sensemaking 0.19.1 → 0.20.0

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 (167) hide show
  1. package/dist/cjs/chunk/extract.d.cts +1 -0
  2. package/dist/cjs/chunk/extract.d.ts +1 -0
  3. package/dist/cjs/chunk/extract.js +23 -5
  4. package/dist/cjs/chunk/extract.js.map +1 -1
  5. package/dist/cjs/chunk/index.d.cts +1 -1
  6. package/dist/cjs/chunk/index.d.ts +1 -1
  7. package/dist/cjs/chunk/index.js +3 -0
  8. package/dist/cjs/chunk/index.js.map +1 -1
  9. package/dist/cjs/index.d.cts +1 -1
  10. package/dist/cjs/index.d.ts +1 -1
  11. package/dist/cjs/index.js +3 -0
  12. package/dist/cjs/index.js.map +1 -1
  13. package/dist/cjs/scan/pool.d.cts +15 -0
  14. package/dist/cjs/scan/pool.d.ts +15 -0
  15. package/dist/cjs/scan/pool.js +263 -0
  16. package/dist/cjs/scan/pool.js.map +1 -0
  17. package/dist/cjs/scan/reparse.d.cts +2 -0
  18. package/dist/cjs/scan/reparse.d.ts +2 -0
  19. package/dist/cjs/scan/reparse.js +15 -76
  20. package/dist/cjs/scan/reparse.js.map +1 -1
  21. package/dist/cjs/store/builder.d.cts +11 -0
  22. package/dist/cjs/store/builder.d.ts +11 -0
  23. package/dist/cjs/store/builder.js +27 -0
  24. package/dist/cjs/store/builder.js.map +1 -0
  25. package/dist/cjs/store/duckdb/fieldStats.js +87 -10
  26. package/dist/cjs/store/duckdb/fieldStats.js.map +1 -1
  27. package/dist/cjs/store/duckdb/lexical.d.cts +1 -0
  28. package/dist/cjs/store/duckdb/lexical.d.ts +1 -0
  29. package/dist/cjs/store/duckdb/lexical.js +21 -5
  30. package/dist/cjs/store/duckdb/lexical.js.map +1 -1
  31. package/dist/cjs/store/duckdb/open.d.cts +9 -8
  32. package/dist/cjs/store/duckdb/open.d.ts +9 -8
  33. package/dist/cjs/store/duckdb/open.js +69 -120
  34. package/dist/cjs/store/duckdb/open.js.map +1 -1
  35. package/dist/cjs/store/duckdb/reconcile.d.cts +2 -6
  36. package/dist/cjs/store/duckdb/reconcile.d.ts +2 -6
  37. package/dist/cjs/store/duckdb/reconcile.js +45 -581
  38. package/dist/cjs/store/duckdb/reconcile.js.map +1 -1
  39. package/dist/cjs/store/duckdb/store.d.cts +1 -2
  40. package/dist/cjs/store/duckdb/store.d.ts +1 -2
  41. package/dist/cjs/store/duckdb/store.js +3 -26
  42. package/dist/cjs/store/duckdb/store.js.map +1 -1
  43. package/dist/cjs/store/index.d.cts +9 -3
  44. package/dist/cjs/store/index.d.ts +9 -3
  45. package/dist/cjs/store/index.js +40 -7
  46. package/dist/cjs/store/index.js.map +1 -1
  47. package/dist/cjs/store/open.d.cts +15 -0
  48. package/dist/cjs/store/open.d.ts +15 -0
  49. package/dist/cjs/store/open.js +516 -0
  50. package/dist/cjs/store/open.js.map +1 -0
  51. package/dist/cjs/store/reconcile.d.cts +7 -0
  52. package/dist/cjs/store/reconcile.d.ts +7 -0
  53. package/dist/cjs/store/reconcile.js +752 -0
  54. package/dist/cjs/store/reconcile.js.map +1 -0
  55. package/dist/cjs/store/shared.d.cts +1 -0
  56. package/dist/cjs/store/shared.d.ts +1 -0
  57. package/dist/cjs/store/shared.js +37 -0
  58. package/dist/cjs/store/shared.js.map +1 -1
  59. package/dist/cjs/store/signature.d.cts +3 -0
  60. package/dist/cjs/store/signature.d.ts +3 -0
  61. package/dist/cjs/store/signature.js +162 -0
  62. package/dist/cjs/store/signature.js.map +1 -0
  63. package/dist/cjs/store/sqlite/open.d.cts +9 -9
  64. package/dist/cjs/store/sqlite/open.d.ts +9 -9
  65. package/dist/cjs/store/sqlite/open.js +188 -198
  66. package/dist/cjs/store/sqlite/open.js.map +1 -1
  67. package/dist/cjs/store/sqlite/reconcile.d.cts +5 -10
  68. package/dist/cjs/store/sqlite/reconcile.d.ts +5 -10
  69. package/dist/cjs/store/sqlite/reconcile.js +48 -850
  70. package/dist/cjs/store/sqlite/reconcile.js.map +1 -1
  71. package/dist/cjs/store/sqlite/store.d.cts +1 -1
  72. package/dist/cjs/store/sqlite/store.d.ts +1 -1
  73. package/dist/cjs/store/sqlite/store.js +3 -13
  74. package/dist/cjs/store/sqlite/store.js.map +1 -1
  75. package/dist/cjs/store/turso/open.d.cts +6 -11
  76. package/dist/cjs/store/turso/open.d.ts +6 -11
  77. package/dist/cjs/store/turso/open.js +78 -169
  78. package/dist/cjs/store/turso/open.js.map +1 -1
  79. package/dist/cjs/store/turso/reconcile.d.cts +5 -6
  80. package/dist/cjs/store/turso/reconcile.d.ts +5 -6
  81. package/dist/cjs/store/turso/reconcile.js +169 -729
  82. package/dist/cjs/store/turso/reconcile.js.map +1 -1
  83. package/dist/cjs/store/turso/store.d.cts +1 -2
  84. package/dist/cjs/store/turso/store.d.ts +1 -2
  85. package/dist/cjs/store/turso/store.js +1 -12
  86. package/dist/cjs/store/turso/store.js.map +1 -1
  87. package/dist/cjs/store/types.d.cts +28 -6
  88. package/dist/cjs/store/types.d.ts +28 -6
  89. package/dist/cjs/store/types.js +14 -0
  90. package/dist/cjs/store/types.js.map +1 -1
  91. package/dist/cjs/text/strip.js +3 -6
  92. package/dist/cjs/text/strip.js.map +1 -1
  93. package/dist/cjs/watch.js +28 -16
  94. package/dist/cjs/watch.js.map +1 -1
  95. package/dist/esm/chunk/extract.d.ts +1 -0
  96. package/dist/esm/chunk/extract.js +10 -2
  97. package/dist/esm/chunk/extract.js.map +1 -1
  98. package/dist/esm/chunk/index.d.ts +1 -1
  99. package/dist/esm/chunk/index.js +1 -1
  100. package/dist/esm/chunk/index.js.map +1 -1
  101. package/dist/esm/index.d.ts +1 -1
  102. package/dist/esm/index.js +1 -1
  103. package/dist/esm/index.js.map +1 -1
  104. package/dist/esm/scan/pool.d.ts +15 -0
  105. package/dist/esm/scan/pool.js +77 -0
  106. package/dist/esm/scan/pool.js.map +1 -0
  107. package/dist/esm/scan/reparse.d.ts +2 -0
  108. package/dist/esm/scan/reparse.js +11 -58
  109. package/dist/esm/scan/reparse.js.map +1 -1
  110. package/dist/esm/store/builder.d.ts +11 -0
  111. package/dist/esm/store/builder.js +14 -0
  112. package/dist/esm/store/builder.js.map +1 -0
  113. package/dist/esm/store/duckdb/fieldStats.js +32 -19
  114. package/dist/esm/store/duckdb/fieldStats.js.map +1 -1
  115. package/dist/esm/store/duckdb/lexical.d.ts +1 -0
  116. package/dist/esm/store/duckdb/lexical.js +22 -9
  117. package/dist/esm/store/duckdb/lexical.js.map +1 -1
  118. package/dist/esm/store/duckdb/open.d.ts +9 -8
  119. package/dist/esm/store/duckdb/open.js +32 -57
  120. package/dist/esm/store/duckdb/open.js.map +1 -1
  121. package/dist/esm/store/duckdb/reconcile.d.ts +2 -6
  122. package/dist/esm/store/duckdb/reconcile.js +24 -117
  123. package/dist/esm/store/duckdb/reconcile.js.map +1 -1
  124. package/dist/esm/store/duckdb/store.d.ts +1 -2
  125. package/dist/esm/store/duckdb/store.js +4 -11
  126. package/dist/esm/store/duckdb/store.js.map +1 -1
  127. package/dist/esm/store/index.d.ts +9 -3
  128. package/dist/esm/store/index.js +23 -7
  129. package/dist/esm/store/index.js.map +1 -1
  130. package/dist/esm/store/open.d.ts +15 -0
  131. package/dist/esm/store/open.js +137 -0
  132. package/dist/esm/store/open.js.map +1 -0
  133. package/dist/esm/store/reconcile.d.ts +7 -0
  134. package/dist/esm/store/reconcile.js +132 -0
  135. package/dist/esm/store/reconcile.js.map +1 -0
  136. package/dist/esm/store/shared.d.ts +1 -0
  137. package/dist/esm/store/shared.js +9 -0
  138. package/dist/esm/store/shared.js.map +1 -1
  139. package/dist/esm/store/signature.d.ts +3 -0
  140. package/dist/esm/store/signature.js +35 -0
  141. package/dist/esm/store/signature.js.map +1 -0
  142. package/dist/esm/store/sqlite/open.d.ts +9 -9
  143. package/dist/esm/store/sqlite/open.js +81 -111
  144. package/dist/esm/store/sqlite/open.js.map +1 -1
  145. package/dist/esm/store/sqlite/reconcile.d.ts +5 -10
  146. package/dist/esm/store/sqlite/reconcile.js +31 -201
  147. package/dist/esm/store/sqlite/reconcile.js.map +1 -1
  148. package/dist/esm/store/sqlite/store.d.ts +1 -1
  149. package/dist/esm/store/sqlite/store.js +4 -7
  150. package/dist/esm/store/sqlite/store.js.map +1 -1
  151. package/dist/esm/store/turso/open.d.ts +6 -11
  152. package/dist/esm/store/turso/open.js +30 -77
  153. package/dist/esm/store/turso/open.js.map +1 -1
  154. package/dist/esm/store/turso/reconcile.d.ts +5 -6
  155. package/dist/esm/store/turso/reconcile.js +42 -146
  156. package/dist/esm/store/turso/reconcile.js.map +1 -1
  157. package/dist/esm/store/turso/store.d.ts +1 -2
  158. package/dist/esm/store/turso/store.js +2 -6
  159. package/dist/esm/store/turso/store.js.map +1 -1
  160. package/dist/esm/store/types.d.ts +28 -6
  161. package/dist/esm/store/types.js +12 -1
  162. package/dist/esm/store/types.js.map +1 -1
  163. package/dist/esm/text/strip.js +3 -4
  164. package/dist/esm/text/strip.js.map +1 -1
  165. package/dist/esm/watch.js +5 -3
  166. package/dist/esm/watch.js.map +1 -1
  167. package/package.json +2 -2
@@ -0,0 +1,15 @@
1
+ import type { ResolvedConfig } from '../config/index.js';
2
+ import type { Builder } from './builder.js';
3
+ import type { OpenDialect, Store } from './types.js';
4
+ export interface OpenResult {
5
+ store: Store;
6
+ cfg: ResolvedConfig;
7
+ dbPath: string;
8
+ parsed: number;
9
+ warnings: string[];
10
+ }
11
+ export declare function openWithDialect<Handle>(cfg: ResolvedConfig, dialect: OpenDialect<Handle>): Promise<OpenResult>;
12
+ export declare function openWithBuilder<Handle>(cfg: ResolvedConfig, dialect: OpenDialect<Handle>): Promise<OpenResult & {
13
+ builder: Builder;
14
+ }>;
15
+ export declare function docCount(store: Store): Promise<number>;
@@ -0,0 +1,137 @@
1
+ import { mkdirSync } from 'node:fs';
2
+ import { join } from 'node:path';
3
+ import { featureSignature, STATE_DIR } from '../config/index.js';
4
+ import { rekeyChunkText } from '../embed/handoff.js';
5
+ import { FEATURES } from '../features/index.js';
6
+ import { createBuilder } from './builder.js';
7
+ import { clearCache } from './cache.js';
8
+ import { getMeta, setMeta } from './shared.js';
9
+ import { changedSignatureKeys, embedIdentityAdopted, signatureDiff } from './signature.js';
10
+ async function connectWithDialect(cfg, dialect, keepBuilderOpen) {
11
+ const stateDir = join(cfg.baseDir, STATE_DIR);
12
+ mkdirSync(stateDir, {
13
+ recursive: true
14
+ });
15
+ const dbPath = join(stateDir, dialect.filename);
16
+ const { handle, conn } = await dialect.connect(dbPath, cfg);
17
+ // A throw below must release this handle, or the leaked WAL makes the cache file undeletable on
18
+ // Windows. `closed` keeps the catch from double-closing a handle a rebuild branch already closed.
19
+ let closed = false;
20
+ // Created here, not inside reconcile(): its pool must survive across every build() call this
21
+ // connection sees (a watcher's repeated ticks), not just the one below.
22
+ const builder = createBuilder(conn, cfg, cfg.baseDir, dialect.reconcileDialect);
23
+ try {
24
+ await conn.exec('CREATE TABLE IF NOT EXISTS meta (key TEXT PRIMARY KEY, value TEXT)');
25
+ // Schema-version or feature-set mismatch: reconcile only reparses changed files, so an
26
+ // old cache can't be patched incrementally -- rebuild instead (cheap: nothing expensive lives here).
27
+ const version = await getMeta(conn, 'schema_version');
28
+ const features = await getMeta(conn, 'features');
29
+ const wantFeatures = featureSignature(cfg, FEATURES);
30
+ let tokenizeOnlyRebuild = false;
31
+ if (version !== null && version !== dialect.schemaVersion || features !== null && features !== wantFeatures) {
32
+ // Indexing derives from presets, so a config edit rebuilding the cache must say so and
33
+ // name what changed -- silent rebuilds make derived indexing look like a hang or a bug.
34
+ if (version !== null && version !== dialect.schemaVersion) {
35
+ console.error('sense: cache format changed (new sensemaking version); rebuilding the index');
36
+ // Close before clearCache deletes the files out from under the still-open handle.
37
+ closed = true;
38
+ await dialect.close(handle);
39
+ clearCache(cfg);
40
+ return connectWithDialect(cfg, dialect, keepBuilderOpen);
41
+ }
42
+ const changedKeys = changedSignatureKeys(features !== null && features !== void 0 ? features : '', wantFeatures);
43
+ // Only a dialect-owned carve-out (sqlite's tokenizer) skips the rebuild below with its own
44
+ // partial DDL swap; every other single- or multi-segment change takes it.
45
+ if (dialect.partialRebuild && await dialect.partialRebuild(handle, conn, cfg, changedKeys)) {
46
+ tokenizeOnlyRebuild = true;
47
+ } else if (changedKeys.size === 1 && changedKeys.has('embed') && embedIdentityAdopted(features !== null && features !== void 0 ? features : '', wantFeatures)) {
48
+ // First sight of a resolved weight identity: the model itself hasn't changed, so
49
+ // adopt it into meta with no rebuild and no re-embed, mirroring the tokenize precedent.
50
+ console.error("sense: recorded the embedding model's resolved identity; vectors are unaffected");
51
+ await setMeta(conn, 'features', wantFeatures);
52
+ } else {
53
+ const changed = signatureDiff(features !== null && features !== void 0 ? features : '', wantFeatures);
54
+ console.error(`sense: config change (${changed}) rebuilds the index`);
55
+ closed = true;
56
+ await dialect.close(handle);
57
+ clearCache(cfg);
58
+ return connectWithDialect(cfg, dialect, keepBuilderOpen);
59
+ }
60
+ }
61
+ // A rebuild trigger independent of the signature comparison above (sqlite: the content
62
+ // table's own DDL no longer matches the resolved tokenizer).
63
+ const extraReason = dialect.extraRebuildReason ? dialect.extraRebuildReason(handle) : null;
64
+ if (extraReason !== null) {
65
+ console.error(`sense: ${extraReason}; rebuilding the index`);
66
+ closed = true;
67
+ await dialect.close(handle);
68
+ clearCache(cfg);
69
+ return connectWithDialect(cfg, dialect, keepBuilderOpen);
70
+ }
71
+ await dialect.ensureSchema(handle, conn, cfg);
72
+ let rebuildWarnings = [];
73
+ if (tokenizeOnlyRebuild && dialect.postSchemaRebuild) {
74
+ rebuildWarnings = await dialect.postSchemaRebuild(handle, conn, cfg, cfg.baseDir, wantFeatures);
75
+ }
76
+ // 3x the largest reconcile this cache has recorded, floored at 30s and capped at 10min.
77
+ // Installed before build() -- that call is the one that races a watcher's transaction.
78
+ if (dialect.setDerivedBusyTimeout) {
79
+ var _ref;
80
+ const recordedMaxMs = Number((_ref = await getMeta(conn, 'reconcile_max_ms')) !== null && _ref !== void 0 ? _ref : '0');
81
+ await dialect.setDerivedBusyTimeout(handle, conn, Math.min(Math.max(30000, 3 * recordedMaxMs), 600000));
82
+ }
83
+ const { parsed, warnings } = await builder.build();
84
+ // A one-shot open is done reconciling for good; a watcher keeps `builder` alive across its run.
85
+ if (!keepBuilderOpen) await builder.close();
86
+ return {
87
+ handle,
88
+ conn,
89
+ cfg,
90
+ dbPath,
91
+ parsed,
92
+ warnings: [
93
+ ...rebuildWarnings,
94
+ ...warnings
95
+ ],
96
+ builder
97
+ };
98
+ } catch (err) {
99
+ await builder.close();
100
+ if (!closed) await dialect.close(handle);
101
+ throw err;
102
+ }
103
+ }
104
+ // A store's open(): connects (see connectWithDialect above), then wraps the resulting connection
105
+ // in the Store interface. The builder's pool closes right after the initial build.
106
+ export async function openWithDialect(cfg, dialect) {
107
+ const { handle, conn, cfg: resolvedCfg, dbPath, parsed, warnings } = await connectWithDialect(cfg, dialect, false);
108
+ const store = dialect.createStore(handle, conn, resolvedCfg);
109
+ // reconcile ran before this object existed, so its chunk text is keyed by the connection.
110
+ rekeyChunkText(conn, store);
111
+ return {
112
+ store,
113
+ cfg: resolvedCfg,
114
+ dbPath,
115
+ parsed,
116
+ warnings
117
+ };
118
+ }
119
+ // Same connect+build as openWithDialect, but keeps the builder's pool alive and hands it back
120
+ // alongside the store, for a caller (a watcher) that reconciles repeatedly on this connection.
121
+ export async function openWithBuilder(cfg, dialect) {
122
+ const { handle, conn, cfg: resolvedCfg, dbPath, parsed, warnings, builder } = await connectWithDialect(cfg, dialect, true);
123
+ const store = dialect.createStore(handle, conn, resolvedCfg);
124
+ rekeyChunkText(conn, store);
125
+ return {
126
+ store,
127
+ cfg: resolvedCfg,
128
+ dbPath,
129
+ parsed,
130
+ warnings,
131
+ builder
132
+ };
133
+ }
134
+ export async function docCount(store) {
135
+ const stmt = await store.prepare('SELECT COUNT(*) AS n FROM frontmatter');
136
+ return (await stmt.get()).n;
137
+ }
@@ -0,0 +1 @@
1
+ {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/open.ts"],"sourcesContent":["import { mkdirSync } from 'node:fs';\nimport { join } from 'node:path';\nimport type { ResolvedConfig } from '../config/index.ts';\nimport { featureSignature, STATE_DIR } from '../config/index.ts';\nimport { rekeyChunkText } from '../embed/handoff.ts';\nimport { FEATURES } from '../features/index.ts';\nimport type { Builder } from './builder.ts';\nimport { createBuilder } from './builder.ts';\nimport { clearCache } from './cache.ts';\nimport { getMeta, setMeta } from './shared.ts';\nimport { changedSignatureKeys, embedIdentityAdopted, signatureDiff } from './signature.ts';\nimport type { Connection, OpenDialect, Store } from './types.ts';\n\n// One open algorithm shared by every store, parameterised by a per-engine OpenDialect (types.ts).\n// Ordering is universal, not a dialect concern: connect, meta table, schema-version and\n// feature-signature comparison (rebuild, adopt, or partial rebuild), any extra rebuild trigger,\n// ensureSchema, the derived busy_timeout PRAGMA, reconcile, then the store itself.\n\nexport interface OpenResult {\n store: Store;\n cfg: ResolvedConfig;\n dbPath: string;\n parsed: number;\n warnings: string[];\n}\n\ninterface ConnectResult<Handle> {\n handle: Handle;\n conn: Connection;\n cfg: ResolvedConfig;\n dbPath: string;\n parsed: number;\n warnings: string[];\n // Closed already unless `keepBuilderOpen` was set: a one-shot open reconciles once and is done\n // with it, a watcher keeps calling build() on the same instance across its whole run.\n builder: Builder;\n}\n\nasync function connectWithDialect<Handle>(cfg: ResolvedConfig, dialect: OpenDialect<Handle>, keepBuilderOpen: boolean): Promise<ConnectResult<Handle>> {\n const stateDir = join(cfg.baseDir, STATE_DIR);\n mkdirSync(stateDir, { recursive: true });\n const dbPath = join(stateDir, dialect.filename);\n\n const { handle, conn } = await dialect.connect(dbPath, cfg);\n // A throw below must release this handle, or the leaked WAL makes the cache file undeletable on\n // Windows. `closed` keeps the catch from double-closing a handle a rebuild branch already closed.\n let closed = false;\n // Created here, not inside reconcile(): its pool must survive across every build() call this\n // connection sees (a watcher's repeated ticks), not just the one below.\n const builder = createBuilder(conn, cfg, cfg.baseDir, dialect.reconcileDialect);\n try {\n await conn.exec('CREATE TABLE IF NOT EXISTS meta (key TEXT PRIMARY KEY, value TEXT)');\n\n // Schema-version or feature-set mismatch: reconcile only reparses changed files, so an\n // old cache can't be patched incrementally -- rebuild instead (cheap: nothing expensive lives here).\n const version = await getMeta(conn, 'schema_version');\n const features = await getMeta(conn, 'features');\n const wantFeatures = featureSignature(cfg, FEATURES);\n let tokenizeOnlyRebuild = false;\n if ((version !== null && version !== dialect.schemaVersion) || (features !== null && features !== wantFeatures)) {\n // Indexing derives from presets, so a config edit rebuilding the cache must say so and\n // name what changed -- silent rebuilds make derived indexing look like a hang or a bug.\n if (version !== null && version !== dialect.schemaVersion) {\n console.error('sense: cache format changed (new sensemaking version); rebuilding the index');\n // Close before clearCache deletes the files out from under the still-open handle.\n closed = true;\n await dialect.close(handle);\n clearCache(cfg);\n return connectWithDialect(cfg, dialect, keepBuilderOpen);\n }\n const changedKeys = changedSignatureKeys(features ?? '', wantFeatures);\n // Only a dialect-owned carve-out (sqlite's tokenizer) skips the rebuild below with its own\n // partial DDL swap; every other single- or multi-segment change takes it.\n if (dialect.partialRebuild && (await dialect.partialRebuild(handle, conn, cfg, changedKeys))) {\n tokenizeOnlyRebuild = true;\n } else if (changedKeys.size === 1 && changedKeys.has('embed') && embedIdentityAdopted(features ?? '', wantFeatures)) {\n // First sight of a resolved weight identity: the model itself hasn't changed, so\n // adopt it into meta with no rebuild and no re-embed, mirroring the tokenize precedent.\n console.error(\"sense: recorded the embedding model's resolved identity; vectors are unaffected\");\n await setMeta(conn, 'features', wantFeatures);\n } else {\n const changed = signatureDiff(features ?? '', wantFeatures);\n console.error(`sense: config change (${changed}) rebuilds the index`);\n closed = true;\n await dialect.close(handle);\n clearCache(cfg);\n return connectWithDialect(cfg, dialect, keepBuilderOpen);\n }\n }\n\n // A rebuild trigger independent of the signature comparison above (sqlite: the content\n // table's own DDL no longer matches the resolved tokenizer).\n const extraReason = dialect.extraRebuildReason ? dialect.extraRebuildReason(handle) : null;\n if (extraReason !== null) {\n console.error(`sense: ${extraReason}; rebuilding the index`);\n closed = true;\n await dialect.close(handle);\n clearCache(cfg);\n return connectWithDialect(cfg, dialect, keepBuilderOpen);\n }\n\n await dialect.ensureSchema(handle, conn, cfg);\n\n let rebuildWarnings: string[] = [];\n if (tokenizeOnlyRebuild && dialect.postSchemaRebuild) {\n rebuildWarnings = await dialect.postSchemaRebuild(handle, conn, cfg, cfg.baseDir, wantFeatures);\n }\n\n // 3x the largest reconcile this cache has recorded, floored at 30s and capped at 10min.\n // Installed before build() -- that call is the one that races a watcher's transaction.\n if (dialect.setDerivedBusyTimeout) {\n const recordedMaxMs = Number((await getMeta(conn, 'reconcile_max_ms')) ?? '0');\n await dialect.setDerivedBusyTimeout(handle, conn, Math.min(Math.max(30000, 3 * recordedMaxMs), 600_000));\n }\n\n const { parsed, warnings } = await builder.build();\n // A one-shot open is done reconciling for good; a watcher keeps `builder` alive across its run.\n if (!keepBuilderOpen) await builder.close();\n\n return { handle, conn, cfg, dbPath, parsed, warnings: [...rebuildWarnings, ...warnings], builder };\n } catch (err) {\n await builder.close();\n if (!closed) await dialect.close(handle);\n throw err;\n }\n}\n\n// A store's open(): connects (see connectWithDialect above), then wraps the resulting connection\n// in the Store interface. The builder's pool closes right after the initial build.\nexport async function openWithDialect<Handle>(cfg: ResolvedConfig, dialect: OpenDialect<Handle>): Promise<OpenResult> {\n const { handle, conn, cfg: resolvedCfg, dbPath, parsed, warnings } = await connectWithDialect(cfg, dialect, false);\n const store = dialect.createStore(handle, conn, resolvedCfg);\n // reconcile ran before this object existed, so its chunk text is keyed by the connection.\n rekeyChunkText(conn, store);\n return { store, cfg: resolvedCfg, dbPath, parsed, warnings };\n}\n\n// Same connect+build as openWithDialect, but keeps the builder's pool alive and hands it back\n// alongside the store, for a caller (a watcher) that reconciles repeatedly on this connection.\nexport async function openWithBuilder<Handle>(cfg: ResolvedConfig, dialect: OpenDialect<Handle>): Promise<OpenResult & { builder: Builder }> {\n const { handle, conn, cfg: resolvedCfg, dbPath, parsed, warnings, builder } = await connectWithDialect(cfg, dialect, true);\n const store = dialect.createStore(handle, conn, resolvedCfg);\n rekeyChunkText(conn, store);\n return { store, cfg: resolvedCfg, dbPath, parsed, warnings, builder };\n}\n\nexport async function docCount(store: Store): Promise<number> {\n const stmt = await store.prepare('SELECT COUNT(*) AS n FROM frontmatter');\n return ((await stmt.get()) as { n: number }).n;\n}\n"],"names":["mkdirSync","join","featureSignature","STATE_DIR","rekeyChunkText","FEATURES","createBuilder","clearCache","getMeta","setMeta","changedSignatureKeys","embedIdentityAdopted","signatureDiff","connectWithDialect","cfg","dialect","keepBuilderOpen","stateDir","baseDir","recursive","dbPath","filename","handle","conn","connect","closed","builder","reconcileDialect","exec","version","features","wantFeatures","tokenizeOnlyRebuild","schemaVersion","console","error","close","changedKeys","partialRebuild","size","has","changed","extraReason","extraRebuildReason","ensureSchema","rebuildWarnings","postSchemaRebuild","setDerivedBusyTimeout","recordedMaxMs","Number","Math","min","max","parsed","warnings","build","err","openWithDialect","resolvedCfg","store","createStore","openWithBuilder","docCount","stmt","prepare","get","n"],"mappings":"AAAA,SAASA,SAAS,QAAQ,UAAU;AACpC,SAASC,IAAI,QAAQ,YAAY;AAEjC,SAASC,gBAAgB,EAAEC,SAAS,QAAQ,qBAAqB;AACjE,SAASC,cAAc,QAAQ,sBAAsB;AACrD,SAASC,QAAQ,QAAQ,uBAAuB;AAEhD,SAASC,aAAa,QAAQ,eAAe;AAC7C,SAASC,UAAU,QAAQ,aAAa;AACxC,SAASC,OAAO,EAAEC,OAAO,QAAQ,cAAc;AAC/C,SAASC,oBAAoB,EAAEC,oBAAoB,EAAEC,aAAa,QAAQ,iBAAiB;AA4B3F,eAAeC,mBAA2BC,GAAmB,EAAEC,OAA4B,EAAEC,eAAwB;IACnH,MAAMC,WAAWhB,KAAKa,IAAII,OAAO,EAAEf;IACnCH,UAAUiB,UAAU;QAAEE,WAAW;IAAK;IACtC,MAAMC,SAASnB,KAAKgB,UAAUF,QAAQM,QAAQ;IAE9C,MAAM,EAAEC,MAAM,EAAEC,IAAI,EAAE,GAAG,MAAMR,QAAQS,OAAO,CAACJ,QAAQN;IACvD,gGAAgG;IAChG,kGAAkG;IAClG,IAAIW,SAAS;IACb,6FAA6F;IAC7F,wEAAwE;IACxE,MAAMC,UAAUpB,cAAciB,MAAMT,KAAKA,IAAII,OAAO,EAAEH,QAAQY,gBAAgB;IAC9E,IAAI;QACF,MAAMJ,KAAKK,IAAI,CAAC;QAEhB,uFAAuF;QACvF,qGAAqG;QACrG,MAAMC,UAAU,MAAMrB,QAAQe,MAAM;QACpC,MAAMO,WAAW,MAAMtB,QAAQe,MAAM;QACrC,MAAMQ,eAAe7B,iBAAiBY,KAAKT;QAC3C,IAAI2B,sBAAsB;QAC1B,IAAI,AAACH,YAAY,QAAQA,YAAYd,QAAQkB,aAAa,IAAMH,aAAa,QAAQA,aAAaC,cAAe;YAC/G,uFAAuF;YACvF,wFAAwF;YACxF,IAAIF,YAAY,QAAQA,YAAYd,QAAQkB,aAAa,EAAE;gBACzDC,QAAQC,KAAK,CAAC;gBACd,kFAAkF;gBAClFV,SAAS;gBACT,MAAMV,QAAQqB,KAAK,CAACd;gBACpBf,WAAWO;gBACX,OAAOD,mBAAmBC,KAAKC,SAASC;YAC1C;YACA,MAAMqB,cAAc3B,qBAAqBoB,qBAAAA,sBAAAA,WAAY,IAAIC;YACzD,2FAA2F;YAC3F,0EAA0E;YAC1E,IAAIhB,QAAQuB,cAAc,IAAK,MAAMvB,QAAQuB,cAAc,CAAChB,QAAQC,MAAMT,KAAKuB,cAAe;gBAC5FL,sBAAsB;YACxB,OAAO,IAAIK,YAAYE,IAAI,KAAK,KAAKF,YAAYG,GAAG,CAAC,YAAY7B,qBAAqBmB,qBAAAA,sBAAAA,WAAY,IAAIC,eAAe;gBACnH,iFAAiF;gBACjF,wFAAwF;gBACxFG,QAAQC,KAAK,CAAC;gBACd,MAAM1B,QAAQc,MAAM,YAAYQ;YAClC,OAAO;gBACL,MAAMU,UAAU7B,cAAckB,qBAAAA,sBAAAA,WAAY,IAAIC;gBAC9CG,QAAQC,KAAK,CAAC,CAAC,sBAAsB,EAAEM,QAAQ,oBAAoB,CAAC;gBACpEhB,SAAS;gBACT,MAAMV,QAAQqB,KAAK,CAACd;gBACpBf,WAAWO;gBACX,OAAOD,mBAAmBC,KAAKC,SAASC;YAC1C;QACF;QAEA,uFAAuF;QACvF,6DAA6D;QAC7D,MAAM0B,cAAc3B,QAAQ4B,kBAAkB,GAAG5B,QAAQ4B,kBAAkB,CAACrB,UAAU;QACtF,IAAIoB,gBAAgB,MAAM;YACxBR,QAAQC,KAAK,CAAC,CAAC,OAAO,EAAEO,YAAY,sBAAsB,CAAC;YAC3DjB,SAAS;YACT,MAAMV,QAAQqB,KAAK,CAACd;YACpBf,WAAWO;YACX,OAAOD,mBAAmBC,KAAKC,SAASC;QAC1C;QAEA,MAAMD,QAAQ6B,YAAY,CAACtB,QAAQC,MAAMT;QAEzC,IAAI+B,kBAA4B,EAAE;QAClC,IAAIb,uBAAuBjB,QAAQ+B,iBAAiB,EAAE;YACpDD,kBAAkB,MAAM9B,QAAQ+B,iBAAiB,CAACxB,QAAQC,MAAMT,KAAKA,IAAII,OAAO,EAAEa;QACpF;QAEA,wFAAwF;QACxF,uFAAuF;QACvF,IAAIhB,QAAQgC,qBAAqB,EAAE;gBACH;YAA9B,MAAMC,gBAAgBC,QAAQ,OAAA,MAAMzC,QAAQe,MAAM,iCAApB,kBAAA,OAA4C;YAC1E,MAAMR,QAAQgC,qBAAqB,CAACzB,QAAQC,MAAM2B,KAAKC,GAAG,CAACD,KAAKE,GAAG,CAAC,OAAO,IAAIJ,gBAAgB;QACjG;QAEA,MAAM,EAAEK,MAAM,EAAEC,QAAQ,EAAE,GAAG,MAAM5B,QAAQ6B,KAAK;QAChD,gGAAgG;QAChG,IAAI,CAACvC,iBAAiB,MAAMU,QAAQU,KAAK;QAEzC,OAAO;YAAEd;YAAQC;YAAMT;YAAKM;YAAQiC;YAAQC,UAAU;mBAAIT;mBAAoBS;aAAS;YAAE5B;QAAQ;IACnG,EAAE,OAAO8B,KAAK;QACZ,MAAM9B,QAAQU,KAAK;QACnB,IAAI,CAACX,QAAQ,MAAMV,QAAQqB,KAAK,CAACd;QACjC,MAAMkC;IACR;AACF;AAEA,iGAAiG;AACjG,mFAAmF;AACnF,OAAO,eAAeC,gBAAwB3C,GAAmB,EAAEC,OAA4B;IAC7F,MAAM,EAAEO,MAAM,EAAEC,IAAI,EAAET,KAAK4C,WAAW,EAAEtC,MAAM,EAAEiC,MAAM,EAAEC,QAAQ,EAAE,GAAG,MAAMzC,mBAAmBC,KAAKC,SAAS;IAC5G,MAAM4C,QAAQ5C,QAAQ6C,WAAW,CAACtC,QAAQC,MAAMmC;IAChD,0FAA0F;IAC1FtD,eAAemB,MAAMoC;IACrB,OAAO;QAAEA;QAAO7C,KAAK4C;QAAatC;QAAQiC;QAAQC;IAAS;AAC7D;AAEA,8FAA8F;AAC9F,+FAA+F;AAC/F,OAAO,eAAeO,gBAAwB/C,GAAmB,EAAEC,OAA4B;IAC7F,MAAM,EAAEO,MAAM,EAAEC,IAAI,EAAET,KAAK4C,WAAW,EAAEtC,MAAM,EAAEiC,MAAM,EAAEC,QAAQ,EAAE5B,OAAO,EAAE,GAAG,MAAMb,mBAAmBC,KAAKC,SAAS;IACrH,MAAM4C,QAAQ5C,QAAQ6C,WAAW,CAACtC,QAAQC,MAAMmC;IAChDtD,eAAemB,MAAMoC;IACrB,OAAO;QAAEA;QAAO7C,KAAK4C;QAAatC;QAAQiC;QAAQC;QAAU5B;IAAQ;AACtE;AAEA,OAAO,eAAeoC,SAASH,KAAY;IACzC,MAAMI,OAAO,MAAMJ,MAAMK,OAAO,CAAC;IACjC,OAAO,AAAE,CAAA,MAAMD,KAAKE,GAAG,EAAC,EAAqBC,CAAC;AAChD"}
@@ -0,0 +1,7 @@
1
+ import type { Config } from '../config/index.js';
2
+ import type { ParsePool } from '../scan/pool.js';
3
+ import type { Connection, ReconcileDialect } from './types.js';
4
+ export declare function reconcile(conn: Connection, cfg: Config, baseDir: string, dialect: ReconcileDialect, pool?: ParsePool): Promise<{
5
+ parsed: number;
6
+ warnings: string[];
7
+ }>;
@@ -0,0 +1,132 @@
1
+ import { activeFeatures } from '../features/index.js';
2
+ import { progress } from '../output/progress.js';
3
+ import { listFiles, RESERVED_COLUMNS } from '../scan/index.js';
4
+ import { reparseFiles } from '../scan/reparse.js';
5
+ import { getColumns, quoteIdent } from './shared.js';
6
+ import { withTransaction } from './transaction.js';
7
+ // One reconcile algorithm shared by every store, parameterised by a per-engine ReconcileDialect
8
+ // (types.ts). Ordering is universal, not a dialect concern: ALTER, then the frontmatter upsert,
9
+ // then reconcileContent, then preset_files, then the vanished-frontmatter delete, then feature hooks
10
+ // -- a vanished path's content delete (inside reconcileContent) must precede its frontmatter
11
+ // delete, since sqlite's delete SQL resolves the row via its frontmatter rowid.
12
+ // Feature-owned columns (`_rank`) must stay out of the upsert: a reparse would null the last
13
+ // computed value on every touch, not just the reconciles that recompute it.
14
+ const CORE_FRONTMATTER_COLUMNS = new Set([
15
+ 'path',
16
+ '_mtime',
17
+ '_ctime',
18
+ '_size',
19
+ '_parse_error'
20
+ ]);
21
+ export async function reconcile(conn, cfg, baseDir, dialect, pool) {
22
+ const files = listFiles(cfg, baseDir);
23
+ const currentSet = new Set(files.map((f)=>f.relPath));
24
+ const existingStmt = await conn.prepare('SELECT "path", "_mtime", "_size" FROM frontmatter');
25
+ const existingRows = await existingStmt.all();
26
+ const existing = new Map(existingRows.map((r)=>[
27
+ r.path,
28
+ r
29
+ ]));
30
+ const vanished = existingRows.filter((r)=>!currentSet.has(r.path)).map((r)=>r.path);
31
+ const toReparse = files.filter((f)=>{
32
+ const row = existing.get(f.relPath);
33
+ return !row || row._mtime !== f.mtimeMs || row._size !== f.size;
34
+ });
35
+ if (vanished.length === 0 && toReparse.length === 0) return {
36
+ parsed: 0,
37
+ warnings: []
38
+ };
39
+ const features = activeFeatures(cfg);
40
+ const seenColumns = await getColumns(conn);
41
+ // Bulk reparses (a sync, a cold build) are the long silences a query can hit; short
42
+ // reconciles stay silent (progress() has a threshold).
43
+ const report = progress('reparsing files', toReparse.length);
44
+ const { docs: parsedDocs, warnings, newColumns } = await reparseFiles(toReparse, features, cfg, seenColumns, report.tick, {
45
+ pool
46
+ });
47
+ report.finish();
48
+ for (const col of newColumns)seenColumns.add(col);
49
+ const allColumns = [
50
+ ...seenColumns
51
+ ];
52
+ // Fence before ALTERing: a store's own failure past this point is a raw, engine-specific
53
+ // error with no indication of the boundary or the levers -- dialect.checkColumnLimit names both.
54
+ dialect.checkColumnLimit(allColumns.length);
55
+ // Columns the frontmatter upsert actually writes: core + parsed frontmatter keys, never a
56
+ // feature-owned reserved column (see CORE_FRONTMATTER_COLUMNS above).
57
+ const writableColumns = allColumns.filter((c)=>CORE_FRONTMATTER_COLUMNS.has(c) || !RESERVED_COLUMNS.has(c));
58
+ // ON CONFLICT DO UPDATE (not OR REPLACE) keeps the row's rowid stable across reparses --
59
+ // sqlite's content rows are coupled to that rowid.
60
+ 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(', ')}`;
61
+ const added = toReparse.filter((f)=>!existing.has(f.relPath)).map((f)=>f.relPath);
62
+ const delta = {
63
+ files,
64
+ reparsed: parsedDocs.map((d)=>d.relPath),
65
+ added,
66
+ vanished
67
+ };
68
+ const addedSet = new Set(added);
69
+ const reparsedExisting = parsedDocs.map((d)=>d.relPath).filter((p)=>!addedSet.has(p));
70
+ // Paths whose content (and, per feature, other rows) need clearing: gone entirely, or about to
71
+ // be reinserted fresh. Disjoint from `added`, which has nothing to clear.
72
+ const touched = [
73
+ ...vanished,
74
+ ...reparsedExisting
75
+ ];
76
+ const txStart = Date.now();
77
+ await withTransaction(conn, async ()=>{
78
+ var _feature_remove, _feature_afterReconcile;
79
+ // Re-read inside the write transaction: newColumns came from a read taken before it opened, so a
80
+ // concurrent reconcile may have added some of them since. ALTER has no IF NOT EXISTS.
81
+ const present = await getColumns(conn);
82
+ for (const col of newColumns)if (!present.has(col)) await conn.exec(`ALTER TABLE frontmatter ADD COLUMN ${quoteIdent(col)}${dialect.columnTypeSuffix()}`);
83
+ if (parsedDocs.length > 0) {
84
+ const rows = parsedDocs.map((doc)=>writableColumns.map((col)=>{
85
+ var _doc_data_col;
86
+ if (col === 'path') return doc.relPath;
87
+ if (col === '_mtime') return doc.mtimeMs;
88
+ if (col === '_ctime') return doc.ctimeMs;
89
+ if (col === '_size') return doc.size;
90
+ // Written per parse, unlike _rank, which a feature pass owns and the upsert skips.
91
+ if (col === '_parse_error') return doc.parseError;
92
+ return (_doc_data_col = doc.data[col]) !== null && _doc_data_col !== void 0 ? _doc_data_col : null;
93
+ }));
94
+ await conn.runBatch(insertSql, rows);
95
+ }
96
+ // After the upsert, so every doc already has the frontmatter row sqlite's content rowid
97
+ // couples to. ON CONFLICT DO UPDATE preserves that rowid, so a reparse keeps its identity.
98
+ await dialect.reconcileContent(conn, touched, parsedDocs, delta, cfg);
99
+ // A preset edit forces a full rebuild, so an unchanged doc's coverage is already correct;
100
+ // new docs have nothing to clear, which keeps cold builds linear.
101
+ if (touched.length > 0) await conn.runBatch('DELETE FROM preset_files WHERE "path" = ?', touched.map((p)=>[
102
+ p
103
+ ]));
104
+ const presetRows = [];
105
+ for (const doc of parsedDocs)for (const presetName of doc.presets)presetRows.push([
106
+ doc.relPath,
107
+ presetName
108
+ ]);
109
+ if (presetRows.length > 0) await conn.runBatch('INSERT INTO preset_files ("path", preset) VALUES (?, ?)', presetRows);
110
+ // Before the feature hooks, never after: rank's afterReconcile reads frontmatter as PageRank's
111
+ // node set, so a lingering vanished row would dilute rank mass across every surviving note.
112
+ if (vanished.length > 0) await conn.runBatch('DELETE FROM frontmatter WHERE "path" = ?', vanished.map((p)=>[
113
+ p
114
+ ]));
115
+ if (touched.length > 0) for (const feature of features)await ((_feature_remove = feature.remove) === null || _feature_remove === void 0 ? void 0 : _feature_remove.call(feature, conn, touched, delta));
116
+ for (const feature of features){
117
+ var _feature_store;
118
+ const docsForFeature = parsedDocs.map((doc)=>({
119
+ path: doc.relPath,
120
+ extracted: doc.extracted[feature.name]
121
+ }));
122
+ await ((_feature_store = feature.store) === null || _feature_store === void 0 ? void 0 : _feature_store.call(feature, conn, docsForFeature, delta));
123
+ }
124
+ for (const feature of features)await ((_feature_afterReconcile = feature.afterReconcile) === null || _feature_afterReconcile === void 0 ? void 0 : _feature_afterReconcile.call(feature, conn, delta));
125
+ }, dialect.beginMode());
126
+ const durationMs = Date.now() - txStart;
127
+ if (dialect.recordDuration) await dialect.recordDuration(conn, durationMs);
128
+ return {
129
+ parsed: parsedDocs.length,
130
+ warnings
131
+ };
132
+ }
@@ -0,0 +1 @@
1
+ {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/reconcile.ts"],"sourcesContent":["import type { Config } from '../config/index.ts';\nimport { activeFeatures } from '../features/index.ts';\nimport type { ExtractedDoc, ReconcileDelta } from '../features/types.ts';\nimport { progress } from '../output/progress.ts';\nimport { listFiles, RESERVED_COLUMNS } from '../scan/index.ts';\nimport type { ParsePool } from '../scan/pool.ts';\nimport { reparseFiles } from '../scan/reparse.ts';\nimport { getColumns, quoteIdent } from './shared.ts';\nimport { withTransaction } from './transaction.ts';\nimport type { Connection, ReconcileDialect } from './types.ts';\n\n// One reconcile algorithm shared by every store, parameterised by a per-engine ReconcileDialect\n// (types.ts). Ordering is universal, not a dialect concern: ALTER, then the frontmatter upsert,\n// then reconcileContent, then preset_files, then the vanished-frontmatter delete, then feature hooks\n// -- a vanished path's content delete (inside reconcileContent) must precede its frontmatter\n// delete, since sqlite's delete SQL resolves the row via its frontmatter rowid.\n\n// Feature-owned columns (`_rank`) must stay out of the upsert: a reparse would null the last\n// computed value on every touch, not just the reconciles that recompute it.\nconst CORE_FRONTMATTER_COLUMNS = new Set(['path', '_mtime', '_ctime', '_size', '_parse_error']);\n\nexport async function reconcile(conn: Connection, cfg: Config, baseDir: string, dialect: ReconcileDialect, pool?: ParsePool): 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, { pool });\n report.finish();\n for (const col of newColumns) seenColumns.add(col);\n\n const allColumns = [...seenColumns];\n // Fence before ALTERing: a store's own failure past this point is a raw, engine-specific\n // error with no indication of the boundary or the levers -- dialect.checkColumnLimit names both.\n dialect.checkColumnLimit(allColumns.length);\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 DO UPDATE (not OR REPLACE) keeps the row's rowid stable across reparses --\n // sqlite's content rows are coupled to that rowid.\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 // Paths whose content (and, per feature, other rows) need clearing: gone entirely, or about to\n // be reinserted fresh. Disjoint from `added`, which has nothing to clear.\n const touched = [...vanished, ...reparsedExisting];\n\n const txStart = Date.now();\n await withTransaction(\n conn,\n async () => {\n // Re-read inside the write transaction: newColumns came from a read taken before it opened, so a\n // concurrent reconcile may have added some of them since. ALTER has no IF NOT EXISTS.\n const present = await getColumns(conn);\n for (const col of newColumns) if (!present.has(col)) await conn.exec(`ALTER TABLE frontmatter ADD COLUMN ${quoteIdent(col)}${dialect.columnTypeSuffix()}`);\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 // After the upsert, so every doc already has the frontmatter row sqlite's content rowid\n // couples to. ON CONFLICT DO UPDATE preserves that rowid, so a reparse keeps its identity.\n await dialect.reconcileContent(conn, touched, parsedDocs, delta, cfg);\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 if (touched.length > 0)\n await conn.runBatch(\n 'DELETE FROM preset_files WHERE \"path\" = ?',\n touched.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 // Before the feature hooks, never after: rank's afterReconcile reads frontmatter as PageRank's\n // node set, so a lingering vanished row would dilute rank mass across every surviving note.\n if (vanished.length > 0)\n await conn.runBatch(\n 'DELETE FROM frontmatter WHERE \"path\" = ?',\n vanished.map((p) => [p])\n );\n\n if (touched.length > 0) for (const feature of features) await feature.remove?.(conn, touched, 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 dialect.beginMode()\n );\n\n const durationMs = Date.now() - txStart;\n if (dialect.recordDuration) await dialect.recordDuration(conn, durationMs);\n\n return { parsed: parsedDocs.length, warnings };\n}\n"],"names":["activeFeatures","progress","listFiles","RESERVED_COLUMNS","reparseFiles","getColumns","quoteIdent","withTransaction","CORE_FRONTMATTER_COLUMNS","Set","reconcile","conn","cfg","baseDir","dialect","pool","files","currentSet","map","f","relPath","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","checkColumnLimit","writableColumns","c","insertSql","join","added","delta","reparsed","d","addedSet","reparsedExisting","p","touched","txStart","Date","now","feature","present","exec","columnTypeSuffix","rows","doc","ctimeMs","parseError","data","runBatch","reconcileContent","presetRows","presetName","presets","push","remove","docsForFeature","extracted","name","store","afterReconcile","beginMode","durationMs","recordDuration"],"mappings":"AACA,SAASA,cAAc,QAAQ,uBAAuB;AAEtD,SAASC,QAAQ,QAAQ,wBAAwB;AACjD,SAASC,SAAS,EAAEC,gBAAgB,QAAQ,mBAAmB;AAE/D,SAASC,YAAY,QAAQ,qBAAqB;AAClD,SAASC,UAAU,EAAEC,UAAU,QAAQ,cAAc;AACrD,SAASC,eAAe,QAAQ,mBAAmB;AAGnD,gGAAgG;AAChG,gGAAgG;AAChG,qGAAqG;AACrG,6FAA6F;AAC7F,gFAAgF;AAEhF,6FAA6F;AAC7F,4EAA4E;AAC5E,MAAMC,2BAA2B,IAAIC,IAAI;IAAC;IAAQ;IAAU;IAAU;IAAS;CAAe;AAE9F,OAAO,eAAeC,UAAUC,IAAgB,EAAEC,GAAW,EAAEC,OAAe,EAAEC,OAAyB,EAAEC,IAAgB;IACzH,MAAMC,QAAQd,UAAUU,KAAKC;IAC7B,MAAMI,aAAa,IAAIR,IAAIO,MAAME,GAAG,CAAC,CAACC,IAAMA,EAAEC,OAAO;IAErD,MAAMC,eAAe,MAAMV,KAAKW,OAAO,CAAC;IACxC,MAAMC,eAAgB,MAAMF,aAAaG,GAAG;IAC5C,MAAMC,WAAW,IAAIC,IAAIH,aAAaL,GAAG,CAAC,CAACS,IAAM;YAACA,EAAEC,IAAI;YAAED;SAAE;IAC5D,MAAME,WAAWN,aAAaO,MAAM,CAAC,CAACH,IAAM,CAACV,WAAWc,GAAG,CAACJ,EAAEC,IAAI,GAAGV,GAAG,CAAC,CAACS,IAAMA,EAAEC,IAAI;IAEtF,MAAMI,YAAYhB,MAAMc,MAAM,CAAC,CAACX;QAC9B,MAAMc,MAAMR,SAASS,GAAG,CAACf,EAAEC,OAAO;QAClC,OAAO,CAACa,OAAOA,IAAIE,MAAM,KAAKhB,EAAEiB,OAAO,IAAIH,IAAII,KAAK,KAAKlB,EAAEmB,IAAI;IACjE;IAEA,IAAIT,SAASU,MAAM,KAAK,KAAKP,UAAUO,MAAM,KAAK,GAAG,OAAO;QAAEC,QAAQ;QAAGC,UAAU,EAAE;IAAC;IAEtF,MAAMC,WAAW1C,eAAeY;IAChC,MAAM+B,cAAc,MAAMtC,WAAWM;IAErC,oFAAoF;IACpF,uDAAuD;IACvD,MAAMiC,SAAS3C,SAAS,mBAAmB+B,UAAUO,MAAM;IAC3D,MAAM,EAAEM,MAAMC,UAAU,EAAEL,QAAQ,EAAEM,UAAU,EAAE,GAAG,MAAM3C,aAAa4B,WAAWU,UAAU9B,KAAK+B,aAAaC,OAAOI,IAAI,EAAE;QAAEjC;IAAK;IACjI6B,OAAOK,MAAM;IACb,KAAK,MAAMC,OAAOH,WAAYJ,YAAYQ,GAAG,CAACD;IAE9C,MAAME,aAAa;WAAIT;KAAY;IACnC,yFAAyF;IACzF,iGAAiG;IACjG7B,QAAQuC,gBAAgB,CAACD,WAAWb,MAAM;IAC1C,0FAA0F;IAC1F,sEAAsE;IACtE,MAAMe,kBAAkBF,WAAWtB,MAAM,CAAC,CAACyB,IAAM/C,yBAAyBuB,GAAG,CAACwB,MAAM,CAACpD,iBAAiB4B,GAAG,CAACwB;IAC1G,yFAAyF;IACzF,mDAAmD;IACnD,MAAMC,YAAY,CAAC,yBAAyB,EAAEF,gBAAgBpC,GAAG,CAACZ,YAAYmD,IAAI,CAAC,MAAM,UAAU,EAAEH,gBAAgBpC,GAAG,CAAC,IAAM,KAAKuC,IAAI,CAAC,MAAM,oCAAoC,EAAEH,gBAClLxB,MAAM,CAAC,CAACyB,IAAMA,MAAM,QACpBrC,GAAG,CAAC,CAACqC,IAAM,GAAGjD,WAAWiD,GAAG,YAAY,EAAEjD,WAAWiD,IAAI,EACzDE,IAAI,CAAC,OAAO;IAEf,MAAMC,QAAQ1B,UAAUF,MAAM,CAAC,CAACX,IAAM,CAACM,SAASM,GAAG,CAACZ,EAAEC,OAAO,GAAGF,GAAG,CAAC,CAACC,IAAMA,EAAEC,OAAO;IACpF,MAAMuC,QAAwB;QAAE3C;QAAO4C,UAAUd,WAAW5B,GAAG,CAAC,CAAC2C,IAAMA,EAAEzC,OAAO;QAAGsC;QAAO7B;IAAS;IACnG,MAAMiC,WAAW,IAAIrD,IAAIiD;IACzB,MAAMK,mBAAmBjB,WAAW5B,GAAG,CAAC,CAAC2C,IAAMA,EAAEzC,OAAO,EAAEU,MAAM,CAAC,CAACkC,IAAM,CAACF,SAAS/B,GAAG,CAACiC;IACtF,+FAA+F;IAC/F,0EAA0E;IAC1E,MAAMC,UAAU;WAAIpC;WAAakC;KAAiB;IAElD,MAAMG,UAAUC,KAAKC,GAAG;IACxB,MAAM7D,gBACJI,MACA;YA4CgE0D,iBAKxBA;QAhDtC,iGAAiG;QACjG,sFAAsF;QACtF,MAAMC,UAAU,MAAMjE,WAAWM;QACjC,KAAK,MAAMuC,OAAOH,WAAY,IAAI,CAACuB,QAAQvC,GAAG,CAACmB,MAAM,MAAMvC,KAAK4D,IAAI,CAAC,CAAC,mCAAmC,EAAEjE,WAAW4C,OAAOpC,QAAQ0D,gBAAgB,IAAI;QAEzJ,IAAI1B,WAAWP,MAAM,GAAG,GAAG;YACzB,MAAMkC,OAAO3B,WAAW5B,GAAG,CAAC,CAACwD,MAC3BpB,gBAAgBpC,GAAG,CAAC,CAACgC;wBAOZwB;oBANP,IAAIxB,QAAQ,QAAQ,OAAOwB,IAAItD,OAAO;oBACtC,IAAI8B,QAAQ,UAAU,OAAOwB,IAAItC,OAAO;oBACxC,IAAIc,QAAQ,UAAU,OAAOwB,IAAIC,OAAO;oBACxC,IAAIzB,QAAQ,SAAS,OAAOwB,IAAIpC,IAAI;oBACpC,mFAAmF;oBACnF,IAAIY,QAAQ,gBAAgB,OAAOwB,IAAIE,UAAU;oBACjD,QAAOF,gBAAAA,IAAIG,IAAI,CAAC3B,IAAI,cAAbwB,2BAAAA,gBAAiB;gBAC1B;YAEF,MAAM/D,KAAKmE,QAAQ,CAACtB,WAAWiB;QACjC;QAEA,wFAAwF;QACxF,2FAA2F;QAC3F,MAAM3D,QAAQiE,gBAAgB,CAACpE,MAAMsD,SAASnB,YAAYa,OAAO/C;QAEjE,0FAA0F;QAC1F,kEAAkE;QAClE,IAAIqD,QAAQ1B,MAAM,GAAG,GACnB,MAAM5B,KAAKmE,QAAQ,CACjB,6CACAb,QAAQ/C,GAAG,CAAC,CAAC8C,IAAM;gBAACA;aAAE;QAE1B,MAAMgB,aAA0B,EAAE;QAClC,KAAK,MAAMN,OAAO5B,WAAY,KAAK,MAAMmC,cAAcP,IAAIQ,OAAO,CAAEF,WAAWG,IAAI,CAAC;YAACT,IAAItD,OAAO;YAAE6D;SAAW;QAC7G,IAAID,WAAWzC,MAAM,GAAG,GAAG,MAAM5B,KAAKmE,QAAQ,CAAC,2DAA2DE;QAE1G,+FAA+F;QAC/F,4FAA4F;QAC5F,IAAInD,SAASU,MAAM,GAAG,GACpB,MAAM5B,KAAKmE,QAAQ,CACjB,4CACAjD,SAASX,GAAG,CAAC,CAAC8C,IAAM;gBAACA;aAAE;QAG3B,IAAIC,QAAQ1B,MAAM,GAAG,GAAG,KAAK,MAAM8B,WAAW3B,SAAU,QAAM2B,kBAAAA,QAAQe,MAAM,cAAdf,sCAAAA,qBAAAA,SAAiB1D,MAAMsD,SAASN;QAC9F,KAAK,MAAMU,WAAW3B,SAAU;gBAExB2B;YADN,MAAMgB,iBAAiCvC,WAAW5B,GAAG,CAAC,CAACwD,MAAS,CAAA;oBAAE9C,MAAM8C,IAAItD,OAAO;oBAAEkE,WAAWZ,IAAIY,SAAS,CAACjB,QAAQkB,IAAI,CAAC;gBAAC,CAAA;YAC5H,QAAMlB,iBAAAA,QAAQmB,KAAK,cAAbnB,qCAAAA,oBAAAA,SAAgB1D,MAAM0E,gBAAgB1B;QAC9C;QACA,KAAK,MAAMU,WAAW3B,SAAU,QAAM2B,0BAAAA,QAAQoB,cAAc,cAAtBpB,8CAAAA,6BAAAA,SAAyB1D,MAAMgD;IACvE,GACA7C,QAAQ4E,SAAS;IAGnB,MAAMC,aAAaxB,KAAKC,GAAG,KAAKF;IAChC,IAAIpD,QAAQ8E,cAAc,EAAE,MAAM9E,QAAQ8E,cAAc,CAACjF,MAAMgF;IAE/D,OAAO;QAAEnD,QAAQM,WAAWP,MAAM;QAAEE;IAAS;AAC/C"}
@@ -3,3 +3,4 @@ export declare function quoteIdent(name: string): string;
3
3
  export declare function getColumns(conn: Connection): Promise<Set<string>>;
4
4
  export declare function getMeta(conn: Connection, key: string): Promise<string | null>;
5
5
  export declare function setMeta(conn: Connection, key: string, value: string | null): Promise<void>;
6
+ export declare function recordReconcileDuration(conn: Connection, ms: number): Promise<void>;
@@ -22,3 +22,12 @@ export async function setMeta(conn, key, value) {
22
22
  const stmt = await conn.prepare('INSERT INTO meta (key, value) VALUES (?, ?) ON CONFLICT(key) DO UPDATE SET value = excluded.value');
23
23
  await stmt.run(key, value);
24
24
  }
25
+ // Reconcile's own write-transaction duration, for open()'s derived busy_timeout: keep the
26
+ // observed max so a big watcher reconcile's lock hold is what the next open bounds its wait against.
27
+ export async function recordReconcileDuration(conn, ms) {
28
+ const prevRaw = await getMeta(conn, 'reconcile_max_ms');
29
+ // -1, not 0, so a genuinely 0ms first reconcile (sub-millisecond, common on a tiny tree)
30
+ // still gets recorded instead of losing to the "nothing recorded yet" default.
31
+ const prevMax = prevRaw === null ? -1 : Number(prevRaw);
32
+ if (ms > prevMax) await setMeta(conn, 'reconcile_max_ms', String(ms));
33
+ }
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/shared.ts"],"sourcesContent":["import type { Connection } from './types.ts';\n\n// SQL and meta-table primitives both stores' open()/reconcile() need. Engine-neutral: both\n// stores' Connection satisfies the same async exec/prepare/runBatch shape (types.ts).\n\nexport function quoteIdent(name: string): string {\n return `\"${name.split('\"').join('\"\"')}\"`;\n}\n\nexport async function getColumns(conn: Connection): Promise<Set<string>> {\n const stmt = await conn.prepare('PRAGMA table_info(frontmatter)');\n const rows = (await stmt.all()) as Array<{ name: string }>;\n return new Set(rows.map((r) => r.name));\n}\n\nexport async function getMeta(conn: Connection, key: string): Promise<string | null> {\n const stmt = await conn.prepare('SELECT value FROM meta WHERE key = ?');\n const row = (await stmt.get(key)) as { value: string } | undefined;\n return row ? row.value : null;\n}\n\nexport async function setMeta(conn: Connection, key: string, value: string | null): Promise<void> {\n if (value === null) {\n const stmt = await conn.prepare('DELETE FROM meta WHERE key = ?');\n await stmt.run(key);\n return;\n }\n const stmt = await conn.prepare('INSERT INTO meta (key, value) VALUES (?, ?) ON CONFLICT(key) DO UPDATE SET value = excluded.value');\n await stmt.run(key, value);\n}\n"],"names":["quoteIdent","name","split","join","getColumns","conn","stmt","prepare","rows","all","Set","map","r","getMeta","key","row","get","value","setMeta","run"],"mappings":"AAEA,2FAA2F;AAC3F,sFAAsF;AAEtF,OAAO,SAASA,WAAWC,IAAY;IACrC,OAAO,CAAC,CAAC,EAAEA,KAAKC,KAAK,CAAC,KAAKC,IAAI,CAAC,MAAM,CAAC,CAAC;AAC1C;AAEA,OAAO,eAAeC,WAAWC,IAAgB;IAC/C,MAAMC,OAAO,MAAMD,KAAKE,OAAO,CAAC;IAChC,MAAMC,OAAQ,MAAMF,KAAKG,GAAG;IAC5B,OAAO,IAAIC,IAAIF,KAAKG,GAAG,CAAC,CAACC,IAAMA,EAAEX,IAAI;AACvC;AAEA,OAAO,eAAeY,QAAQR,IAAgB,EAAES,GAAW;IACzD,MAAMR,OAAO,MAAMD,KAAKE,OAAO,CAAC;IAChC,MAAMQ,MAAO,MAAMT,KAAKU,GAAG,CAACF;IAC5B,OAAOC,MAAMA,IAAIE,KAAK,GAAG;AAC3B;AAEA,OAAO,eAAeC,QAAQb,IAAgB,EAAES,GAAW,EAAEG,KAAoB;IAC/E,IAAIA,UAAU,MAAM;QAClB,MAAMX,OAAO,MAAMD,KAAKE,OAAO,CAAC;QAChC,MAAMD,KAAKa,GAAG,CAACL;QACf;IACF;IACA,MAAMR,OAAO,MAAMD,KAAKE,OAAO,CAAC;IAChC,MAAMD,KAAKa,GAAG,CAACL,KAAKG;AACtB"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/shared.ts"],"sourcesContent":["import type { Connection } from './types.ts';\n\n// SQL and meta-table primitives both stores' open()/reconcile() need. Engine-neutral: both\n// stores' Connection satisfies the same async exec/prepare/runBatch shape (types.ts).\n\nexport function quoteIdent(name: string): string {\n return `\"${name.split('\"').join('\"\"')}\"`;\n}\n\nexport async function getColumns(conn: Connection): Promise<Set<string>> {\n const stmt = await conn.prepare('PRAGMA table_info(frontmatter)');\n const rows = (await stmt.all()) as Array<{ name: string }>;\n return new Set(rows.map((r) => r.name));\n}\n\nexport async function getMeta(conn: Connection, key: string): Promise<string | null> {\n const stmt = await conn.prepare('SELECT value FROM meta WHERE key = ?');\n const row = (await stmt.get(key)) as { value: string } | undefined;\n return row ? row.value : null;\n}\n\nexport async function setMeta(conn: Connection, key: string, value: string | null): Promise<void> {\n if (value === null) {\n const stmt = await conn.prepare('DELETE FROM meta WHERE key = ?');\n await stmt.run(key);\n return;\n }\n const stmt = await conn.prepare('INSERT INTO meta (key, value) VALUES (?, ?) ON CONFLICT(key) DO UPDATE SET value = excluded.value');\n await stmt.run(key, value);\n}\n\n// Reconcile's own write-transaction duration, for open()'s derived busy_timeout: keep the\n// observed max so a big watcher reconcile's lock hold is what the next open bounds its wait against.\nexport async function recordReconcileDuration(conn: Connection, ms: number): Promise<void> {\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 // still gets recorded instead of losing to the \"nothing recorded yet\" default.\n const prevMax = prevRaw === null ? -1 : Number(prevRaw);\n if (ms > prevMax) await setMeta(conn, 'reconcile_max_ms', String(ms));\n}\n"],"names":["quoteIdent","name","split","join","getColumns","conn","stmt","prepare","rows","all","Set","map","r","getMeta","key","row","get","value","setMeta","run","recordReconcileDuration","ms","prevRaw","prevMax","Number","String"],"mappings":"AAEA,2FAA2F;AAC3F,sFAAsF;AAEtF,OAAO,SAASA,WAAWC,IAAY;IACrC,OAAO,CAAC,CAAC,EAAEA,KAAKC,KAAK,CAAC,KAAKC,IAAI,CAAC,MAAM,CAAC,CAAC;AAC1C;AAEA,OAAO,eAAeC,WAAWC,IAAgB;IAC/C,MAAMC,OAAO,MAAMD,KAAKE,OAAO,CAAC;IAChC,MAAMC,OAAQ,MAAMF,KAAKG,GAAG;IAC5B,OAAO,IAAIC,IAAIF,KAAKG,GAAG,CAAC,CAACC,IAAMA,EAAEX,IAAI;AACvC;AAEA,OAAO,eAAeY,QAAQR,IAAgB,EAAES,GAAW;IACzD,MAAMR,OAAO,MAAMD,KAAKE,OAAO,CAAC;IAChC,MAAMQ,MAAO,MAAMT,KAAKU,GAAG,CAACF;IAC5B,OAAOC,MAAMA,IAAIE,KAAK,GAAG;AAC3B;AAEA,OAAO,eAAeC,QAAQb,IAAgB,EAAES,GAAW,EAAEG,KAAoB;IAC/E,IAAIA,UAAU,MAAM;QAClB,MAAMX,OAAO,MAAMD,KAAKE,OAAO,CAAC;QAChC,MAAMD,KAAKa,GAAG,CAACL;QACf;IACF;IACA,MAAMR,OAAO,MAAMD,KAAKE,OAAO,CAAC;IAChC,MAAMD,KAAKa,GAAG,CAACL,KAAKG;AACtB;AAEA,0FAA0F;AAC1F,qGAAqG;AACrG,OAAO,eAAeG,wBAAwBf,IAAgB,EAAEgB,EAAU;IACxE,MAAMC,UAAU,MAAMT,QAAQR,MAAM;IACpC,yFAAyF;IACzF,+EAA+E;IAC/E,MAAMkB,UAAUD,YAAY,OAAO,CAAC,IAAIE,OAAOF;IAC/C,IAAID,KAAKE,SAAS,MAAML,QAAQb,MAAM,oBAAoBoB,OAAOJ;AACnE"}
@@ -0,0 +1,3 @@
1
+ export declare function changedSignatureKeys(before: string, after: string): Set<string>;
2
+ export declare function embedIdentityAdopted(before: string, after: string): boolean;
3
+ export declare function signatureDiff(before: string, after: string): string;
@@ -0,0 +1,35 @@
1
+ // Feature-signature comparison for open()'s rebuild-vs-adopt decision (config.featureSignature's
2
+ // format: global features, embed provider, tokenize, then one segment per preset). Pure string
3
+ // comparison, no SQL, shared by every store's open dialect.
4
+ // Segment keys that moved between two feature signatures.
5
+ export function changedSignatureKeys(before, after) {
6
+ const keyOf = (part)=>part.startsWith('preset:') ? part.split(':').slice(0, 2).join(':') : part.split(':')[0];
7
+ const parse = (sig)=>new Map(sig.split('|').map((part)=>[
8
+ keyOf(part),
9
+ part
10
+ ]));
11
+ const a = parse(before);
12
+ const b = parse(after);
13
+ const changed = new Set();
14
+ for (const [key, val] of b)if (a.get(key) !== val) changed.add(key);
15
+ for (const key of a.keys())if (!b.has(key)) changed.add(key);
16
+ return changed;
17
+ }
18
+ // Whether the embed segment only gained its resolved weight identity: same provider and
19
+ // model, no identity recorded before, one now -- adopted into meta without a rebuild.
20
+ export function embedIdentityAdopted(before, after) {
21
+ const embedPart = (sig)=>sig.split('|').find((p)=>p.startsWith('embed:'));
22
+ const b = embedPart(before);
23
+ const a = embedPart(after);
24
+ if (b === undefined || a === undefined) return false;
25
+ const at = a.indexOf('@');
26
+ return b.indexOf('@') === -1 && at !== -1 && a.slice(0, at) === b;
27
+ }
28
+ // Names what moved, for the rebuild notice.
29
+ export function signatureDiff(before, after) {
30
+ const changed = changedSignatureKeys(before, after);
31
+ const label = (key)=>key === 'embed' ? 'embed settings' : key === 'tokenize' ? 'content tokenizer' : key.startsWith('preset:') ? `preset "${key.slice(7)}"` : 'features';
32
+ return changed.size === 0 ? 'features' : [
33
+ ...changed
34
+ ].map(label).join(', ');
35
+ }
@@ -0,0 +1 @@
1
+ {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/signature.ts"],"sourcesContent":["// Feature-signature comparison for open()'s rebuild-vs-adopt decision (config.featureSignature's\n// format: global features, embed provider, tokenize, then one segment per preset). Pure string\n// comparison, no SQL, shared by every store's open dialect.\n\n// Segment keys that moved between two feature signatures.\nexport function changedSignatureKeys(before: string, after: string): Set<string> {\n const keyOf = (part: string) => (part.startsWith('preset:') ? part.split(':').slice(0, 2).join(':') : part.split(':')[0]);\n const parse = (sig: string) => new Map(sig.split('|').map((part) => [keyOf(part), part]));\n const a = parse(before);\n const b = parse(after);\n const changed = new Set<string>();\n for (const [key, val] of b) if (a.get(key) !== val) changed.add(key);\n for (const key of a.keys()) if (!b.has(key)) changed.add(key);\n return changed;\n}\n\n// Whether the embed segment only gained its resolved weight identity: same provider and\n// model, no identity recorded before, one now -- adopted into meta without a rebuild.\nexport function embedIdentityAdopted(before: string, after: string): boolean {\n const embedPart = (sig: string) => sig.split('|').find((p) => p.startsWith('embed:'));\n const b = embedPart(before);\n const a = embedPart(after);\n if (b === undefined || a === undefined) return false;\n const at = a.indexOf('@');\n return b.indexOf('@') === -1 && at !== -1 && a.slice(0, at) === b;\n}\n\n// Names what moved, for the rebuild notice.\nexport function signatureDiff(before: string, after: string): string {\n const changed = changedSignatureKeys(before, after);\n const label = (key: string) => (key === 'embed' ? 'embed settings' : key === 'tokenize' ? 'content tokenizer' : key.startsWith('preset:') ? `preset \"${key.slice(7)}\"` : 'features');\n return changed.size === 0 ? 'features' : [...changed].map(label).join(', ');\n}\n"],"names":["changedSignatureKeys","before","after","keyOf","part","startsWith","split","slice","join","parse","sig","Map","map","a","b","changed","Set","key","val","get","add","keys","has","embedIdentityAdopted","embedPart","find","p","undefined","at","indexOf","signatureDiff","label","size"],"mappings":"AAAA,iGAAiG;AACjG,+FAA+F;AAC/F,4DAA4D;AAE5D,0DAA0D;AAC1D,OAAO,SAASA,qBAAqBC,MAAc,EAAEC,KAAa;IAChE,MAAMC,QAAQ,CAACC,OAAkBA,KAAKC,UAAU,CAAC,aAAaD,KAAKE,KAAK,CAAC,KAAKC,KAAK,CAAC,GAAG,GAAGC,IAAI,CAAC,OAAOJ,KAAKE,KAAK,CAAC,IAAI,CAAC,EAAE;IACxH,MAAMG,QAAQ,CAACC,MAAgB,IAAIC,IAAID,IAAIJ,KAAK,CAAC,KAAKM,GAAG,CAAC,CAACR,OAAS;gBAACD,MAAMC;gBAAOA;aAAK;IACvF,MAAMS,IAAIJ,MAAMR;IAChB,MAAMa,IAAIL,MAAMP;IAChB,MAAMa,UAAU,IAAIC;IACpB,KAAK,MAAM,CAACC,KAAKC,IAAI,IAAIJ,EAAG,IAAID,EAAEM,GAAG,CAACF,SAASC,KAAKH,QAAQK,GAAG,CAACH;IAChE,KAAK,MAAMA,OAAOJ,EAAEQ,IAAI,GAAI,IAAI,CAACP,EAAEQ,GAAG,CAACL,MAAMF,QAAQK,GAAG,CAACH;IACzD,OAAOF;AACT;AAEA,wFAAwF;AACxF,sFAAsF;AACtF,OAAO,SAASQ,qBAAqBtB,MAAc,EAAEC,KAAa;IAChE,MAAMsB,YAAY,CAACd,MAAgBA,IAAIJ,KAAK,CAAC,KAAKmB,IAAI,CAAC,CAACC,IAAMA,EAAErB,UAAU,CAAC;IAC3E,MAAMS,IAAIU,UAAUvB;IACpB,MAAMY,IAAIW,UAAUtB;IACpB,IAAIY,MAAMa,aAAad,MAAMc,WAAW,OAAO;IAC/C,MAAMC,KAAKf,EAAEgB,OAAO,CAAC;IACrB,OAAOf,EAAEe,OAAO,CAAC,SAAS,CAAC,KAAKD,OAAO,CAAC,KAAKf,EAAEN,KAAK,CAAC,GAAGqB,QAAQd;AAClE;AAEA,4CAA4C;AAC5C,OAAO,SAASgB,cAAc7B,MAAc,EAAEC,KAAa;IACzD,MAAMa,UAAUf,qBAAqBC,QAAQC;IAC7C,MAAM6B,QAAQ,CAACd,MAAiBA,QAAQ,UAAU,mBAAmBA,QAAQ,aAAa,sBAAsBA,IAAIZ,UAAU,CAAC,aAAa,CAAC,QAAQ,EAAEY,IAAIV,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG;IACzK,OAAOQ,QAAQiB,IAAI,KAAK,IAAI,aAAa;WAAIjB;KAAQ,CAACH,GAAG,CAACmB,OAAOvB,IAAI,CAAC;AACxE"}
@@ -1,13 +1,13 @@
1
+ import { DatabaseSync } from 'node:sqlite';
1
2
  import type { ResolvedConfig } from '../../config/index.js';
2
- import type { Store } from '../types.js';
3
+ import type { OpenResult } from '../open.js';
4
+ import type { OpenDialect } from '../types.js';
3
5
  export declare const DB_FILENAME = "cache.db";
4
- export declare const SCHEMA_VERSION = "18";
5
- export interface OpenResult {
6
- store: Store;
7
- cfg: ResolvedConfig;
8
- dbPath: string;
9
- parsed: number;
10
- warnings: string[];
6
+ export declare const SCHEMA_VERSION = "19";
7
+ export type { OpenResult };
8
+ interface SqliteHandle {
9
+ db: DatabaseSync;
10
+ tokenize: string;
11
11
  }
12
+ export declare const sqliteOpenDialect: OpenDialect<SqliteHandle>;
12
13
  export declare function openSqlite(cfg: ResolvedConfig): Promise<OpenResult>;
13
- export declare function docCount(store: Store): Promise<number>;