE,EAAE,EACtFO,IAAI,CAAC;AACV;AAEA,kGAAkG;AAClG,iGAAiG;AACjG,SAAS6B,gBAAgBd,KAAa;IACpC,MAAMC,QAAQF,gBAAgBC;IAC9B,IAAInB,MAAM;IACV,IAAInC,SAAS;IACb,KAAK,MAAMN,SAAS4D,MAAM3D,QAAQ,CAACiD,kBAAmB;QACpD,MAAMhD,QAAQF,MAAMG,KAAK;QACzB,MAAMI,MAAMP,KAAK,CAAC,EAAE;QACpB,IAAIqE,iBAAiBnE,OAAO2D,UAAUlC,YAAY9B,IAAI,CAACU,MAAM;QAC7DkC,OAAOmB,MAAMpD,KAAK,CAACF,QAAQJ;QAC3B,MAAMyE,YAAYlC,IAAIzC,KAAK,CAAC+C;QAC5B,IAAI4B,WAAW;YACblC,MAAM,GAAGA,IAAIjC,KAAK,CAAC,GAAGmE,UAAUxE,KAAK,IAAIwE,SAAS,CAAC,EAAE,GAAGJ,SAAShE,KAAK,GAAGoE,SAAS,CAAC,EAAE,GAAGA,SAAS,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,OAAO;QACtH,OAAO;YACLlC,OAAO8B,SAAShE,KAAKyC,iBAAiB;QACxC;QACA1C,SAASJ,QAAQF,KAAK,CAAC,EAAE,CAACK,MAAM;IAClC;IACA,OAAOC,WAAW,IAAIsD,QAAQnB,MAAMmB,MAAMpD,KAAK,CAACF;AAClD;AAEA,2FAA2F;AAC3F,wFAAwF;AACxF,OAAO,SAASsE,aAAahB,KAAa;IACxC,IAAI,CAAClE,QAAQG,IAAI,CAAC+D,QAAQ,OAAOA;IACjCA,QAAQO,qBAAqBP;IAC7BA,QAAQc,gBAAgBd;IACxB,MAAMC,QAAQF,gBAAgBC;IAC9B,IAAInB,MAAM;IACV,IAAIiB,SAAS;IACb,IAAImB,SAAS;IACb,MAAMC,SAASlB,MAAMmB,KAAK,CAACvF,MAAM,2CAA2C;IAC5E,IAAK,IAAIyB,IAAI,GAAGA,IAAI6D,OAAOzE,MAAM,EAAEY,IAAK;QACtC,IAAIA,IAAI,MAAM,GAAG;YACf,KAAK,MAAM+D,MAAMF,MAAM,CAAC7D,EAAE,CAAE,IAAI+D,OAAO,KAAKtB,SAAS,CAACA;YACtDjB,OAAOqC,MAAM,CAAC7D,EAAE;YAChB4D,UAAUC,MAAM,CAAC7D,EAAE,CAACZ,MAAM;YAC1B;QACF;QACA,MAAM4E,eAAeZ,iBAAiBQ,QAAQhB;QAC9C,IAAIH,UAAUuB,cAAc;YAC1BxC,OAAOqC,MAAM,CAAC7D,EAAE,EAAE,2CAA2C;YAC7D4D,UAAUC,MAAM,CAAC7D,EAAE,CAACZ,MAAM;YAC1B;QACF;QACA,MAAMoB,IAAIgB,IAAIzC,KAAK,CAAC+C;QACpB,IAAItB,GAAG;YACLgB,MAAM,GAAGA,IAAIjC,KAAK,CAAC,GAAGiB,EAAEtB,KAAK,IAAIsB,CAAC,CAAC,EAAE,GAAG8C,SAASO,MAAM,CAAC7D,EAAE,EAAE,GAAGQ,CAAC,CAAC,EAAE,GAAGA,CAAC,CAAC,EAAE,CAAC,KAAK,CAAC,GAAG;QACtF,OAAO;YACLgB,OAAO8B,SAASO,MAAM,CAAC7D,EAAE,EAAE+B;QAC7B;QACA6B,UAAUC,MAAM,CAAC7D,EAAE,CAACZ,MAAM;IAC5B;IACA,OAAOoC;AACT;AAEA,kGAAkG;AAClG,2FAA2F;AAC3F,OAAO,SAASpB,cAAczB,IAAY;IACxC,OAAOA,KAAKsF,SAAS,CAAC,OAAOxC,OAAO,CAAC,WAAW,IAAIyC,WAAW;AACjE"}

package/dist/esm/watch-claim.d.ts

@@ -0,0 +1,39 @@
+import { type WorkerErrorPayload } from './scan/worker-error.js';
+export declare const WATCH_CLAIM_FILENAME = ".sense-watch.db";
+export declare const WATCH_HEARTBEAT_INTERVAL_MS = 5000;
+export declare const WATCH_STALE_HEARTBEAT_MS = 15000;
+export interface WatchClaimRecord {
+    token: string;
+    pid: number;
+    heartbeatMs: number;
+}
+export interface WatchClaimWorkerData {
+    configDir: string;
+    force: boolean;
+    heartbeatIntervalMs: number;
+    pid: number;
+    token: string;
+}
+export type WatchClaimWorkerMessage = {
+    type: 'acquired';
+} | {
+    type: 'stopped';
+} | {
+    type: 'failure';
+    error: WorkerErrorPayload;
+};
+export interface WatchClaimController {
+    close(): Promise<void>;
+}
+export declare class WatchClaimDatabase {
+    private readonly db;
+    private readonly conn;
+    constructor(configDir: string);
+    acquire(token: string, pid: number, force: boolean): Promise<void>;
+    renew(token: string): boolean;
+    release(token: string): boolean;
+    read(): WatchClaimRecord | null;
+    close(): void;
+}
+export declare function readWatchClaim(configDir: string): WatchClaimRecord | null;
+export declare function startWatchClaim(configDir: string, force: boolean, heartbeatIntervalMs: number, onFailure: (err: Error) => void): Promise<WatchClaimController>;

package/dist/esm/watch-claim.js

@@ -0,0 +1,176 @@
+import { randomUUID } from 'node:crypto';
+import { existsSync } from 'node:fs';
+import { join } from 'node:path';
+import { DatabaseSync } from 'node:sqlite';
+import { Worker } from 'node:worker_threads';
+import { SenseError } from './errors.js';
+import { resolveWorkerFile } from './lib/worker-file.js';
+import { reviveError } from './scan/worker-error.js';
+import { createConnection } from './store/sqlite/connection.js';
+import { BEGIN_WRITE, withTransaction } from './store/transaction.js';
+export const WATCH_CLAIM_FILENAME = '.sense-watch.db';
+export const WATCH_HEARTBEAT_INTERVAL_MS = 5000;
+export const WATCH_STALE_HEARTBEAT_MS = 15000;
+const CLAIM_BUSY_TIMEOUT_MS = 30000;
+function claimPath(configDir) {
+    return join(configDir, WATCH_CLAIM_FILENAME);
+}
+function changedOne(changes) {
+    return changes === 1 || changes === BigInt(1);
+}
+// This database is separate from the selected search store so its heartbeat remains writable while
+// a DuckDB, Turso, or SQLite reconciliation owns the cache connection or blocks the main thread.
+export class WatchClaimDatabase {
+    async acquire(token, pid, force) {
+        await withTransaction(this.conn, async ()=>{
+            const current = this.db.prepare('SELECT token, pid, heartbeat_ms FROM watch_claim WHERE singleton = 1').get();
+            if (current && !force) {
+                const age = Date.now() - current.heartbeat_ms;
+                if (age >= 0 && age < WATCH_STALE_HEARTBEAT_MS) throw new SenseError('WATCH_ACTIVE', `another watcher appears active (heartbeat ${Math.round(age / 1000)}s ago); use --force to override`);
+            }
+            this.db.prepare('INSERT INTO watch_claim (singleton, token, pid, heartbeat_ms) VALUES (1, ?, ?, ?) ON CONFLICT(singleton) DO UPDATE SET token = excluded.token, pid = excluded.pid, heartbeat_ms = excluded.heartbeat_ms').run(token, pid, Date.now());
+        }, BEGIN_WRITE);
+    }
+    renew(token) {
+        return changedOne(this.db.prepare('UPDATE watch_claim SET heartbeat_ms = ? WHERE singleton = 1 AND token = ?').run(Date.now(), token).changes);
+    }
+    release(token) {
+        return changedOne(this.db.prepare('DELETE FROM watch_claim WHERE singleton = 1 AND token = ?').run(token).changes);
+    }
+    read() {
+        const row = this.db.prepare('SELECT token, pid, heartbeat_ms FROM watch_claim WHERE singleton = 1').get();
+        return row ? {
+            token: row.token,
+            pid: row.pid,
+            heartbeatMs: row.heartbeat_ms
+        } : null;
+    }
+    close() {
+        this.db.close();
+    }
+    constructor(configDir){
+        const db = new DatabaseSync(claimPath(configDir));
+        try {
+            db.exec(`PRAGMA busy_timeout = ${CLAIM_BUSY_TIMEOUT_MS}`);
+            db.exec('CREATE TABLE IF NOT EXISTS watch_claim (singleton INTEGER PRIMARY KEY CHECK (singleton = 1), token TEXT NOT NULL, pid INTEGER NOT NULL, heartbeat_ms INTEGER NOT NULL)');
+        } catch (err) {
+            try {
+                db.close();
+            } catch (closeError) {
+                throw new AggregateError([
+                    err,
+                    closeError
+                ], 'watch claim setup and database cleanup both failed');
+            }
+            throw err;
+        }
+        this.db = db;
+        this.conn = createConnection(db);
+    }
+}
+// Status opens the coordinator read-only and therefore cannot create either the file or its schema.
+export function readWatchClaim(configDir) {
+    const path = claimPath(configDir);
+    if (!existsSync(path)) return null;
+    let db;
+    try {
+        db = new DatabaseSync(path, {
+            readOnly: true
+        });
+    } catch (err) {
+        if (!existsSync(path)) return null;
+        throw err;
+    }
+    try {
+        db.exec(`PRAGMA busy_timeout = ${CLAIM_BUSY_TIMEOUT_MS}`);
+        const row = db.prepare('SELECT token, pid, heartbeat_ms FROM watch_claim WHERE singleton = 1').get();
+        return row ? {
+            token: row.token,
+            pid: row.pid,
+            heartbeatMs: row.heartbeat_ms
+        } : null;
+    } catch (err) {
+        if (/no such table: watch_claim/.test(err.message)) return null;
+        throw err;
+    } finally{
+        db.close();
+    }
+}
+export async function startWatchClaim(configDir, force, heartbeatIntervalMs, onFailure) {
+    const workerData = {
+        configDir,
+        force,
+        heartbeatIntervalMs,
+        pid: process.pid,
+        token: randomUUID()
+    };
+    const worker = new Worker(resolveWorkerFile('watch-heartbeat'), {
+        workerData
+    });
+    let acquired = false;
+    let closing = false;
+    let exited = false;
+    let failureReported = false;
+    let closeError = null;
+    let closePromise;
+    let resolveExit;
+    const exit = new Promise((resolve)=>{
+        resolveExit = resolve;
+    });
+    let resolveStart;
+    let rejectStart;
+    const start = new Promise((resolve, reject)=>{
+        resolveStart = resolve;
+        rejectStart = reject;
+    });
+    const fail = (err)=>{
+        if (failureReported) return;
+        failureReported = true;
+        if (!acquired) rejectStart(err);
+        else if (closing) closeError !== null && closeError !== void 0 ? closeError : closeError = err;
+        else onFailure(err);
+    };
+    worker.on('message', (message)=>{
+        if (message.type === 'acquired') {
+            acquired = true;
+            resolveStart();
+        } else if (message.type === 'failure') {
+            fail(reviveError(message.error));
+        }
+    });
+    worker.on('error', fail);
+    worker.on('exit', (code)=>{
+        exited = true;
+        resolveExit();
+        if (closing && code !== 0) closeError !== null && closeError !== void 0 ? closeError : closeError = new Error(`watch heartbeat worker stopped with exit code ${code}`);
+        else if (!closing && !failureReported) fail(new Error(`watch heartbeat worker stopped unexpectedly with exit code ${code}`));
+    });
+    try {
+        await start;
+    } catch (err) {
+        closing = true;
+        if (!exited) await worker.terminate();
+        await exit;
+        throw err;
+    }
+    return {
+        close () {
+            closePromise !== null && closePromise !== void 0 ? closePromise : closePromise = (async ()=>{
+                closing = true;
+                if (!exited) {
+                    try {
+                        worker.postMessage({
+                            type: 'stop'
+                        });
+                    } catch (err) {
+                        closeError !== null && closeError !== void 0 ? closeError : closeError = err instanceof Error ? err : new Error(String(err));
+                        await worker.terminate();
+                    }
+                }
+                await exit;
+                if (closeError) throw closeError;
+            })();
+            return closePromise;
+        }
+    };
+}

package/dist/esm/watch-claim.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/watch-claim.ts"],"sourcesContent":["import { randomUUID } from 'node:crypto';\nimport { existsSync } from 'node:fs';\nimport { join } from 'node:path';\nimport { DatabaseSync } from 'node:sqlite';\nimport { Worker } from 'node:worker_threads';\nimport { SenseError } from './errors.ts';\nimport { resolveWorkerFile } from './lib/worker-file.ts';\nimport { reviveError, type WorkerErrorPayload } from './scan/worker-error.ts';\nimport { createConnection } from './store/sqlite/connection.ts';\nimport { BEGIN_WRITE, withTransaction } from './store/transaction.ts';\n\nexport const WATCH_CLAIM_FILENAME = '.sense-watch.db';\nexport const WATCH_HEARTBEAT_INTERVAL_MS = 5_000;\nexport const WATCH_STALE_HEARTBEAT_MS = 15_000;\nconst CLAIM_BUSY_TIMEOUT_MS = 30_000;\n\nexport interface WatchClaimRecord {\n  token: string;\n  pid: number;\n  heartbeatMs: number;\n}\n\ninterface ClaimRow {\n  token: string;\n  pid: number;\n  heartbeat_ms: number;\n}\n\nexport interface WatchClaimWorkerData {\n  configDir: string;\n  force: boolean;\n  heartbeatIntervalMs: number;\n  pid: number;\n  token: string;\n}\n\nexport type WatchClaimWorkerMessage = { type: 'acquired' } | { type: 'stopped' } | { type: 'failure'; error: WorkerErrorPayload };\n\nexport interface WatchClaimController {\n  close(): Promise<void>;\n}\n\nfunction claimPath(configDir: string): string {\n  return join(configDir, WATCH_CLAIM_FILENAME);\n}\n\nfunction changedOne(changes: number | bigint): boolean {\n  return changes === 1 || changes === BigInt(1);\n}\n\n// This database is separate from the selected search store so its heartbeat remains writable while\n// a DuckDB, Turso, or SQLite reconciliation owns the cache connection or blocks the main thread.\nexport class WatchClaimDatabase {\n  private readonly db: DatabaseSync;\n  private readonly conn: ReturnType<typeof createConnection>;\n\n  constructor(configDir: string) {\n    const db = new DatabaseSync(claimPath(configDir));\n    try {\n      db.exec(`PRAGMA busy_timeout = ${CLAIM_BUSY_TIMEOUT_MS}`);\n      db.exec('CREATE TABLE IF NOT EXISTS watch_claim (singleton INTEGER PRIMARY KEY CHECK (singleton = 1), token TEXT NOT NULL, pid INTEGER NOT NULL, heartbeat_ms INTEGER NOT NULL)');\n    } catch (err) {\n      try {\n        db.close();\n      } catch (closeError) {\n        throw new AggregateError([err, closeError], 'watch claim setup and database cleanup both failed');\n      }\n      throw err;\n    }\n    this.db = db;\n    this.conn = createConnection(db);\n  }\n\n  async acquire(token: string, pid: number, force: boolean): Promise<void> {\n    await withTransaction(\n      this.conn,\n      async () => {\n        const current = this.db.prepare('SELECT token, pid, heartbeat_ms FROM watch_claim WHERE singleton = 1').get() as ClaimRow | undefined;\n        if (current && !force) {\n          const age = Date.now() - current.heartbeat_ms;\n          if (age >= 0 && age < WATCH_STALE_HEARTBEAT_MS) throw new SenseError('WATCH_ACTIVE', `another watcher appears active (heartbeat ${Math.round(age / 1000)}s ago); use --force to override`);\n        }\n        this.db.prepare('INSERT INTO watch_claim (singleton, token, pid, heartbeat_ms) VALUES (1, ?, ?, ?) ON CONFLICT(singleton) DO UPDATE SET token = excluded.token, pid = excluded.pid, heartbeat_ms = excluded.heartbeat_ms').run(token, pid, Date.now());\n      },\n      BEGIN_WRITE\n    );\n  }\n\n  renew(token: string): boolean {\n    return changedOne(this.db.prepare('UPDATE watch_claim SET heartbeat_ms = ? WHERE singleton = 1 AND token = ?').run(Date.now(), token).changes);\n  }\n\n  release(token: string): boolean {\n    return changedOne(this.db.prepare('DELETE FROM watch_claim WHERE singleton = 1 AND token = ?').run(token).changes);\n  }\n\n  read(): WatchClaimRecord | null {\n    const row = this.db.prepare('SELECT token, pid, heartbeat_ms FROM watch_claim WHERE singleton = 1').get() as ClaimRow | undefined;\n    return row ? { token: row.token, pid: row.pid, heartbeatMs: row.heartbeat_ms } : null;\n  }\n\n  close(): void {\n    this.db.close();\n  }\n}\n\n// Status opens the coordinator read-only and therefore cannot create either the file or its schema.\nexport function readWatchClaim(configDir: string): WatchClaimRecord | null {\n  const path = claimPath(configDir);\n  if (!existsSync(path)) return null;\n  let db: DatabaseSync;\n  try {\n    db = new DatabaseSync(path, { readOnly: true });\n  } catch (err) {\n    if (!existsSync(path)) return null;\n    throw err;\n  }\n  try {\n    db.exec(`PRAGMA busy_timeout = ${CLAIM_BUSY_TIMEOUT_MS}`);\n    const row = db.prepare('SELECT token, pid, heartbeat_ms FROM watch_claim WHERE singleton = 1').get() as ClaimRow | undefined;\n    return row ? { token: row.token, pid: row.pid, heartbeatMs: row.heartbeat_ms } : null;\n  } catch (err) {\n    if (/no such table: watch_claim/.test((err as Error).message)) return null;\n    throw err;\n  } finally {\n    db.close();\n  }\n}\n\nexport async function startWatchClaim(configDir: string, force: boolean, heartbeatIntervalMs: number, onFailure: (err: Error) => void): Promise<WatchClaimController> {\n  const workerData: WatchClaimWorkerData = { configDir, force, heartbeatIntervalMs, pid: process.pid, token: randomUUID() };\n  const worker = new Worker(resolveWorkerFile('watch-heartbeat'), { workerData });\n  let acquired = false;\n  let closing = false;\n  let exited = false;\n  let failureReported = false;\n  let closeError: Error | null = null;\n  let closePromise: Promise<void> | undefined;\n  let resolveExit!: () => void;\n  const exit = new Promise<void>((resolve) => {\n    resolveExit = resolve;\n  });\n  let resolveStart!: () => void;\n  let rejectStart!: (err: Error) => void;\n  const start = new Promise<void>((resolve, reject) => {\n    resolveStart = resolve;\n    rejectStart = reject;\n  });\n\n  const fail = (err: Error): void => {\n    if (failureReported) return;\n    failureReported = true;\n    if (!acquired) rejectStart(err);\n    else if (closing) closeError ??= err;\n    else onFailure(err);\n  };\n  worker.on('message', (message: WatchClaimWorkerMessage) => {\n    if (message.type === 'acquired') {\n      acquired = true;\n      resolveStart();\n    } else if (message.type === 'failure') {\n      fail(reviveError(message.error));\n    }\n  });\n  worker.on('error', fail);\n  worker.on('exit', (code) => {\n    exited = true;\n    resolveExit();\n    if (closing && code !== 0) closeError ??= new Error(`watch heartbeat worker stopped with exit code ${code}`);\n    else if (!closing && !failureReported) fail(new Error(`watch heartbeat worker stopped unexpectedly with exit code ${code}`));\n  });\n\n  try {\n    await start;\n  } catch (err) {\n    closing = true;\n    if (!exited) await worker.terminate();\n    await exit;\n    throw err;\n  }\n\n  return {\n    close(): Promise<void> {\n      closePromise ??= (async () => {\n        closing = true;\n        if (!exited) {\n          try {\n            worker.postMessage({ type: 'stop' });\n          } catch (err) {\n            closeError ??= err instanceof Error ? err : new Error(String(err));\n            await worker.terminate();\n          }\n        }\n        await exit;\n        if (closeError) throw closeError;\n      })();\n      return closePromise;\n    },\n  };\n}\n"],"names":["randomUUID","existsSync","join","DatabaseSync","Worker","SenseError","resolveWorkerFile","reviveError","createConnection","BEGIN_WRITE","withTransaction","WATCH_CLAIM_FILENAME","WATCH_HEARTBEAT_INTERVAL_MS","WATCH_STALE_HEARTBEAT_MS","CLAIM_BUSY_TIMEOUT_MS","claimPath","configDir","changedOne","changes","BigInt","WatchClaimDatabase","acquire","token","pid","force","conn","current","db","prepare","get","age","Date","now","heartbeat_ms","Math","round","run","renew","release","read","row","heartbeatMs","close","exec","err","closeError","AggregateError","readWatchClaim","path","readOnly","test","message","startWatchClaim","heartbeatIntervalMs","onFailure","workerData","process","worker","acquired","closing","exited","failureReported","closePromise","resolveExit","exit","Promise","resolve","resolveStart","rejectStart","start","reject","fail","on","type","error","code","Error","terminate","postMessage","String"],"mappings":"AAAA,SAASA,UAAU,QAAQ,cAAc;AACzC,SAASC,UAAU,QAAQ,UAAU;AACrC,SAASC,IAAI,QAAQ,YAAY;AACjC,SAASC,YAAY,QAAQ,cAAc;AAC3C,SAASC,MAAM,QAAQ,sBAAsB;AAC7C,SAASC,UAAU,QAAQ,cAAc;AACzC,SAASC,iBAAiB,QAAQ,uBAAuB;AACzD,SAASC,WAAW,QAAiC,yBAAyB;AAC9E,SAASC,gBAAgB,QAAQ,+BAA+B;AAChE,SAASC,WAAW,EAAEC,eAAe,QAAQ,yBAAyB;AAEtE,OAAO,MAAMC,uBAAuB,kBAAkB;AACtD,OAAO,MAAMC,8BAA8B,KAAM;AACjD,OAAO,MAAMC,2BAA2B,MAAO;AAC/C,MAAMC,wBAAwB;AA4B9B,SAASC,UAAUC,SAAiB;IAClC,OAAOd,KAAKc,WAAWL;AACzB;AAEA,SAASM,WAAWC,OAAwB;IAC1C,OAAOA,YAAY,KAAKA,YAAYC,OAAO;AAC7C;AAEA,mGAAmG;AACnG,iGAAiG;AACjG,OAAO,MAAMC;IAqBX,MAAMC,QAAQC,KAAa,EAAEC,GAAW,EAAEC,KAAc,EAAiB;QACvE,MAAMd,gBACJ,IAAI,CAACe,IAAI,EACT;YACE,MAAMC,UAAU,IAAI,CAACC,EAAE,CAACC,OAAO,CAAC,wEAAwEC,GAAG;YAC3G,IAAIH,WAAW,CAACF,OAAO;gBACrB,MAAMM,MAAMC,KAAKC,GAAG,KAAKN,QAAQO,YAAY;gBAC7C,IAAIH,OAAO,KAAKA,MAAMjB,0BAA0B,MAAM,IAAIR,WAAW,gBAAgB,CAAC,0CAA0C,EAAE6B,KAAKC,KAAK,CAACL,MAAM,MAAM,+BAA+B,CAAC;YAC3L;YACA,IAAI,CAACH,EAAE,CAACC,OAAO,CAAC,2MAA2MQ,GAAG,CAACd,OAAOC,KAAKQ,KAAKC,GAAG;QACrP,GACAvB;IAEJ;IAEA4B,MAAMf,KAAa,EAAW;QAC5B,OAAOL,WAAW,IAAI,CAACU,EAAE,CAACC,OAAO,CAAC,6EAA6EQ,GAAG,CAACL,KAAKC,GAAG,IAAIV,OAAOJ,OAAO;IAC/I;IAEAoB,QAAQhB,KAAa,EAAW;QAC9B,OAAOL,WAAW,IAAI,CAACU,EAAE,CAACC,OAAO,CAAC,6DAA6DQ,GAAG,CAACd,OAAOJ,OAAO;IACnH;IAEAqB,OAAgC;QAC9B,MAAMC,MAAM,IAAI,CAACb,EAAE,CAACC,OAAO,CAAC,wEAAwEC,GAAG;QACvG,OAAOW,MAAM;YAAElB,OAAOkB,IAAIlB,KAAK;YAAEC,KAAKiB,IAAIjB,GAAG;YAAEkB,aAAaD,IAAIP,YAAY;QAAC,IAAI;IACnF;IAEAS,QAAc;QACZ,IAAI,CAACf,EAAE,CAACe,KAAK;IACf;IA/CA,YAAY1B,SAAiB,CAAE;QAC7B,MAAMW,KAAK,IAAIxB,aAAaY,UAAUC;QACtC,IAAI;YACFW,GAAGgB,IAAI,CAAC,CAAC,sBAAsB,EAAE7B,uBAAuB;YACxDa,GAAGgB,IAAI,CAAC;QACV,EAAE,OAAOC,KAAK;YACZ,IAAI;gBACFjB,GAAGe,KAAK;YACV,EAAE,OAAOG,YAAY;gBACnB,MAAM,IAAIC,eAAe;oBAACF;oBAAKC;iBAAW,EAAE;YAC9C;YACA,MAAMD;QACR;QACA,IAAI,CAACjB,EAAE,GAAGA;QACV,IAAI,CAACF,IAAI,GAAGjB,iBAAiBmB;IAC/B;AAiCF;AAEA,oGAAoG;AACpG,OAAO,SAASoB,eAAe/B,SAAiB;IAC9C,MAAMgC,OAAOjC,UAAUC;IACvB,IAAI,CAACf,WAAW+C,OAAO,OAAO;IAC9B,IAAIrB;IACJ,IAAI;QACFA,KAAK,IAAIxB,aAAa6C,MAAM;YAAEC,UAAU;QAAK;IAC/C,EAAE,OAAOL,KAAK;QACZ,IAAI,CAAC3C,WAAW+C,OAAO,OAAO;QAC9B,MAAMJ;IACR;IACA,IAAI;QACFjB,GAAGgB,IAAI,CAAC,CAAC,sBAAsB,EAAE7B,uBAAuB;QACxD,MAAM0B,MAAMb,GAAGC,OAAO,CAAC,wEAAwEC,GAAG;QAClG,OAAOW,MAAM;YAAElB,OAAOkB,IAAIlB,KAAK;YAAEC,KAAKiB,IAAIjB,GAAG;YAAEkB,aAAaD,IAAIP,YAAY;QAAC,IAAI;IACnF,EAAE,OAAOW,KAAK;QACZ,IAAI,6BAA6BM,IAAI,CAAC,AAACN,IAAcO,OAAO,GAAG,OAAO;QACtE,MAAMP;IACR,SAAU;QACRjB,GAAGe,KAAK;IACV;AACF;AAEA,OAAO,eAAeU,gBAAgBpC,SAAiB,EAAEQ,KAAc,EAAE6B,mBAA2B,EAAEC,SAA+B;IACnI,MAAMC,aAAmC;QAAEvC;QAAWQ;QAAO6B;QAAqB9B,KAAKiC,QAAQjC,GAAG;QAAED,OAAOtB;IAAa;IACxH,MAAMyD,SAAS,IAAIrD,OAAOE,kBAAkB,oBAAoB;QAAEiD;IAAW;IAC7E,IAAIG,WAAW;IACf,IAAIC,UAAU;IACd,IAAIC,SAAS;IACb,IAAIC,kBAAkB;IACtB,IAAIhB,aAA2B;IAC/B,IAAIiB;IACJ,IAAIC;IACJ,MAAMC,OAAO,IAAIC,QAAc,CAACC;QAC9BH,cAAcG;IAChB;IACA,IAAIC;IACJ,IAAIC;IACJ,MAAMC,QAAQ,IAAIJ,QAAc,CAACC,SAASI;QACxCH,eAAeD;QACfE,cAAcE;IAChB;IAEA,MAAMC,OAAO,CAAC3B;QACZ,IAAIiB,iBAAiB;QACrBA,kBAAkB;QAClB,IAAI,CAACH,UAAUU,YAAYxB;aACtB,IAAIe,SAASd,uBAAAA,wBAAAA,aAAAA,aAAeD;aAC5BU,UAAUV;IACjB;IACAa,OAAOe,EAAE,CAAC,WAAW,CAACrB;QACpB,IAAIA,QAAQsB,IAAI,KAAK,YAAY;YAC/Bf,WAAW;YACXS;QACF,OAAO,IAAIhB,QAAQsB,IAAI,KAAK,WAAW;YACrCF,KAAKhE,YAAY4C,QAAQuB,KAAK;QAChC;IACF;IACAjB,OAAOe,EAAE,CAAC,SAASD;IACnBd,OAAOe,EAAE,CAAC,QAAQ,CAACG;QACjBf,SAAS;QACTG;QACA,IAAIJ,WAAWgB,SAAS,GAAG9B,uBAAAA,wBAAAA,aAAAA,aAAe,IAAI+B,MAAM,CAAC,8CAA8C,EAAED,MAAM;aACtG,IAAI,CAAChB,WAAW,CAACE,iBAAiBU,KAAK,IAAIK,MAAM,CAAC,2DAA2D,EAAED,MAAM;IAC5H;IAEA,IAAI;QACF,MAAMN;IACR,EAAE,OAAOzB,KAAK;QACZe,UAAU;QACV,IAAI,CAACC,QAAQ,MAAMH,OAAOoB,SAAS;QACnC,MAAMb;QACN,MAAMpB;IACR;IAEA,OAAO;QACLF;YACEoB,yBAAAA,0BAAAA,eAAAA,eAAiB,AAAC,CAAA;gBAChBH,UAAU;gBACV,IAAI,CAACC,QAAQ;oBACX,IAAI;wBACFH,OAAOqB,WAAW,CAAC;4BAAEL,MAAM;wBAAO;oBACpC,EAAE,OAAO7B,KAAK;wBACZC,uBAAAA,wBAAAA,aAAAA,aAAeD,eAAegC,QAAQhC,MAAM,IAAIgC,MAAMG,OAAOnC;wBAC7D,MAAMa,OAAOoB,SAAS;oBACxB;gBACF;gBACA,MAAMb;gBACN,IAAInB,YAAY,MAAMA;YACxB,CAAA;YACA,OAAOiB;QACT;IACF;AACF"}