sensemaking 0.19.0 → 0.19.2

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (110) hide show
  1. package/dist/cjs/index.d.cts +1 -1
  2. package/dist/cjs/index.d.ts +1 -1
  3. package/dist/cjs/index.js +3 -0
  4. package/dist/cjs/index.js.map +1 -1
  5. package/dist/cjs/store/duckdb/fieldStats.js +87 -10
  6. package/dist/cjs/store/duckdb/fieldStats.js.map +1 -1
  7. package/dist/cjs/store/duckdb/open.js +3 -2
  8. package/dist/cjs/store/duckdb/open.js.map +1 -1
  9. package/dist/cjs/store/duckdb/reconcile.d.cts +2 -6
  10. package/dist/cjs/store/duckdb/reconcile.d.ts +2 -6
  11. package/dist/cjs/store/duckdb/reconcile.js +41 -581
  12. package/dist/cjs/store/duckdb/reconcile.js.map +1 -1
  13. package/dist/cjs/store/duckdb/store.js +5 -3
  14. package/dist/cjs/store/duckdb/store.js.map +1 -1
  15. package/dist/cjs/store/index.d.cts +1 -0
  16. package/dist/cjs/store/index.d.ts +1 -0
  17. package/dist/cjs/store/index.js +4 -0
  18. package/dist/cjs/store/index.js.map +1 -1
  19. package/dist/cjs/store/reconcile.d.cts +6 -0
  20. package/dist/cjs/store/reconcile.d.ts +6 -0
  21. package/dist/cjs/store/reconcile.js +750 -0
  22. package/dist/cjs/store/reconcile.js.map +1 -0
  23. package/dist/cjs/store/shared.d.cts +1 -0
  24. package/dist/cjs/store/shared.d.ts +1 -0
  25. package/dist/cjs/store/shared.js +37 -0
  26. package/dist/cjs/store/shared.js.map +1 -1
  27. package/dist/cjs/store/sqlite/connection.js +1 -1
  28. package/dist/cjs/store/sqlite/connection.js.map +1 -1
  29. package/dist/cjs/store/sqlite/open.js +33 -11
  30. package/dist/cjs/store/sqlite/open.js.map +1 -1
  31. package/dist/cjs/store/sqlite/reconcile.d.cts +5 -10
  32. package/dist/cjs/store/sqlite/reconcile.d.ts +5 -10
  33. package/dist/cjs/store/sqlite/reconcile.js +48 -840
  34. package/dist/cjs/store/sqlite/reconcile.js.map +1 -1
  35. package/dist/cjs/store/sqlite/signature.d.cts +6 -0
  36. package/dist/cjs/store/sqlite/signature.d.ts +6 -0
  37. package/dist/cjs/store/sqlite/signature.js +380 -0
  38. package/dist/cjs/store/sqlite/signature.js.map +1 -0
  39. package/dist/cjs/store/sqlite/store.js +5 -3
  40. package/dist/cjs/store/sqlite/store.js.map +1 -1
  41. package/dist/cjs/store/transaction.d.cts +2 -1
  42. package/dist/cjs/store/transaction.d.ts +2 -1
  43. package/dist/cjs/store/transaction.js +16 -6
  44. package/dist/cjs/store/transaction.js.map +1 -1
  45. package/dist/cjs/store/turso/connection.js +1 -1
  46. package/dist/cjs/store/turso/connection.js.map +1 -1
  47. package/dist/cjs/store/turso/open.d.cts +2 -0
  48. package/dist/cjs/store/turso/open.d.ts +2 -0
  49. package/dist/cjs/store/turso/open.js +115 -57
  50. package/dist/cjs/store/turso/open.js.map +1 -1
  51. package/dist/cjs/store/turso/reconcile.d.cts +3 -6
  52. package/dist/cjs/store/turso/reconcile.d.ts +3 -6
  53. package/dist/cjs/store/turso/reconcile.js +159 -596
  54. package/dist/cjs/store/turso/reconcile.js.map +1 -1
  55. package/dist/cjs/store/turso/store.js +3 -2
  56. package/dist/cjs/store/turso/store.js.map +1 -1
  57. package/dist/cjs/store/types.d.cts +12 -2
  58. package/dist/cjs/store/types.d.ts +12 -2
  59. package/dist/cjs/store/types.js +14 -0
  60. package/dist/cjs/store/types.js.map +1 -1
  61. package/dist/esm/index.d.ts +1 -1
  62. package/dist/esm/index.js +1 -1
  63. package/dist/esm/index.js.map +1 -1
  64. package/dist/esm/store/duckdb/fieldStats.js +32 -19
  65. package/dist/esm/store/duckdb/fieldStats.js.map +1 -1
  66. package/dist/esm/store/duckdb/open.js +3 -2
  67. package/dist/esm/store/duckdb/open.js.map +1 -1
  68. package/dist/esm/store/duckdb/reconcile.d.ts +2 -6
  69. package/dist/esm/store/duckdb/reconcile.js +20 -117
  70. package/dist/esm/store/duckdb/reconcile.js.map +1 -1
  71. package/dist/esm/store/duckdb/store.js +5 -3
  72. package/dist/esm/store/duckdb/store.js.map +1 -1
  73. package/dist/esm/store/index.d.ts +1 -0
  74. package/dist/esm/store/index.js +1 -0
  75. package/dist/esm/store/index.js.map +1 -1
  76. package/dist/esm/store/reconcile.d.ts +6 -0
  77. package/dist/esm/store/reconcile.js +130 -0
  78. package/dist/esm/store/reconcile.js.map +1 -0
  79. package/dist/esm/store/shared.d.ts +1 -0
  80. package/dist/esm/store/shared.js +9 -0
  81. package/dist/esm/store/shared.js.map +1 -1
  82. package/dist/esm/store/sqlite/connection.js +2 -2
  83. package/dist/esm/store/sqlite/connection.js.map +1 -1
  84. package/dist/esm/store/sqlite/open.js +28 -8
  85. package/dist/esm/store/sqlite/open.js.map +1 -1
  86. package/dist/esm/store/sqlite/reconcile.d.ts +5 -10
  87. package/dist/esm/store/sqlite/reconcile.js +31 -198
  88. package/dist/esm/store/sqlite/reconcile.js.map +1 -1
  89. package/dist/esm/store/sqlite/signature.d.ts +6 -0
  90. package/dist/esm/store/sqlite/signature.js +58 -0
  91. package/dist/esm/store/sqlite/signature.js.map +1 -0
  92. package/dist/esm/store/sqlite/store.js +5 -3
  93. package/dist/esm/store/sqlite/store.js.map +1 -1
  94. package/dist/esm/store/transaction.d.ts +2 -1
  95. package/dist/esm/store/transaction.js +11 -6
  96. package/dist/esm/store/transaction.js.map +1 -1
  97. package/dist/esm/store/turso/connection.js +2 -2
  98. package/dist/esm/store/turso/connection.js.map +1 -1
  99. package/dist/esm/store/turso/open.d.ts +2 -0
  100. package/dist/esm/store/turso/open.js +14 -4
  101. package/dist/esm/store/turso/open.js.map +1 -1
  102. package/dist/esm/store/turso/reconcile.d.ts +3 -6
  103. package/dist/esm/store/turso/reconcile.js +31 -132
  104. package/dist/esm/store/turso/reconcile.js.map +1 -1
  105. package/dist/esm/store/turso/store.js +3 -2
  106. package/dist/esm/store/turso/store.js.map +1 -1
  107. package/dist/esm/store/types.d.ts +12 -2
  108. package/dist/esm/store/types.js +12 -1
  109. package/dist/esm/store/types.js.map +1 -1
  110. package/package.json +1 -1
@@ -6,10 +6,11 @@ import { STORE_DIMS } from '../../embed/types.js';
6
6
  import { SenseError } from '../../errors.js';
7
7
  import { activeFeatures, FEATURES } from '../../features/index.js';
8
8
  import { clearCache } from '../cache.js';
9
+ import { reconcile } from '../reconcile.js';
9
10
  import { getMeta, setMeta } from '../shared.js';
10
11
  import { createConnection } from './connection.js';
11
12
  import { TURSO_PACKAGE, tursoApi } from './native.js';
12
- import { reconcile } from './reconcile.js';
13
+ import { tursoDialect } from './reconcile.js';
13
14
  import { createStore } from './store.js';
14
15
  export const DB_FILENAME = 'cache.turso.db';
15
16
  // Independent of sqlite's and duckdb's SCHEMA_VERSION: each store's cache shape evolves
@@ -17,11 +18,20 @@ export const DB_FILENAME = 'cache.turso.db';
17
18
  export const SCHEMA_VERSION = '3';
18
19
  // The ngram index is scoped to disjoint "_ngram" sidecar columns: a second index over the same
19
20
  // columns makes a bare substring match a whole word, defeating the prefix-query rejection.
21
+ // The two FTS indexes, named here because reconcile drops and rebuilds them around a bulk load:
22
+ // Tantivy maintains them per inserted row, which is quadratic in what is already indexed.
23
+ export const CONTENT_FTS_DDL = [
24
+ `CREATE INDEX IF NOT EXISTS content_fts ON content USING fts (title, summary, text) WITH (weights = 'title=10.0,summary=5.0,text=1.0')`,
25
+ `CREATE INDEX IF NOT EXISTS content_fts_ngram ON content USING fts (title_ngram, summary_ngram, text_ngram) WITH (tokenizer='ngram', weights='title_ngram=10.0,summary_ngram=5.0,text_ngram=1.0')`
26
+ ];
27
+ export const CONTENT_FTS_NAMES = [
28
+ 'content_fts',
29
+ 'content_fts_ngram'
30
+ ];
20
31
  async function ensureSchema(conn, cfg) {
21
32
  await conn.exec(`CREATE TABLE IF NOT EXISTS frontmatter ("path" TEXT PRIMARY KEY, "_mtime" REAL, "_ctime" REAL, "_size" INTEGER, "_parse_error" TEXT)`);
22
33
  await conn.exec(`CREATE TABLE IF NOT EXISTS content ("path" TEXT PRIMARY KEY, title TEXT, summary TEXT, text TEXT, title_ngram TEXT, summary_ngram TEXT, text_ngram TEXT)`);
23
- await conn.exec(`CREATE INDEX IF NOT EXISTS content_fts ON content USING fts (title, summary, text) WITH (weights = 'title=10.0,summary=5.0,text=1.0')`);
24
- await conn.exec(`CREATE INDEX IF NOT EXISTS content_fts_ngram ON content USING fts (title_ngram, summary_ngram, text_ngram) WITH (tokenizer='ngram', weights='title_ngram=10.0,summary_ngram=5.0,text_ngram=1.0')`);
34
+ for (const ddl of CONTENT_FTS_DDL)await conn.exec(ddl);
25
35
  await conn.exec(`CREATE TABLE IF NOT EXISTS preset_files ("path" TEXT, preset TEXT, PRIMARY KEY ("path", preset))`);
26
36
  await conn.exec('CREATE INDEX IF NOT EXISTS preset_files_preset ON preset_files(preset)');
27
37
  for (const feature of activeFeatures(cfg)){
@@ -89,7 +99,7 @@ async function connect(cfg) {
89
99
  // Installed before reconcile() -- that call is the one that races a watcher's transaction.
90
100
  const recordedMaxMs = Number((_ref = await getMeta(conn, 'reconcile_max_ms')) !== null && _ref !== void 0 ? _ref : '0');
91
101
  await conn.exec(`PRAGMA busy_timeout = ${Math.min(Math.max(30000, 3 * recordedMaxMs), 600000)}`);
92
- const { parsed, warnings } = await reconcile(conn, cfg, cfg.baseDir);
102
+ const { parsed, warnings } = await reconcile(conn, cfg, cfg.baseDir, tursoDialect);
93
103
  return {
94
104
  db,
95
105
  conn,
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/turso/open.ts"],"sourcesContent":["import { mkdirSync } from 'node:fs';\nimport { join } from 'node:path';\nimport type { Database } from '@tursodatabase/database';\nimport type { Config, ResolvedConfig } from '../../config/index.ts';\nimport { featureSignature, STATE_DIR } from '../../config/index.ts';\nimport { rekeyChunkText } from '../../embed/handoff.ts';\nimport { STORE_DIMS } from '../../embed/types.ts';\nimport { SenseError } from '../../errors.ts';\nimport { activeFeatures, FEATURES } from '../../features/index.ts';\nimport { clearCache } from '../cache.ts';\nimport { getMeta, setMeta } from '../shared.ts';\nimport type { Connection, Store } from '../types.ts';\nimport { createConnection } from './connection.ts';\nimport { TURSO_PACKAGE, tursoApi } from './native.ts';\nimport { reconcile } from './reconcile.ts';\nimport { createStore } from './store.ts';\n\nexport const DB_FILENAME = 'cache.turso.db';\n// Independent of sqlite's and duckdb's SCHEMA_VERSION: each store's cache shape evolves\n// separately. Covers the FTS indexes, the \"_ngram\" sidecar columns, and embeddings.vector's width.\nexport const SCHEMA_VERSION = '3';\n\nexport interface OpenResult {\n store: Store;\n cfg: ResolvedConfig;\n dbPath: string;\n parsed: number;\n warnings: string[];\n}\n\n// The ngram index is scoped to disjoint \"_ngram\" sidecar columns: a second index over the same\n// columns makes a bare substring match a whole word, defeating the prefix-query rejection.\nasync function ensureSchema(conn: Connection, cfg: Config): Promise<void> {\n await conn.exec(`CREATE TABLE IF NOT EXISTS frontmatter (\"path\" TEXT PRIMARY KEY, \"_mtime\" REAL, \"_ctime\" REAL, \"_size\" INTEGER, \"_parse_error\" TEXT)`);\n await conn.exec(`CREATE TABLE IF NOT EXISTS content (\"path\" TEXT PRIMARY KEY, title TEXT, summary TEXT, text TEXT, title_ngram TEXT, summary_ngram TEXT, text_ngram TEXT)`);\n await conn.exec(`CREATE INDEX IF NOT EXISTS content_fts ON content USING fts (title, summary, text) WITH (weights = 'title=10.0,summary=5.0,text=1.0')`);\n await conn.exec(`CREATE INDEX IF NOT EXISTS content_fts_ngram ON content USING fts (title_ngram, summary_ngram, text_ngram) WITH (tokenizer='ngram', weights='title_ngram=10.0,summary_ngram=5.0,text_ngram=1.0')`);\n await conn.exec(`CREATE TABLE IF NOT EXISTS preset_files (\"path\" TEXT, preset TEXT, PRIMARY KEY (\"path\", preset))`);\n await conn.exec('CREATE INDEX IF NOT EXISTS preset_files_preset ON preset_files(preset)');\n for (const feature of activeFeatures(cfg)) {\n // Native F32_BLOB(STORE_DIMS) instead of the embed feature's engine-neutral BLOB DDL. `scale`\n // is kept unused, so the shared reconcile-time INSERT/DELETE names a column both stores have.\n if (feature.name === 'embed') {\n await conn.exec(`CREATE TABLE IF NOT EXISTS embeddings (\"path\" TEXT, chunk INTEGER, start_line INTEGER, end_line INTEGER, scale REAL, vector F32_BLOB(${STORE_DIMS}), PRIMARY KEY (\"path\", chunk))`);\n continue;\n }\n await feature.schema(conn);\n }\n if ((await getMeta(conn, 'schema_version')) === null) await setMeta(conn, 'schema_version', SCHEMA_VERSION);\n if ((await getMeta(conn, 'features')) === null) await setMeta(conn, 'features', featureSignature(cfg, FEATURES));\n}\n\ninterface ConnectResult {\n db: Database;\n conn: Connection;\n cfg: ResolvedConfig;\n dbPath: string;\n parsed: number;\n warnings: string[];\n}\n\nasync function connect(cfg: ResolvedConfig): Promise<ConnectResult> {\n const stateDir = join(cfg.baseDir, STATE_DIR);\n mkdirSync(stateDir, { recursive: true });\n const dbPath = join(stateDir, DB_FILENAME);\n\n // Dynamic, not a top-level import: a sqlite or duckdb tree must never attempt to resolve this\n // optional dependency until a turso tree is actually opened. Installed on first use if missing.\n let turso: Awaited<ReturnType<typeof tursoApi>>;\n try {\n turso = await tursoApi();\n } catch (err) {\n if (err instanceof SenseError) throw err;\n throw new SenseError('STORE_DEPENDENCY_MISSING', `store \"turso\" needs the ${TURSO_PACKAGE} package (${(err as Error).message})`);\n }\n\n let db: Database;\n try {\n // Floored at the same 30s sqlite opens with. `timeout` is connect-time only in this client;\n // the derived value is set via runtime PRAGMA below. `index_method` is required for ensureSchema()'s FTS indexes (T1).\n db = await turso.connect(dbPath, { timeout: 30_000, experimental: ['index_method'] });\n } catch (err) {\n throw new SenseError('STORE_DEPENDENCY_MISSING', `store \"turso\" failed to open ${dbPath}: ${(err as Error).message}`);\n }\n // A throw below must release this handle, or the leaked WAL makes the db file undeletable on\n // Windows. closed guards the catch against double-closing.\n let closed = false;\n try {\n const conn = createConnection(db);\n\n await conn.exec('CREATE TABLE IF NOT EXISTS meta (key TEXT PRIMARY KEY, value TEXT)');\n\n const version = await getMeta(conn, 'schema_version');\n const features = await getMeta(conn, 'features');\n const wantFeatures = featureSignature(cfg, FEATURES);\n if ((version !== null && version !== SCHEMA_VERSION) || (features !== null && features !== wantFeatures)) {\n // Reconcile only reparses changed files, so a stale cache is rebuilt rather than patched\n // incrementally. No tokenize-only or embed-identity-adopted partial path: any mismatch rebuilds.\n const reason = version !== null && version !== SCHEMA_VERSION ? 'cache format changed (new sensemaking version)' : 'config change (features, tokenizer, or presets)';\n console.error(`sense: ${reason}; rebuilding the index`);\n closed = true;\n await db.close();\n clearCache(cfg);\n return connect(cfg);\n }\n\n await ensureSchema(conn, cfg);\n\n // 3x the largest reconcile this cache has recorded, floored at 30s and capped at 10min.\n // Installed before reconcile() -- that call is the one that races a watcher's transaction.\n const recordedMaxMs = Number((await getMeta(conn, 'reconcile_max_ms')) ?? '0');\n await conn.exec(`PRAGMA busy_timeout = ${Math.min(Math.max(30000, 3 * recordedMaxMs), 600_000)}`);\n\n const { parsed, warnings } = await reconcile(conn, cfg, cfg.baseDir);\n\n return { db, conn, cfg, dbPath, parsed, warnings };\n } catch (err) {\n if (!closed) await db.close();\n throw err;\n }\n}\n\n// The turso store's open: connects (fully async, see connect() above), then wraps the resulting\n// connection in the Store interface.\nexport async function openTurso(cfg: ResolvedConfig): Promise<OpenResult> {\n const { db, conn, cfg: resolvedCfg, dbPath, parsed, warnings } = await connect(cfg);\n const store = createStore(db, conn, resolvedCfg, resolvedCfg.baseDir);\n // reconcile ran before this object existed, so its chunk text is keyed by the connection.\n rekeyChunkText(conn, store);\n return { store: store, cfg: resolvedCfg, dbPath, parsed, warnings };\n}\n"],"names":["mkdirSync","join","featureSignature","STATE_DIR","rekeyChunkText","STORE_DIMS","SenseError","activeFeatures","FEATURES","clearCache","getMeta","setMeta","createConnection","TURSO_PACKAGE","tursoApi","reconcile","createStore","DB_FILENAME","SCHEMA_VERSION","ensureSchema","conn","cfg","exec","feature","name","schema","connect","stateDir","baseDir","recursive","dbPath","turso","err","message","db","timeout","experimental","closed","version","features","wantFeatures","reason","console","error","close","recordedMaxMs","Number","Math","min","max","parsed","warnings","openTurso","resolvedCfg","store"],"mappings":"AAAA,SAASA,SAAS,QAAQ,UAAU;AACpC,SAASC,IAAI,QAAQ,YAAY;AAGjC,SAASC,gBAAgB,EAAEC,SAAS,QAAQ,wBAAwB;AACpE,SAASC,cAAc,QAAQ,yBAAyB;AACxD,SAASC,UAAU,QAAQ,uBAAuB;AAClD,SAASC,UAAU,QAAQ,kBAAkB;AAC7C,SAASC,cAAc,EAAEC,QAAQ,QAAQ,0BAA0B;AACnE,SAASC,UAAU,QAAQ,cAAc;AACzC,SAASC,OAAO,EAAEC,OAAO,QAAQ,eAAe;AAEhD,SAASC,gBAAgB,QAAQ,kBAAkB;AACnD,SAASC,aAAa,EAAEC,QAAQ,QAAQ,cAAc;AACtD,SAASC,SAAS,QAAQ,iBAAiB;AAC3C,SAASC,WAAW,QAAQ,aAAa;AAEzC,OAAO,MAAMC,cAAc,iBAAiB;AAC5C,wFAAwF;AACxF,mGAAmG;AACnG,OAAO,MAAMC,iBAAiB,IAAI;AAUlC,+FAA+F;AAC/F,2FAA2F;AAC3F,eAAeC,aAAaC,IAAgB,EAAEC,GAAW;IACvD,MAAMD,KAAKE,IAAI,CAAC,CAAC,oIAAoI,CAAC;IACtJ,MAAMF,KAAKE,IAAI,CAAC,CAAC,wJAAwJ,CAAC;IAC1K,MAAMF,KAAKE,IAAI,CAAC,CAAC,qIAAqI,CAAC;IACvJ,MAAMF,KAAKE,IAAI,CAAC,CAAC,gMAAgM,CAAC;IAClN,MAAMF,KAAKE,IAAI,CAAC,CAAC,gGAAgG,CAAC;IAClH,MAAMF,KAAKE,IAAI,CAAC;IAChB,KAAK,MAAMC,WAAWhB,eAAec,KAAM;QACzC,8FAA8F;QAC9F,8FAA8F;QAC9F,IAAIE,QAAQC,IAAI,KAAK,SAAS;YAC5B,MAAMJ,KAAKE,IAAI,CAAC,CAAC,qIAAqI,EAAEjB,WAAW,+BAA+B,CAAC;YACnM;QACF;QACA,MAAMkB,QAAQE,MAAM,CAACL;IACvB;IACA,IAAI,AAAC,MAAMV,QAAQU,MAAM,sBAAuB,MAAM,MAAMT,QAAQS,MAAM,kBAAkBF;IAC5F,IAAI,AAAC,MAAMR,QAAQU,MAAM,gBAAiB,MAAM,MAAMT,QAAQS,MAAM,YAAYlB,iBAAiBmB,KAAKb;AACxG;AAWA,eAAekB,QAAQL,GAAmB;IACxC,MAAMM,WAAW1B,KAAKoB,IAAIO,OAAO,EAAEzB;IACnCH,UAAU2B,UAAU;QAAEE,WAAW;IAAK;IACtC,MAAMC,SAAS7B,KAAK0B,UAAUV;IAE9B,8FAA8F;IAC9F,gGAAgG;IAChG,IAAIc;IACJ,IAAI;QACFA,QAAQ,MAAMjB;IAChB,EAAE,OAAOkB,KAAK;QACZ,IAAIA,eAAe1B,YAAY,MAAM0B;QACrC,MAAM,IAAI1B,WAAW,4BAA4B,CAAC,wBAAwB,EAAEO,cAAc,UAAU,EAAE,AAACmB,IAAcC,OAAO,CAAC,CAAC,CAAC;IACjI;IAEA,IAAIC;IACJ,IAAI;QACF,4FAA4F;QAC5F,uHAAuH;QACvHA,KAAK,MAAMH,MAAML,OAAO,CAACI,QAAQ;YAAEK,SAAS;YAAQC,cAAc;gBAAC;aAAe;QAAC;IACrF,EAAE,OAAOJ,KAAK;QACZ,MAAM,IAAI1B,WAAW,4BAA4B,CAAC,6BAA6B,EAAEwB,OAAO,EAAE,EAAE,AAACE,IAAcC,OAAO,EAAE;IACtH;IACA,6FAA6F;IAC7F,2DAA2D;IAC3D,IAAII,SAAS;IACb,IAAI;YAuB4B;QAtB9B,MAAMjB,OAAOR,iBAAiBsB;QAE9B,MAAMd,KAAKE,IAAI,CAAC;QAEhB,MAAMgB,UAAU,MAAM5B,QAAQU,MAAM;QACpC,MAAMmB,WAAW,MAAM7B,QAAQU,MAAM;QACrC,MAAMoB,eAAetC,iBAAiBmB,KAAKb;QAC3C,IAAI,AAAC8B,YAAY,QAAQA,YAAYpB,kBAAoBqB,aAAa,QAAQA,aAAaC,cAAe;YACxG,yFAAyF;YACzF,iGAAiG;YACjG,MAAMC,SAASH,YAAY,QAAQA,YAAYpB,iBAAiB,mDAAmD;YACnHwB,QAAQC,KAAK,CAAC,CAAC,OAAO,EAAEF,OAAO,sBAAsB,CAAC;YACtDJ,SAAS;YACT,MAAMH,GAAGU,KAAK;YACdnC,WAAWY;YACX,OAAOK,QAAQL;QACjB;QAEA,MAAMF,aAAaC,MAAMC;QAEzB,wFAAwF;QACxF,2FAA2F;QAC3F,MAAMwB,gBAAgBC,QAAQ,OAAA,MAAMpC,QAAQU,MAAM,iCAApB,kBAAA,OAA4C;QAC1E,MAAMA,KAAKE,IAAI,CAAC,CAAC,sBAAsB,EAAEyB,KAAKC,GAAG,CAACD,KAAKE,GAAG,CAAC,OAAO,IAAIJ,gBAAgB,SAAU;QAEhG,MAAM,EAAEK,MAAM,EAAEC,QAAQ,EAAE,GAAG,MAAMpC,UAAUK,MAAMC,KAAKA,IAAIO,OAAO;QAEnE,OAAO;YAAEM;YAAId;YAAMC;YAAKS;YAAQoB;YAAQC;QAAS;IACnD,EAAE,OAAOnB,KAAK;QACZ,IAAI,CAACK,QAAQ,MAAMH,GAAGU,KAAK;QAC3B,MAAMZ;IACR;AACF;AAEA,gGAAgG;AAChG,qCAAqC;AACrC,OAAO,eAAeoB,UAAU/B,GAAmB;IACjD,MAAM,EAAEa,EAAE,EAAEd,IAAI,EAAEC,KAAKgC,WAAW,EAAEvB,MAAM,EAAEoB,MAAM,EAAEC,QAAQ,EAAE,GAAG,MAAMzB,QAAQL;IAC/E,MAAMiC,QAAQtC,YAAYkB,IAAId,MAAMiC,aAAaA,YAAYzB,OAAO;IACpE,0FAA0F;IAC1FxB,eAAegB,MAAMkC;IACrB,OAAO;QAAEA,OAAOA;QAAOjC,KAAKgC;QAAavB;QAAQoB;QAAQC;IAAS;AACpE"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/turso/open.ts"],"sourcesContent":["import { mkdirSync } from 'node:fs';\nimport { join } from 'node:path';\nimport type { Database } from '@tursodatabase/database';\nimport type { Config, ResolvedConfig } from '../../config/index.ts';\nimport { featureSignature, STATE_DIR } from '../../config/index.ts';\nimport { rekeyChunkText } from '../../embed/handoff.ts';\nimport { STORE_DIMS } from '../../embed/types.ts';\nimport { SenseError } from '../../errors.ts';\nimport { activeFeatures, FEATURES } from '../../features/index.ts';\nimport { clearCache } from '../cache.ts';\nimport { reconcile } from '../reconcile.ts';\nimport { getMeta, setMeta } from '../shared.ts';\nimport type { Connection, Store } from '../types.ts';\nimport { createConnection } from './connection.ts';\nimport { TURSO_PACKAGE, tursoApi } from './native.ts';\nimport { tursoDialect } from './reconcile.ts';\nimport { createStore } from './store.ts';\n\nexport const DB_FILENAME = 'cache.turso.db';\n// Independent of sqlite's and duckdb's SCHEMA_VERSION: each store's cache shape evolves\n// separately. Covers the FTS indexes, the \"_ngram\" sidecar columns, and embeddings.vector's width.\nexport const SCHEMA_VERSION = '3';\n\nexport interface OpenResult {\n store: Store;\n cfg: ResolvedConfig;\n dbPath: string;\n parsed: number;\n warnings: string[];\n}\n\n// The ngram index is scoped to disjoint \"_ngram\" sidecar columns: a second index over the same\n// columns makes a bare substring match a whole word, defeating the prefix-query rejection.\n// The two FTS indexes, named here because reconcile drops and rebuilds them around a bulk load:\n// Tantivy maintains them per inserted row, which is quadratic in what is already indexed.\nexport const CONTENT_FTS_DDL = [\n `CREATE INDEX IF NOT EXISTS content_fts ON content USING fts (title, summary, text) WITH (weights = 'title=10.0,summary=5.0,text=1.0')`,\n `CREATE INDEX IF NOT EXISTS content_fts_ngram ON content USING fts (title_ngram, summary_ngram, text_ngram) WITH (tokenizer='ngram', weights='title_ngram=10.0,summary_ngram=5.0,text_ngram=1.0')`,\n] as const;\nexport const CONTENT_FTS_NAMES = ['content_fts', 'content_fts_ngram'] as const;\n\nasync function ensureSchema(conn: Connection, cfg: Config): Promise<void> {\n await conn.exec(`CREATE TABLE IF NOT EXISTS frontmatter (\"path\" TEXT PRIMARY KEY, \"_mtime\" REAL, \"_ctime\" REAL, \"_size\" INTEGER, \"_parse_error\" TEXT)`);\n await conn.exec(`CREATE TABLE IF NOT EXISTS content (\"path\" TEXT PRIMARY KEY, title TEXT, summary TEXT, text TEXT, title_ngram TEXT, summary_ngram TEXT, text_ngram TEXT)`);\n for (const ddl of CONTENT_FTS_DDL) await conn.exec(ddl);\n await conn.exec(`CREATE TABLE IF NOT EXISTS preset_files (\"path\" TEXT, preset TEXT, PRIMARY KEY (\"path\", preset))`);\n await conn.exec('CREATE INDEX IF NOT EXISTS preset_files_preset ON preset_files(preset)');\n for (const feature of activeFeatures(cfg)) {\n // Native F32_BLOB(STORE_DIMS) instead of the embed feature's engine-neutral BLOB DDL. `scale`\n // is kept unused, so the shared reconcile-time INSERT/DELETE names a column both stores have.\n if (feature.name === 'embed') {\n await conn.exec(`CREATE TABLE IF NOT EXISTS embeddings (\"path\" TEXT, chunk INTEGER, start_line INTEGER, end_line INTEGER, scale REAL, vector F32_BLOB(${STORE_DIMS}), PRIMARY KEY (\"path\", chunk))`);\n continue;\n }\n await feature.schema(conn);\n }\n if ((await getMeta(conn, 'schema_version')) === null) await setMeta(conn, 'schema_version', SCHEMA_VERSION);\n if ((await getMeta(conn, 'features')) === null) await setMeta(conn, 'features', featureSignature(cfg, FEATURES));\n}\n\ninterface ConnectResult {\n db: Database;\n conn: Connection;\n cfg: ResolvedConfig;\n dbPath: string;\n parsed: number;\n warnings: string[];\n}\n\nasync function connect(cfg: ResolvedConfig): Promise<ConnectResult> {\n const stateDir = join(cfg.baseDir, STATE_DIR);\n mkdirSync(stateDir, { recursive: true });\n const dbPath = join(stateDir, DB_FILENAME);\n\n // Dynamic, not a top-level import: a sqlite or duckdb tree must never attempt to resolve this\n // optional dependency until a turso tree is actually opened. Installed on first use if missing.\n let turso: Awaited<ReturnType<typeof tursoApi>>;\n try {\n turso = await tursoApi();\n } catch (err) {\n if (err instanceof SenseError) throw err;\n throw new SenseError('STORE_DEPENDENCY_MISSING', `store \"turso\" needs the ${TURSO_PACKAGE} package (${(err as Error).message})`);\n }\n\n let db: Database;\n try {\n // Floored at the same 30s sqlite opens with. `timeout` is connect-time only in this client;\n // the derived value is set via runtime PRAGMA below. `index_method` is required for ensureSchema()'s FTS indexes (T1).\n db = await turso.connect(dbPath, { timeout: 30_000, experimental: ['index_method'] });\n } catch (err) {\n throw new SenseError('STORE_DEPENDENCY_MISSING', `store \"turso\" failed to open ${dbPath}: ${(err as Error).message}`);\n }\n // A throw below must release this handle, or the leaked WAL makes the db file undeletable on\n // Windows. closed guards the catch against double-closing.\n let closed = false;\n try {\n const conn = createConnection(db);\n\n await conn.exec('CREATE TABLE IF NOT EXISTS meta (key TEXT PRIMARY KEY, value TEXT)');\n\n const version = await getMeta(conn, 'schema_version');\n const features = await getMeta(conn, 'features');\n const wantFeatures = featureSignature(cfg, FEATURES);\n if ((version !== null && version !== SCHEMA_VERSION) || (features !== null && features !== wantFeatures)) {\n // Reconcile only reparses changed files, so a stale cache is rebuilt rather than patched\n // incrementally. No tokenize-only or embed-identity-adopted partial path: any mismatch rebuilds.\n const reason = version !== null && version !== SCHEMA_VERSION ? 'cache format changed (new sensemaking version)' : 'config change (features, tokenizer, or presets)';\n console.error(`sense: ${reason}; rebuilding the index`);\n closed = true;\n await db.close();\n clearCache(cfg);\n return connect(cfg);\n }\n\n await ensureSchema(conn, cfg);\n\n // 3x the largest reconcile this cache has recorded, floored at 30s and capped at 10min.\n // Installed before reconcile() -- that call is the one that races a watcher's transaction.\n const recordedMaxMs = Number((await getMeta(conn, 'reconcile_max_ms')) ?? '0');\n await conn.exec(`PRAGMA busy_timeout = ${Math.min(Math.max(30000, 3 * recordedMaxMs), 600_000)}`);\n\n const { parsed, warnings } = await reconcile(conn, cfg, cfg.baseDir, tursoDialect);\n\n return { db, conn, cfg, dbPath, parsed, warnings };\n } catch (err) {\n if (!closed) await db.close();\n throw err;\n }\n}\n\n// The turso store's open: connects (fully async, see connect() above), then wraps the resulting\n// connection in the Store interface.\nexport async function openTurso(cfg: ResolvedConfig): Promise<OpenResult> {\n const { db, conn, cfg: resolvedCfg, dbPath, parsed, warnings } = await connect(cfg);\n const store = createStore(db, conn, resolvedCfg, resolvedCfg.baseDir);\n // reconcile ran before this object existed, so its chunk text is keyed by the connection.\n rekeyChunkText(conn, store);\n return { store: store, cfg: resolvedCfg, dbPath, parsed, warnings };\n}\n"],"names":["mkdirSync","join","featureSignature","STATE_DIR","rekeyChunkText","STORE_DIMS","SenseError","activeFeatures","FEATURES","clearCache","reconcile","getMeta","setMeta","createConnection","TURSO_PACKAGE","tursoApi","tursoDialect","createStore","DB_FILENAME","SCHEMA_VERSION","CONTENT_FTS_DDL","CONTENT_FTS_NAMES","ensureSchema","conn","cfg","exec","ddl","feature","name","schema","connect","stateDir","baseDir","recursive","dbPath","turso","err","message","db","timeout","experimental","closed","version","features","wantFeatures","reason","console","error","close","recordedMaxMs","Number","Math","min","max","parsed","warnings","openTurso","resolvedCfg","store"],"mappings":"AAAA,SAASA,SAAS,QAAQ,UAAU;AACpC,SAASC,IAAI,QAAQ,YAAY;AAGjC,SAASC,gBAAgB,EAAEC,SAAS,QAAQ,wBAAwB;AACpE,SAASC,cAAc,QAAQ,yBAAyB;AACxD,SAASC,UAAU,QAAQ,uBAAuB;AAClD,SAASC,UAAU,QAAQ,kBAAkB;AAC7C,SAASC,cAAc,EAAEC,QAAQ,QAAQ,0BAA0B;AACnE,SAASC,UAAU,QAAQ,cAAc;AACzC,SAASC,SAAS,QAAQ,kBAAkB;AAC5C,SAASC,OAAO,EAAEC,OAAO,QAAQ,eAAe;AAEhD,SAASC,gBAAgB,QAAQ,kBAAkB;AACnD,SAASC,aAAa,EAAEC,QAAQ,QAAQ,cAAc;AACtD,SAASC,YAAY,QAAQ,iBAAiB;AAC9C,SAASC,WAAW,QAAQ,aAAa;AAEzC,OAAO,MAAMC,cAAc,iBAAiB;AAC5C,wFAAwF;AACxF,mGAAmG;AACnG,OAAO,MAAMC,iBAAiB,IAAI;AAUlC,+FAA+F;AAC/F,2FAA2F;AAC3F,gGAAgG;AAChG,0FAA0F;AAC1F,OAAO,MAAMC,kBAAkB;IAC7B,CAAC,qIAAqI,CAAC;IACvI,CAAC,gMAAgM,CAAC;CACnM,CAAU;AACX,OAAO,MAAMC,oBAAoB;IAAC;IAAe;CAAoB,CAAU;AAE/E,eAAeC,aAAaC,IAAgB,EAAEC,GAAW;IACvD,MAAMD,KAAKE,IAAI,CAAC,CAAC,oIAAoI,CAAC;IACtJ,MAAMF,KAAKE,IAAI,CAAC,CAAC,wJAAwJ,CAAC;IAC1K,KAAK,MAAMC,OAAON,gBAAiB,MAAMG,KAAKE,IAAI,CAACC;IACnD,MAAMH,KAAKE,IAAI,CAAC,CAAC,gGAAgG,CAAC;IAClH,MAAMF,KAAKE,IAAI,CAAC;IAChB,KAAK,MAAME,WAAWpB,eAAeiB,KAAM;QACzC,8FAA8F;QAC9F,8FAA8F;QAC9F,IAAIG,QAAQC,IAAI,KAAK,SAAS;YAC5B,MAAML,KAAKE,IAAI,CAAC,CAAC,qIAAqI,EAAEpB,WAAW,+BAA+B,CAAC;YACnM;QACF;QACA,MAAMsB,QAAQE,MAAM,CAACN;IACvB;IACA,IAAI,AAAC,MAAMZ,QAAQY,MAAM,sBAAuB,MAAM,MAAMX,QAAQW,MAAM,kBAAkBJ;IAC5F,IAAI,AAAC,MAAMR,QAAQY,MAAM,gBAAiB,MAAM,MAAMX,QAAQW,MAAM,YAAYrB,iBAAiBsB,KAAKhB;AACxG;AAWA,eAAesB,QAAQN,GAAmB;IACxC,MAAMO,WAAW9B,KAAKuB,IAAIQ,OAAO,EAAE7B;IACnCH,UAAU+B,UAAU;QAAEE,WAAW;IAAK;IACtC,MAAMC,SAASjC,KAAK8B,UAAUb;IAE9B,8FAA8F;IAC9F,gGAAgG;IAChG,IAAIiB;IACJ,IAAI;QACFA,QAAQ,MAAMpB;IAChB,EAAE,OAAOqB,KAAK;QACZ,IAAIA,eAAe9B,YAAY,MAAM8B;QACrC,MAAM,IAAI9B,WAAW,4BAA4B,CAAC,wBAAwB,EAAEQ,cAAc,UAAU,EAAE,AAACsB,IAAcC,OAAO,CAAC,CAAC,CAAC;IACjI;IAEA,IAAIC;IACJ,IAAI;QACF,4FAA4F;QAC5F,uHAAuH;QACvHA,KAAK,MAAMH,MAAML,OAAO,CAACI,QAAQ;YAAEK,SAAS;YAAQC,cAAc;gBAAC;aAAe;QAAC;IACrF,EAAE,OAAOJ,KAAK;QACZ,MAAM,IAAI9B,WAAW,4BAA4B,CAAC,6BAA6B,EAAE4B,OAAO,EAAE,EAAE,AAACE,IAAcC,OAAO,EAAE;IACtH;IACA,6FAA6F;IAC7F,2DAA2D;IAC3D,IAAII,SAAS;IACb,IAAI;YAuB4B;QAtB9B,MAAMlB,OAAOV,iBAAiByB;QAE9B,MAAMf,KAAKE,IAAI,CAAC;QAEhB,MAAMiB,UAAU,MAAM/B,QAAQY,MAAM;QACpC,MAAMoB,WAAW,MAAMhC,QAAQY,MAAM;QACrC,MAAMqB,eAAe1C,iBAAiBsB,KAAKhB;QAC3C,IAAI,AAACkC,YAAY,QAAQA,YAAYvB,kBAAoBwB,aAAa,QAAQA,aAAaC,cAAe;YACxG,yFAAyF;YACzF,iGAAiG;YACjG,MAAMC,SAASH,YAAY,QAAQA,YAAYvB,iBAAiB,mDAAmD;YACnH2B,QAAQC,KAAK,CAAC,CAAC,OAAO,EAAEF,OAAO,sBAAsB,CAAC;YACtDJ,SAAS;YACT,MAAMH,GAAGU,KAAK;YACdvC,WAAWe;YACX,OAAOM,QAAQN;QACjB;QAEA,MAAMF,aAAaC,MAAMC;QAEzB,wFAAwF;QACxF,2FAA2F;QAC3F,MAAMyB,gBAAgBC,QAAQ,OAAA,MAAMvC,QAAQY,MAAM,iCAApB,kBAAA,OAA4C;QAC1E,MAAMA,KAAKE,IAAI,CAAC,CAAC,sBAAsB,EAAE0B,KAAKC,GAAG,CAACD,KAAKE,GAAG,CAAC,OAAO,IAAIJ,gBAAgB,SAAU;QAEhG,MAAM,EAAEK,MAAM,EAAEC,QAAQ,EAAE,GAAG,MAAM7C,UAAUa,MAAMC,KAAKA,IAAIQ,OAAO,EAAEhB;QAErE,OAAO;YAAEsB;YAAIf;YAAMC;YAAKU;YAAQoB;YAAQC;QAAS;IACnD,EAAE,OAAOnB,KAAK;QACZ,IAAI,CAACK,QAAQ,MAAMH,GAAGU,KAAK;QAC3B,MAAMZ;IACR;AACF;AAEA,gGAAgG;AAChG,qCAAqC;AACrC,OAAO,eAAeoB,UAAUhC,GAAmB;IACjD,MAAM,EAAEc,EAAE,EAAEf,IAAI,EAAEC,KAAKiC,WAAW,EAAEvB,MAAM,EAAEoB,MAAM,EAAEC,QAAQ,EAAE,GAAG,MAAMzB,QAAQN;IAC/E,MAAMkC,QAAQzC,YAAYqB,IAAIf,MAAMkC,aAAaA,YAAYzB,OAAO;IACpE,0FAA0F;IAC1F5B,eAAemB,MAAMmC;IACrB,OAAO;QAAEA,OAAOA;QAAOlC,KAAKiC;QAAavB;QAAQoB;QAAQC;IAAS;AACpE"}
@@ -1,6 +1,3 @@
1
- import type { Config } from '../../config/index.js';
2
- import type { Connection } from '../types.js';
3
- export declare function reconcile(conn: Connection, cfg: Config, baseDir: string): Promise<{
4
- parsed: number;
5
- warnings: string[];
6
- }>;
1
+ import type { ReconcileDialect } from '../types.js';
2
+ export declare const FTS_REBUILD_THRESHOLD = 250;
3
+ export declare const tursoDialect: ReconcileDialect;
@@ -1,20 +1,10 @@
1
1
  import { SenseError } from '../../errors.js';
2
- import { activeFeatures } from '../../features/index.js';
3
- import { progress } from '../../output/progress.js';
4
- import { listFiles, RESERVED_COLUMNS } from '../../scan/index.js';
5
- import { reparseFiles } from '../../scan/reparse.js';
6
2
  import { hasUnspacedRun } from '../../text/segment.js';
7
- import { getColumns, getMeta, quoteIdent, setMeta } from '../shared.js';
8
- import { withTransaction } from '../transaction.js';
9
- // Fork of sqlite/reconcile.ts, not duckdb's, so this store keeps reconcile_max_ms bookkeeping
10
- // (open.ts's busy timeout derives from it). content is a plain table with "_ngram" sidecars for lexical.ts.
11
- const CORE_FRONTMATTER_COLUMNS = new Set([
12
- 'path',
13
- '_mtime',
14
- '_ctime',
15
- '_size',
16
- '_parse_error'
17
- ]);
3
+ import { recordReconcileDuration } from '../shared.js';
4
+ import { BEGIN_WRITE } from '../transaction.js';
5
+ import { CONTENT_FTS_DDL, CONTENT_FTS_NAMES } from './open.js';
6
+ // This store's dialect (types.ts's ReconcileDialect) for the shared orchestration in
7
+ // store/reconcile.ts. content is a plain table with "_ngram" sidecars for lexical.ts.
18
8
  // '' when the field has no unspaced-script run (the common case), so the ngram index carries
19
9
  // nothing for it -- same "pay for nothing when absent" shape as sqlite's segmentField sidecars.
20
10
  function ngramSidecar(text) {
@@ -23,6 +13,9 @@ function ngramSidecar(text) {
23
13
  // Not a compile-time cap on ALTER TABLE ADD COLUMN (spike-measured: turso accepts 10,000 with
24
14
  // no error); the real fence is a SELECT projecting more than this many result columns, which fails to prepare.
25
15
  const MAX_FRONTMATTER_COLUMNS = 2000;
16
+ // Changed files above which rebuilding the FTS index beats maintaining it per row. Measured
17
+ // 2026-08-30; the derivation and its bounds are pinned in this file's spec.
18
+ export const FTS_REBUILD_THRESHOLD = 250;
26
19
  // No rowid coupling (unlike sqlite's FTS5 content): `path` is content's own primary key.
27
20
  const INSERT_CONTENT_SQL = `INSERT INTO content ("path", title, summary, text, title_ngram, summary_ngram, text_ngram) VALUES (?, ?, ?, ?, ?, ?, ?)`;
28
21
  function contentRow(doc) {
@@ -37,122 +30,28 @@ function contentRow(doc) {
37
30
  ngramSidecar(text)
38
31
  ];
39
32
  }
40
- export async function reconcile(conn, cfg, baseDir) {
41
- const files = listFiles(cfg, baseDir);
42
- const currentSet = new Set(files.map((f)=>f.relPath));
43
- const existingStmt = await conn.prepare('SELECT "path", "_mtime", "_size" FROM frontmatter');
44
- const existingRows = await existingStmt.all();
45
- const existing = new Map(existingRows.map((r)=>[
46
- r.path,
47
- r
33
+ async function reconcileContent(conn, touched, docs, delta) {
34
+ const churn = delta.reparsed.length + delta.vanished.length;
35
+ // Tantivy indexes per inserted row at a cost that grows with the batch, so a large insert is
36
+ // superlinear and a rebuild wins past the threshold.
37
+ const bulk = delta.files.length === 0 || churn > FTS_REBUILD_THRESHOLD;
38
+ if (bulk) for (const name of CONTENT_FTS_NAMES)await conn.exec(`DROP INDEX IF EXISTS ${name}`);
39
+ // content is a plain table keyed by its own path (no rowid subquery, unlike sqlite's FTS5
40
+ // content), so vanished and reparsed docs delete in one pass.
41
+ if (touched.length > 0) await conn.runBatch('DELETE FROM content WHERE "path" = ?', touched.map((p)=>[
42
+ p
48
43
  ]));
49
- const vanished = existingRows.filter((r)=>!currentSet.has(r.path)).map((r)=>r.path);
50
- const toReparse = files.filter((f)=>{
51
- const row = existing.get(f.relPath);
52
- return !row || row._mtime !== f.mtimeMs || row._size !== f.size;
53
- });
54
- if (vanished.length === 0 && toReparse.length === 0) return {
55
- parsed: 0,
56
- warnings: []
57
- };
58
- const features = activeFeatures(cfg);
59
- const seenColumns = await getColumns(conn);
60
- // Bulk reparses (a sync, a cold build) are the long silences a query can hit; short
61
- // reconciles stay silent (progress() has a threshold).
62
- const report = progress('reparsing files', toReparse.length);
63
- const { docs: parsedDocs, warnings, newColumns } = await reparseFiles(toReparse, features, cfg, seenColumns, report.tick);
64
- report.finish();
65
- for (const col of newColumns)seenColumns.add(col);
66
- const allColumns = [
67
- ...seenColumns
68
- ];
69
- if (allColumns.length > MAX_FRONTMATTER_COLUMNS) {
70
- throw new SenseError('COLUMN_LIMIT', `frontmatter would need ${allColumns.length} columns, crossing turso's SELECT result-set column limit (${MAX_FRONTMATTER_COLUMNS}; ALTER TABLE ADD COLUMN itself accepts far more, but a query projecting past this many columns fails to prepare). Narrow the presets' include globs so fewer/other files are indexed, or fix whatever is generating unbounded frontmatter keys.`);
71
- }
72
- // Columns the frontmatter upsert actually writes: core + parsed frontmatter keys, never a
73
- // feature-owned reserved column (see CORE_FRONTMATTER_COLUMNS above).
74
- const writableColumns = allColumns.filter((c)=>CORE_FRONTMATTER_COLUMNS.has(c) || !RESERVED_COLUMNS.has(c));
75
- // ON CONFLICT UPDATE, rather than the delete-and-insert `OR` clause, keeps the row's identity stable across reparses.
76
- const insertSql = `INSERT INTO frontmatter (${writableColumns.map(quoteIdent).join(', ')}) VALUES (${writableColumns.map(()=>'?').join(', ')}) ON CONFLICT("path") DO UPDATE SET ${writableColumns.filter((c)=>c !== 'path').map((c)=>`${quoteIdent(c)} = excluded.${quoteIdent(c)}`).join(', ')}`;
77
- const added = toReparse.filter((f)=>!existing.has(f.relPath)).map((f)=>f.relPath);
78
- const delta = {
79
- files,
80
- reparsed: parsedDocs.map((d)=>d.relPath),
81
- added,
82
- vanished
83
- };
84
- const addedSet = new Set(added);
85
- const reparsedExisting = parsedDocs.map((d)=>d.relPath).filter((p)=>!addedSet.has(p));
86
- const txStart = Date.now();
87
- await withTransaction(conn, async ()=>{
88
- var _feature_remove, _feature_afterReconcile;
89
- for (const col of newColumns)await conn.exec(`ALTER TABLE frontmatter ADD COLUMN ${quoteIdent(col)}`);
90
- if (parsedDocs.length > 0) {
91
- const rows = parsedDocs.map((doc)=>writableColumns.map((col)=>{
92
- var _doc_data_col;
93
- if (col === 'path') return doc.relPath;
94
- if (col === '_mtime') return doc.mtimeMs;
95
- if (col === '_ctime') return doc.ctimeMs;
96
- if (col === '_size') return doc.size;
97
- // Written per parse, unlike _rank, which a feature pass owns and the upsert skips.
98
- if (col === '_parse_error') return doc.parseError;
99
- return (_doc_data_col = doc.data[col]) !== null && _doc_data_col !== void 0 ? _doc_data_col : null;
100
- }));
101
- await conn.runBatch(insertSql, rows);
102
- }
103
- // content is a plain table keyed by its own path (no rowid subquery, unlike sqlite's FTS5
104
- // content), so vanished and reparsed docs delete in one pass.
105
- const contentTouched = [
106
- ...vanished,
107
- ...reparsedExisting
108
- ];
109
- if (contentTouched.length > 0) await conn.runBatch('DELETE FROM content WHERE "path" = ?', contentTouched.map((p)=>[
110
- p
111
- ]));
112
- if (parsedDocs.length > 0) await conn.runBatch(INSERT_CONTENT_SQL, parsedDocs.map(contentRow));
113
- if (vanished.length > 0) await conn.runBatch('DELETE FROM frontmatter WHERE "path" = ?', vanished.map((p)=>[
114
- p
115
- ]));
116
- // A preset edit forces a full rebuild, so an unchanged doc's coverage is already correct;
117
- // new docs have nothing to clear, which keeps cold builds linear.
118
- const presetTouched = [
119
- ...vanished,
120
- ...reparsedExisting
121
- ];
122
- if (presetTouched.length > 0) await conn.runBatch('DELETE FROM preset_files WHERE "path" = ?', presetTouched.map((p)=>[
123
- p
124
- ]));
125
- const presetRows = [];
126
- for (const doc of parsedDocs)for (const presetName of doc.presets)presetRows.push([
127
- doc.relPath,
128
- presetName
129
- ]);
130
- if (presetRows.length > 0) await conn.runBatch('INSERT INTO preset_files ("path", preset) VALUES (?, ?)', presetRows);
131
- const removedPaths = [
132
- ...vanished,
133
- ...reparsedExisting
134
- ];
135
- if (removedPaths.length > 0) for (const feature of features)await ((_feature_remove = feature.remove) === null || _feature_remove === void 0 ? void 0 : _feature_remove.call(feature, conn, removedPaths, delta));
136
- for (const feature of features){
137
- var _feature_store;
138
- const docsForFeature = parsedDocs.map((doc)=>({
139
- path: doc.relPath,
140
- extracted: doc.extracted[feature.name]
141
- }));
142
- await ((_feature_store = feature.store) === null || _feature_store === void 0 ? void 0 : _feature_store.call(feature, conn, docsForFeature, delta));
143
- }
144
- for (const feature of features)await ((_feature_afterReconcile = feature.afterReconcile) === null || _feature_afterReconcile === void 0 ? void 0 : _feature_afterReconcile.call(feature, conn, delta));
145
- });
146
- // Reconcile's own write-transaction duration, for open()'s derived busy_timeout: the observed
147
- // max is what the next open bounds its wait against.
148
- const durationMs = Date.now() - txStart;
149
- const prevRaw = await getMeta(conn, 'reconcile_max_ms');
150
- // -1, not 0, so a genuinely 0ms first reconcile (sub-millisecond, common on a tiny tree)
151
- // gets recorded rather than losing to the "nothing recorded yet" default.
152
- const prevMax = prevRaw === null ? -1 : Number(prevRaw);
153
- if (durationMs > prevMax) await setMeta(conn, 'reconcile_max_ms', String(durationMs));
154
- return {
155
- parsed: parsedDocs.length,
156
- warnings
157
- };
44
+ if (docs.length > 0) await conn.runBatch(INSERT_CONTENT_SQL, docs.map(contentRow));
45
+ if (bulk) for (const ddl of CONTENT_FTS_DDL)await conn.exec(ddl);
158
46
  }
47
+ export const tursoDialect = {
48
+ beginMode: ()=>BEGIN_WRITE,
49
+ checkColumnLimit (count) {
50
+ if (count > MAX_FRONTMATTER_COLUMNS) {
51
+ throw new SenseError('COLUMN_LIMIT', `frontmatter would need ${count} columns, crossing turso's SELECT result-set column limit (${MAX_FRONTMATTER_COLUMNS}; ALTER TABLE ADD COLUMN itself accepts far more, but a query projecting past this many columns fails to prepare). Narrow the presets' include globs so fewer/other files are indexed, or fix whatever is generating unbounded frontmatter keys.`);
52
+ }
53
+ },
54
+ columnTypeSuffix: ()=>'',
55
+ reconcileContent,
56
+ recordDuration: recordReconcileDuration
57
+ };
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/turso/reconcile.ts"],"sourcesContent":["import type { Config } from '../../config/index.ts';\nimport { SenseError } from '../../errors.ts';\nimport { activeFeatures } from '../../features/index.ts';\nimport type { ExtractedDoc, ReconcileDelta } from '../../features/types.ts';\nimport { progress } from '../../output/progress.ts';\nimport type { ParsedDoc } from '../../scan/index.ts';\nimport { listFiles, RESERVED_COLUMNS } from '../../scan/index.ts';\nimport { reparseFiles } from '../../scan/reparse.ts';\nimport { hasUnspacedRun } from '../../text/segment.ts';\nimport { getColumns, getMeta, quoteIdent, setMeta } from '../shared.ts';\nimport { withTransaction } from '../transaction.ts';\nimport type { Connection } from '../types.ts';\n\n// Fork of sqlite/reconcile.ts, not duckdb's, so this store keeps reconcile_max_ms bookkeeping\n// (open.ts's busy timeout derives from it). content is a plain table with \"_ngram\" sidecars for lexical.ts.\nconst CORE_FRONTMATTER_COLUMNS = new Set(['path', '_mtime', '_ctime', '_size', '_parse_error']);\n\n// '' when the field has no unspaced-script run (the common case), so the ngram index carries\n// nothing for it -- same \"pay for nothing when absent\" shape as sqlite's segmentField sidecars.\nfunction ngramSidecar(text: string): string {\n return hasUnspacedRun(text) ? text : '';\n}\n\n// Not a compile-time cap on ALTER TABLE ADD COLUMN (spike-measured: turso accepts 10,000 with\n// no error); the real fence is a SELECT projecting more than this many result columns, which fails to prepare.\nconst MAX_FRONTMATTER_COLUMNS = 2000;\n\n// No rowid coupling (unlike sqlite's FTS5 content): `path` is content's own primary key.\nconst INSERT_CONTENT_SQL = `INSERT INTO content (\"path\", title, summary, text, title_ngram, summary_ngram, text_ngram) VALUES (?, ?, ?, ?, ?, ?, ?)`;\n\nfunction contentRow(doc: ParsedDoc): unknown[] {\n const { title, summary, text } = doc.search;\n return [doc.relPath, title, summary, text, ngramSidecar(title), ngramSidecar(summary), ngramSidecar(text)];\n}\n\nexport async function reconcile(conn: Connection, cfg: Config, baseDir: string): Promise<{ parsed: number; warnings: string[] }> {\n const files = listFiles(cfg, baseDir);\n const currentSet = new Set(files.map((f) => f.relPath));\n\n const existingStmt = await conn.prepare('SELECT \"path\", \"_mtime\", \"_size\" FROM frontmatter');\n const existingRows = (await existingStmt.all()) as Array<{ path: string; _mtime: number; _size: number }>;\n const existing = new Map(existingRows.map((r) => [r.path, r]));\n const vanished = existingRows.filter((r) => !currentSet.has(r.path)).map((r) => r.path);\n\n const toReparse = files.filter((f) => {\n const row = existing.get(f.relPath);\n return !row || row._mtime !== f.mtimeMs || row._size !== f.size;\n });\n\n if (vanished.length === 0 && toReparse.length === 0) return { parsed: 0, warnings: [] };\n\n const features = activeFeatures(cfg);\n const seenColumns = await getColumns(conn);\n\n // Bulk reparses (a sync, a cold build) are the long silences a query can hit; short\n // reconciles stay silent (progress() has a threshold).\n const report = progress('reparsing files', toReparse.length);\n const { docs: parsedDocs, warnings, newColumns } = await reparseFiles(toReparse, features, cfg, seenColumns, report.tick);\n report.finish();\n for (const col of newColumns) seenColumns.add(col);\n\n const allColumns = [...seenColumns];\n if (allColumns.length > MAX_FRONTMATTER_COLUMNS) {\n throw new SenseError(\n 'COLUMN_LIMIT',\n `frontmatter would need ${allColumns.length} columns, crossing turso's SELECT result-set column limit (${MAX_FRONTMATTER_COLUMNS}; ALTER TABLE ADD COLUMN itself accepts far more, but a query projecting past this many columns fails to prepare). Narrow the presets' include globs so fewer/other files are indexed, or fix whatever is generating unbounded frontmatter keys.`\n );\n }\n // Columns the frontmatter upsert actually writes: core + parsed frontmatter keys, never a\n // feature-owned reserved column (see CORE_FRONTMATTER_COLUMNS above).\n const writableColumns = allColumns.filter((c) => CORE_FRONTMATTER_COLUMNS.has(c) || !RESERVED_COLUMNS.has(c));\n // ON CONFLICT UPDATE, rather than the delete-and-insert `OR` clause, keeps the row's identity stable across reparses.\n const insertSql = `INSERT INTO frontmatter (${writableColumns.map(quoteIdent).join(', ')}) VALUES (${writableColumns.map(() => '?').join(', ')}) ON CONFLICT(\"path\") DO UPDATE SET ${writableColumns\n .filter((c) => c !== 'path')\n .map((c) => `${quoteIdent(c)} = excluded.${quoteIdent(c)}`)\n .join(', ')}`;\n\n const added = toReparse.filter((f) => !existing.has(f.relPath)).map((f) => f.relPath);\n const delta: ReconcileDelta = { files, reparsed: parsedDocs.map((d) => d.relPath), added, vanished };\n const addedSet = new Set(added);\n const reparsedExisting = parsedDocs.map((d) => d.relPath).filter((p) => !addedSet.has(p));\n\n const txStart = Date.now();\n await withTransaction(conn, async () => {\n for (const col of newColumns) await conn.exec(`ALTER TABLE frontmatter ADD COLUMN ${quoteIdent(col)}`);\n\n if (parsedDocs.length > 0) {\n const rows = parsedDocs.map((doc) =>\n writableColumns.map((col) => {\n if (col === 'path') return doc.relPath;\n if (col === '_mtime') return doc.mtimeMs;\n if (col === '_ctime') return doc.ctimeMs;\n if (col === '_size') return doc.size;\n // Written per parse, unlike _rank, which a feature pass owns and the upsert skips.\n if (col === '_parse_error') return doc.parseError;\n return doc.data[col] ?? null;\n })\n );\n await conn.runBatch(insertSql, rows);\n }\n\n // content is a plain table keyed by its own path (no rowid subquery, unlike sqlite's FTS5\n // content), so vanished and reparsed docs delete in one pass.\n const contentTouched = [...vanished, ...reparsedExisting];\n if (contentTouched.length > 0)\n await conn.runBatch(\n 'DELETE FROM content WHERE \"path\" = ?',\n contentTouched.map((p) => [p])\n );\n if (parsedDocs.length > 0) await conn.runBatch(INSERT_CONTENT_SQL, parsedDocs.map(contentRow));\n\n if (vanished.length > 0)\n await conn.runBatch(\n 'DELETE FROM frontmatter WHERE \"path\" = ?',\n vanished.map((p) => [p])\n );\n\n // A preset edit forces a full rebuild, so an unchanged doc's coverage is already correct;\n // new docs have nothing to clear, which keeps cold builds linear.\n const presetTouched = [...vanished, ...reparsedExisting];\n if (presetTouched.length > 0)\n await conn.runBatch(\n 'DELETE FROM preset_files WHERE \"path\" = ?',\n presetTouched.map((p) => [p])\n );\n const presetRows: unknown[][] = [];\n for (const doc of parsedDocs) for (const presetName of doc.presets) presetRows.push([doc.relPath, presetName]);\n if (presetRows.length > 0) await conn.runBatch('INSERT INTO preset_files (\"path\", preset) VALUES (?, ?)', presetRows);\n\n const removedPaths = [...vanished, ...reparsedExisting];\n if (removedPaths.length > 0) for (const feature of features) await feature.remove?.(conn, removedPaths, delta);\n for (const feature of features) {\n const docsForFeature: ExtractedDoc[] = parsedDocs.map((doc) => ({ path: doc.relPath, extracted: doc.extracted[feature.name] }));\n await feature.store?.(conn, docsForFeature, delta);\n }\n for (const feature of features) await feature.afterReconcile?.(conn, delta);\n });\n\n // Reconcile's own write-transaction duration, for open()'s derived busy_timeout: the observed\n // max is what the next open bounds its wait against.\n const durationMs = Date.now() - txStart;\n const prevRaw = await getMeta(conn, 'reconcile_max_ms');\n // -1, not 0, so a genuinely 0ms first reconcile (sub-millisecond, common on a tiny tree)\n // gets recorded rather than losing to the \"nothing recorded yet\" default.\n const prevMax = prevRaw === null ? -1 : Number(prevRaw);\n if (durationMs > prevMax) await setMeta(conn, 'reconcile_max_ms', String(durationMs));\n\n return { parsed: parsedDocs.length, warnings };\n}\n"],"names":["SenseError","activeFeatures","progress","listFiles","RESERVED_COLUMNS","reparseFiles","hasUnspacedRun","getColumns","getMeta","quoteIdent","setMeta","withTransaction","CORE_FRONTMATTER_COLUMNS","Set","ngramSidecar","text","MAX_FRONTMATTER_COLUMNS","INSERT_CONTENT_SQL","contentRow","doc","title","summary","search","relPath","reconcile","conn","cfg","baseDir","files","currentSet","map","f","existingStmt","prepare","existingRows","all","existing","Map","r","path","vanished","filter","has","toReparse","row","get","_mtime","mtimeMs","_size","size","length","parsed","warnings","features","seenColumns","report","docs","parsedDocs","newColumns","tick","finish","col","add","allColumns","writableColumns","c","insertSql","join","added","delta","reparsed","d","addedSet","reparsedExisting","p","txStart","Date","now","feature","exec","rows","ctimeMs","parseError","data","runBatch","contentTouched","presetTouched","presetRows","presetName","presets","push","removedPaths","remove","docsForFeature","extracted","name","store","afterReconcile","durationMs","prevRaw","prevMax","Number","String"],"mappings":"AACA,SAASA,UAAU,QAAQ,kBAAkB;AAC7C,SAASC,cAAc,QAAQ,0BAA0B;AAEzD,SAASC,QAAQ,QAAQ,2BAA2B;AAEpD,SAASC,SAAS,EAAEC,gBAAgB,QAAQ,sBAAsB;AAClE,SAASC,YAAY,QAAQ,wBAAwB;AACrD,SAASC,cAAc,QAAQ,wBAAwB;AACvD,SAASC,UAAU,EAAEC,OAAO,EAAEC,UAAU,EAAEC,OAAO,QAAQ,eAAe;AACxE,SAASC,eAAe,QAAQ,oBAAoB;AAGpD,8FAA8F;AAC9F,4GAA4G;AAC5G,MAAMC,2BAA2B,IAAIC,IAAI;IAAC;IAAQ;IAAU;IAAU;IAAS;CAAe;AAE9F,6FAA6F;AAC7F,gGAAgG;AAChG,SAASC,aAAaC,IAAY;IAChC,OAAOT,eAAeS,QAAQA,OAAO;AACvC;AAEA,8FAA8F;AAC9F,+GAA+G;AAC/G,MAAMC,0BAA0B;AAEhC,yFAAyF;AACzF,MAAMC,qBAAqB,CAAC,uHAAuH,CAAC;AAEpJ,SAASC,WAAWC,GAAc;IAChC,MAAM,EAAEC,KAAK,EAAEC,OAAO,EAAEN,IAAI,EAAE,GAAGI,IAAIG,MAAM;IAC3C,OAAO;QAACH,IAAII,OAAO;QAAEH;QAAOC;QAASN;QAAMD,aAAaM;QAAQN,aAAaO;QAAUP,aAAaC;KAAM;AAC5G;AAEA,OAAO,eAAeS,UAAUC,IAAgB,EAAEC,GAAW,EAAEC,OAAe;IAC5E,MAAMC,QAAQzB,UAAUuB,KAAKC;IAC7B,MAAME,aAAa,IAAIhB,IAAIe,MAAME,GAAG,CAAC,CAACC,IAAMA,EAAER,OAAO;IAErD,MAAMS,eAAe,MAAMP,KAAKQ,OAAO,CAAC;IACxC,MAAMC,eAAgB,MAAMF,aAAaG,GAAG;IAC5C,MAAMC,WAAW,IAAIC,IAAIH,aAAaJ,GAAG,CAAC,CAACQ,IAAM;YAACA,EAAEC,IAAI;YAAED;SAAE;IAC5D,MAAME,WAAWN,aAAaO,MAAM,CAAC,CAACH,IAAM,CAACT,WAAWa,GAAG,CAACJ,EAAEC,IAAI,GAAGT,GAAG,CAAC,CAACQ,IAAMA,EAAEC,IAAI;IAEtF,MAAMI,YAAYf,MAAMa,MAAM,CAAC,CAACV;QAC9B,MAAMa,MAAMR,SAASS,GAAG,CAACd,EAAER,OAAO;QAClC,OAAO,CAACqB,OAAOA,IAAIE,MAAM,KAAKf,EAAEgB,OAAO,IAAIH,IAAII,KAAK,KAAKjB,EAAEkB,IAAI;IACjE;IAEA,IAAIT,SAASU,MAAM,KAAK,KAAKP,UAAUO,MAAM,KAAK,GAAG,OAAO;QAAEC,QAAQ;QAAGC,UAAU,EAAE;IAAC;IAEtF,MAAMC,WAAWpD,eAAeyB;IAChC,MAAM4B,cAAc,MAAM/C,WAAWkB;IAErC,oFAAoF;IACpF,uDAAuD;IACvD,MAAM8B,SAASrD,SAAS,mBAAmByC,UAAUO,MAAM;IAC3D,MAAM,EAAEM,MAAMC,UAAU,EAAEL,QAAQ,EAAEM,UAAU,EAAE,GAAG,MAAMrD,aAAasC,WAAWU,UAAU3B,KAAK4B,aAAaC,OAAOI,IAAI;IACxHJ,OAAOK,MAAM;IACb,KAAK,MAAMC,OAAOH,WAAYJ,YAAYQ,GAAG,CAACD;IAE9C,MAAME,aAAa;WAAIT;KAAY;IACnC,IAAIS,WAAWb,MAAM,GAAGlC,yBAAyB;QAC/C,MAAM,IAAIhB,WACR,gBACA,CAAC,uBAAuB,EAAE+D,WAAWb,MAAM,CAAC,2DAA2D,EAAElC,wBAAwB,gPAAgP,CAAC;IAEtX;IACA,0FAA0F;IAC1F,sEAAsE;IACtE,MAAMgD,kBAAkBD,WAAWtB,MAAM,CAAC,CAACwB,IAAMrD,yBAAyB8B,GAAG,CAACuB,MAAM,CAAC7D,iBAAiBsC,GAAG,CAACuB;IAC1G,sHAAsH;IACtH,MAAMC,YAAY,CAAC,yBAAyB,EAAEF,gBAAgBlC,GAAG,CAACrB,YAAY0D,IAAI,CAAC,MAAM,UAAU,EAAEH,gBAAgBlC,GAAG,CAAC,IAAM,KAAKqC,IAAI,CAAC,MAAM,oCAAoC,EAAEH,gBAClLvB,MAAM,CAAC,CAACwB,IAAMA,MAAM,QACpBnC,GAAG,CAAC,CAACmC,IAAM,GAAGxD,WAAWwD,GAAG,YAAY,EAAExD,WAAWwD,IAAI,EACzDE,IAAI,CAAC,OAAO;IAEf,MAAMC,QAAQzB,UAAUF,MAAM,CAAC,CAACV,IAAM,CAACK,SAASM,GAAG,CAACX,EAAER,OAAO,GAAGO,GAAG,CAAC,CAACC,IAAMA,EAAER,OAAO;IACpF,MAAM8C,QAAwB;QAAEzC;QAAO0C,UAAUb,WAAW3B,GAAG,CAAC,CAACyC,IAAMA,EAAEhD,OAAO;QAAG6C;QAAO5B;IAAS;IACnG,MAAMgC,WAAW,IAAI3D,IAAIuD;IACzB,MAAMK,mBAAmBhB,WAAW3B,GAAG,CAAC,CAACyC,IAAMA,EAAEhD,OAAO,EAAEkB,MAAM,CAAC,CAACiC,IAAM,CAACF,SAAS9B,GAAG,CAACgC;IAEtF,MAAMC,UAAUC,KAAKC,GAAG;IACxB,MAAMlE,gBAAgBc,MAAM;YA+CyCqD,iBAK7BA;QAnDtC,KAAK,MAAMjB,OAAOH,WAAY,MAAMjC,KAAKsD,IAAI,CAAC,CAAC,mCAAmC,EAAEtE,WAAWoD,MAAM;QAErG,IAAIJ,WAAWP,MAAM,GAAG,GAAG;YACzB,MAAM8B,OAAOvB,WAAW3B,GAAG,CAAC,CAACX,MAC3B6C,gBAAgBlC,GAAG,CAAC,CAAC+B;wBAOZ1C;oBANP,IAAI0C,QAAQ,QAAQ,OAAO1C,IAAII,OAAO;oBACtC,IAAIsC,QAAQ,UAAU,OAAO1C,IAAI4B,OAAO;oBACxC,IAAIc,QAAQ,UAAU,OAAO1C,IAAI8D,OAAO;oBACxC,IAAIpB,QAAQ,SAAS,OAAO1C,IAAI8B,IAAI;oBACpC,mFAAmF;oBACnF,IAAIY,QAAQ,gBAAgB,OAAO1C,IAAI+D,UAAU;oBACjD,QAAO/D,gBAAAA,IAAIgE,IAAI,CAACtB,IAAI,cAAb1C,2BAAAA,gBAAiB;gBAC1B;YAEF,MAAMM,KAAK2D,QAAQ,CAAClB,WAAWc;QACjC;QAEA,0FAA0F;QAC1F,8DAA8D;QAC9D,MAAMK,iBAAiB;eAAI7C;eAAaiC;SAAiB;QACzD,IAAIY,eAAenC,MAAM,GAAG,GAC1B,MAAMzB,KAAK2D,QAAQ,CACjB,wCACAC,eAAevD,GAAG,CAAC,CAAC4C,IAAM;gBAACA;aAAE;QAEjC,IAAIjB,WAAWP,MAAM,GAAG,GAAG,MAAMzB,KAAK2D,QAAQ,CAACnE,oBAAoBwC,WAAW3B,GAAG,CAACZ;QAElF,IAAIsB,SAASU,MAAM,GAAG,GACpB,MAAMzB,KAAK2D,QAAQ,CACjB,4CACA5C,SAASV,GAAG,CAAC,CAAC4C,IAAM;gBAACA;aAAE;QAG3B,0FAA0F;QAC1F,kEAAkE;QAClE,MAAMY,gBAAgB;eAAI9C;eAAaiC;SAAiB;QACxD,IAAIa,cAAcpC,MAAM,GAAG,GACzB,MAAMzB,KAAK2D,QAAQ,CACjB,6CACAE,cAAcxD,GAAG,CAAC,CAAC4C,IAAM;gBAACA;aAAE;QAEhC,MAAMa,aAA0B,EAAE;QAClC,KAAK,MAAMpE,OAAOsC,WAAY,KAAK,MAAM+B,cAAcrE,IAAIsE,OAAO,CAAEF,WAAWG,IAAI,CAAC;YAACvE,IAAII,OAAO;YAAEiE;SAAW;QAC7G,IAAID,WAAWrC,MAAM,GAAG,GAAG,MAAMzB,KAAK2D,QAAQ,CAAC,2DAA2DG;QAE1G,MAAMI,eAAe;eAAInD;eAAaiC;SAAiB;QACvD,IAAIkB,aAAazC,MAAM,GAAG,GAAG,KAAK,MAAM4B,WAAWzB,SAAU,QAAMyB,kBAAAA,QAAQc,MAAM,cAAdd,sCAAAA,qBAAAA,SAAiBrD,MAAMkE,cAActB;QACxG,KAAK,MAAMS,WAAWzB,SAAU;gBAExByB;YADN,MAAMe,iBAAiCpC,WAAW3B,GAAG,CAAC,CAACX,MAAS,CAAA;oBAAEoB,MAAMpB,IAAII,OAAO;oBAAEuE,WAAW3E,IAAI2E,SAAS,CAAChB,QAAQiB,IAAI,CAAC;gBAAC,CAAA;YAC5H,QAAMjB,iBAAAA,QAAQkB,KAAK,cAAblB,qCAAAA,oBAAAA,SAAgBrD,MAAMoE,gBAAgBxB;QAC9C;QACA,KAAK,MAAMS,WAAWzB,SAAU,QAAMyB,0BAAAA,QAAQmB,cAAc,cAAtBnB,8CAAAA,6BAAAA,SAAyBrD,MAAM4C;IACvE;IAEA,8FAA8F;IAC9F,qDAAqD;IACrD,MAAM6B,aAAatB,KAAKC,GAAG,KAAKF;IAChC,MAAMwB,UAAU,MAAM3F,QAAQiB,MAAM;IACpC,yFAAyF;IACzF,0EAA0E;IAC1E,MAAM2E,UAAUD,YAAY,OAAO,CAAC,IAAIE,OAAOF;IAC/C,IAAID,aAAaE,SAAS,MAAM1F,QAAQe,MAAM,oBAAoB6E,OAAOJ;IAEzE,OAAO;QAAE/C,QAAQM,WAAWP,MAAM;QAAEE;IAAS;AAC/C"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/turso/reconcile.ts"],"sourcesContent":["import { SenseError } from '../../errors.ts';\nimport type { ReconcileDelta } from '../../features/types.ts';\nimport type { ParsedDoc } from '../../scan/index.ts';\nimport { hasUnspacedRun } from '../../text/segment.ts';\nimport { recordReconcileDuration } from '../shared.ts';\nimport { BEGIN_WRITE } from '../transaction.ts';\nimport type { Connection, ReconcileDialect } from '../types.ts';\nimport { CONTENT_FTS_DDL, CONTENT_FTS_NAMES } from './open.ts';\n\n// This store's dialect (types.ts's ReconcileDialect) for the shared orchestration in\n// store/reconcile.ts. content is a plain table with \"_ngram\" sidecars for lexical.ts.\n\n// '' when the field has no unspaced-script run (the common case), so the ngram index carries\n// nothing for it -- same \"pay for nothing when absent\" shape as sqlite's segmentField sidecars.\nfunction ngramSidecar(text: string): string {\n return hasUnspacedRun(text) ? text : '';\n}\n\n// Not a compile-time cap on ALTER TABLE ADD COLUMN (spike-measured: turso accepts 10,000 with\n// no error); the real fence is a SELECT projecting more than this many result columns, which fails to prepare.\nconst MAX_FRONTMATTER_COLUMNS = 2000;\n\n// Changed files above which rebuilding the FTS index beats maintaining it per row. Measured\n// 2026-08-30; the derivation and its bounds are pinned in this file's spec.\nexport const FTS_REBUILD_THRESHOLD = 250;\n\n// No rowid coupling (unlike sqlite's FTS5 content): `path` is content's own primary key.\nconst INSERT_CONTENT_SQL = `INSERT INTO content (\"path\", title, summary, text, title_ngram, summary_ngram, text_ngram) VALUES (?, ?, ?, ?, ?, ?, ?)`;\n\nfunction contentRow(doc: ParsedDoc): unknown[] {\n const { title, summary, text } = doc.search;\n return [doc.relPath, title, summary, text, ngramSidecar(title), ngramSidecar(summary), ngramSidecar(text)];\n}\n\nasync function reconcileContent(conn: Connection, touched: string[], docs: ParsedDoc[], delta: ReconcileDelta): Promise<void> {\n const churn = delta.reparsed.length + delta.vanished.length;\n\n // Tantivy indexes per inserted row at a cost that grows with the batch, so a large insert is\n // superlinear and a rebuild wins past the threshold.\n const bulk = delta.files.length === 0 || churn > FTS_REBUILD_THRESHOLD;\n if (bulk) for (const name of CONTENT_FTS_NAMES) await conn.exec(`DROP INDEX IF EXISTS ${name}`);\n\n // content is a plain table keyed by its own path (no rowid subquery, unlike sqlite's FTS5\n // content), so vanished and reparsed docs delete in one pass.\n if (touched.length > 0)\n await conn.runBatch(\n 'DELETE FROM content WHERE \"path\" = ?',\n touched.map((p) => [p])\n );\n if (docs.length > 0) await conn.runBatch(INSERT_CONTENT_SQL, docs.map(contentRow));\n if (bulk) for (const ddl of CONTENT_FTS_DDL) await conn.exec(ddl);\n}\n\nexport const tursoDialect: ReconcileDialect = {\n beginMode: () => BEGIN_WRITE,\n checkColumnLimit(count) {\n if (count > MAX_FRONTMATTER_COLUMNS) {\n throw new SenseError(\n 'COLUMN_LIMIT',\n `frontmatter would need ${count} columns, crossing turso's SELECT result-set column limit (${MAX_FRONTMATTER_COLUMNS}; ALTER TABLE ADD COLUMN itself accepts far more, but a query projecting past this many columns fails to prepare). Narrow the presets' include globs so fewer/other files are indexed, or fix whatever is generating unbounded frontmatter keys.`\n );\n }\n },\n columnTypeSuffix: () => '',\n reconcileContent,\n recordDuration: recordReconcileDuration,\n};\n"],"names":["SenseError","hasUnspacedRun","recordReconcileDuration","BEGIN_WRITE","CONTENT_FTS_DDL","CONTENT_FTS_NAMES","ngramSidecar","text","MAX_FRONTMATTER_COLUMNS","FTS_REBUILD_THRESHOLD","INSERT_CONTENT_SQL","contentRow","doc","title","summary","search","relPath","reconcileContent","conn","touched","docs","delta","churn","reparsed","length","vanished","bulk","files","name","exec","runBatch","map","p","ddl","tursoDialect","beginMode","checkColumnLimit","count","columnTypeSuffix","recordDuration"],"mappings":"AAAA,SAASA,UAAU,QAAQ,kBAAkB;AAG7C,SAASC,cAAc,QAAQ,wBAAwB;AACvD,SAASC,uBAAuB,QAAQ,eAAe;AACvD,SAASC,WAAW,QAAQ,oBAAoB;AAEhD,SAASC,eAAe,EAAEC,iBAAiB,QAAQ,YAAY;AAE/D,qFAAqF;AACrF,sFAAsF;AAEtF,6FAA6F;AAC7F,gGAAgG;AAChG,SAASC,aAAaC,IAAY;IAChC,OAAON,eAAeM,QAAQA,OAAO;AACvC;AAEA,8FAA8F;AAC9F,+GAA+G;AAC/G,MAAMC,0BAA0B;AAEhC,4FAA4F;AAC5F,4EAA4E;AAC5E,OAAO,MAAMC,wBAAwB,IAAI;AAEzC,yFAAyF;AACzF,MAAMC,qBAAqB,CAAC,uHAAuH,CAAC;AAEpJ,SAASC,WAAWC,GAAc;IAChC,MAAM,EAAEC,KAAK,EAAEC,OAAO,EAAEP,IAAI,EAAE,GAAGK,IAAIG,MAAM;IAC3C,OAAO;QAACH,IAAII,OAAO;QAAEH;QAAOC;QAASP;QAAMD,aAAaO;QAAQP,aAAaQ;QAAUR,aAAaC;KAAM;AAC5G;AAEA,eAAeU,iBAAiBC,IAAgB,EAAEC,OAAiB,EAAEC,IAAiB,EAAEC,KAAqB;IAC3G,MAAMC,QAAQD,MAAME,QAAQ,CAACC,MAAM,GAAGH,MAAMI,QAAQ,CAACD,MAAM;IAE3D,6FAA6F;IAC7F,qDAAqD;IACrD,MAAME,OAAOL,MAAMM,KAAK,CAACH,MAAM,KAAK,KAAKF,QAAQb;IACjD,IAAIiB,MAAM,KAAK,MAAME,QAAQvB,kBAAmB,MAAMa,KAAKW,IAAI,CAAC,CAAC,qBAAqB,EAAED,MAAM;IAE9F,0FAA0F;IAC1F,8DAA8D;IAC9D,IAAIT,QAAQK,MAAM,GAAG,GACnB,MAAMN,KAAKY,QAAQ,CACjB,wCACAX,QAAQY,GAAG,CAAC,CAACC,IAAM;YAACA;SAAE;IAE1B,IAAIZ,KAAKI,MAAM,GAAG,GAAG,MAAMN,KAAKY,QAAQ,CAACpB,oBAAoBU,KAAKW,GAAG,CAACpB;IACtE,IAAIe,MAAM,KAAK,MAAMO,OAAO7B,gBAAiB,MAAMc,KAAKW,IAAI,CAACI;AAC/D;AAEA,OAAO,MAAMC,eAAiC;IAC5CC,WAAW,IAAMhC;IACjBiC,kBAAiBC,KAAK;QACpB,IAAIA,QAAQ7B,yBAAyB;YACnC,MAAM,IAAIR,WACR,gBACA,CAAC,uBAAuB,EAAEqC,MAAM,2DAA2D,EAAE7B,wBAAwB,gPAAgP,CAAC;QAE1W;IACF;IACA8B,kBAAkB,IAAM;IACxBrB;IACAsB,gBAAgBrC;AAClB,EAAE"}
@@ -1,10 +1,11 @@
1
1
  import { STORE_DIMS } from '../../embed/types.js';
2
+ import { reconcile } from '../reconcile.js';
2
3
  import { getColumns } from '../shared.js';
3
4
  import { withTransaction } from '../transaction.js';
4
5
  import { hasVectorRow, pendingRows } from '../vectors.js';
5
6
  import { fieldStats } from './fieldStats.js';
6
7
  import { queryLexical } from './lexical.js';
7
- import { reconcile } from './reconcile.js';
8
+ import { tursoDialect } from './reconcile.js';
8
9
  import { scanCandidates, scanSimilar, writeVectorBatch } from './vectors.js';
9
10
  // No 'snippets': fts_highlight returns the whole column, not a bounded window, so hits use
10
11
  // the caller's JS excerpt. No 'watch-concurrency': single-process by default, as on duckdb.
@@ -32,7 +33,7 @@ export function createStore(db, conn, cfg, baseDir) {
32
33
  return withTransaction(conn, fn);
33
34
  },
34
35
  async reconcile () {
35
- return reconcile(conn, cfg, baseDir);
36
+ return reconcile(conn, cfg, baseDir, tursoDialect);
36
37
  },
37
38
  docs: {
38
39
  async columns () {
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/turso/store.ts"],"sourcesContent":["import type { Database } from '@tursodatabase/database';\nimport type { Config } from '../../config/index.ts';\nimport { STORE_DIMS } from '../../embed/types.ts';\nimport { getColumns } from '../shared.ts';\nimport { withTransaction } from '../transaction.ts';\nimport type { Capability, Connection, Statement, Store } from '../types.ts';\nimport { hasVectorRow, pendingRows } from '../vectors.ts';\nimport { fieldStats } from './fieldStats.ts';\nimport { queryLexical } from './lexical.ts';\nimport { reconcile } from './reconcile.ts';\nimport { scanCandidates, scanSimilar, writeVectorBatch } from './vectors.ts';\n\n// No 'snippets': fts_highlight returns the whole column, not a bounded window, so hits use\n// the caller's JS excerpt. No 'watch-concurrency': single-process by default, as on duckdb.\nexport const CAPABILITIES: ReadonlySet<Capability> = new Set(['lexical', 'phrases', 'vectors']);\n\n// Shares one Connection instance (conn) with open()'s own reconcile call so transaction depth\n// (see transaction.ts) is tracked against the same object everywhere.\nexport function createStore(db: Database, conn: Connection, cfg: Config, baseDir: string): Store {\n return {\n name: 'turso',\n capabilities: CAPABILITIES,\n async exec(sql: string): Promise<void> {\n await conn.exec(sql);\n },\n async prepare(sql: string): Promise<Statement> {\n return conn.prepare(sql);\n },\n async runBatch(sql: string, paramRows: unknown[][]): Promise<void> {\n await conn.runBatch(sql, paramRows);\n },\n async transaction<T>(fn: () => Promise<T>): Promise<T> {\n return withTransaction(conn, fn);\n },\n async reconcile() {\n return reconcile(conn, cfg, baseDir);\n },\n docs: {\n async columns() {\n return [...(await getColumns(conn))];\n },\n fieldStats: (columns, scopeWhere) => fieldStats(conn, columns, scopeWhere),\n },\n lexical: {\n query: (terms, opts) => queryLexical(conn, terms, opts),\n },\n // The column's fixed DDL width (STORE_DIMS) is what every scan binds against, not the\n // interface's per-call storeDims -- see vectors.ts's padded() for why a shorter vector is still correct against a wider column.\n vectors: {\n pending: () => pendingRows(conn),\n writeVectors: (rows) => writeVectorBatch(conn, STORE_DIMS, rows),\n candidates: (qv, _storeDims, fetch, allowed) => scanCandidates(conn, qv, STORE_DIMS, fetch, allowed),\n similar: (path, opts) => scanSimilar(conn, STORE_DIMS, path, opts),\n hasVector: (path) => hasVectorRow(conn, path),\n },\n async engineStatus() {\n // Read back rather than recomputed: this is what open() actually set (3x the largest\n // recorded reconcile, floored at 30s, capped at 10min). Turso's PRAGMA busy_timeout names its column \"busy_timeout\" (spike-verified), not \"timeout\" like real SQLite.\n const row = (await (await db.prepare('PRAGMA busy_timeout')).get()) as { busy_timeout: number };\n return { busy_timeout: `${row.busy_timeout}ms (derived: 3x the largest reconcile this cache has recorded, floored at 30000ms)` };\n },\n raw: {\n async prepare(sql: string) {\n const stmt = await db.prepare(sql);\n stmt.safeIntegers(true); // int64 past 2^53 arrives as BigInt instead of losing precision\n return {\n columns: () => stmt.columns(),\n // sense sql streams through the client's own async generator.\n iterate: async function* (...params: unknown[]) {\n yield* stmt.iterate(...params);\n },\n };\n },\n },\n async close() {\n await db.close();\n },\n };\n}\n"],"names":["STORE_DIMS","getColumns","withTransaction","hasVectorRow","pendingRows","fieldStats","queryLexical","reconcile","scanCandidates","scanSimilar","writeVectorBatch","CAPABILITIES","Set","createStore","db","conn","cfg","baseDir","name","capabilities","exec","sql","prepare","runBatch","paramRows","transaction","fn","docs","columns","scopeWhere","lexical","query","terms","opts","vectors","pending","writeVectors","rows","candidates","qv","_storeDims","fetch","allowed","similar","path","hasVector","engineStatus","row","get","busy_timeout","raw","stmt","safeIntegers","iterate","params","close"],"mappings":"AAEA,SAASA,UAAU,QAAQ,uBAAuB;AAClD,SAASC,UAAU,QAAQ,eAAe;AAC1C,SAASC,eAAe,QAAQ,oBAAoB;AAEpD,SAASC,YAAY,EAAEC,WAAW,QAAQ,gBAAgB;AAC1D,SAASC,UAAU,QAAQ,kBAAkB;AAC7C,SAASC,YAAY,QAAQ,eAAe;AAC5C,SAASC,SAAS,QAAQ,iBAAiB;AAC3C,SAASC,cAAc,EAAEC,WAAW,EAAEC,gBAAgB,QAAQ,eAAe;AAE7E,2FAA2F;AAC3F,4FAA4F;AAC5F,OAAO,MAAMC,eAAwC,IAAIC,IAAI;IAAC;IAAW;IAAW;CAAU,EAAE;AAEhG,8FAA8F;AAC9F,sEAAsE;AACtE,OAAO,SAASC,YAAYC,EAAY,EAAEC,IAAgB,EAAEC,GAAW,EAAEC,OAAe;IACtF,OAAO;QACLC,MAAM;QACNC,cAAcR;QACd,MAAMS,MAAKC,GAAW;YACpB,MAAMN,KAAKK,IAAI,CAACC;QAClB;QACA,MAAMC,SAAQD,GAAW;YACvB,OAAON,KAAKO,OAAO,CAACD;QACtB;QACA,MAAME,UAASF,GAAW,EAAEG,SAAsB;YAChD,MAAMT,KAAKQ,QAAQ,CAACF,KAAKG;QAC3B;QACA,MAAMC,aAAeC,EAAoB;YACvC,OAAOxB,gBAAgBa,MAAMW;QAC/B;QACA,MAAMnB;YACJ,OAAOA,UAAUQ,MAAMC,KAAKC;QAC9B;QACAU,MAAM;YACJ,MAAMC;gBACJ,OAAO;uBAAK,MAAM3B,WAAWc;iBAAO;YACtC;YACAV,YAAY,CAACuB,SAASC,aAAexB,WAAWU,MAAMa,SAASC;QACjE;QACAC,SAAS;YACPC,OAAO,CAACC,OAAOC,OAAS3B,aAAaS,MAAMiB,OAAOC;QACpD;QACA,sFAAsF;QACtF,gIAAgI;QAChIC,SAAS;YACPC,SAAS,IAAM/B,YAAYW;YAC3BqB,cAAc,CAACC,OAAS3B,iBAAiBK,MAAMf,YAAYqC;YAC3DC,YAAY,CAACC,IAAIC,YAAYC,OAAOC,UAAYlC,eAAeO,MAAMwB,IAAIvC,YAAYyC,OAAOC;YAC5FC,SAAS,CAACC,MAAMX,OAASxB,YAAYM,MAAMf,YAAY4C,MAAMX;YAC7DY,WAAW,CAACD,OAASzC,aAAaY,MAAM6B;QAC1C;QACA,MAAME;YACJ,qFAAqF;YACrF,sKAAsK;YACtK,MAAMC,MAAO,MAAM,AAAC,CAAA,MAAMjC,GAAGQ,OAAO,CAAC,sBAAqB,EAAG0B,GAAG;YAChE,OAAO;gBAAEC,cAAc,GAAGF,IAAIE,YAAY,CAAC,kFAAkF,CAAC;YAAC;QACjI;QACAC,KAAK;YACH,MAAM5B,SAAQD,GAAW;gBACvB,MAAM8B,OAAO,MAAMrC,GAAGQ,OAAO,CAACD;gBAC9B8B,KAAKC,YAAY,CAAC,OAAO,gEAAgE;gBACzF,OAAO;oBACLxB,SAAS,IAAMuB,KAAKvB,OAAO;oBAC3B,8DAA8D;oBAC9DyB,SAAS,gBAAiB,GAAGC,MAAiB;wBAC5C,OAAOH,KAAKE,OAAO,IAAIC;oBACzB;gBACF;YACF;QACF;QACA,MAAMC;YACJ,MAAMzC,GAAGyC,KAAK;QAChB;IACF;AACF"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/turso/store.ts"],"sourcesContent":["import type { Database } from '@tursodatabase/database';\nimport type { Config } from '../../config/index.ts';\nimport { STORE_DIMS } from '../../embed/types.ts';\nimport { reconcile } from '../reconcile.ts';\nimport { getColumns } from '../shared.ts';\nimport { withTransaction } from '../transaction.ts';\nimport type { Capability, Connection, Statement, Store } from '../types.ts';\nimport { hasVectorRow, pendingRows } from '../vectors.ts';\nimport { fieldStats } from './fieldStats.ts';\nimport { queryLexical } from './lexical.ts';\nimport { tursoDialect } from './reconcile.ts';\nimport { scanCandidates, scanSimilar, writeVectorBatch } from './vectors.ts';\n\n// No 'snippets': fts_highlight returns the whole column, not a bounded window, so hits use\n// the caller's JS excerpt. No 'watch-concurrency': single-process by default, as on duckdb.\nexport const CAPABILITIES: ReadonlySet<Capability> = new Set(['lexical', 'phrases', 'vectors']);\n\n// Shares one Connection instance (conn) with open()'s own reconcile call so transaction depth\n// (see transaction.ts) is tracked against the same object everywhere.\nexport function createStore(db: Database, conn: Connection, cfg: Config, baseDir: string): Store {\n return {\n name: 'turso',\n capabilities: CAPABILITIES,\n async exec(sql: string): Promise<void> {\n await conn.exec(sql);\n },\n async prepare(sql: string): Promise<Statement> {\n return conn.prepare(sql);\n },\n async runBatch(sql: string, paramRows: unknown[][]): Promise<void> {\n await conn.runBatch(sql, paramRows);\n },\n async transaction<T>(fn: () => Promise<T>): Promise<T> {\n return withTransaction(conn, fn);\n },\n async reconcile() {\n return reconcile(conn, cfg, baseDir, tursoDialect);\n },\n docs: {\n async columns() {\n return [...(await getColumns(conn))];\n },\n fieldStats: (columns, scopeWhere) => fieldStats(conn, columns, scopeWhere),\n },\n lexical: {\n query: (terms, opts) => queryLexical(conn, terms, opts),\n },\n // The column's fixed DDL width (STORE_DIMS) is what every scan binds against, not the\n // interface's per-call storeDims -- see vectors.ts's padded() for why a shorter vector is still correct against a wider column.\n vectors: {\n pending: () => pendingRows(conn),\n writeVectors: (rows) => writeVectorBatch(conn, STORE_DIMS, rows),\n candidates: (qv, _storeDims, fetch, allowed) => scanCandidates(conn, qv, STORE_DIMS, fetch, allowed),\n similar: (path, opts) => scanSimilar(conn, STORE_DIMS, path, opts),\n hasVector: (path) => hasVectorRow(conn, path),\n },\n async engineStatus() {\n // Read back rather than recomputed: this is what open() actually set (3x the largest\n // recorded reconcile, floored at 30s, capped at 10min). Turso's PRAGMA busy_timeout names its column \"busy_timeout\" (spike-verified), not \"timeout\" like real SQLite.\n const row = (await (await db.prepare('PRAGMA busy_timeout')).get()) as { busy_timeout: number };\n return { busy_timeout: `${row.busy_timeout}ms (derived: 3x the largest reconcile this cache has recorded, floored at 30000ms)` };\n },\n raw: {\n async prepare(sql: string) {\n const stmt = await db.prepare(sql);\n stmt.safeIntegers(true); // int64 past 2^53 arrives as BigInt instead of losing precision\n return {\n columns: () => stmt.columns(),\n // sense sql streams through the client's own async generator.\n iterate: async function* (...params: unknown[]) {\n yield* stmt.iterate(...params);\n },\n };\n },\n },\n async close() {\n await db.close();\n },\n };\n}\n"],"names":["STORE_DIMS","reconcile","getColumns","withTransaction","hasVectorRow","pendingRows","fieldStats","queryLexical","tursoDialect","scanCandidates","scanSimilar","writeVectorBatch","CAPABILITIES","Set","createStore","db","conn","cfg","baseDir","name","capabilities","exec","sql","prepare","runBatch","paramRows","transaction","fn","docs","columns","scopeWhere","lexical","query","terms","opts","vectors","pending","writeVectors","rows","candidates","qv","_storeDims","fetch","allowed","similar","path","hasVector","engineStatus","row","get","busy_timeout","raw","stmt","safeIntegers","iterate","params","close"],"mappings":"AAEA,SAASA,UAAU,QAAQ,uBAAuB;AAClD,SAASC,SAAS,QAAQ,kBAAkB;AAC5C,SAASC,UAAU,QAAQ,eAAe;AAC1C,SAASC,eAAe,QAAQ,oBAAoB;AAEpD,SAASC,YAAY,EAAEC,WAAW,QAAQ,gBAAgB;AAC1D,SAASC,UAAU,QAAQ,kBAAkB;AAC7C,SAASC,YAAY,QAAQ,eAAe;AAC5C,SAASC,YAAY,QAAQ,iBAAiB;AAC9C,SAASC,cAAc,EAAEC,WAAW,EAAEC,gBAAgB,QAAQ,eAAe;AAE7E,2FAA2F;AAC3F,4FAA4F;AAC5F,OAAO,MAAMC,eAAwC,IAAIC,IAAI;IAAC;IAAW;IAAW;CAAU,EAAE;AAEhG,8FAA8F;AAC9F,sEAAsE;AACtE,OAAO,SAASC,YAAYC,EAAY,EAAEC,IAAgB,EAAEC,GAAW,EAAEC,OAAe;IACtF,OAAO;QACLC,MAAM;QACNC,cAAcR;QACd,MAAMS,MAAKC,GAAW;YACpB,MAAMN,KAAKK,IAAI,CAACC;QAClB;QACA,MAAMC,SAAQD,GAAW;YACvB,OAAON,KAAKO,OAAO,CAACD;QACtB;QACA,MAAME,UAASF,GAAW,EAAEG,SAAsB;YAChD,MAAMT,KAAKQ,QAAQ,CAACF,KAAKG;QAC3B;QACA,MAAMC,aAAeC,EAAoB;YACvC,OAAOxB,gBAAgBa,MAAMW;QAC/B;QACA,MAAM1B;YACJ,OAAOA,UAAUe,MAAMC,KAAKC,SAASV;QACvC;QACAoB,MAAM;YACJ,MAAMC;gBACJ,OAAO;uBAAK,MAAM3B,WAAWc;iBAAO;YACtC;YACAV,YAAY,CAACuB,SAASC,aAAexB,WAAWU,MAAMa,SAASC;QACjE;QACAC,SAAS;YACPC,OAAO,CAACC,OAAOC,OAAS3B,aAAaS,MAAMiB,OAAOC;QACpD;QACA,sFAAsF;QACtF,gIAAgI;QAChIC,SAAS;YACPC,SAAS,IAAM/B,YAAYW;YAC3BqB,cAAc,CAACC,OAAS3B,iBAAiBK,MAAMhB,YAAYsC;YAC3DC,YAAY,CAACC,IAAIC,YAAYC,OAAOC,UAAYlC,eAAeO,MAAMwB,IAAIxC,YAAY0C,OAAOC;YAC5FC,SAAS,CAACC,MAAMX,OAASxB,YAAYM,MAAMhB,YAAY6C,MAAMX;YAC7DY,WAAW,CAACD,OAASzC,aAAaY,MAAM6B;QAC1C;QACA,MAAME;YACJ,qFAAqF;YACrF,sKAAsK;YACtK,MAAMC,MAAO,MAAM,AAAC,CAAA,MAAMjC,GAAGQ,OAAO,CAAC,sBAAqB,EAAG0B,GAAG;YAChE,OAAO;gBAAEC,cAAc,GAAGF,IAAIE,YAAY,CAAC,kFAAkF,CAAC;YAAC;QACjI;QACAC,KAAK;YACH,MAAM5B,SAAQD,GAAW;gBACvB,MAAM8B,OAAO,MAAMrC,GAAGQ,OAAO,CAACD;gBAC9B8B,KAAKC,YAAY,CAAC,OAAO,gEAAgE;gBACzF,OAAO;oBACLxB,SAAS,IAAMuB,KAAKvB,OAAO;oBAC3B,8DAA8D;oBAC9DyB,SAAS,gBAAiB,GAAGC,MAAiB;wBAC5C,OAAOH,KAAKE,OAAO,IAAIC;oBACzB;gBACF;YACF;QACF;QACA,MAAMC;YACJ,MAAMzC,GAAGyC,KAAK;QAChB;IACF;AACF"}
@@ -1,5 +1,8 @@
1
- import type { StoreName } from '../config/index.js';
2
- export type Capability = 'phrases' | 'snippets' | 'watch-concurrency' | 'lexical' | 'vectors';
1
+ import type { Config, StoreName } from '../config/index.js';
2
+ import type { ReconcileDelta } from '../features/types.js';
3
+ import type { ParsedDoc } from '../scan/index.js';
4
+ export declare const CAPABILITY_NAMES: readonly ["phrases", "snippets", "watch-concurrency", "lexical", "vectors", "sql-functions"];
5
+ export type Capability = (typeof CAPABILITY_NAMES)[number];
3
6
  export interface RunResult {
4
7
  changes: number | bigint;
5
8
  lastInsertRowid: number | bigint;
@@ -19,6 +22,13 @@ export interface Connection {
19
22
  prepare(sql: string): Promise<Statement>;
20
23
  runBatch(sql: string, paramRows: unknown[][]): Promise<void>;
21
24
  }
25
+ export interface ReconcileDialect {
26
+ beginMode(): string;
27
+ checkColumnLimit(count: number): void;
28
+ columnTypeSuffix(): string;
29
+ reconcileContent(conn: Connection, touched: string[], docs: ParsedDoc[], delta: ReconcileDelta, cfg: Config): Promise<void>;
30
+ recordDuration?(conn: Connection, ms: number): Promise<void>;
31
+ }
22
32
  export interface FieldStat {
23
33
  field: string;
24
34
  coverage: number;
@@ -1,3 +1,14 @@
1
1
  // The backing-store interface: a minimal portable statement surface (exec/prepare), plus
2
2
  // dedicated interfaces exactly where engines diverge (lexical index, vector scan, raw sql).
3
- export { };
3
+ // 'lexical'/'vectors': the store's LexicalIndex/VectorStore is functionally implemented, not
4
+ // present-but-inert. 'sql-functions': the engine can register has/basename/segment as UDFs at all,
5
+ // which turso's client cannot. The rest are finer FTS5-only behaviors; a missing one fails at open
6
+ // or first use. An array, not a bare union, so a runtime check reads the same list the type does.
7
+ export const CAPABILITY_NAMES = [
8
+ 'phrases',
9
+ 'snippets',
10
+ 'watch-concurrency',
11
+ 'lexical',
12
+ 'vectors',
13
+ 'sql-functions'
14
+ ];
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/types.ts"],"sourcesContent":["// The backing-store interface: a minimal portable statement surface (exec/prepare), plus\n// dedicated interfaces exactly where engines diverge (lexical index, vector scan, raw sql).\n\nimport type { StoreName } from '../config/index.ts';\n\n// 'lexical'/'vectors': the store's LexicalIndex/VectorStore is functionally implemented, not\n// present-but-inert. The rest are finer FTS5-only behaviors; a missing one fails at open or first use.\nexport type Capability = 'phrases' | 'snippets' | 'watch-concurrency' | 'lexical' | 'vectors';\n\nexport interface RunResult {\n changes: number | bigint;\n lastInsertRowid: number | bigint;\n}\n\n// A prepared statement's async surface: every supported engine crosses a real async boundary\n// for a query (DuckDB has no synchronous client), so run/get/all return Promises.\nexport interface Statement {\n run(...params: unknown[]): Promise<RunResult>;\n get(...params: unknown[]): Promise<unknown>;\n all(...params: unknown[]): Promise<unknown[]>;\n iterate(...params: unknown[]): AsyncIterable<unknown>;\n columns(): Array<{ name: string }>;\n setReadBigInts(enabled: boolean): void;\n}\n\n// Connection surface feature-owned SQL runs against inside a store's hot loops (schema,\n// reconcile), portable so a feature never imports an engine-specific client type.\nexport interface Connection {\n exec(sql: string): Promise<void>;\n prepare(sql: string): Promise<Statement>;\n runBatch(sql: string, paramRows: unknown[][]): Promise<void>;\n}\n\nexport interface FieldStat {\n field: string;\n coverage: number;\n type: string;\n}\n\nexport interface DocumentStore {\n // Frontmatter column names (including internal ones; callers filter).\n columns(): Promise<string[]>;\n // Per-column coverage (non-null count) and observed type set, aggregated in one SQL query,\n // never one row per note. `scopeWhere` is a caller-built WHERE fragment, per LexicalQueryOptions.\n fieldStats(columns: string[], scopeWhere: string): Promise<FieldStat[]>;\n}\n\nexport interface LexicalHit {\n path: string;\n hit: string | null;\n}\n\nexport interface LexicalQueryOptions {\n whereJoin: string;\n whereCond: string;\n scopeCond: string;\n limit: number;\n}\n\nexport interface LexicalIndex {\n // Ranked word-match query with excerpt, scoped by the caller-built SQL fragments (the same\n // fragments narrowByWhere/materializeScope produce elsewhere).\n query(terms: string, opts: LexicalQueryOptions): Promise<LexicalHit[]>;\n}\n\nexport interface VectorCandidate {\n path: string;\n lines: string;\n similarity: number;\n}\n\nexport interface VectorSimilar {\n path: string;\n similarity: number;\n}\n\nexport interface VectorWriteRow {\n path: string;\n chunk: number;\n scale: number;\n vector: Buffer;\n}\n\nexport interface VectorStore {\n // Rows whose vector is NULL: never embedded, or added since.\n pending(): Promise<Array<{ path: string; chunk: number }>>;\n // One batch write per call (never per row) so a provider's embedding batch stays inside a\n // single store method.\n writeVectors(rows: VectorWriteRow[]): Promise<void>;\n candidates(queryVector: Float32Array, storeDims: number, fetch: number, allowed?: Set<string>): Promise<VectorCandidate[]>;\n similar(path: string, opts: { exclude: Set<string>; allowed?: Set<string>; k: number }): Promise<VectorSimilar[]>;\n hasVector(path: string): Promise<boolean>;\n}\n\n// The `sense sql` passthrough: string in, streamed rows out. Each store registers the same\n// sense-supplied functions and applies its own read-bigints/error-translation behavior.\nexport interface RawStatement {\n columns(): Array<{ name: string }>;\n iterate(...params: unknown[]): AsyncIterable<unknown>;\n}\n\nexport interface SqlSession {\n prepare(sql: string): Promise<RawStatement>;\n}\n\nexport interface Store {\n readonly name: StoreName;\n readonly capabilities: ReadonlySet<Capability>;\n exec(sql: string): Promise<void>;\n prepare(sql: string): Promise<Statement>;\n // One crossing, N rows: every external write loop (search's candidate insert, graph ring's\n // temp-table writes) goes through this instead of looping over `run()`. Same contract as Connection.runBatch.\n runBatch(sql: string, paramRows: unknown[][]): Promise<void>;\n // Pins one snapshot across multi-statement reads; `map` and `peek` use it. A network-bound\n // write (search's embedding top-up) must stay outside it. Nesting joins the enclosing transaction.\n transaction<T>(fn: () => Promise<T>): Promise<T>;\n // The whole file-sync pass (parse changed files, run every feature hook, resolve links,\n // recompute rank) behind one method: per-file iteration never crosses the async boundary.\n reconcile(): Promise<{ parsed: number; warnings: string[] }>;\n docs: DocumentStore;\n lexical: LexicalIndex;\n vectors: VectorStore;\n raw: SqlSession;\n // Engine-level facts for `sense status` (e.g. a derived busy_timeout PRAGMA reading); each\n // store owns what it reports and how it is worded. Empty when there is nothing to report.\n engineStatus(): Promise<Record<string, string>>;\n close(): Promise<void>;\n}\n"],"names":[],"mappings":"AAAA,yFAAyF;AACzF,4FAA4F;AAwG5F,WAsBC"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/types.ts"],"sourcesContent":["// The backing-store interface: a minimal portable statement surface (exec/prepare), plus\n// dedicated interfaces exactly where engines diverge (lexical index, vector scan, raw sql).\n\nimport type { Config, StoreName } from '../config/index.ts';\nimport type { ReconcileDelta } from '../features/types.ts';\nimport type { ParsedDoc } from '../scan/index.ts';\n\n// 'lexical'/'vectors': the store's LexicalIndex/VectorStore is functionally implemented, not\n// present-but-inert. 'sql-functions': the engine can register has/basename/segment as UDFs at all,\n// which turso's client cannot. The rest are finer FTS5-only behaviors; a missing one fails at open\n// or first use. An array, not a bare union, so a runtime check reads the same list the type does.\nexport const CAPABILITY_NAMES = ['phrases', 'snippets', 'watch-concurrency', 'lexical', 'vectors', 'sql-functions'] as const;\nexport type Capability = (typeof CAPABILITY_NAMES)[number];\n\nexport interface RunResult {\n changes: number | bigint;\n lastInsertRowid: number | bigint;\n}\n\n// A prepared statement's async surface: every supported engine crosses a real async boundary\n// for a query (DuckDB has no synchronous client), so run/get/all return Promises.\nexport interface Statement {\n run(...params: unknown[]): Promise<RunResult>;\n get(...params: unknown[]): Promise<unknown>;\n all(...params: unknown[]): Promise<unknown[]>;\n iterate(...params: unknown[]): AsyncIterable<unknown>;\n columns(): Array<{ name: string }>;\n setReadBigInts(enabled: boolean): void;\n}\n\n// Connection surface feature-owned SQL runs against inside a store's hot loops (schema,\n// reconcile), portable so a feature never imports an engine-specific client type.\nexport interface Connection {\n exec(sql: string): Promise<void>;\n prepare(sql: string): Promise<Statement>;\n runBatch(sql: string, paramRows: unknown[][]): Promise<void>;\n}\n\n// One reconcile algorithm (src/store/reconcile.ts), parameterised per engine. reconcileContent is\n// the load-bearing member: it lands every content-table change and owns its own multi-step\n// strategy (sqlite FTS5 incremental, duckdb combined delete/insert, turso incremental-or-rebuild).\nexport interface ReconcileDialect {\n // BEGIN mode for the whole reconcile: sqlite/turso 'BEGIN IMMEDIATE', duckdb 'BEGIN'.\n beginMode(): string;\n // Throws SenseError('COLUMN_LIMIT', ...) past this store's own column ceiling and reasoning.\n checkColumnLimit(count: number): void;\n // Type suffix for ALTER TABLE ADD COLUMN: '' for sqlite/turso, ' VARIANT' for duckdb.\n columnTypeSuffix(): string;\n // Deletes content rows for `touched`, inserts rows for `docs`; `delta` carries the tree state a\n // strategy may need. Must not open its own transaction, and must not return before its own\n // multi-step strategy (e.g. turso's DROP/rebuild) completes.\n reconcileContent(conn: Connection, touched: string[], docs: ParsedDoc[], delta: ReconcileDelta, cfg: Config): Promise<void>;\n // Records this reconcile's write-transaction duration. sqlite/turso use it for open()'s derived\n // busy_timeout; duckdb has no such PRAGMA and omits it.\n recordDuration?(conn: Connection, ms: number): Promise<void>;\n}\n\nexport interface FieldStat {\n field: string;\n coverage: number;\n type: string;\n}\n\nexport interface DocumentStore {\n // Frontmatter column names (including internal ones; callers filter).\n columns(): Promise<string[]>;\n // Per-column coverage (non-null count) and observed type set, aggregated in one SQL query,\n // never one row per note. `scopeWhere` is a caller-built WHERE fragment, per LexicalQueryOptions.\n fieldStats(columns: string[], scopeWhere: string): Promise<FieldStat[]>;\n}\n\nexport interface LexicalHit {\n path: string;\n hit: string | null;\n}\n\nexport interface LexicalQueryOptions {\n whereJoin: string;\n whereCond: string;\n scopeCond: string;\n limit: number;\n}\n\nexport interface LexicalIndex {\n // Ranked word-match query with excerpt, scoped by the caller-built SQL fragments (the same\n // fragments narrowByWhere/materializeScope produce elsewhere).\n query(terms: string, opts: LexicalQueryOptions): Promise<LexicalHit[]>;\n}\n\nexport interface VectorCandidate {\n path: string;\n lines: string;\n similarity: number;\n}\n\nexport interface VectorSimilar {\n path: string;\n similarity: number;\n}\n\nexport interface VectorWriteRow {\n path: string;\n chunk: number;\n scale: number;\n vector: Buffer;\n}\n\nexport interface VectorStore {\n // Rows whose vector is NULL: never embedded, or added since.\n pending(): Promise<Array<{ path: string; chunk: number }>>;\n // One batch write per call (never per row) so a provider's embedding batch stays inside a\n // single store method.\n writeVectors(rows: VectorWriteRow[]): Promise<void>;\n candidates(queryVector: Float32Array, storeDims: number, fetch: number, allowed?: Set<string>): Promise<VectorCandidate[]>;\n similar(path: string, opts: { exclude: Set<string>; allowed?: Set<string>; k: number }): Promise<VectorSimilar[]>;\n hasVector(path: string): Promise<boolean>;\n}\n\n// The `sense sql` passthrough: string in, streamed rows out. Each store registers the same\n// sense-supplied functions and applies its own read-bigints/error-translation behavior.\nexport interface RawStatement {\n columns(): Array<{ name: string }>;\n iterate(...params: unknown[]): AsyncIterable<unknown>;\n}\n\nexport interface SqlSession {\n prepare(sql: string): Promise<RawStatement>;\n}\n\nexport interface Store {\n readonly name: StoreName;\n readonly capabilities: ReadonlySet<Capability>;\n exec(sql: string): Promise<void>;\n prepare(sql: string): Promise<Statement>;\n // One crossing, N rows: every external write loop (search's candidate insert, graph ring's\n // temp-table writes) goes through this instead of looping over `run()`. Same contract as Connection.runBatch.\n runBatch(sql: string, paramRows: unknown[][]): Promise<void>;\n // Pins one snapshot across multi-statement reads; `map` and `peek` use it. A network-bound\n // write (search's embedding top-up) must stay outside it. Nesting joins the enclosing transaction.\n transaction<T>(fn: () => Promise<T>): Promise<T>;\n // The whole file-sync pass (parse changed files, run every feature hook, resolve links,\n // recompute rank) behind one method: per-file iteration never crosses the async boundary.\n reconcile(): Promise<{ parsed: number; warnings: string[] }>;\n docs: DocumentStore;\n lexical: LexicalIndex;\n vectors: VectorStore;\n raw: SqlSession;\n // Engine-level facts for `sense status` (e.g. a derived busy_timeout PRAGMA reading); each\n // store owns what it reports and how it is worded. Empty when there is nothing to report.\n engineStatus(): Promise<Record<string, string>>;\n close(): Promise<void>;\n}\n"],"names":["CAPABILITY_NAMES"],"mappings":"AAAA,yFAAyF;AACzF,4FAA4F;AAM5F,6FAA6F;AAC7F,mGAAmG;AACnG,mGAAmG;AACnG,kGAAkG;AAClG,OAAO,MAAMA,mBAAmB;IAAC;IAAW;IAAY;IAAqB;IAAW;IAAW;CAAgB,CAAU"}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "sensemaking",
3
- "version": "0.19.0",
3
+ "version": "0.19.2",
4
4
  "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",
5
5
  "keywords": [
6
6
  "markdown",