sensemaking 0.15.0 → 0.15.2

package/dist/esm/features/links.js.map

@@ -1 +1 @@
-{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/features/links.ts"],"sourcesContent":["import posix from 'node:path/posix';\nimport type { DatabaseSync } from 'node:sqlite';\nimport type { FileStat } from '../scan.ts';\nimport type { Feature, ReconcileDelta } from './types.ts';\n\n// links(src, target, target_base, dst, embed): target as written, target_base its baseKey\n// (indexed, for the incremental resolve below), dst the resolved path or NULL (a\n// queryable dead link), embed whether it's `![[x]]`/`![](x.md)` rather than `[[x]]`/`[](x.md)`\n// -- Obsidian's grain: a target both linked and embedded in the same note is two rows.\n\n// Wikilinks ([[target]], [[target#anchor|alias]], embeds) plus relative markdown links to .md files.\nfunction extract(_raw: string, body: string): Array<{ target: string; embed: boolean }> {\n  const seen = new Set<string>();\n  const results: Array<{ target: string; embed: boolean }> = [];\n  const add = (target: string, embed: boolean) => {\n    const key = `${target}\\0${embed ? '1' : '0'}`;\n    if (seen.has(key)) return;\n    seen.add(key);\n    results.push({ target, embed });\n  };\n  // To the first ]], as Obsidian parses it, so a heading or alias holding a lone ] still\n  // matches; anchor and alias split off in code, since a class-based regex stops at the ].\n  for (const m of body.matchAll(/\\[\\[(.*?)\\]\\]/g)) {\n    const target = m[1].split('|')[0].split('#')[0].trim();\n    if (target) add(target, body[(m.index ?? 0) - 1] === '!');\n  }\n  for (const m of body.matchAll(/(!)?\\[[^\\]]*\\]\\(([^)]+\\.md)(?:#[^)]*)?\\)/g)) {\n    const target = m[2].trim();\n    if (target && !/^[a-z]+:\\/\\//i.test(target)) add(target, m[1] === '!');\n  }\n  return results;\n}\n\nfunction cleanTarget(target: string): string {\n  return target.replace(/\\\\/g, '/').replace(/^\\.\\//, '');\n}\n\nfunction baseKey(path: string): string {\n  return posix.basename(path).replace(/\\.md$/i, '').toLowerCase();\n}\n\n// Obsidian-style: exact relative path (with/without .md), path relative to the linking\n// note's directory, then basename match (lexicographically first on ties).\nfunction resolveTarget(src: string, target: string, pathSet: Set<string>, byBase: Map<string, string[]>): string | null {\n  const clean = cleanTarget(target);\n  const fromSrc = posix.normalize(posix.join(posix.dirname(src), clean));\n  for (const candidate of [clean, `${clean}.md`, fromSrc, `${fromSrc}.md`]) {\n    if (pathSet.has(candidate)) return candidate;\n  }\n  return byBase.get(baseKey(clean))?.[0] ?? null;\n}\n\nfunction buildByBase(files: FileStat[]): Map<string, string[]> {\n  const byBase = new Map<string, string[]>();\n  for (const path of files.map((f) => f.relPath).sort()) {\n    const key = baseKey(path);\n    const list = byBase.get(key);\n    if (list) list.push(path);\n    else byBase.set(key, [path]);\n  }\n  return byBase;\n}\n\ninterface LinkRow {\n  src: string;\n  target: string;\n  dst: string | null;\n  embed: number;\n}\n\n// Applies resolveTarget to `rows`, writing only rows whose dst actually changes. The UPDATE\n// keys on (src, target): a link/embed sibling pair shares one resolution, so either row's\n// mismatch rewrites both.\nfunction resolveRows(db: DatabaseSync, rows: LinkRow[], pathSet: Set<string>, byBase: Map<string, string[]>): boolean {\n  const update = db.prepare('UPDATE links SET dst = ? WHERE src = ? AND target = ?');\n  let changed = false;\n  for (const row of rows) {\n    const dst = resolveTarget(row.src, row.target, pathSet, byBase);\n    if (dst !== row.dst) {\n      update.run(dst, row.src, row.target);\n      changed = true;\n    }\n  }\n  return changed;\n}\n\n// A new or deleted file can change any note's resolution, so re-resolve the whole table.\n// Fallback for cold builds and large deltas -- see afterReconcile's threshold.\nfunction resolveAll(db: DatabaseSync, files: FileStat[]): boolean {\n  const pathSet = new Set(files.map((f) => f.relPath));\n  const byBase = buildByBase(files);\n  const rows = db.prepare('SELECT src, target, dst, embed FROM links').all() as unknown as LinkRow[];\n  return resolveRows(db, rows, pathSet, byBase);\n}\n\n// Re-resolve only what this reconcile could have touched: re-stored sources, links whose\n// target basename matches an added/vanished file, and links whose dst just vanished. All\n// three are indexed lookups, so this never reads the whole table.\nfunction resolveIncremental(db: DatabaseSync, delta: ReconcileDelta): boolean {\n  const pathSet = new Set(delta.files.map((f) => f.relPath));\n  const byBase = buildByBase(delta.files);\n\n  const changedBasenames = new Set<string>();\n  for (const p of delta.added) changedBasenames.add(baseKey(p));\n  for (const p of delta.vanished) changedBasenames.add(baseKey(p));\n\n  // \\0 can't appear in a path or target, so the key never collides; a space can.\n  const candidates = new Map<string, LinkRow>();\n  const collect = (rows: LinkRow[]) => {\n    for (const row of rows) candidates.set(`${row.src}\\0${row.target}\\0${row.embed}`, row);\n  };\n  // Chunked so a large-but-under-threshold delta never exceeds SQLite's bound-variable\n  // limit (SQLITE_MAX_VARIABLE_NUMBER, 32766).\n  const collectIn = (column: string, keys: string[]) => {\n    for (let i = 0; i < keys.length; i += 500) {\n      const chunk = keys.slice(i, i + 500);\n      const placeholders = chunk.map(() => '?').join(', ');\n      collect(db.prepare(`SELECT src, target, dst, embed FROM links WHERE ${column} IN (${placeholders})`).all(...chunk) as unknown as LinkRow[]);\n    }\n  };\n\n  collectIn('src', delta.reparsed);\n  collectIn('target_base', [...changedBasenames]);\n  collectIn('dst', delta.vanished);\n\n  return resolveRows(db, [...candidates.values()], pathSet, byBase);\n}\n\n// Resolved edges, for rank and for search's graph expansion.\nexport function linkEdges(db: DatabaseSync): [string, string][] {\n  // One edge per row, as 0.12's grain always had it -- two written targets resolving to one\n  // dst stay two edges (a weight the fusion evals were gated on). The only exclusion is an\n  // embed whose exact (src, target) also exists as a link: that second row is new with the\n  // embed grain and would double an edge that used to be one row. The NOT EXISTS probes the\n  // primary key and fires only for embed rows; both branches still scan the table.\n  const sql = `SELECT src, dst FROM links WHERE dst IS NOT NULL AND embed = 0\n    UNION ALL\n    SELECT src, dst FROM links l WHERE dst IS NOT NULL AND embed = 1\n      AND NOT EXISTS (SELECT 1 FROM links l0 WHERE l0.src = l.src AND l0.target = l.target AND l0.embed = 0)`;\n  return (db.prepare(sql).all() as Array<{ src: string; dst: string }>).map((r) => [r.src, r.dst]);\n}\n\n// remove() has already deleted the rows by the time afterReconcile runs, so it records here\n// whether they carried edges. Keyed on the delta object, so the state dies with the reconcile.\nconst removedWithLinks = new WeakMap<ReconcileDelta, Set<string>>();\n\nfunction recordRemoved(delta: ReconcileDelta, path: string): void {\n  let set = removedWithLinks.get(delta);\n  if (!set) {\n    set = new Set();\n    removedWithLinks.set(delta, set);\n  }\n  set.add(path);\n}\n\n// remove() runs per path; the vanished list is an array, so cache the Set per delta.\nconst vanishedSets = new WeakMap<ReconcileDelta, Set<string>>();\nfunction vanishedSet(delta: ReconcileDelta): Set<string> {\n  let set = vanishedSets.get(delta);\n  if (!set) {\n    set = new Set(delta.vanished);\n    vanishedSets.set(delta, set);\n  }\n  return set;\n}\n\nexport const links: Feature = {\n  name: 'links',\n  schema(db) {\n    db.exec('CREATE TABLE IF NOT EXISTS links (src TEXT, target TEXT, target_base TEXT, dst TEXT, embed INTEGER, PRIMARY KEY (src, target, embed))');\n    db.exec('CREATE INDEX IF NOT EXISTS links_dst ON links(dst)');\n    db.exec('CREATE INDEX IF NOT EXISTS links_target_base ON links(target_base)');\n  },\n  extract,\n  // Reparsed docs are diffed in store() below instead of wiped here: deleting and\n  // re-inserting resets dst to NULL, which made every touch-only reparse read as an edge\n  // change and recompute PageRank -- measured as most of the remaining update cost.\n  remove(db, path, delta) {\n    if (!vanishedSet(delta).has(path)) return;\n    const result = db.prepare('DELETE FROM links WHERE src = ?').run(path);\n    if (Number(result.changes) > 0) recordRemoved(delta, path);\n  },\n  store(db, path, extracted, delta) {\n    const targets = extracted as Array<{ target: string; embed: boolean }>;\n    // Stale rows (target/embed pairs the new parse no longer contains) are real edge removals.\n    // \\0 can't appear in a target, so the composite key never collides.\n    let stale: ReturnType<ReturnType<DatabaseSync['prepare']>['run']>;\n    if (targets.length === 0) {\n      stale = db.prepare('DELETE FROM links WHERE src = ?').run(path);\n    } else {\n      const placeholders = targets.map(() => '?').join(', ');\n      const keys = targets.map((t) => `${t.target}\\0${t.embed ? '1' : '0'}`);\n      stale = db.prepare(`DELETE FROM links WHERE src = ? AND (target || char(0) || embed) NOT IN (${placeholders})`).run(path, ...keys);\n    }\n    if (Number(stale.changes) > 0) recordRemoved(delta, path);\n    // Upsert preserves dst on surviving rows, so an unchanged link resolves to the same\n    // value and reports no change; only genuinely new rows start at NULL.\n    const insert = db.prepare('INSERT INTO links (src, target, target_base, dst, embed) VALUES (?, ?, ?, NULL, ?) ON CONFLICT(src, target, embed) DO UPDATE SET target_base = excluded.target_base');\n    for (const { target, embed } of targets) insert.run(path, target, baseKey(cleanTarget(target)), embed ? 1 : 0);\n  },\n  afterReconcile(db, delta) {\n    // Any deleted rows that carried edges mean the edge set shrank -- whether the file\n    // vanished or was reparsed down to fewer/no links. dstChanged alone misses the\n    // reparsed-to-zero-links case (no surviving rows to re-resolve).\n    const removeHadLinks = (removedWithLinks.get(delta)?.size ?? 0) > 0;\n\n    const churn = delta.reparsed.length + delta.vanished.length;\n    // Past this share of the tree, a full pass is the only one guaranteed to match\n    // resolveTarget's ambiguity rules. A cold build clears the threshold on its own.\n    const large = delta.files.length === 0 || churn > 0.2 * delta.files.length;\n\n    const dstChanged = large ? resolveAll(db, delta.files) : resolveIncremental(db, delta);\n    delta.linksChanged = dstChanged || removeHadLinks;\n  },\n};\n"],"names":["posix","extract","_raw","body","seen","Set","results","add","target","embed","key","has","push","m","matchAll","split","trim","index","test","cleanTarget","replace","baseKey","path","basename","toLowerCase","resolveTarget","src","pathSet","byBase","clean","fromSrc","normalize","join","dirname","candidate","get","buildByBase","files","Map","map","f","relPath","sort","list","set","resolveRows","db","rows","update","prepare","changed","row","dst","run","resolveAll","all","resolveIncremental","delta","changedBasenames","p","added","vanished","candidates","collect","collectIn","column","keys","i","length","chunk","slice","placeholders","reparsed","values","linkEdges","sql","r","removedWithLinks","WeakMap","recordRemoved","vanishedSets","vanishedSet","links","name","schema","exec","remove","result","Number","changes","store","extracted","targets","stale","t","insert","afterReconcile","removeHadLinks","size","churn","large","dstChanged","linksChanged"],"mappings":"AAAA,OAAOA,WAAW,kBAAkB;AAKpC,0FAA0F;AAC1F,iFAAiF;AACjF,+FAA+F;AAC/F,uFAAuF;AAEvF,qGAAqG;AACrG,SAASC,QAAQC,IAAY,EAAEC,IAAY;IACzC,MAAMC,OAAO,IAAIC;IACjB,MAAMC,UAAqD,EAAE;IAC7D,MAAMC,MAAM,CAACC,QAAgBC;QAC3B,MAAMC,MAAM,GAAGF,OAAO,EAAE,EAAEC,QAAQ,MAAM,KAAK;QAC7C,IAAIL,KAAKO,GAAG,CAACD,MAAM;QACnBN,KAAKG,GAAG,CAACG;QACTJ,QAAQM,IAAI,CAAC;YAAEJ;YAAQC;QAAM;IAC/B;IACA,uFAAuF;IACvF,yFAAyF;IACzF,KAAK,MAAMI,KAAKV,KAAKW,QAAQ,CAAC,kBAAmB;YAEjBD;QAD9B,MAAML,SAASK,CAAC,CAAC,EAAE,CAACE,KAAK,CAAC,IAAI,CAAC,EAAE,CAACA,KAAK,CAAC,IAAI,CAAC,EAAE,CAACC,IAAI;QACpD,IAAIR,QAAQD,IAAIC,QAAQL,IAAI,CAAC,EAACU,WAAAA,EAAEI,KAAK,cAAPJ,sBAAAA,WAAW,KAAK,EAAE,KAAK;IACvD;IACA,KAAK,MAAMA,KAAKV,KAAKW,QAAQ,CAAC,6CAA8C;QAC1E,MAAMN,SAASK,CAAC,CAAC,EAAE,CAACG,IAAI;QACxB,IAAIR,UAAU,CAAC,gBAAgBU,IAAI,CAACV,SAASD,IAAIC,QAAQK,CAAC,CAAC,EAAE,KAAK;IACpE;IACA,OAAOP;AACT;AAEA,SAASa,YAAYX,MAAc;IACjC,OAAOA,OAAOY,OAAO,CAAC,OAAO,KAAKA,OAAO,CAAC,SAAS;AACrD;AAEA,SAASC,QAAQC,IAAY;IAC3B,OAAOtB,MAAMuB,QAAQ,CAACD,MAAMF,OAAO,CAAC,UAAU,IAAII,WAAW;AAC/D;AAEA,uFAAuF;AACvF,2EAA2E;AAC3E,SAASC,cAAcC,GAAW,EAAElB,MAAc,EAAEmB,OAAoB,EAAEC,MAA6B;;QAM9FA;IALP,MAAMC,QAAQV,YAAYX;IAC1B,MAAMsB,UAAU9B,MAAM+B,SAAS,CAAC/B,MAAMgC,IAAI,CAAChC,MAAMiC,OAAO,CAACP,MAAMG;IAC/D,KAAK,MAAMK,aAAa;QAACL;QAAO,GAAGA,MAAM,GAAG,CAAC;QAAEC;QAAS,GAAGA,QAAQ,GAAG,CAAC;KAAC,CAAE;QACxE,IAAIH,QAAQhB,GAAG,CAACuB,YAAY,OAAOA;IACrC;IACA,gBAAON,cAAAA,OAAOO,GAAG,CAACd,QAAQQ,qBAAnBD,kCAAAA,WAA4B,CAAC,EAAE,uCAAI;AAC5C;AAEA,SAASQ,YAAYC,KAAiB;IACpC,MAAMT,SAAS,IAAIU;IACnB,KAAK,MAAMhB,QAAQe,MAAME,GAAG,CAAC,CAACC,IAAMA,EAAEC,OAAO,EAAEC,IAAI,GAAI;QACrD,MAAMhC,MAAMW,QAAQC;QACpB,MAAMqB,OAAOf,OAAOO,GAAG,CAACzB;QACxB,IAAIiC,MAAMA,KAAK/B,IAAI,CAACU;aACfM,OAAOgB,GAAG,CAAClC,KAAK;YAACY;SAAK;IAC7B;IACA,OAAOM;AACT;AASA,4FAA4F;AAC5F,0FAA0F;AAC1F,0BAA0B;AAC1B,SAASiB,YAAYC,EAAgB,EAAEC,IAAe,EAAEpB,OAAoB,EAAEC,MAA6B;IACzG,MAAMoB,SAASF,GAAGG,OAAO,CAAC;IAC1B,IAAIC,UAAU;IACd,KAAK,MAAMC,OAAOJ,KAAM;QACtB,MAAMK,MAAM3B,cAAc0B,IAAIzB,GAAG,EAAEyB,IAAI3C,MAAM,EAAEmB,SAASC;QACxD,IAAIwB,QAAQD,IAAIC,GAAG,EAAE;YACnBJ,OAAOK,GAAG,CAACD,KAAKD,IAAIzB,GAAG,EAAEyB,IAAI3C,MAAM;YACnC0C,UAAU;QACZ;IACF;IACA,OAAOA;AACT;AAEA,yFAAyF;AACzF,+EAA+E;AAC/E,SAASI,WAAWR,EAAgB,EAAET,KAAiB;IACrD,MAAMV,UAAU,IAAItB,IAAIgC,MAAME,GAAG,CAAC,CAACC,IAAMA,EAAEC,OAAO;IAClD,MAAMb,SAASQ,YAAYC;IAC3B,MAAMU,OAAOD,GAAGG,OAAO,CAAC,6CAA6CM,GAAG;IACxE,OAAOV,YAAYC,IAAIC,MAAMpB,SAASC;AACxC;AAEA,yFAAyF;AACzF,yFAAyF;AACzF,kEAAkE;AAClE,SAAS4B,mBAAmBV,EAAgB,EAAEW,KAAqB;IACjE,MAAM9B,UAAU,IAAItB,IAAIoD,MAAMpB,KAAK,CAACE,GAAG,CAAC,CAACC,IAAMA,EAAEC,OAAO;IACxD,MAAMb,SAASQ,YAAYqB,MAAMpB,KAAK;IAEtC,MAAMqB,mBAAmB,IAAIrD;IAC7B,KAAK,MAAMsD,KAAKF,MAAMG,KAAK,CAAEF,iBAAiBnD,GAAG,CAACc,QAAQsC;IAC1D,KAAK,MAAMA,KAAKF,MAAMI,QAAQ,CAAEH,iBAAiBnD,GAAG,CAACc,QAAQsC;IAE7D,+EAA+E;IAC/E,MAAMG,aAAa,IAAIxB;IACvB,MAAMyB,UAAU,CAAChB;QACf,KAAK,MAAMI,OAAOJ,KAAMe,WAAWlB,GAAG,CAAC,GAAGO,IAAIzB,GAAG,CAAC,EAAE,EAAEyB,IAAI3C,MAAM,CAAC,EAAE,EAAE2C,IAAI1C,KAAK,EAAE,EAAE0C;IACpF;IACA,qFAAqF;IACrF,6CAA6C;IAC7C,MAAMa,YAAY,CAACC,QAAgBC;QACjC,IAAK,IAAIC,IAAI,GAAGA,IAAID,KAAKE,MAAM,EAAED,KAAK,IAAK;YACzC,MAAME,QAAQH,KAAKI,KAAK,CAACH,GAAGA,IAAI;YAChC,MAAMI,eAAeF,MAAM9B,GAAG,CAAC,IAAM,KAAKP,IAAI,CAAC;YAC/C+B,QAAQjB,GAAGG,OAAO,CAAC,CAAC,gDAAgD,EAAEgB,OAAO,KAAK,EAAEM,aAAa,CAAC,CAAC,EAAEhB,GAAG,IAAIc;QAC9G;IACF;IAEAL,UAAU,OAAOP,MAAMe,QAAQ;IAC/BR,UAAU,eAAe;WAAIN;KAAiB;IAC9CM,UAAU,OAAOP,MAAMI,QAAQ;IAE/B,OAAOhB,YAAYC,IAAI;WAAIgB,WAAWW,MAAM;KAAG,EAAE9C,SAASC;AAC5D;AAEA,6DAA6D;AAC7D,OAAO,SAAS8C,UAAU5B,EAAgB;IACxC,0FAA0F;IAC1F,yFAAyF;IACzF,yFAAyF;IACzF,0FAA0F;IAC1F,iFAAiF;IACjF,MAAM6B,MAAM,CAAC;;;4GAG6F,CAAC;IAC3G,OAAO,AAAC7B,GAAGG,OAAO,CAAC0B,KAAKpB,GAAG,GAA2ChB,GAAG,CAAC,CAACqC,IAAM;YAACA,EAAElD,GAAG;YAAEkD,EAAExB,GAAG;SAAC;AACjG;AAEA,4FAA4F;AAC5F,+FAA+F;AAC/F,MAAMyB,mBAAmB,IAAIC;AAE7B,SAASC,cAActB,KAAqB,EAAEnC,IAAY;IACxD,IAAIsB,MAAMiC,iBAAiB1C,GAAG,CAACsB;IAC/B,IAAI,CAACb,KAAK;QACRA,MAAM,IAAIvC;QACVwE,iBAAiBjC,GAAG,CAACa,OAAOb;IAC9B;IACAA,IAAIrC,GAAG,CAACe;AACV;AAEA,qFAAqF;AACrF,MAAM0D,eAAe,IAAIF;AACzB,SAASG,YAAYxB,KAAqB;IACxC,IAAIb,MAAMoC,aAAa7C,GAAG,CAACsB;IAC3B,IAAI,CAACb,KAAK;QACRA,MAAM,IAAIvC,IAAIoD,MAAMI,QAAQ;QAC5BmB,aAAapC,GAAG,CAACa,OAAOb;IAC1B;IACA,OAAOA;AACT;AAEA,OAAO,MAAMsC,QAAiB;IAC5BC,MAAM;IACNC,QAAOtC,EAAE;QACPA,GAAGuC,IAAI,CAAC;QACRvC,GAAGuC,IAAI,CAAC;QACRvC,GAAGuC,IAAI,CAAC;IACV;IACApF;IACA,gFAAgF;IAChF,uFAAuF;IACvF,kFAAkF;IAClFqF,QAAOxC,EAAE,EAAExB,IAAI,EAAEmC,KAAK;QACpB,IAAI,CAACwB,YAAYxB,OAAO9C,GAAG,CAACW,OAAO;QACnC,MAAMiE,SAASzC,GAAGG,OAAO,CAAC,mCAAmCI,GAAG,CAAC/B;QACjE,IAAIkE,OAAOD,OAAOE,OAAO,IAAI,GAAGV,cAActB,OAAOnC;IACvD;IACAoE,OAAM5C,EAAE,EAAExB,IAAI,EAAEqE,SAAS,EAAElC,KAAK;QAC9B,MAAMmC,UAAUD;QAChB,2FAA2F;QAC3F,oEAAoE;QACpE,IAAIE;QACJ,IAAID,QAAQxB,MAAM,KAAK,GAAG;YACxByB,QAAQ/C,GAAGG,OAAO,CAAC,mCAAmCI,GAAG,CAAC/B;QAC5D,OAAO;YACL,MAAMiD,eAAeqB,QAAQrD,GAAG,CAAC,IAAM,KAAKP,IAAI,CAAC;YACjD,MAAMkC,OAAO0B,QAAQrD,GAAG,CAAC,CAACuD,IAAM,GAAGA,EAAEtF,MAAM,CAAC,EAAE,EAAEsF,EAAErF,KAAK,GAAG,MAAM,KAAK;YACrEoF,QAAQ/C,GAAGG,OAAO,CAAC,CAAC,yEAAyE,EAAEsB,aAAa,CAAC,CAAC,EAAElB,GAAG,CAAC/B,SAAS4C;QAC/H;QACA,IAAIsB,OAAOK,MAAMJ,OAAO,IAAI,GAAGV,cAActB,OAAOnC;QACpD,oFAAoF;QACpF,sEAAsE;QACtE,MAAMyE,SAASjD,GAAGG,OAAO,CAAC;QAC1B,KAAK,MAAM,EAAEzC,MAAM,EAAEC,KAAK,EAAE,IAAImF,QAASG,OAAO1C,GAAG,CAAC/B,MAAMd,QAAQa,QAAQF,YAAYX,UAAUC,QAAQ,IAAI;IAC9G;IACAuF,gBAAelD,EAAE,EAAEW,KAAK;;YAIEoB;QAHxB,mFAAmF;QACnF,+EAA+E;QAC/E,iEAAiE;QACjE,MAAMoB,iBAAiB,UAACpB,wBAAAA,iBAAiB1C,GAAG,CAACsB,oBAArBoB,4CAAAA,sBAA6BqB,IAAI,uCAAI,KAAK;QAElE,MAAMC,QAAQ1C,MAAMe,QAAQ,CAACJ,MAAM,GAAGX,MAAMI,QAAQ,CAACO,MAAM;QAC3D,+EAA+E;QAC/E,iFAAiF;QACjF,MAAMgC,QAAQ3C,MAAMpB,KAAK,CAAC+B,MAAM,KAAK,KAAK+B,QAAQ,MAAM1C,MAAMpB,KAAK,CAAC+B,MAAM;QAE1E,MAAMiC,aAAaD,QAAQ9C,WAAWR,IAAIW,MAAMpB,KAAK,IAAImB,mBAAmBV,IAAIW;QAChFA,MAAM6C,YAAY,GAAGD,cAAcJ;IACrC;AACF,EAAE"}
+{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/features/links.ts"],"sourcesContent":["import posix from 'node:path/posix';\nimport type { DatabaseSync } from 'node:sqlite';\nimport { maskRegions } from '../fences.ts';\nimport type { FileStat } from '../scan.ts';\nimport type { Feature, ReconcileDelta } from './types.ts';\n\n// links(src, target, target_base, dst, embed): target as written, target_base its baseKey\n// (indexed, for the incremental resolve below), dst the resolved path or NULL (a\n// queryable dead link), embed whether it's `![[x]]`/`![](x.md)` rather than `[[x]]`/`[](x.md)`\n// -- Obsidian's grain: a target both linked and embedded in the same note is two rows.\n\n// Wikilinks ([[target]], [[target#anchor|alias]], embeds), internal markdown links, and\n// frontmatter values that are exactly a wikilink (\"[[X]]\"; mid-string and ![[...]] forms are\n// not links there -- Obsidian's rule, probe-verified).\n// One rule for [[inner]] text wherever it appears: alias stripped after |, a leading #\n// keeps the anchor as the target (a same-note self-link needs a heading name), otherwise\n// the anchor splits off. Null = not a link.\nfunction parseWikilinkInner(inner: string): string | null {\n  const beforeAlias = inner.split('|')[0].trim();\n  if (beforeAlias.startsWith('#')) return beforeAlias.length > 1 ? beforeAlias : null;\n  const target = beforeAlias.split('#')[0].trim();\n  return target || null;\n}\n\nfunction extract(_raw: string, rawBody: string, _search?: { title: string; summary: string }, data?: Record<string, unknown>): Array<{ target: string; embed: boolean }> {\n  const body = /\\[\\[|\\]\\(/.test(rawBody) ? maskRegions(rawBody) : rawBody;\n  const seen = new Set<string>();\n  const results: Array<{ target: string; embed: boolean }> = [];\n  const add = (target: string, embed: boolean) => {\n    const key = `${target}\\0${embed ? '1' : '0'}`;\n    if (seen.has(key)) return;\n    seen.add(key);\n    results.push({ target, embed });\n  };\n  // To the first ]], as Obsidian parses it, so a heading or alias holding a lone ] still\n  // matches; anchor and alias split off in parseWikilinkInner.\n  for (const m of body.matchAll(/\\[\\[(.*?)\\]\\]/g)) {\n    const target = parseWikilinkInner(m[1]);\n    if (target) add(target, body[(m.index ?? 0) - 1] === '!');\n  }\n  // Any internal destination, not only .md-suffixed: Obsidian gives markdown links full\n  // linkpath resolution, so \\](Zektor) resolves like [[Zektor]] and \\](#anchor) is a\n  // self-edge. External URLs (a scheme anywhere survives the malformed double-paren case)\n  // and titled links (dest stops at whitespace) are skipped.\n  // A space in the destination is only valid ahead of a quoted title; a domain-shaped first\n  // segment (www.example.com/...) is external even without a scheme.\n  for (const m of body.matchAll(/(!)?\\[[^\\]]*\\]\\(((?:[^()\\s]|\\([^()\\s]*\\))+)(?:\\s+(?:\"[^)]*\"|'[^)]*'))?\\)/g)) {\n    const dest = m[2].trim();\n    // External: a URI scheme prefix (mailto:, tel:, data:, https:...), protocol-relative\n    // //host, a www. shorthand, or a scheme anywhere (the malformed double-paren case).\n    if (!dest || dest.includes('://')) continue;\n    if (/^([a-z][a-z0-9+.-]*:|\\/\\/|www\\.)/i.test(dest)) continue;\n    const target = dest.startsWith('#') ? dest : dest.split('#')[0];\n    if (target) add(target, m[1] === '!');\n  }\n  for (const value of Object.values(data ?? {})) {\n    for (const item of Array.isArray(value) ? value : [value]) {\n      if (typeof item !== 'string') continue;\n      const m = /^\\[\\[(.*)\\]\\]$/.exec(item.trim());\n      if (!m) continue;\n      const target = parseWikilinkInner(m[1]);\n      if (target) add(target, false);\n    }\n  }\n  return results;\n}\n\nfunction cleanTarget(target: string): string {\n  return target.replace(/\\\\/g, '/').replace(/^\\.\\//, '');\n}\n\nfunction baseKey(path: string): string {\n  return posix.basename(path).replace(/\\.md$/i, '').toLowerCase();\n}\n\n// Obsidian-style: exact relative path (with/without .md), path relative to the linking\n// note's directory, then basename match (shortest path wins on ties -- verified against a\n// cold-loaded Obsidian vault on 6+ real collision pairs; byBase's lists are pre-sorted that way).\nfunction resolveTarget(src: string, target: string, pathSet: Set<string>, byBase: Map<string, string[]>): string | null {\n  // [[#Heading]]: a same-note anchor, always the note itself -- never depends on other files.\n  if (target.startsWith('#')) return src;\n  const clean = cleanTarget(target);\n  const fromSrc = posix.normalize(posix.join(posix.dirname(src), clean));\n  for (const candidate of [clean, `${clean}.md`, fromSrc, `${fromSrc}.md`]) {\n    if (pathSet.has(candidate)) return candidate;\n  }\n  const candidates = byBase.get(baseKey(clean));\n  if (!candidates) return null;\n  // The linking note itself wins a basename collision (Obsidian resolves to self before the\n  // shortest-path rule -- cold-load verified on the hub corpus).\n  return candidates.includes(src) ? src : candidates[0];\n}\n\n// Shortest path wins a basename collision; equal lengths fall back to lexicographic, our own\n// deterministic tiebreak for a case Obsidian itself leaves registration-order-dependent.\nfunction buildByBase(files: FileStat[]): Map<string, string[]> {\n  const byBase = new Map<string, string[]>();\n  const paths = files.map((f) => f.relPath).sort((a, b) => a.length - b.length || (a < b ? -1 : a > b ? 1 : 0));\n  for (const path of paths) {\n    const key = baseKey(path);\n    const list = byBase.get(key);\n    if (list) list.push(path);\n    else byBase.set(key, [path]);\n  }\n  return byBase;\n}\n\ninterface LinkRow {\n  src: string;\n  target: string;\n  dst: string | null;\n  embed: number;\n}\n\n// Applies resolveTarget to `rows`, writing only rows whose dst actually changes. The UPDATE\n// keys on (src, target): a link/embed sibling pair shares one resolution, so either row's\n// mismatch rewrites both.\nfunction resolveRows(db: DatabaseSync, rows: LinkRow[], pathSet: Set<string>, byBase: Map<string, string[]>): boolean {\n  const update = db.prepare('UPDATE links SET dst = ? WHERE src = ? AND target = ?');\n  let changed = false;\n  for (const row of rows) {\n    const dst = resolveTarget(row.src, row.target, pathSet, byBase);\n    if (dst !== row.dst) {\n      update.run(dst, row.src, row.target);\n      changed = true;\n    }\n  }\n  return changed;\n}\n\n// A new or deleted file can change any note's resolution, so re-resolve the whole table.\n// Fallback for cold builds and large deltas -- see afterReconcile's threshold.\nfunction resolveAll(db: DatabaseSync, files: FileStat[]): boolean {\n  const pathSet = new Set(files.map((f) => f.relPath));\n  const byBase = buildByBase(files);\n  const rows = db.prepare('SELECT src, target, dst, embed FROM links').all() as unknown as LinkRow[];\n  return resolveRows(db, rows, pathSet, byBase);\n}\n\n// Re-resolve only what this reconcile could have touched: re-stored sources, links whose\n// target basename matches an added/vanished file, and links whose dst just vanished. All\n// three are indexed lookups, so this never reads the whole table.\nfunction resolveIncremental(db: DatabaseSync, delta: ReconcileDelta): boolean {\n  const pathSet = new Set(delta.files.map((f) => f.relPath));\n  const byBase = buildByBase(delta.files);\n\n  const changedBasenames = new Set<string>();\n  for (const p of delta.added) changedBasenames.add(baseKey(p));\n  for (const p of delta.vanished) changedBasenames.add(baseKey(p));\n\n  // \\0 can't appear in a path or target, so the key never collides; a space can.\n  const candidates = new Map<string, LinkRow>();\n  const collect = (rows: LinkRow[]) => {\n    for (const row of rows) candidates.set(`${row.src}\\0${row.target}\\0${row.embed}`, row);\n  };\n  // Chunked so a large-but-under-threshold delta never exceeds SQLite's bound-variable\n  // limit (SQLITE_MAX_VARIABLE_NUMBER, 32766).\n  const collectIn = (column: string, keys: string[]) => {\n    for (let i = 0; i < keys.length; i += 500) {\n      const chunk = keys.slice(i, i + 500);\n      const placeholders = chunk.map(() => '?').join(', ');\n      collect(db.prepare(`SELECT src, target, dst, embed FROM links WHERE ${column} IN (${placeholders})`).all(...chunk) as unknown as LinkRow[]);\n    }\n  };\n\n  collectIn('src', delta.reparsed);\n  collectIn('target_base', [...changedBasenames]);\n  collectIn('dst', delta.vanished);\n\n  return resolveRows(db, [...candidates.values()], pathSet, byBase);\n}\n\n// Resolved edges, for rank and for search's graph expansion.\nexport function linkEdges(db: DatabaseSync): [string, string][] {\n  // One edge per row, as 0.12's grain always had it -- two written targets resolving to one\n  // dst stay two edges (a weight the fusion evals were gated on). The only exclusion is an\n  // embed whose exact (src, target) also exists as a link: that second row is new with the\n  // embed grain and would double an edge that used to be one row. The NOT EXISTS probes the\n  // primary key and fires only for embed rows; both branches still scan the table.\n  // Self-edges (same-note anchors) are excluded here so PageRank mass is not self-recycled --\n  // Obsidian's own graph view hides self-loops too. The rows themselves stay in the table for\n  // backlinks/peek.\n  const sql = `SELECT src, dst FROM links WHERE dst IS NOT NULL AND embed = 0 AND src != dst\n    UNION ALL\n    SELECT src, dst FROM links l WHERE dst IS NOT NULL AND embed = 1 AND src != dst\n      AND NOT EXISTS (SELECT 1 FROM links l0 WHERE l0.src = l.src AND l0.target = l.target AND l0.embed = 0)`;\n  return (db.prepare(sql).all() as Array<{ src: string; dst: string }>).map((r) => [r.src, r.dst]);\n}\n\n// remove() has already deleted the rows by the time afterReconcile runs, so it records here\n// whether they carried edges. Keyed on the delta object, so the state dies with the reconcile.\nconst removedWithLinks = new WeakMap<ReconcileDelta, Set<string>>();\n\nfunction recordRemoved(delta: ReconcileDelta, path: string): void {\n  let set = removedWithLinks.get(delta);\n  if (!set) {\n    set = new Set();\n    removedWithLinks.set(delta, set);\n  }\n  set.add(path);\n}\n\n// remove() runs per path; the vanished list is an array, so cache the Set per delta.\nconst vanishedSets = new WeakMap<ReconcileDelta, Set<string>>();\nfunction vanishedSet(delta: ReconcileDelta): Set<string> {\n  let set = vanishedSets.get(delta);\n  if (!set) {\n    set = new Set(delta.vanished);\n    vanishedSets.set(delta, set);\n  }\n  return set;\n}\n\nexport const links: Feature = {\n  name: 'links',\n  schema(db) {\n    db.exec('CREATE TABLE IF NOT EXISTS links (src TEXT, target TEXT, target_base TEXT, dst TEXT, embed INTEGER, PRIMARY KEY (src, target, embed))');\n    db.exec('CREATE INDEX IF NOT EXISTS links_dst ON links(dst)');\n    db.exec('CREATE INDEX IF NOT EXISTS links_target_base ON links(target_base)');\n  },\n  extract,\n  // Reparsed docs are diffed in store() below instead of wiped here: deleting and\n  // re-inserting resets dst to NULL, which made every touch-only reparse read as an edge\n  // change and recompute PageRank -- measured as most of the remaining update cost.\n  remove(db, path, delta) {\n    if (!vanishedSet(delta).has(path)) return;\n    const result = db.prepare('DELETE FROM links WHERE src = ?').run(path);\n    if (Number(result.changes) > 0) recordRemoved(delta, path);\n  },\n  store(db, path, extracted, delta) {\n    const targets = extracted as Array<{ target: string; embed: boolean }>;\n    // Stale rows (target/embed pairs the new parse no longer contains) are real edge removals.\n    // \\0 can't appear in a target, so the composite key never collides.\n    let stale: ReturnType<ReturnType<DatabaseSync['prepare']>['run']>;\n    if (targets.length === 0) {\n      stale = db.prepare('DELETE FROM links WHERE src = ?').run(path);\n    } else {\n      const placeholders = targets.map(() => '?').join(', ');\n      const keys = targets.map((t) => `${t.target}\\0${t.embed ? '1' : '0'}`);\n      stale = db.prepare(`DELETE FROM links WHERE src = ? AND (target || char(0) || embed) NOT IN (${placeholders})`).run(path, ...keys);\n    }\n    if (Number(stale.changes) > 0) recordRemoved(delta, path);\n    // Upsert preserves dst on surviving rows, so an unchanged link resolves to the same\n    // value and reports no change; only genuinely new rows start at NULL.\n    const insert = db.prepare('INSERT INTO links (src, target, target_base, dst, embed) VALUES (?, ?, ?, NULL, ?) ON CONFLICT(src, target, embed) DO UPDATE SET target_base = excluded.target_base');\n    // A same-note anchor's dst is always its own src, never another file's basename, so it\n    // gets no target_base -- SQL's `IN` never matches NULL, keeping it out of\n    // resolveIncremental's cross-file collectIn('target_base', ...) entirely.\n    for (const { target, embed } of targets) insert.run(path, target, target.startsWith('#') ? null : baseKey(cleanTarget(target)), embed ? 1 : 0);\n  },\n  afterReconcile(db, delta) {\n    // Any deleted rows that carried edges mean the edge set shrank -- whether the file\n    // vanished or was reparsed down to fewer/no links. dstChanged alone misses the\n    // reparsed-to-zero-links case (no surviving rows to re-resolve).\n    const removeHadLinks = (removedWithLinks.get(delta)?.size ?? 0) > 0;\n\n    const churn = delta.reparsed.length + delta.vanished.length;\n    // Past this share of the tree, a full pass is the only one guaranteed to match\n    // resolveTarget's ambiguity rules. A cold build clears the threshold on its own.\n    const large = delta.files.length === 0 || churn > 0.2 * delta.files.length;\n\n    const dstChanged = large ? resolveAll(db, delta.files) : resolveIncremental(db, delta);\n    delta.linksChanged = dstChanged || removeHadLinks;\n  },\n};\n"],"names":["posix","maskRegions","parseWikilinkInner","inner","beforeAlias","split","trim","startsWith","length","target","extract","_raw","rawBody","_search","data","body","test","seen","Set","results","add","embed","key","has","push","m","matchAll","index","dest","includes","value","Object","values","item","Array","isArray","exec","cleanTarget","replace","baseKey","path","basename","toLowerCase","resolveTarget","src","pathSet","byBase","clean","fromSrc","normalize","join","dirname","candidate","candidates","get","buildByBase","files","Map","paths","map","f","relPath","sort","a","b","list","set","resolveRows","db","rows","update","prepare","changed","row","dst","run","resolveAll","all","resolveIncremental","delta","changedBasenames","p","added","vanished","collect","collectIn","column","keys","i","chunk","slice","placeholders","reparsed","linkEdges","sql","r","removedWithLinks","WeakMap","recordRemoved","vanishedSets","vanishedSet","links","name","schema","remove","result","Number","changes","store","extracted","targets","stale","t","insert","afterReconcile","removeHadLinks","size","churn","large","dstChanged","linksChanged"],"mappings":"AAAA,OAAOA,WAAW,kBAAkB;AAEpC,SAASC,WAAW,QAAQ,eAAe;AAI3C,0FAA0F;AAC1F,iFAAiF;AACjF,+FAA+F;AAC/F,uFAAuF;AAEvF,wFAAwF;AACxF,6FAA6F;AAC7F,uDAAuD;AACvD,uFAAuF;AACvF,yFAAyF;AACzF,4CAA4C;AAC5C,SAASC,mBAAmBC,KAAa;IACvC,MAAMC,cAAcD,MAAME,KAAK,CAAC,IAAI,CAAC,EAAE,CAACC,IAAI;IAC5C,IAAIF,YAAYG,UAAU,CAAC,MAAM,OAAOH,YAAYI,MAAM,GAAG,IAAIJ,cAAc;IAC/E,MAAMK,SAASL,YAAYC,KAAK,CAAC,IAAI,CAAC,EAAE,CAACC,IAAI;IAC7C,OAAOG,UAAU;AACnB;AAEA,SAASC,QAAQC,IAAY,EAAEC,OAAe,EAAEC,OAA4C,EAAEC,IAA8B;IAC1H,MAAMC,OAAO,YAAYC,IAAI,CAACJ,WAAWX,YAAYW,WAAWA;IAChE,MAAMK,OAAO,IAAIC;IACjB,MAAMC,UAAqD,EAAE;IAC7D,MAAMC,MAAM,CAACX,QAAgBY;QAC3B,MAAMC,MAAM,GAAGb,OAAO,EAAE,EAAEY,QAAQ,MAAM,KAAK;QAC7C,IAAIJ,KAAKM,GAAG,CAACD,MAAM;QACnBL,KAAKG,GAAG,CAACE;QACTH,QAAQK,IAAI,CAAC;YAAEf;YAAQY;QAAM;IAC/B;IACA,uFAAuF;IACvF,6DAA6D;IAC7D,KAAK,MAAMI,KAAKV,KAAKW,QAAQ,CAAC,kBAAmB;YAEjBD;QAD9B,MAAMhB,SAASP,mBAAmBuB,CAAC,CAAC,EAAE;QACtC,IAAIhB,QAAQW,IAAIX,QAAQM,IAAI,CAAC,EAACU,WAAAA,EAAEE,KAAK,cAAPF,sBAAAA,WAAW,KAAK,EAAE,KAAK;IACvD;IACA,sFAAsF;IACtF,mFAAmF;IACnF,wFAAwF;IACxF,2DAA2D;IAC3D,0FAA0F;IAC1F,mEAAmE;IACnE,KAAK,MAAMA,KAAKV,KAAKW,QAAQ,CAAC,6EAA8E;QAC1G,MAAME,OAAOH,CAAC,CAAC,EAAE,CAACnB,IAAI;QACtB,qFAAqF;QACrF,oFAAoF;QACpF,IAAI,CAACsB,QAAQA,KAAKC,QAAQ,CAAC,QAAQ;QACnC,IAAI,oCAAoCb,IAAI,CAACY,OAAO;QACpD,MAAMnB,SAASmB,KAAKrB,UAAU,CAAC,OAAOqB,OAAOA,KAAKvB,KAAK,CAAC,IAAI,CAAC,EAAE;QAC/D,IAAII,QAAQW,IAAIX,QAAQgB,CAAC,CAAC,EAAE,KAAK;IACnC;IACA,KAAK,MAAMK,SAASC,OAAOC,MAAM,CAAClB,iBAAAA,kBAAAA,OAAQ,CAAC,GAAI;QAC7C,KAAK,MAAMmB,QAAQC,MAAMC,OAAO,CAACL,SAASA,QAAQ;YAACA;SAAM,CAAE;YACzD,IAAI,OAAOG,SAAS,UAAU;YAC9B,MAAMR,IAAI,iBAAiBW,IAAI,CAACH,KAAK3B,IAAI;YACzC,IAAI,CAACmB,GAAG;YACR,MAAMhB,SAASP,mBAAmBuB,CAAC,CAAC,EAAE;YACtC,IAAIhB,QAAQW,IAAIX,QAAQ;QAC1B;IACF;IACA,OAAOU;AACT;AAEA,SAASkB,YAAY5B,MAAc;IACjC,OAAOA,OAAO6B,OAAO,CAAC,OAAO,KAAKA,OAAO,CAAC,SAAS;AACrD;AAEA,SAASC,QAAQC,IAAY;IAC3B,OAAOxC,MAAMyC,QAAQ,CAACD,MAAMF,OAAO,CAAC,UAAU,IAAII,WAAW;AAC/D;AAEA,uFAAuF;AACvF,0FAA0F;AAC1F,kGAAkG;AAClG,SAASC,cAAcC,GAAW,EAAEnC,MAAc,EAAEoC,OAAoB,EAAEC,MAA6B;IACrG,4FAA4F;IAC5F,IAAIrC,OAAOF,UAAU,CAAC,MAAM,OAAOqC;IACnC,MAAMG,QAAQV,YAAY5B;IAC1B,MAAMuC,UAAUhD,MAAMiD,SAAS,CAACjD,MAAMkD,IAAI,CAAClD,MAAMmD,OAAO,CAACP,MAAMG;IAC/D,KAAK,MAAMK,aAAa;QAACL;QAAO,GAAGA,MAAM,GAAG,CAAC;QAAEC;QAAS,GAAGA,QAAQ,GAAG,CAAC;KAAC,CAAE;QACxE,IAAIH,QAAQtB,GAAG,CAAC6B,YAAY,OAAOA;IACrC;IACA,MAAMC,aAAaP,OAAOQ,GAAG,CAACf,QAAQQ;IACtC,IAAI,CAACM,YAAY,OAAO;IACxB,0FAA0F;IAC1F,+DAA+D;IAC/D,OAAOA,WAAWxB,QAAQ,CAACe,OAAOA,MAAMS,UAAU,CAAC,EAAE;AACvD;AAEA,6FAA6F;AAC7F,yFAAyF;AACzF,SAASE,YAAYC,KAAiB;IACpC,MAAMV,SAAS,IAAIW;IACnB,MAAMC,QAAQF,MAAMG,GAAG,CAAC,CAACC,IAAMA,EAAEC,OAAO,EAAEC,IAAI,CAAC,CAACC,GAAGC,IAAMD,EAAEvD,MAAM,GAAGwD,EAAExD,MAAM,IAAKuD,CAAAA,IAAIC,IAAI,CAAC,IAAID,IAAIC,IAAI,IAAI,CAAA;IAC1G,KAAK,MAAMxB,QAAQkB,MAAO;QACxB,MAAMpC,MAAMiB,QAAQC;QACpB,MAAMyB,OAAOnB,OAAOQ,GAAG,CAAChC;QACxB,IAAI2C,MAAMA,KAAKzC,IAAI,CAACgB;aACfM,OAAOoB,GAAG,CAAC5C,KAAK;YAACkB;SAAK;IAC7B;IACA,OAAOM;AACT;AASA,4FAA4F;AAC5F,0FAA0F;AAC1F,0BAA0B;AAC1B,SAASqB,YAAYC,EAAgB,EAAEC,IAAe,EAAExB,OAAoB,EAAEC,MAA6B;IACzG,MAAMwB,SAASF,GAAGG,OAAO,CAAC;IAC1B,IAAIC,UAAU;IACd,KAAK,MAAMC,OAAOJ,KAAM;QACtB,MAAMK,MAAM/B,cAAc8B,IAAI7B,GAAG,EAAE6B,IAAIhE,MAAM,EAAEoC,SAASC;QACxD,IAAI4B,QAAQD,IAAIC,GAAG,EAAE;YACnBJ,OAAOK,GAAG,CAACD,KAAKD,IAAI7B,GAAG,EAAE6B,IAAIhE,MAAM;YACnC+D,UAAU;QACZ;IACF;IACA,OAAOA;AACT;AAEA,yFAAyF;AACzF,+EAA+E;AAC/E,SAASI,WAAWR,EAAgB,EAAEZ,KAAiB;IACrD,MAAMX,UAAU,IAAI3B,IAAIsC,MAAMG,GAAG,CAAC,CAACC,IAAMA,EAAEC,OAAO;IAClD,MAAMf,SAASS,YAAYC;IAC3B,MAAMa,OAAOD,GAAGG,OAAO,CAAC,6CAA6CM,GAAG;IACxE,OAAOV,YAAYC,IAAIC,MAAMxB,SAASC;AACxC;AAEA,yFAAyF;AACzF,yFAAyF;AACzF,kEAAkE;AAClE,SAASgC,mBAAmBV,EAAgB,EAAEW,KAAqB;IACjE,MAAMlC,UAAU,IAAI3B,IAAI6D,MAAMvB,KAAK,CAACG,GAAG,CAAC,CAACC,IAAMA,EAAEC,OAAO;IACxD,MAAMf,SAASS,YAAYwB,MAAMvB,KAAK;IAEtC,MAAMwB,mBAAmB,IAAI9D;IAC7B,KAAK,MAAM+D,KAAKF,MAAMG,KAAK,CAAEF,iBAAiB5D,GAAG,CAACmB,QAAQ0C;IAC1D,KAAK,MAAMA,KAAKF,MAAMI,QAAQ,CAAEH,iBAAiB5D,GAAG,CAACmB,QAAQ0C;IAE7D,+EAA+E;IAC/E,MAAM5B,aAAa,IAAII;IACvB,MAAM2B,UAAU,CAACf;QACf,KAAK,MAAMI,OAAOJ,KAAMhB,WAAWa,GAAG,CAAC,GAAGO,IAAI7B,GAAG,CAAC,EAAE,EAAE6B,IAAIhE,MAAM,CAAC,EAAE,EAAEgE,IAAIpD,KAAK,EAAE,EAAEoD;IACpF;IACA,qFAAqF;IACrF,6CAA6C;IAC7C,MAAMY,YAAY,CAACC,QAAgBC;QACjC,IAAK,IAAIC,IAAI,GAAGA,IAAID,KAAK/E,MAAM,EAAEgF,KAAK,IAAK;YACzC,MAAMC,QAAQF,KAAKG,KAAK,CAACF,GAAGA,IAAI;YAChC,MAAMG,eAAeF,MAAM9B,GAAG,CAAC,IAAM,KAAKT,IAAI,CAAC;YAC/CkC,QAAQhB,GAAGG,OAAO,CAAC,CAAC,gDAAgD,EAAEe,OAAO,KAAK,EAAEK,aAAa,CAAC,CAAC,EAAEd,GAAG,IAAIY;QAC9G;IACF;IAEAJ,UAAU,OAAON,MAAMa,QAAQ;IAC/BP,UAAU,eAAe;WAAIL;KAAiB;IAC9CK,UAAU,OAAON,MAAMI,QAAQ;IAE/B,OAAOhB,YAAYC,IAAI;WAAIf,WAAWrB,MAAM;KAAG,EAAEa,SAASC;AAC5D;AAEA,6DAA6D;AAC7D,OAAO,SAAS+C,UAAUzB,EAAgB;IACxC,0FAA0F;IAC1F,yFAAyF;IACzF,yFAAyF;IACzF,0FAA0F;IAC1F,iFAAiF;IACjF,4FAA4F;IAC5F,4FAA4F;IAC5F,kBAAkB;IAClB,MAAM0B,MAAM,CAAC;;;4GAG6F,CAAC;IAC3G,OAAO,AAAC1B,GAAGG,OAAO,CAACuB,KAAKjB,GAAG,GAA2ClB,GAAG,CAAC,CAACoC,IAAM;YAACA,EAAEnD,GAAG;YAAEmD,EAAErB,GAAG;SAAC;AACjG;AAEA,4FAA4F;AAC5F,+FAA+F;AAC/F,MAAMsB,mBAAmB,IAAIC;AAE7B,SAASC,cAAcnB,KAAqB,EAAEvC,IAAY;IACxD,IAAI0B,MAAM8B,iBAAiB1C,GAAG,CAACyB;IAC/B,IAAI,CAACb,KAAK;QACRA,MAAM,IAAIhD;QACV8E,iBAAiB9B,GAAG,CAACa,OAAOb;IAC9B;IACAA,IAAI9C,GAAG,CAACoB;AACV;AAEA,qFAAqF;AACrF,MAAM2D,eAAe,IAAIF;AACzB,SAASG,YAAYrB,KAAqB;IACxC,IAAIb,MAAMiC,aAAa7C,GAAG,CAACyB;IAC3B,IAAI,CAACb,KAAK;QACRA,MAAM,IAAIhD,IAAI6D,MAAMI,QAAQ;QAC5BgB,aAAajC,GAAG,CAACa,OAAOb;IAC1B;IACA,OAAOA;AACT;AAEA,OAAO,MAAMmC,QAAiB;IAC5BC,MAAM;IACNC,QAAOnC,EAAE;QACPA,GAAGhC,IAAI,CAAC;QACRgC,GAAGhC,IAAI,CAAC;QACRgC,GAAGhC,IAAI,CAAC;IACV;IACA1B;IACA,gFAAgF;IAChF,uFAAuF;IACvF,kFAAkF;IAClF8F,QAAOpC,EAAE,EAAE5B,IAAI,EAAEuC,KAAK;QACpB,IAAI,CAACqB,YAAYrB,OAAOxD,GAAG,CAACiB,OAAO;QACnC,MAAMiE,SAASrC,GAAGG,OAAO,CAAC,mCAAmCI,GAAG,CAACnC;QACjE,IAAIkE,OAAOD,OAAOE,OAAO,IAAI,GAAGT,cAAcnB,OAAOvC;IACvD;IACAoE,OAAMxC,EAAE,EAAE5B,IAAI,EAAEqE,SAAS,EAAE9B,KAAK;QAC9B,MAAM+B,UAAUD;QAChB,2FAA2F;QAC3F,oEAAoE;QACpE,IAAIE;QACJ,IAAID,QAAQtG,MAAM,KAAK,GAAG;YACxBuG,QAAQ3C,GAAGG,OAAO,CAAC,mCAAmCI,GAAG,CAACnC;QAC5D,OAAO;YACL,MAAMmD,eAAemB,QAAQnD,GAAG,CAAC,IAAM,KAAKT,IAAI,CAAC;YACjD,MAAMqC,OAAOuB,QAAQnD,GAAG,CAAC,CAACqD,IAAM,GAAGA,EAAEvG,MAAM,CAAC,EAAE,EAAEuG,EAAE3F,KAAK,GAAG,MAAM,KAAK;YACrE0F,QAAQ3C,GAAGG,OAAO,CAAC,CAAC,yEAAyE,EAAEoB,aAAa,CAAC,CAAC,EAAEhB,GAAG,CAACnC,SAAS+C;QAC/H;QACA,IAAImB,OAAOK,MAAMJ,OAAO,IAAI,GAAGT,cAAcnB,OAAOvC;QACpD,oFAAoF;QACpF,sEAAsE;QACtE,MAAMyE,SAAS7C,GAAGG,OAAO,CAAC;QAC1B,uFAAuF;QACvF,0EAA0E;QAC1E,0EAA0E;QAC1E,KAAK,MAAM,EAAE9D,MAAM,EAAEY,KAAK,EAAE,IAAIyF,QAASG,OAAOtC,GAAG,CAACnC,MAAM/B,QAAQA,OAAOF,UAAU,CAAC,OAAO,OAAOgC,QAAQF,YAAY5B,UAAUY,QAAQ,IAAI;IAC9I;IACA6F,gBAAe9C,EAAE,EAAEW,KAAK;;YAIEiB;QAHxB,mFAAmF;QACnF,+EAA+E;QAC/E,iEAAiE;QACjE,MAAMmB,iBAAiB,UAACnB,wBAAAA,iBAAiB1C,GAAG,CAACyB,oBAArBiB,4CAAAA,sBAA6BoB,IAAI,uCAAI,KAAK;QAElE,MAAMC,QAAQtC,MAAMa,QAAQ,CAACpF,MAAM,GAAGuE,MAAMI,QAAQ,CAAC3E,MAAM;QAC3D,+EAA+E;QAC/E,iFAAiF;QACjF,MAAM8G,QAAQvC,MAAMvB,KAAK,CAAChD,MAAM,KAAK,KAAK6G,QAAQ,MAAMtC,MAAMvB,KAAK,CAAChD,MAAM;QAE1E,MAAM+G,aAAaD,QAAQ1C,WAAWR,IAAIW,MAAMvB,KAAK,IAAIsB,mBAAmBV,IAAIW;QAChFA,MAAMyC,YAAY,GAAGD,cAAcJ;IACrC;AACF,EAAE"}

package/dist/esm/features/tags.js

@@ -1,4 +1,4 @@
-import { fenceTracker } from '../fences.js';
+import { commentTracker, fenceTracker, maskCodeSpans } from '../fences.js';
 // tags(path, tag): Obsidian's file.tags grain -- frontmatter list/string tags plus inline
 // #tags from the prose, deduplicated, source not distinguished. Nested tags store full
 // (book/scifi); `tag = 'book' OR tag LIKE 'book/%'` is how a caller matches the parent too.
@@ -109,53 +109,12 @@ function frontmatterTags(data) {
     }
     return found;
 }
-// A code span opens on a run of N backticks and closes at the next run of exactly N -- a
-// shorter or longer run in between is literal text, not a delimiter (CommonMark code spans).
-// Masked with spaces so column positions and tag-boundary whitespace are unaffected.
-function maskCodeSpans(line) {
-    let out = '';
-    let i = 0;
-    while(i < line.length){
-        if (line[i] !== '`') {
-            out += line[i];
-            i++;
-            continue;
-        }
-        let j = i;
-        while(line[j] === '`')j++;
-        const n = j - i;
-        let k = j;
-        let closeStart = -1;
-        let closeEnd = -1;
-        while(k < line.length){
-            if (line[k] !== '`') {
-                k++;
-                continue;
-            }
-            let m = k;
-            while(line[m] === '`')m++;
-            if (m - k === n) {
-                closeStart = k;
-                closeEnd = m;
-                break;
-            }
-            k = m;
-        }
-        if (closeStart >= 0) {
-            out += ' '.repeat(closeEnd - i);
-            i = closeEnd;
-        } else {
-            out += line.slice(i, j);
-            i = j;
-        }
-    }
-    return out;
-}
 // #tag tokens outside fenced code blocks, inline code spans, wikilinks, HTML tags, HTML blocks,
-// and link destinations.
+// <!-- --> comments, and link destinations.
 function inlineTags(body) {
     const found = [];
     const fence = fenceTracker();
+    const comment = commentTracker();
     let inHtmlBlock = false;
     let htmlBlockClose = null; // set while inside a type-1 (script/pre/style/textarea) block
     for (const line of body.split('\n')){
@@ -171,11 +130,17 @@ function inlineTags(body) {
             }
             continue;
         }
-        if (fence.feed(line)) continue;
-        if (fence.inFence) continue;
+        if (!comment.inComment) {
+            if (fence.feed(line)) continue;
+            if (fence.inFence) continue;
+        }
+        // Obsidian parses nothing inside a <!-- --> comment; mask its span (which may open, close,
+        // or run the whole line) before any other check sees this line's text. Skipped when there
+        // is no comment marker anywhere near this line -- most lines, so worth the branch.
+        const masked = comment.inComment || line.includes('<!--') ? comment.mask(line) : line;
         // Indented or blockquoted HTML blocks still swallow their content in Obsidian, so the
         // opener test runs after stripping leading whitespace and > markers.
-        const stripped = line.replace(/^[ \t>]*/, '');
+        const stripped = masked.replace(/^[ \t>]*/, '');
         if (stripped[0] === '<') {
             const openMatch = HTML_BLOCK_OPEN_RE.exec(stripped);
             if (openMatch) {
@@ -193,8 +158,8 @@ function inlineTags(body) {
                 }
             }
         }
-        if (!line.includes('#')) continue; // most lines; skip the regex work
-        let cleaned = line.includes('`') ? maskCodeSpans(line) : line;
+        if (!masked.includes('#')) continue; // most lines; skip the regex work
+        let cleaned = masked.includes('`') ? maskCodeSpans(masked) : masked;
         if (cleaned.includes('[[')) cleaned = cleaned.replace(WIKILINK_RE, (m)=>' '.repeat(m.length));
         if (cleaned.includes('](')) cleaned = cleaned.replace(LINK_DEST_RE, (m)=>' '.repeat(m.length));
         if (cleaned.includes('<')) cleaned = cleaned.replace(HTML_TAG_RE, (m)=>' '.repeat(m.length));

package/dist/esm/features/tags.js.map

@@ -1 +1 @@
-{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/features/tags.ts"],"sourcesContent":["import { fenceTracker } from '../fences.ts';\nimport type { Feature } from './types.ts';\n\n// tags(path, tag): Obsidian's file.tags grain -- frontmatter list/string tags plus inline\n// #tags from the prose, deduplicated, source not distinguished. Nested tags store full\n// (book/scifi); `tag = 'book' OR tag LIKE 'book/%'` is how a caller matches the parent too.\n\n// Obsidian treats [[#Heading]] as a same-note link, not a tag.\nconst WIKILINK_RE = /\\[\\[.*?\\]\\]/g; // to the first ]], so a heading holding a lone ] still masks\n// Obsidian doesn't read tags inside HTML markup.\nconst HTML_TAG_RE = /<\\/?[a-zA-Z][^>]*>/g; // tag-shaped only: a comparison's `< 5` must not open a span\n// Anchors on start-of-line or a preceding whitespace/(/[ so `a#b` and URL fragments don't count.\nconst INLINE_TAG_RE = /(?:^|[\\s([])#([\\p{L}\\p{N}_/-]+)/gu;\n// A markdown link destination `](...)` -- `[text](#anchor)` is a same-page link, not a tag.\nconst LINK_DEST_RE = /\\]\\((?:[^()]|\\([^()]*\\))*\\)/g; // one paren-nesting level, as CommonMark destinations allow: (https://x/a_(b)#frag)\n\n// CommonMark's HTML-block type-6 list (fixed by the spec, not a drifting enumeration): a line\n// starting with an open or close tag of one of these, at column 0, opens a block that swallows\n// following lines -- including any #tag in them -- until a blank line closes it.\nconst HTML_BLOCK_TAGS = new Set([\n  'address',\n  'article',\n  'aside',\n  'base',\n  'basefont',\n  'blockquote',\n  'body',\n  'caption',\n  'center',\n  'col',\n  'colgroup',\n  'dd',\n  'details',\n  'dialog',\n  'dir',\n  'div',\n  'dl',\n  'dt',\n  'fieldset',\n  'figcaption',\n  'figure',\n  'footer',\n  'form',\n  'frame',\n  'frameset',\n  'h1',\n  'h2',\n  'h3',\n  'h4',\n  'h5',\n  'h6',\n  'head',\n  'header',\n  'hr',\n  'html',\n  'iframe',\n  'legend',\n  'li',\n  'link',\n  'main',\n  'menu',\n  'menuitem',\n  'nav',\n  'noframes',\n  'ol',\n  'optgroup',\n  'option',\n  'p',\n  'param',\n  'search',\n  'section',\n  'summary',\n  'table',\n  'tbody',\n  'td',\n  'tfoot',\n  'th',\n  'thead',\n  'title',\n  'tr',\n  'track',\n  'ul',\n]);\n// Type-1 blocks (script/pre/style/textarea): closes on the line holding the matching end tag,\n// not on a blank line, and that line is the last one skipped.\nconst HTML_PRE_TAGS = new Set(['script', 'pre', 'style', 'textarea']);\n// An opening or closing tag at column 0, tag name captured for the lookups above.\nconst HTML_BLOCK_OPEN_RE = /^<\\/?([a-zA-Z][a-zA-Z0-9]*)(?:[ \\t]|\\/?>|$)/;\n\n// Strips a leading # (frontmatter entries may carry one) and a trailing /; rejects an\n// all-digit result -- a tag needs at least one non-digit character.\nfunction normalizeTag(raw: string): string | null {\n  const stripped = raw.replace(/^#/, '').replace(/\\/+$/, '');\n  if (!stripped || /^\\d+$/.test(stripped)) return null;\n  return stripped;\n}\n\n// data.tags: a YAML list (Obsidian also accepts a bare string). Null members and non-string\n// members are skipped rather than throwing -- `tags:\\n  -` parses to [null].\nfunction frontmatterTags(data?: Record<string, unknown>): string[] {\n  const raw = data?.tags;\n  const list = Array.isArray(raw) ? raw : typeof raw === 'string' ? [raw] : [];\n  const found: string[] = [];\n  for (const item of list) {\n    if (typeof item !== 'string') continue;\n    const tag = normalizeTag(item);\n    if (tag) found.push(tag);\n  }\n  return found;\n}\n\n// A code span opens on a run of N backticks and closes at the next run of exactly N -- a\n// shorter or longer run in between is literal text, not a delimiter (CommonMark code spans).\n// Masked with spaces so column positions and tag-boundary whitespace are unaffected.\nfunction maskCodeSpans(line: string): string {\n  let out = '';\n  let i = 0;\n  while (i < line.length) {\n    if (line[i] !== '`') {\n      out += line[i];\n      i++;\n      continue;\n    }\n    let j = i;\n    while (line[j] === '`') j++;\n    const n = j - i;\n    let k = j;\n    let closeStart = -1;\n    let closeEnd = -1;\n    while (k < line.length) {\n      if (line[k] !== '`') {\n        k++;\n        continue;\n      }\n      let m = k;\n      while (line[m] === '`') m++;\n      if (m - k === n) {\n        closeStart = k;\n        closeEnd = m;\n        break;\n      }\n      k = m;\n    }\n    if (closeStart >= 0) {\n      out += ' '.repeat(closeEnd - i);\n      i = closeEnd;\n    } else {\n      out += line.slice(i, j);\n      i = j;\n    }\n  }\n  return out;\n}\n\n// #tag tokens outside fenced code blocks, inline code spans, wikilinks, HTML tags, HTML blocks,\n// and link destinations.\nfunction inlineTags(body: string): string[] {\n  const found: string[] = [];\n  const fence = fenceTracker();\n  let inHtmlBlock = false;\n  let htmlBlockClose: RegExp | null = null; // set while inside a type-1 (script/pre/style/textarea) block\n  for (const line of body.split('\\n')) {\n    if (inHtmlBlock) {\n      // A fence-like line here is still HTML-block content -- the block wins until it closes.\n      if (htmlBlockClose) {\n        if (htmlBlockClose.test(line)) {\n          inHtmlBlock = false;\n          htmlBlockClose = null;\n        }\n      } else if (/^[ \\t>]*$/.test(line)) {\n        inHtmlBlock = false;\n      }\n      continue;\n    }\n    if (fence.feed(line)) continue;\n    if (fence.inFence) continue;\n    // Indented or blockquoted HTML blocks still swallow their content in Obsidian, so the\n    // opener test runs after stripping leading whitespace and > markers.\n    const stripped = line.replace(/^[ \\t>]*/, '');\n    if (stripped[0] === '<') {\n      const openMatch = HTML_BLOCK_OPEN_RE.exec(stripped);\n      if (openMatch) {\n        const tagName = openMatch[1].toLowerCase();\n        const isClosingTag = stripped[1] === '/';\n        if (!isClosingTag && HTML_PRE_TAGS.has(tagName)) {\n          inHtmlBlock = true;\n          // Any of the four type-1 closers ends the block, not only the tag that opened it.\n          htmlBlockClose = /<\\/(?:script|pre|style|textarea)>/i;\n          continue;\n        }\n        if (HTML_BLOCK_TAGS.has(tagName)) {\n          inHtmlBlock = true;\n          continue;\n        }\n      }\n    }\n    if (!line.includes('#')) continue; // most lines; skip the regex work\n    let cleaned = line.includes('`') ? maskCodeSpans(line) : line;\n    if (cleaned.includes('[[')) cleaned = cleaned.replace(WIKILINK_RE, (m) => ' '.repeat(m.length));\n    if (cleaned.includes('](')) cleaned = cleaned.replace(LINK_DEST_RE, (m) => ' '.repeat(m.length));\n    if (cleaned.includes('<')) cleaned = cleaned.replace(HTML_TAG_RE, (m) => ' '.repeat(m.length));\n    for (const m of cleaned.matchAll(INLINE_TAG_RE)) {\n      const tag = normalizeTag(m[1]);\n      if (tag) found.push(tag);\n    }\n  }\n  return found;\n}\n\nfunction extract(_raw: string, body: string, _search?: { title: string; summary: string }, data?: Record<string, unknown>): string[] {\n  return [...new Set([...frontmatterTags(data), ...inlineTags(body)])].sort();\n}\n\nexport const tags: Feature = {\n  name: 'tags',\n  schema(db) {\n    db.exec('CREATE TABLE IF NOT EXISTS tags (\"path\" TEXT, tag TEXT, PRIMARY KEY (\"path\", tag))');\n    db.exec('CREATE INDEX IF NOT EXISTS tags_tag ON tags(tag)');\n  },\n  extract,\n  // Per-file rows with nothing else to resolve, so only a vanished file needs a delete here;\n  // store() below handles a reparse's stale rows itself.\n  remove(db, path, delta) {\n    if (!delta.vanished.includes(path)) return;\n    db.prepare('DELETE FROM tags WHERE \"path\" = ?').run(path);\n  },\n  store(db, path, extracted) {\n    const found = extracted as string[];\n    if (found.length === 0) {\n      db.prepare('DELETE FROM tags WHERE \"path\" = ?').run(path);\n    } else {\n      const placeholders = found.map(() => '?').join(', ');\n      db.prepare(`DELETE FROM tags WHERE \"path\" = ? AND tag NOT IN (${placeholders})`).run(path, ...found);\n    }\n    const insert = db.prepare('INSERT OR IGNORE INTO tags (\"path\", tag) VALUES (?, ?)');\n    for (const tag of found) insert.run(path, tag);\n  },\n};\n"],"names":["fenceTracker","WIKILINK_RE","HTML_TAG_RE","INLINE_TAG_RE","LINK_DEST_RE","HTML_BLOCK_TAGS","Set","HTML_PRE_TAGS","HTML_BLOCK_OPEN_RE","normalizeTag","raw","stripped","replace","test","frontmatterTags","data","tags","list","Array","isArray","found","item","tag","push","maskCodeSpans","line","out","i","length","j","n","k","closeStart","closeEnd","m","repeat","slice","inlineTags","body","fence","inHtmlBlock","htmlBlockClose","split","feed","inFence","openMatch","exec","tagName","toLowerCase","isClosingTag","has","includes","cleaned","matchAll","extract","_raw","_search","sort","name","schema","db","remove","path","delta","vanished","prepare","run","store","extracted","placeholders","map","join","insert"],"mappings":"AAAA,SAASA,YAAY,QAAQ,eAAe;AAG5C,0FAA0F;AAC1F,uFAAuF;AACvF,4FAA4F;AAE5F,+DAA+D;AAC/D,MAAMC,cAAc,gBAAgB,6DAA6D;AACjG,iDAAiD;AACjD,MAAMC,cAAc,uBAAuB,6DAA6D;AACxG,iGAAiG;AACjG,MAAMC,gBAAgB;AACtB,4FAA4F;AAC5F,MAAMC,eAAe,gCAAgC,oFAAoF;AAEzI,8FAA8F;AAC9F,+FAA+F;AAC/F,iFAAiF;AACjF,MAAMC,kBAAkB,IAAIC,IAAI;IAC9B;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;CACD;AACD,8FAA8F;AAC9F,8DAA8D;AAC9D,MAAMC,gBAAgB,IAAID,IAAI;IAAC;IAAU;IAAO;IAAS;CAAW;AACpE,kFAAkF;AAClF,MAAME,qBAAqB;AAE3B,sFAAsF;AACtF,oEAAoE;AACpE,SAASC,aAAaC,GAAW;IAC/B,MAAMC,WAAWD,IAAIE,OAAO,CAAC,MAAM,IAAIA,OAAO,CAAC,QAAQ;IACvD,IAAI,CAACD,YAAY,QAAQE,IAAI,CAACF,WAAW,OAAO;IAChD,OAAOA;AACT;AAEA,4FAA4F;AAC5F,6EAA6E;AAC7E,SAASG,gBAAgBC,IAA8B;IACrD,MAAML,MAAMK,iBAAAA,2BAAAA,KAAMC,IAAI;IACtB,MAAMC,OAAOC,MAAMC,OAAO,CAACT,OAAOA,MAAM,OAAOA,QAAQ,WAAW;QAACA;KAAI,GAAG,EAAE;IAC5E,MAAMU,QAAkB,EAAE;IAC1B,KAAK,MAAMC,QAAQJ,KAAM;QACvB,IAAI,OAAOI,SAAS,UAAU;QAC9B,MAAMC,MAAMb,aAAaY;QACzB,IAAIC,KAAKF,MAAMG,IAAI,CAACD;IACtB;IACA,OAAOF;AACT;AAEA,yFAAyF;AACzF,6FAA6F;AAC7F,qFAAqF;AACrF,SAASI,cAAcC,IAAY;IACjC,IAAIC,MAAM;IACV,IAAIC,IAAI;IACR,MAAOA,IAAIF,KAAKG,MAAM,CAAE;QACtB,IAAIH,IAAI,CAACE,EAAE,KAAK,KAAK;YACnBD,OAAOD,IAAI,CAACE,EAAE;YACdA;YACA;QACF;QACA,IAAIE,IAAIF;QACR,MAAOF,IAAI,CAACI,EAAE,KAAK,IAAKA;QACxB,MAAMC,IAAID,IAAIF;QACd,IAAII,IAAIF;QACR,IAAIG,aAAa,CAAC;QAClB,IAAIC,WAAW,CAAC;QAChB,MAAOF,IAAIN,KAAKG,MAAM,CAAE;YACtB,IAAIH,IAAI,CAACM,EAAE,KAAK,KAAK;gBACnBA;gBACA;YACF;YACA,IAAIG,IAAIH;YACR,MAAON,IAAI,CAACS,EAAE,KAAK,IAAKA;YACxB,IAAIA,IAAIH,MAAMD,GAAG;gBACfE,aAAaD;gBACbE,WAAWC;gBACX;YACF;YACAH,IAAIG;QACN;QACA,IAAIF,cAAc,GAAG;YACnBN,OAAO,IAAIS,MAAM,CAACF,WAAWN;YAC7BA,IAAIM;QACN,OAAO;YACLP,OAAOD,KAAKW,KAAK,CAACT,GAAGE;YACrBF,IAAIE;QACN;IACF;IACA,OAAOH;AACT;AAEA,gGAAgG;AAChG,yBAAyB;AACzB,SAASW,WAAWC,IAAY;IAC9B,MAAMlB,QAAkB,EAAE;IAC1B,MAAMmB,QAAQvC;IACd,IAAIwC,cAAc;IAClB,IAAIC,iBAAgC,MAAM,8DAA8D;IACxG,KAAK,MAAMhB,QAAQa,KAAKI,KAAK,CAAC,MAAO;QACnC,IAAIF,aAAa;YACf,wFAAwF;YACxF,IAAIC,gBAAgB;gBAClB,IAAIA,eAAe5B,IAAI,CAACY,OAAO;oBAC7Be,cAAc;oBACdC,iBAAiB;gBACnB;YACF,OAAO,IAAI,YAAY5B,IAAI,CAACY,OAAO;gBACjCe,cAAc;YAChB;YACA;QACF;QACA,IAAID,MAAMI,IAAI,CAAClB,OAAO;QACtB,IAAIc,MAAMK,OAAO,EAAE;QACnB,sFAAsF;QACtF,qEAAqE;QACrE,MAAMjC,WAAWc,KAAKb,OAAO,CAAC,YAAY;QAC1C,IAAID,QAAQ,CAAC,EAAE,KAAK,KAAK;YACvB,MAAMkC,YAAYrC,mBAAmBsC,IAAI,CAACnC;YAC1C,IAAIkC,WAAW;gBACb,MAAME,UAAUF,SAAS,CAAC,EAAE,CAACG,WAAW;gBACxC,MAAMC,eAAetC,QAAQ,CAAC,EAAE,KAAK;gBACrC,IAAI,CAACsC,gBAAgB1C,cAAc2C,GAAG,CAACH,UAAU;oBAC/CP,cAAc;oBACd,kFAAkF;oBAClFC,iBAAiB;oBACjB;gBACF;gBACA,IAAIpC,gBAAgB6C,GAAG,CAACH,UAAU;oBAChCP,cAAc;oBACd;gBACF;YACF;QACF;QACA,IAAI,CAACf,KAAK0B,QAAQ,CAAC,MAAM,UAAU,kCAAkC;QACrE,IAAIC,UAAU3B,KAAK0B,QAAQ,CAAC,OAAO3B,cAAcC,QAAQA;QACzD,IAAI2B,QAAQD,QAAQ,CAAC,OAAOC,UAAUA,QAAQxC,OAAO,CAACX,aAAa,CAACiC,IAAM,IAAIC,MAAM,CAACD,EAAEN,MAAM;QAC7F,IAAIwB,QAAQD,QAAQ,CAAC,OAAOC,UAAUA,QAAQxC,OAAO,CAACR,cAAc,CAAC8B,IAAM,IAAIC,MAAM,CAACD,EAAEN,MAAM;QAC9F,IAAIwB,QAAQD,QAAQ,CAAC,MAAMC,UAAUA,QAAQxC,OAAO,CAACV,aAAa,CAACgC,IAAM,IAAIC,MAAM,CAACD,EAAEN,MAAM;QAC5F,KAAK,MAAMM,KAAKkB,QAAQC,QAAQ,CAAClD,eAAgB;YAC/C,MAAMmB,MAAMb,aAAayB,CAAC,CAAC,EAAE;YAC7B,IAAIZ,KAAKF,MAAMG,IAAI,CAACD;QACtB;IACF;IACA,OAAOF;AACT;AAEA,SAASkC,QAAQC,IAAY,EAAEjB,IAAY,EAAEkB,OAA4C,EAAEzC,IAA8B;IACvH,OAAO;WAAI,IAAIT,IAAI;eAAIQ,gBAAgBC;eAAUsB,WAAWC;SAAM;KAAE,CAACmB,IAAI;AAC3E;AAEA,OAAO,MAAMzC,OAAgB;IAC3B0C,MAAM;IACNC,QAAOC,EAAE;QACPA,GAAGd,IAAI,CAAC;QACRc,GAAGd,IAAI,CAAC;IACV;IACAQ;IACA,2FAA2F;IAC3F,uDAAuD;IACvDO,QAAOD,EAAE,EAAEE,IAAI,EAAEC,KAAK;QACpB,IAAI,CAACA,MAAMC,QAAQ,CAACb,QAAQ,CAACW,OAAO;QACpCF,GAAGK,OAAO,CAAC,qCAAqCC,GAAG,CAACJ;IACtD;IACAK,OAAMP,EAAE,EAAEE,IAAI,EAAEM,SAAS;QACvB,MAAMhD,QAAQgD;QACd,IAAIhD,MAAMQ,MAAM,KAAK,GAAG;YACtBgC,GAAGK,OAAO,CAAC,qCAAqCC,GAAG,CAACJ;QACtD,OAAO;YACL,MAAMO,eAAejD,MAAMkD,GAAG,CAAC,IAAM,KAAKC,IAAI,CAAC;YAC/CX,GAAGK,OAAO,CAAC,CAAC,kDAAkD,EAAEI,aAAa,CAAC,CAAC,EAAEH,GAAG,CAACJ,SAAS1C;QAChG;QACA,MAAMoD,SAASZ,GAAGK,OAAO,CAAC;QAC1B,KAAK,MAAM3C,OAAOF,MAAOoD,OAAON,GAAG,CAACJ,MAAMxC;IAC5C;AACF,EAAE"}
+{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/features/tags.ts"],"sourcesContent":["import { commentTracker, fenceTracker, maskCodeSpans } from '../fences.ts';\nimport type { Feature } from './types.ts';\n\n// tags(path, tag): Obsidian's file.tags grain -- frontmatter list/string tags plus inline\n// #tags from the prose, deduplicated, source not distinguished. Nested tags store full\n// (book/scifi); `tag = 'book' OR tag LIKE 'book/%'` is how a caller matches the parent too.\n\n// Obsidian treats [[#Heading]] as a same-note link, not a tag.\nconst WIKILINK_RE = /\\[\\[.*?\\]\\]/g; // to the first ]], so a heading holding a lone ] still masks\n// Obsidian doesn't read tags inside HTML markup.\nconst HTML_TAG_RE = /<\\/?[a-zA-Z][^>]*>/g; // tag-shaped only: a comparison's `< 5` must not open a span\n// Anchors on start-of-line or a preceding whitespace/(/[ so `a#b` and URL fragments don't count.\nconst INLINE_TAG_RE = /(?:^|[\\s([])#([\\p{L}\\p{N}_/-]+)/gu;\n// A markdown link destination `](...)` -- `[text](#anchor)` is a same-page link, not a tag.\nconst LINK_DEST_RE = /\\]\\((?:[^()]|\\([^()]*\\))*\\)/g; // one paren-nesting level, as CommonMark destinations allow: (https://x/a_(b)#frag)\n\n// CommonMark's HTML-block type-6 list (fixed by the spec, not a drifting enumeration): a line\n// starting with an open or close tag of one of these, at column 0, opens a block that swallows\n// following lines -- including any #tag in them -- until a blank line closes it.\nconst HTML_BLOCK_TAGS = new Set([\n  'address',\n  'article',\n  'aside',\n  'base',\n  'basefont',\n  'blockquote',\n  'body',\n  'caption',\n  'center',\n  'col',\n  'colgroup',\n  'dd',\n  'details',\n  'dialog',\n  'dir',\n  'div',\n  'dl',\n  'dt',\n  'fieldset',\n  'figcaption',\n  'figure',\n  'footer',\n  'form',\n  'frame',\n  'frameset',\n  'h1',\n  'h2',\n  'h3',\n  'h4',\n  'h5',\n  'h6',\n  'head',\n  'header',\n  'hr',\n  'html',\n  'iframe',\n  'legend',\n  'li',\n  'link',\n  'main',\n  'menu',\n  'menuitem',\n  'nav',\n  'noframes',\n  'ol',\n  'optgroup',\n  'option',\n  'p',\n  'param',\n  'search',\n  'section',\n  'summary',\n  'table',\n  'tbody',\n  'td',\n  'tfoot',\n  'th',\n  'thead',\n  'title',\n  'tr',\n  'track',\n  'ul',\n]);\n// Type-1 blocks (script/pre/style/textarea): closes on the line holding the matching end tag,\n// not on a blank line, and that line is the last one skipped.\nconst HTML_PRE_TAGS = new Set(['script', 'pre', 'style', 'textarea']);\n// An opening or closing tag at column 0, tag name captured for the lookups above.\nconst HTML_BLOCK_OPEN_RE = /^<\\/?([a-zA-Z][a-zA-Z0-9]*)(?:[ \\t]|\\/?>|$)/;\n\n// Strips a leading # (frontmatter entries may carry one) and a trailing /; rejects an\n// all-digit result -- a tag needs at least one non-digit character.\nfunction normalizeTag(raw: string): string | null {\n  const stripped = raw.replace(/^#/, '').replace(/\\/+$/, '');\n  if (!stripped || /^\\d+$/.test(stripped)) return null;\n  return stripped;\n}\n\n// data.tags: a YAML list (Obsidian also accepts a bare string). Null members and non-string\n// members are skipped rather than throwing -- `tags:\\n  -` parses to [null].\nfunction frontmatterTags(data?: Record<string, unknown>): string[] {\n  const raw = data?.tags;\n  const list = Array.isArray(raw) ? raw : typeof raw === 'string' ? [raw] : [];\n  const found: string[] = [];\n  for (const item of list) {\n    if (typeof item !== 'string') continue;\n    const tag = normalizeTag(item);\n    if (tag) found.push(tag);\n  }\n  return found;\n}\n\n// #tag tokens outside fenced code blocks, inline code spans, wikilinks, HTML tags, HTML blocks,\n// <!-- --> comments, and link destinations.\nfunction inlineTags(body: string): string[] {\n  const found: string[] = [];\n  const fence = fenceTracker();\n  const comment = commentTracker();\n  let inHtmlBlock = false;\n  let htmlBlockClose: RegExp | null = null; // set while inside a type-1 (script/pre/style/textarea) block\n  for (const line of body.split('\\n')) {\n    if (inHtmlBlock) {\n      // A fence-like line here is still HTML-block content -- the block wins until it closes.\n      if (htmlBlockClose) {\n        if (htmlBlockClose.test(line)) {\n          inHtmlBlock = false;\n          htmlBlockClose = null;\n        }\n      } else if (/^[ \\t>]*$/.test(line)) {\n        inHtmlBlock = false;\n      }\n      continue;\n    }\n    if (!comment.inComment) {\n      if (fence.feed(line)) continue;\n      if (fence.inFence) continue;\n    }\n    // Obsidian parses nothing inside a <!-- --> comment; mask its span (which may open, close,\n    // or run the whole line) before any other check sees this line's text. Skipped when there\n    // is no comment marker anywhere near this line -- most lines, so worth the branch.\n    const masked = comment.inComment || line.includes('<!--') ? comment.mask(line) : line;\n    // Indented or blockquoted HTML blocks still swallow their content in Obsidian, so the\n    // opener test runs after stripping leading whitespace and > markers.\n    const stripped = masked.replace(/^[ \\t>]*/, '');\n    if (stripped[0] === '<') {\n      const openMatch = HTML_BLOCK_OPEN_RE.exec(stripped);\n      if (openMatch) {\n        const tagName = openMatch[1].toLowerCase();\n        const isClosingTag = stripped[1] === '/';\n        if (!isClosingTag && HTML_PRE_TAGS.has(tagName)) {\n          inHtmlBlock = true;\n          // Any of the four type-1 closers ends the block, not only the tag that opened it.\n          htmlBlockClose = /<\\/(?:script|pre|style|textarea)>/i;\n          continue;\n        }\n        if (HTML_BLOCK_TAGS.has(tagName)) {\n          inHtmlBlock = true;\n          continue;\n        }\n      }\n    }\n    if (!masked.includes('#')) continue; // most lines; skip the regex work\n    let cleaned = masked.includes('`') ? maskCodeSpans(masked) : masked;\n    if (cleaned.includes('[[')) cleaned = cleaned.replace(WIKILINK_RE, (m) => ' '.repeat(m.length));\n    if (cleaned.includes('](')) cleaned = cleaned.replace(LINK_DEST_RE, (m) => ' '.repeat(m.length));\n    if (cleaned.includes('<')) cleaned = cleaned.replace(HTML_TAG_RE, (m) => ' '.repeat(m.length));\n    for (const m of cleaned.matchAll(INLINE_TAG_RE)) {\n      const tag = normalizeTag(m[1]);\n      if (tag) found.push(tag);\n    }\n  }\n  return found;\n}\n\nfunction extract(_raw: string, body: string, _search?: { title: string; summary: string }, data?: Record<string, unknown>): string[] {\n  return [...new Set([...frontmatterTags(data), ...inlineTags(body)])].sort();\n}\n\nexport const tags: Feature = {\n  name: 'tags',\n  schema(db) {\n    db.exec('CREATE TABLE IF NOT EXISTS tags (\"path\" TEXT, tag TEXT, PRIMARY KEY (\"path\", tag))');\n    db.exec('CREATE INDEX IF NOT EXISTS tags_tag ON tags(tag)');\n  },\n  extract,\n  // Per-file rows with nothing else to resolve, so only a vanished file needs a delete here;\n  // store() below handles a reparse's stale rows itself.\n  remove(db, path, delta) {\n    if (!delta.vanished.includes(path)) return;\n    db.prepare('DELETE FROM tags WHERE \"path\" = ?').run(path);\n  },\n  store(db, path, extracted) {\n    const found = extracted as string[];\n    if (found.length === 0) {\n      db.prepare('DELETE FROM tags WHERE \"path\" = ?').run(path);\n    } else {\n      const placeholders = found.map(() => '?').join(', ');\n      db.prepare(`DELETE FROM tags WHERE \"path\" = ? AND tag NOT IN (${placeholders})`).run(path, ...found);\n    }\n    const insert = db.prepare('INSERT OR IGNORE INTO tags (\"path\", tag) VALUES (?, ?)');\n    for (const tag of found) insert.run(path, tag);\n  },\n};\n"],"names":["commentTracker","fenceTracker","maskCodeSpans","WIKILINK_RE","HTML_TAG_RE","INLINE_TAG_RE","LINK_DEST_RE","HTML_BLOCK_TAGS","Set","HTML_PRE_TAGS","HTML_BLOCK_OPEN_RE","normalizeTag","raw","stripped","replace","test","frontmatterTags","data","tags","list","Array","isArray","found","item","tag","push","inlineTags","body","fence","comment","inHtmlBlock","htmlBlockClose","line","split","inComment","feed","inFence","masked","includes","mask","openMatch","exec","tagName","toLowerCase","isClosingTag","has","cleaned","m","repeat","length","matchAll","extract","_raw","_search","sort","name","schema","db","remove","path","delta","vanished","prepare","run","store","extracted","placeholders","map","join","insert"],"mappings":"AAAA,SAASA,cAAc,EAAEC,YAAY,EAAEC,aAAa,QAAQ,eAAe;AAG3E,0FAA0F;AAC1F,uFAAuF;AACvF,4FAA4F;AAE5F,+DAA+D;AAC/D,MAAMC,cAAc,gBAAgB,6DAA6D;AACjG,iDAAiD;AACjD,MAAMC,cAAc,uBAAuB,6DAA6D;AACxG,iGAAiG;AACjG,MAAMC,gBAAgB;AACtB,4FAA4F;AAC5F,MAAMC,eAAe,gCAAgC,oFAAoF;AAEzI,8FAA8F;AAC9F,+FAA+F;AAC/F,iFAAiF;AACjF,MAAMC,kBAAkB,IAAIC,IAAI;IAC9B;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;IACA;CACD;AACD,8FAA8F;AAC9F,8DAA8D;AAC9D,MAAMC,gBAAgB,IAAID,IAAI;IAAC;IAAU;IAAO;IAAS;CAAW;AACpE,kFAAkF;AAClF,MAAME,qBAAqB;AAE3B,sFAAsF;AACtF,oEAAoE;AACpE,SAASC,aAAaC,GAAW;IAC/B,MAAMC,WAAWD,IAAIE,OAAO,CAAC,MAAM,IAAIA,OAAO,CAAC,QAAQ;IACvD,IAAI,CAACD,YAAY,QAAQE,IAAI,CAACF,WAAW,OAAO;IAChD,OAAOA;AACT;AAEA,4FAA4F;AAC5F,6EAA6E;AAC7E,SAASG,gBAAgBC,IAA8B;IACrD,MAAML,MAAMK,iBAAAA,2BAAAA,KAAMC,IAAI;IACtB,MAAMC,OAAOC,MAAMC,OAAO,CAACT,OAAOA,MAAM,OAAOA,QAAQ,WAAW;QAACA;KAAI,GAAG,EAAE;IAC5E,MAAMU,QAAkB,EAAE;IAC1B,KAAK,MAAMC,QAAQJ,KAAM;QACvB,IAAI,OAAOI,SAAS,UAAU;QAC9B,MAAMC,MAAMb,aAAaY;QACzB,IAAIC,KAAKF,MAAMG,IAAI,CAACD;IACtB;IACA,OAAOF;AACT;AAEA,gGAAgG;AAChG,4CAA4C;AAC5C,SAASI,WAAWC,IAAY;IAC9B,MAAML,QAAkB,EAAE;IAC1B,MAAMM,QAAQ3B;IACd,MAAM4B,UAAU7B;IAChB,IAAI8B,cAAc;IAClB,IAAIC,iBAAgC,MAAM,8DAA8D;IACxG,KAAK,MAAMC,QAAQL,KAAKM,KAAK,CAAC,MAAO;QACnC,IAAIH,aAAa;YACf,wFAAwF;YACxF,IAAIC,gBAAgB;gBAClB,IAAIA,eAAehB,IAAI,CAACiB,OAAO;oBAC7BF,cAAc;oBACdC,iBAAiB;gBACnB;YACF,OAAO,IAAI,YAAYhB,IAAI,CAACiB,OAAO;gBACjCF,cAAc;YAChB;YACA;QACF;QACA,IAAI,CAACD,QAAQK,SAAS,EAAE;YACtB,IAAIN,MAAMO,IAAI,CAACH,OAAO;YACtB,IAAIJ,MAAMQ,OAAO,EAAE;QACrB;QACA,2FAA2F;QAC3F,0FAA0F;QAC1F,mFAAmF;QACnF,MAAMC,SAASR,QAAQK,SAAS,IAAIF,KAAKM,QAAQ,CAAC,UAAUT,QAAQU,IAAI,CAACP,QAAQA;QACjF,sFAAsF;QACtF,qEAAqE;QACrE,MAAMnB,WAAWwB,OAAOvB,OAAO,CAAC,YAAY;QAC5C,IAAID,QAAQ,CAAC,EAAE,KAAK,KAAK;YACvB,MAAM2B,YAAY9B,mBAAmB+B,IAAI,CAAC5B;YAC1C,IAAI2B,WAAW;gBACb,MAAME,UAAUF,SAAS,CAAC,EAAE,CAACG,WAAW;gBACxC,MAAMC,eAAe/B,QAAQ,CAAC,EAAE,KAAK;gBACrC,IAAI,CAAC+B,gBAAgBnC,cAAcoC,GAAG,CAACH,UAAU;oBAC/CZ,cAAc;oBACd,kFAAkF;oBAClFC,iBAAiB;oBACjB;gBACF;gBACA,IAAIxB,gBAAgBsC,GAAG,CAACH,UAAU;oBAChCZ,cAAc;oBACd;gBACF;YACF;QACF;QACA,IAAI,CAACO,OAAOC,QAAQ,CAAC,MAAM,UAAU,kCAAkC;QACvE,IAAIQ,UAAUT,OAAOC,QAAQ,CAAC,OAAOpC,cAAcmC,UAAUA;QAC7D,IAAIS,QAAQR,QAAQ,CAAC,OAAOQ,UAAUA,QAAQhC,OAAO,CAACX,aAAa,CAAC4C,IAAM,IAAIC,MAAM,CAACD,EAAEE,MAAM;QAC7F,IAAIH,QAAQR,QAAQ,CAAC,OAAOQ,UAAUA,QAAQhC,OAAO,CAACR,cAAc,CAACyC,IAAM,IAAIC,MAAM,CAACD,EAAEE,MAAM;QAC9F,IAAIH,QAAQR,QAAQ,CAAC,MAAMQ,UAAUA,QAAQhC,OAAO,CAACV,aAAa,CAAC2C,IAAM,IAAIC,MAAM,CAACD,EAAEE,MAAM;QAC5F,KAAK,MAAMF,KAAKD,QAAQI,QAAQ,CAAC7C,eAAgB;YAC/C,MAAMmB,MAAMb,aAAaoC,CAAC,CAAC,EAAE;YAC7B,IAAIvB,KAAKF,MAAMG,IAAI,CAACD;QACtB;IACF;IACA,OAAOF;AACT;AAEA,SAAS6B,QAAQC,IAAY,EAAEzB,IAAY,EAAE0B,OAA4C,EAAEpC,IAA8B;IACvH,OAAO;WAAI,IAAIT,IAAI;eAAIQ,gBAAgBC;eAAUS,WAAWC;SAAM;KAAE,CAAC2B,IAAI;AAC3E;AAEA,OAAO,MAAMpC,OAAgB;IAC3BqC,MAAM;IACNC,QAAOC,EAAE;QACPA,GAAGhB,IAAI,CAAC;QACRgB,GAAGhB,IAAI,CAAC;IACV;IACAU;IACA,2FAA2F;IAC3F,uDAAuD;IACvDO,QAAOD,EAAE,EAAEE,IAAI,EAAEC,KAAK;QACpB,IAAI,CAACA,MAAMC,QAAQ,CAACvB,QAAQ,CAACqB,OAAO;QACpCF,GAAGK,OAAO,CAAC,qCAAqCC,GAAG,CAACJ;IACtD;IACAK,OAAMP,EAAE,EAAEE,IAAI,EAAEM,SAAS;QACvB,MAAM3C,QAAQ2C;QACd,IAAI3C,MAAM2B,MAAM,KAAK,GAAG;YACtBQ,GAAGK,OAAO,CAAC,qCAAqCC,GAAG,CAACJ;QACtD,OAAO;YACL,MAAMO,eAAe5C,MAAM6C,GAAG,CAAC,IAAM,KAAKC,IAAI,CAAC;YAC/CX,GAAGK,OAAO,CAAC,CAAC,kDAAkD,EAAEI,aAAa,CAAC,CAAC,EAAEH,GAAG,CAACJ,SAASrC;QAChG;QACA,MAAM+C,SAASZ,GAAGK,OAAO,CAAC;QAC1B,KAAK,MAAMtC,OAAOF,MAAO+C,OAAON,GAAG,CAACJ,MAAMnC;IAC5C;AACF,EAAE"}

package/dist/esm/fences.d.ts

@@ -3,3 +3,10 @@ export interface FenceTracker {
     readonly inFence: boolean;
 }
 export declare function fenceTracker(): FenceTracker;
+export interface CommentTracker {
+    mask(line: string): string;
+    readonly inComment: boolean;
+}
+export declare function commentTracker(): CommentTracker;
+export declare function maskRegions(body: string): string;
+export declare function maskCodeSpans(line: string): string;

package/dist/esm/fences.js

@@ -41,3 +41,98 @@ export function fenceTracker() {
         }
     };
 }
+export function commentTracker() {
+    let inComment = false;
+    return {
+        mask (line) {
+            let out = '';
+            let i = 0;
+            while(i < line.length){
+                if (inComment) {
+                    const end = line.indexOf('-->', i);
+                    if (end === -1) {
+                        out += ' '.repeat(line.length - i);
+                        i = line.length;
+                    } else {
+                        out += ' '.repeat(end + 3 - i);
+                        i = end + 3;
+                        inComment = false;
+                    }
+                    continue;
+                }
+                const start = line.indexOf('<!--', i);
+                if (start === -1) {
+                    out += line.slice(i);
+                    i = line.length;
+                } else {
+                    out += `${line.slice(i, start)}    `;
+                    i = start + 4;
+                    inComment = true;
+                }
+            }
+            return out;
+        },
+        get inComment () {
+            return inComment;
+        }
+    };
+}
+// Body with fenced-code regions (delimiter and content lines) and <!-- --> comment spans
+// replaced by spaces, newlines preserved so line numbers and offsets survive. Bodies without a
+// backtick/tilde/`<!--` fast-path out untouched.
+export function maskRegions(body) {
+    if (!/[`~]/.test(body) && !body.includes('<!--')) return body;
+    const fence = fenceTracker();
+    const comment = commentTracker();
+    return body.split('\n').map((line)=>{
+        if (!comment.inComment) {
+            const isDelim = fence.feed(line);
+            if (isDelim || fence.inFence) return ' '.repeat(line.length);
+        }
+        const uncommented = comment.mask(line);
+        // Inline code spans hide their content too: `[[x]]` in prose is not a link.
+        return uncommented.includes('`') ? maskCodeSpans(uncommented) : uncommented;
+    }).join('\n');
+}
+// A code span opens on a run of N backticks and closes at the next run of exactly N -- a
+// shorter or longer run in between is literal text, not a delimiter (CommonMark code spans).
+// Masked with spaces so column positions and tag-boundary whitespace are unaffected.
+export function maskCodeSpans(line) {
+    let out = '';
+    let i = 0;
+    while(i < line.length){
+        if (line[i] !== '`') {
+            out += line[i];
+            i++;
+            continue;
+        }
+        let j = i;
+        while(line[j] === '`')j++;
+        const n = j - i;
+        let k = j;
+        let closeStart = -1;
+        let closeEnd = -1;
+        while(k < line.length){
+            if (line[k] !== '`') {
+                k++;
+                continue;
+            }
+            let m = k;
+            while(line[m] === '`')m++;
+            if (m - k === n) {
+                closeStart = k;
+                closeEnd = m;
+                break;
+            }
+            k = m;
+        }
+        if (closeStart >= 0) {
+            out += ' '.repeat(closeEnd - i);
+            i = closeEnd;
+        } else {
+            out += line.slice(i, j);
+            i = j;
+        }
+    }
+    return out;
+}

package/dist/esm/fences.js.map

@@ -1 +1 @@
-{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/fences.ts"],"sourcesContent":["// Shared line-fence tracker for tags/sections/embed. Opener: >=3 backticks or tildes at column\n// 0 -- no indent allowance, a deliberate divergence from CommonMark's up-to-3-space rule (see\n// test/unit/fences.test.ts DIVERGENCES table). A backtick opener's info string must not itself\n// contain a backtick (spec rule). A closer is a run of the SAME character, length >= the\n// opener's, holding nothing but trailing spaces after the run.\n\nexport interface FenceTracker {\n  feed(line: string): boolean; // true iff this line is a fence delimiter (open or close)\n  readonly inFence: boolean;\n}\n\nconst BACKTICK_OPEN = /^(`{3,})(.*)$/;\nconst TILDE_OPEN = /^(~{3,})(.*)$/;\n\nexport function fenceTracker(): FenceTracker {\n  let inFence = false;\n  let fenceChar = '';\n  let fenceLen = 0;\n  return {\n    feed(line: string): boolean {\n      if (!inFence) {\n        const bt = BACKTICK_OPEN.exec(line);\n        if (bt && !bt[2].includes('`')) {\n          inFence = true;\n          fenceChar = '`';\n          fenceLen = bt[1].length;\n          return true;\n        }\n        const td = TILDE_OPEN.exec(line);\n        if (td) {\n          inFence = true;\n          fenceChar = '~';\n          fenceLen = td[1].length;\n          return true;\n        }\n        return false;\n      }\n      const run = fenceChar === '`' ? /^(`+)(.*)$/ : /^(~+)(.*)$/;\n      const m = run.exec(line);\n      if (m && m[1].length >= fenceLen && m[2].trim() === '') {\n        inFence = false;\n        return true;\n      }\n      return false;\n    },\n    get inFence() {\n      return inFence;\n    },\n  };\n}\n"],"names":["BACKTICK_OPEN","TILDE_OPEN","fenceTracker","inFence","fenceChar","fenceLen","feed","line","bt","exec","includes","length","td","run","m","trim"],"mappings":"AAAA,+FAA+F;AAC/F,8FAA8F;AAC9F,+FAA+F;AAC/F,yFAAyF;AACzF,+DAA+D;AAO/D,MAAMA,gBAAgB;AACtB,MAAMC,aAAa;AAEnB,OAAO,SAASC;IACd,IAAIC,UAAU;IACd,IAAIC,YAAY;IAChB,IAAIC,WAAW;IACf,OAAO;QACLC,MAAKC,IAAY;YACf,IAAI,CAACJ,SAAS;gBACZ,MAAMK,KAAKR,cAAcS,IAAI,CAACF;gBAC9B,IAAIC,MAAM,CAACA,EAAE,CAAC,EAAE,CAACE,QAAQ,CAAC,MAAM;oBAC9BP,UAAU;oBACVC,YAAY;oBACZC,WAAWG,EAAE,CAAC,EAAE,CAACG,MAAM;oBACvB,OAAO;gBACT;gBACA,MAAMC,KAAKX,WAAWQ,IAAI,CAACF;gBAC3B,IAAIK,IAAI;oBACNT,UAAU;oBACVC,YAAY;oBACZC,WAAWO,EAAE,CAAC,EAAE,CAACD,MAAM;oBACvB,OAAO;gBACT;gBACA,OAAO;YACT;YACA,MAAME,MAAMT,cAAc,MAAM,eAAe;YAC/C,MAAMU,IAAID,IAAIJ,IAAI,CAACF;YACnB,IAAIO,KAAKA,CAAC,CAAC,EAAE,CAACH,MAAM,IAAIN,YAAYS,CAAC,CAAC,EAAE,CAACC,IAAI,OAAO,IAAI;gBACtDZ,UAAU;gBACV,OAAO;YACT;YACA,OAAO;QACT;QACA,IAAIA,WAAU;YACZ,OAAOA;QACT;IACF;AACF"}
+{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/fences.ts"],"sourcesContent":["// Shared line-fence tracker for tags/sections/embed. Opener: >=3 backticks or tildes at column\n// 0 -- no indent allowance, a deliberate divergence from CommonMark's up-to-3-space rule (see\n// test/unit/fences.test.ts DIVERGENCES table). A backtick opener's info string must not itself\n// contain a backtick (spec rule). A closer is a run of the SAME character, length >= the\n// opener's, holding nothing but trailing spaces after the run.\n\nexport interface FenceTracker {\n  feed(line: string): boolean; // true iff this line is a fence delimiter (open or close)\n  readonly inFence: boolean;\n}\n\nconst BACKTICK_OPEN = /^(`{3,})(.*)$/;\nconst TILDE_OPEN = /^(~{3,})(.*)$/;\n\nexport function fenceTracker(): FenceTracker {\n  let inFence = false;\n  let fenceChar = '';\n  let fenceLen = 0;\n  return {\n    feed(line: string): boolean {\n      if (!inFence) {\n        const bt = BACKTICK_OPEN.exec(line);\n        if (bt && !bt[2].includes('`')) {\n          inFence = true;\n          fenceChar = '`';\n          fenceLen = bt[1].length;\n          return true;\n        }\n        const td = TILDE_OPEN.exec(line);\n        if (td) {\n          inFence = true;\n          fenceChar = '~';\n          fenceLen = td[1].length;\n          return true;\n        }\n        return false;\n      }\n      const run = fenceChar === '`' ? /^(`+)(.*)$/ : /^(~+)(.*)$/;\n      const m = run.exec(line);\n      if (m && m[1].length >= fenceLen && m[2].trim() === '') {\n        inFence = false;\n        return true;\n      }\n      return false;\n    },\n    get inFence() {\n      return inFence;\n    },\n  };\n}\n\n// Masks <!-- ... --> spans that may cross line boundaries, one line at a time so callers can\n// interleave it with their own per-line state. An unclosed <!-- masks to end of input, matching\n// CommonMark type-2 HTML comments' practical effect (everything after is swallowed).\nexport interface CommentTracker {\n  mask(line: string): string;\n  readonly inComment: boolean;\n}\n\nexport function commentTracker(): CommentTracker {\n  let inComment = false;\n  return {\n    mask(line: string): string {\n      let out = '';\n      let i = 0;\n      while (i < line.length) {\n        if (inComment) {\n          const end = line.indexOf('-->', i);\n          if (end === -1) {\n            out += ' '.repeat(line.length - i);\n            i = line.length;\n          } else {\n            out += ' '.repeat(end + 3 - i);\n            i = end + 3;\n            inComment = false;\n          }\n          continue;\n        }\n        const start = line.indexOf('<!--', i);\n        if (start === -1) {\n          out += line.slice(i);\n          i = line.length;\n        } else {\n          out += `${line.slice(i, start)}    `;\n          i = start + 4;\n          inComment = true;\n        }\n      }\n      return out;\n    },\n    get inComment() {\n      return inComment;\n    },\n  };\n}\n\n// Body with fenced-code regions (delimiter and content lines) and <!-- --> comment spans\n// replaced by spaces, newlines preserved so line numbers and offsets survive. Bodies without a\n// backtick/tilde/`<!--` fast-path out untouched.\nexport function maskRegions(body: string): string {\n  if (!/[`~]/.test(body) && !body.includes('<!--')) return body;\n  const fence = fenceTracker();\n  const comment = commentTracker();\n  return body\n    .split('\\n')\n    .map((line) => {\n      if (!comment.inComment) {\n        const isDelim = fence.feed(line);\n        if (isDelim || fence.inFence) return ' '.repeat(line.length);\n      }\n      const uncommented = comment.mask(line);\n      // Inline code spans hide their content too: `[[x]]` in prose is not a link.\n      return uncommented.includes('`') ? maskCodeSpans(uncommented) : uncommented;\n    })\n    .join('\\n');\n}\n\n// A code span opens on a run of N backticks and closes at the next run of exactly N -- a\n// shorter or longer run in between is literal text, not a delimiter (CommonMark code spans).\n// Masked with spaces so column positions and tag-boundary whitespace are unaffected.\nexport function maskCodeSpans(line: string): string {\n  let out = '';\n  let i = 0;\n  while (i < line.length) {\n    if (line[i] !== '`') {\n      out += line[i];\n      i++;\n      continue;\n    }\n    let j = i;\n    while (line[j] === '`') j++;\n    const n = j - i;\n    let k = j;\n    let closeStart = -1;\n    let closeEnd = -1;\n    while (k < line.length) {\n      if (line[k] !== '`') {\n        k++;\n        continue;\n      }\n      let m = k;\n      while (line[m] === '`') m++;\n      if (m - k === n) {\n        closeStart = k;\n        closeEnd = m;\n        break;\n      }\n      k = m;\n    }\n    if (closeStart >= 0) {\n      out += ' '.repeat(closeEnd - i);\n      i = closeEnd;\n    } else {\n      out += line.slice(i, j);\n      i = j;\n    }\n  }\n  return out;\n}\n"],"names":["BACKTICK_OPEN","TILDE_OPEN","fenceTracker","inFence","fenceChar","fenceLen","feed","line","bt","exec","includes","length","td","run","m","trim","commentTracker","inComment","mask","out","i","end","indexOf","repeat","start","slice","maskRegions","body","test","fence","comment","split","map","isDelim","uncommented","maskCodeSpans","join","j","n","k","closeStart","closeEnd"],"mappings":"AAAA,+FAA+F;AAC/F,8FAA8F;AAC9F,+FAA+F;AAC/F,yFAAyF;AACzF,+DAA+D;AAO/D,MAAMA,gBAAgB;AACtB,MAAMC,aAAa;AAEnB,OAAO,SAASC;IACd,IAAIC,UAAU;IACd,IAAIC,YAAY;IAChB,IAAIC,WAAW;IACf,OAAO;QACLC,MAAKC,IAAY;YACf,IAAI,CAACJ,SAAS;gBACZ,MAAMK,KAAKR,cAAcS,IAAI,CAACF;gBAC9B,IAAIC,MAAM,CAACA,EAAE,CAAC,EAAE,CAACE,QAAQ,CAAC,MAAM;oBAC9BP,UAAU;oBACVC,YAAY;oBACZC,WAAWG,EAAE,CAAC,EAAE,CAACG,MAAM;oBACvB,OAAO;gBACT;gBACA,MAAMC,KAAKX,WAAWQ,IAAI,CAACF;gBAC3B,IAAIK,IAAI;oBACNT,UAAU;oBACVC,YAAY;oBACZC,WAAWO,EAAE,CAAC,EAAE,CAACD,MAAM;oBACvB,OAAO;gBACT;gBACA,OAAO;YACT;YACA,MAAME,MAAMT,cAAc,MAAM,eAAe;YAC/C,MAAMU,IAAID,IAAIJ,IAAI,CAACF;YACnB,IAAIO,KAAKA,CAAC,CAAC,EAAE,CAACH,MAAM,IAAIN,YAAYS,CAAC,CAAC,EAAE,CAACC,IAAI,OAAO,IAAI;gBACtDZ,UAAU;gBACV,OAAO;YACT;YACA,OAAO;QACT;QACA,IAAIA,WAAU;YACZ,OAAOA;QACT;IACF;AACF;AAUA,OAAO,SAASa;IACd,IAAIC,YAAY;IAChB,OAAO;QACLC,MAAKX,IAAY;YACf,IAAIY,MAAM;YACV,IAAIC,IAAI;YACR,MAAOA,IAAIb,KAAKI,MAAM,CAAE;gBACtB,IAAIM,WAAW;oBACb,MAAMI,MAAMd,KAAKe,OAAO,CAAC,OAAOF;oBAChC,IAAIC,QAAQ,CAAC,GAAG;wBACdF,OAAO,IAAII,MAAM,CAAChB,KAAKI,MAAM,GAAGS;wBAChCA,IAAIb,KAAKI,MAAM;oBACjB,OAAO;wBACLQ,OAAO,IAAII,MAAM,CAACF,MAAM,IAAID;wBAC5BA,IAAIC,MAAM;wBACVJ,YAAY;oBACd;oBACA;gBACF;gBACA,MAAMO,QAAQjB,KAAKe,OAAO,CAAC,QAAQF;gBACnC,IAAII,UAAU,CAAC,GAAG;oBAChBL,OAAOZ,KAAKkB,KAAK,CAACL;oBAClBA,IAAIb,KAAKI,MAAM;gBACjB,OAAO;oBACLQ,OAAO,GAAGZ,KAAKkB,KAAK,CAACL,GAAGI,OAAO,IAAI,CAAC;oBACpCJ,IAAII,QAAQ;oBACZP,YAAY;gBACd;YACF;YACA,OAAOE;QACT;QACA,IAAIF,aAAY;YACd,OAAOA;QACT;IACF;AACF;AAEA,yFAAyF;AACzF,+FAA+F;AAC/F,iDAAiD;AACjD,OAAO,SAASS,YAAYC,IAAY;IACtC,IAAI,CAAC,OAAOC,IAAI,CAACD,SAAS,CAACA,KAAKjB,QAAQ,CAAC,SAAS,OAAOiB;IACzD,MAAME,QAAQ3B;IACd,MAAM4B,UAAUd;IAChB,OAAOW,KACJI,KAAK,CAAC,MACNC,GAAG,CAAC,CAACzB;QACJ,IAAI,CAACuB,QAAQb,SAAS,EAAE;YACtB,MAAMgB,UAAUJ,MAAMvB,IAAI,CAACC;YAC3B,IAAI0B,WAAWJ,MAAM1B,OAAO,EAAE,OAAO,IAAIoB,MAAM,CAAChB,KAAKI,MAAM;QAC7D;QACA,MAAMuB,cAAcJ,QAAQZ,IAAI,CAACX;QACjC,4EAA4E;QAC5E,OAAO2B,YAAYxB,QAAQ,CAAC,OAAOyB,cAAcD,eAAeA;IAClE,GACCE,IAAI,CAAC;AACV;AAEA,yFAAyF;AACzF,6FAA6F;AAC7F,qFAAqF;AACrF,OAAO,SAASD,cAAc5B,IAAY;IACxC,IAAIY,MAAM;IACV,IAAIC,IAAI;IACR,MAAOA,IAAIb,KAAKI,MAAM,CAAE;QACtB,IAAIJ,IAAI,CAACa,EAAE,KAAK,KAAK;YACnBD,OAAOZ,IAAI,CAACa,EAAE;YACdA;YACA;QACF;QACA,IAAIiB,IAAIjB;QACR,MAAOb,IAAI,CAAC8B,EAAE,KAAK,IAAKA;QACxB,MAAMC,IAAID,IAAIjB;QACd,IAAImB,IAAIF;QACR,IAAIG,aAAa,CAAC;QAClB,IAAIC,WAAW,CAAC;QAChB,MAAOF,IAAIhC,KAAKI,MAAM,CAAE;YACtB,IAAIJ,IAAI,CAACgC,EAAE,KAAK,KAAK;gBACnBA;gBACA;YACF;YACA,IAAIzB,IAAIyB;YACR,MAAOhC,IAAI,CAACO,EAAE,KAAK,IAAKA;YACxB,IAAIA,IAAIyB,MAAMD,GAAG;gBACfE,aAAaD;gBACbE,WAAW3B;gBACX;YACF;YACAyB,IAAIzB;QACN;QACA,IAAI0B,cAAc,GAAG;YACnBrB,OAAO,IAAII,MAAM,CAACkB,WAAWrB;YAC7BA,IAAIqB;QACN,OAAO;YACLtB,OAAOZ,KAAKkB,KAAK,CAACL,GAAGiB;YACrBjB,IAAIiB;QACN;IACF;IACA,OAAOlB;AACT"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "sensemaking",
-  "version": "0.15.0",
+  "version": "0.15.2",
   "description": "Query and search your markdown notes with context-aware progressive disclosure: SQL over frontmatter, links, and text, plus semantic search and link-graph ranking. No server, no build step",
   "keywords": [
     "markdown",

package/skills/sense/SKILL.md

@@ -5,7 +5,7 @@ description: "Query a markdown tree with the sense CLI: filter notes by frontmat
 
 # sense
 
-SQL over a markdown tree, kept fresh by a filesystem check on every query. Every file becomes rows in `frontmatter` (one column per key, plus `path`/`_mtime`/`_ctime`/`_size`/`_rank`/`_parse_error`; `_ctime` is filesystem birthtime, which a clone or copy resets just like `_mtime`), `content` (FTS5: `title`, `summary`, `text`, `path`, plus machine-written `title_seg`/`summary_seg`/`text_seg` sidecars used for matching Chinese, Japanese, Thai, Khmer, Lao, and Burmese text, not for reading), `links` (`src`, `target`, `dst`, `embed`; `NULL` dst = dead link; `embed` 1 for `![[...]]` embeds, 0 for links; one row per distinct written target and kind with alias and anchor stripped, so `[[Foo]]` and `[[Foo|alias]]` are one row while `[[Foo]]` and `[[notes/Foo]]` are two rows that can share a `dst`, and a target both linked and embedded is a row of each kind), `tags` (`path`, `tag`: frontmatter and inline `#tags` merged and deduplicated; nested tags stored full, so `book/scifi` matches `tag = 'book' OR tag LIKE 'book/%'`), `sections` (heading outline with line ranges and token estimates), and `preset_files` (`path`, `preset`: which presets cover which files). Features add their own storage; `map` and `status` report which are on.
+SQL over a markdown tree, kept fresh by a filesystem check on every query. Every file becomes rows in `frontmatter` (one column per key, plus `path`/`_mtime`/`_ctime`/`_size`/`_rank`/`_parse_error`; `_ctime` is filesystem birthtime, which a clone or copy resets just like `_mtime`), `content` (FTS5: `title`, `summary`, `text`, `path`, plus machine-written `title_seg`/`summary_seg`/`text_seg` sidecars used for matching Chinese, Japanese, Thai, Khmer, Lao, and Burmese text, not for reading), `links` (`src`, `target`, `dst`, `embed`; `NULL` dst = dead link; `embed` 1 for `![[...]]` embeds, 0 for links; one row per distinct written target and kind with alias and anchor stripped, so `[[Foo]]` and `[[Foo|alias]]` are one row while `[[Foo]]` and `[[notes/Foo]]` are two rows that can share a `dst`, and a target both linked and embedded is a row of each kind; extraction matches Obsidian's own graph: comments, code, and link-syntax text yield no rows, `[[#Anchor]]` is a self-edge, a frontmatter value that is exactly `[[X]]` is a link, and a basename collision resolves to the linking note itself, else the shortest path), `tags` (`path`, `tag`: frontmatter and inline `#tags` merged and deduplicated; nested tags stored full, so `book/scifi` matches `tag = 'book' OR tag LIKE 'book/%'`), `sections` (heading outline with line ranges and token estimates), and `preset_files` (`path`, `preset`: which presets cover which files). Features add their own storage; `map` and `status` report which are on.
 
 ## What each tool is for