sensemaking 0.9.4 → 0.10.0
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.
- package/README.md +50 -100
- package/dist/cjs/cli/check.js +7 -13
- package/dist/cjs/cli/check.js.map +1 -1
- package/dist/cjs/cli/download.d.cts +3 -0
- package/dist/cjs/cli/download.d.ts +3 -0
- package/dist/cjs/cli/download.js +214 -0
- package/dist/cjs/cli/download.js.map +1 -0
- package/dist/cjs/cli/index.d.cts +6 -3
- package/dist/cjs/cli/index.d.ts +6 -3
- package/dist/cjs/cli/index.js +23 -5
- package/dist/cjs/cli/index.js.map +1 -1
- package/dist/cjs/cli/init.js +4 -1
- package/dist/cjs/cli/init.js.map +1 -1
- package/dist/cjs/cli/map.js +2 -2
- package/dist/cjs/cli/map.js.map +1 -1
- package/dist/cjs/cli/named.js +10 -18
- package/dist/cjs/cli/named.js.map +1 -1
- package/dist/cjs/cli/path.d.cts +3 -0
- package/dist/cjs/cli/path.d.ts +3 -0
- package/dist/cjs/cli/path.js +161 -0
- package/dist/cjs/cli/path.js.map +1 -0
- package/dist/cjs/cli/peek.js +3 -2
- package/dist/cjs/cli/peek.js.map +1 -1
- package/dist/cjs/cli/related.d.cts +3 -0
- package/dist/cjs/cli/related.d.ts +3 -0
- package/dist/cjs/cli/related.js +279 -0
- package/dist/cjs/cli/related.js.map +1 -0
- package/dist/cjs/cli/search.js +3 -7
- package/dist/cjs/cli/search.js.map +1 -1
- package/dist/cjs/cli/shared.d.cts +3 -1
- package/dist/cjs/cli/shared.d.ts +3 -1
- package/dist/cjs/cli/shared.js +28 -5
- package/dist/cjs/cli/shared.js.map +1 -1
- package/dist/cjs/cli/sql.d.cts +3 -0
- package/dist/cjs/cli/sql.d.ts +3 -0
- package/dist/cjs/cli/{query.js → sql.js} +4 -4
- package/dist/cjs/cli/sql.js.map +1 -0
- package/dist/cjs/cli/status.js +9 -2
- package/dist/cjs/cli/status.js.map +1 -1
- package/dist/cjs/cli.js +14 -15
- package/dist/cjs/cli.js.map +1 -1
- package/dist/cjs/commands.d.cts +12 -4
- package/dist/cjs/commands.d.ts +12 -4
- package/dist/cjs/commands.js +138 -58
- package/dist/cjs/commands.js.map +1 -1
- package/dist/cjs/config.d.cts +6 -4
- package/dist/cjs/config.d.ts +6 -4
- package/dist/cjs/config.js +137 -58
- package/dist/cjs/config.js.map +1 -1
- package/dist/cjs/db.js +14 -34
- package/dist/cjs/db.js.map +1 -1
- package/dist/cjs/errors.d.cts +1 -1
- package/dist/cjs/errors.d.ts +1 -1
- package/dist/cjs/errors.js.map +1 -1
- package/dist/cjs/features/embed.d.cts +14 -0
- package/dist/cjs/features/embed.d.ts +14 -0
- package/dist/cjs/features/embed.js +248 -54
- package/dist/cjs/features/embed.js.map +1 -1
- package/dist/cjs/features/links.js +7 -16
- package/dist/cjs/features/links.js.map +1 -1
- package/dist/cjs/features/rank.js +2 -6
- package/dist/cjs/features/rank.js.map +1 -1
- package/dist/cjs/index.d.cts +1 -1
- package/dist/cjs/index.d.ts +1 -1
- package/dist/cjs/index.js.map +1 -1
- package/dist/cjs/output.js +11 -14
- package/dist/cjs/output.js.map +1 -1
- package/dist/cjs/progress.js +2 -4
- package/dist/cjs/progress.js.map +1 -1
- package/dist/cjs/scan.d.cts +1 -0
- package/dist/cjs/scan.d.ts +1 -0
- package/dist/cjs/scan.js +67 -32
- package/dist/cjs/scan.js.map +1 -1
- package/dist/cjs/search-error.js +4 -6
- package/dist/cjs/search-error.js.map +1 -1
- package/dist/cjs/traverse.d.cts +8 -0
- package/dist/cjs/traverse.d.ts +8 -0
- package/dist/cjs/traverse.js +97 -0
- package/dist/cjs/traverse.js.map +1 -0
- package/dist/esm/cli/check.js +7 -13
- package/dist/esm/cli/check.js.map +1 -1
- package/dist/esm/cli/download.d.ts +3 -0
- package/dist/esm/cli/download.js +27 -0
- package/dist/esm/cli/download.js.map +1 -0
- package/dist/esm/cli/index.d.ts +6 -3
- package/dist/esm/cli/index.js +13 -7
- package/dist/esm/cli/index.js.map +1 -1
- package/dist/esm/cli/init.js +4 -1
- package/dist/esm/cli/init.js.map +1 -1
- package/dist/esm/cli/map.js +3 -2
- package/dist/esm/cli/map.js.map +1 -1
- package/dist/esm/cli/named.js +10 -16
- package/dist/esm/cli/named.js.map +1 -1
- package/dist/esm/cli/path.d.ts +3 -0
- package/dist/esm/cli/path.js +51 -0
- package/dist/esm/cli/path.js.map +1 -0
- package/dist/esm/cli/peek.js +4 -2
- package/dist/esm/cli/peek.js.map +1 -1
- package/dist/esm/cli/related.d.ts +3 -0
- package/dist/esm/cli/related.js +26 -0
- package/dist/esm/cli/related.js.map +1 -0
- package/dist/esm/cli/search.js +2 -5
- package/dist/esm/cli/search.js.map +1 -1
- package/dist/esm/cli/shared.d.ts +3 -1
- package/dist/esm/cli/shared.js +28 -5
- package/dist/esm/cli/shared.js.map +1 -1
- package/dist/esm/cli/sql.d.ts +3 -0
- package/dist/esm/cli/{query.js → sql.js} +4 -4
- package/dist/esm/cli/sql.js.map +1 -0
- package/dist/esm/cli/status.js +8 -1
- package/dist/esm/cli/status.js.map +1 -1
- package/dist/esm/cli.js +13 -10
- package/dist/esm/cli.js.map +1 -1
- package/dist/esm/commands.d.ts +12 -4
- package/dist/esm/commands.js +111 -68
- package/dist/esm/commands.js.map +1 -1
- package/dist/esm/config.d.ts +6 -4
- package/dist/esm/config.js +126 -75
- package/dist/esm/config.js.map +1 -1
- package/dist/esm/db.js +16 -38
- package/dist/esm/db.js.map +1 -1
- package/dist/esm/errors.d.ts +1 -1
- package/dist/esm/errors.js.map +1 -1
- package/dist/esm/features/embed.d.ts +14 -0
- package/dist/esm/features/embed.js +110 -35
- package/dist/esm/features/embed.js.map +1 -1
- package/dist/esm/features/links.js +8 -17
- package/dist/esm/features/links.js.map +1 -1
- package/dist/esm/features/rank.js +2 -6
- package/dist/esm/features/rank.js.map +1 -1
- package/dist/esm/features/types.js.map +1 -1
- package/dist/esm/index.d.ts +1 -1
- package/dist/esm/index.js.map +1 -1
- package/dist/esm/output.js +13 -16
- package/dist/esm/output.js.map +1 -1
- package/dist/esm/progress.js +2 -4
- package/dist/esm/progress.js.map +1 -1
- package/dist/esm/scan.d.ts +1 -0
- package/dist/esm/scan.js +41 -34
- package/dist/esm/scan.js.map +1 -1
- package/dist/esm/search-error.js +4 -6
- package/dist/esm/search-error.js.map +1 -1
- package/dist/esm/traverse.d.ts +8 -0
- package/dist/esm/traverse.js +71 -0
- package/dist/esm/traverse.js.map +1 -0
- package/package.json +18 -5
- package/schema.json +12 -16
- package/skills/sense/EXAMPLES.md +13 -22
- package/skills/sense/SKILL.md +65 -140
- package/skills/sense-setup/EXAMPLES.md +30 -46
- package/skills/sense-setup/SKILL.md +22 -83
- package/dist/cjs/cli/query.d.cts +0 -3
- package/dist/cjs/cli/query.d.ts +0 -3
- package/dist/cjs/cli/query.js.map +0 -1
- package/dist/esm/cli/query.d.ts +0 -3
- package/dist/esm/cli/query.js.map +0 -1
package/dist/esm/db.js
CHANGED
|
@@ -1,8 +1,5 @@
|
|
|
1
|
-
// The
|
|
2
|
-
//
|
|
3
|
-
// StatementSync.columns() both landed in 22.16. 22.15 fails with "no such module: fts5".
|
|
4
|
-
// Nothing outside this file, features/, and commands.ts needs anything past Node 12, so
|
|
5
|
-
// raise the floor only for a sqlite capability, and lower it for nothing.
|
|
1
|
+
// The Node floor (>=22.20) is explained here and nowhere else: 22.20 is the first release with
|
|
2
|
+
// both FTS5 and row-returning INSERT ... RETURNING. Raise it only for a load-bearing capability.
|
|
6
3
|
import { mkdirSync, rmSync } from 'node:fs';
|
|
7
4
|
import { join } from 'node:path';
|
|
8
5
|
import { DatabaseSync } from 'node:sqlite';
|
|
@@ -11,11 +8,8 @@ import { SenseError } from './errors.js';
|
|
|
11
8
|
import { activeFeatures } from './features/index.js';
|
|
12
9
|
import { progress } from './progress.js';
|
|
13
10
|
import { listFiles, parseFile, RESERVED_COLUMNS } from './scan.js';
|
|
14
|
-
//
|
|
15
|
-
//
|
|
16
|
-
// rank's `_rank`) is feature-owned -- scan.ts already refuses to let frontmatter set it,
|
|
17
|
-
// so it must never appear in the upsert below, or a reparse would blow its last computed
|
|
18
|
-
// value away with NULL on every touch, not just the reconciles that recompute it.
|
|
11
|
+
// Feature-owned columns (`_rank`) must stay out of the upsert: a reparse would null the last
|
|
12
|
+
// computed value on every touch, not just the reconciles that recompute it.
|
|
19
13
|
const CORE_FRONTMATTER_COLUMNS = new Set([
|
|
20
14
|
'path',
|
|
21
15
|
'_mtime',
|
|
@@ -24,10 +18,8 @@ const CORE_FRONTMATTER_COLUMNS = new Set([
|
|
|
24
18
|
// Rows -> SQLite: core schema, reconcile loop, has(). Parsing lives in scan.ts;
|
|
25
19
|
// everything beyond frontmatter + content lives in src/features/.
|
|
26
20
|
export const DB_FILENAME = 'cache.db';
|
|
27
|
-
// Cache shape version, independent of the config's own `version`.
|
|
28
|
-
//
|
|
29
|
-
// preset_files(preset, path) table tracks per-preset coverage for status/map (was: 6,
|
|
30
|
-
// frontmatter gains the `layer` column).
|
|
21
|
+
// Cache shape version, independent of the config's own `version`. Bumping it rebuilds
|
|
22
|
+
// existing trees on first query.
|
|
31
23
|
export const SCHEMA_VERSION = '8';
|
|
32
24
|
// SQLite's compile-time SQLITE_MAX_COLUMN, default 2000 (https://www.sqlite.org/limits.html).
|
|
33
25
|
const MAX_FRONTMATTER_COLUMNS = 2000;
|
|
@@ -65,12 +57,8 @@ function getColumns(db) {
|
|
|
65
57
|
function ensureSchema(db, cfg) {
|
|
66
58
|
db.exec(`CREATE TABLE IF NOT EXISTS frontmatter ("path" TEXT PRIMARY KEY, "_mtime" REAL, "_size" INTEGER)`);
|
|
67
59
|
db.exec(`CREATE VIRTUAL TABLE IF NOT EXISTS content USING fts5(title, summary, text, path UNINDEXED, tokenize = 'porter unicode61')`);
|
|
68
|
-
// Coverage, not ownership: a path can appear under several presets.
|
|
69
|
-
//
|
|
70
|
-
// counts per preset without recomputing globs at read time.
|
|
71
|
-
// path leads the PK so the per-doc delete in reconcile is an index hit -- keyed the other
|
|
72
|
-
// way it scans the whole table per doc, which made cold builds quadratic (measured 3x cost
|
|
73
|
-
// per note-count doubling at 13k/26k). Coverage-by-preset reads get their own index.
|
|
60
|
+
// Coverage, not ownership: a path can appear under several presets. path leads the PK so the
|
|
61
|
+
// per-doc delete is an index hit -- keyed the other way, cold builds went quadratic.
|
|
74
62
|
db.exec(`CREATE TABLE IF NOT EXISTS preset_files ("path" TEXT, preset TEXT, PRIMARY KEY ("path", preset))`);
|
|
75
63
|
db.exec('CREATE INDEX IF NOT EXISTS preset_files_preset ON preset_files(preset)');
|
|
76
64
|
for (const feature of activeFeatures(cfg))feature.schema(db);
|
|
@@ -120,7 +108,7 @@ export function reconcile(db, cfg, baseDir) {
|
|
|
120
108
|
let parsedCount = 0;
|
|
121
109
|
for (const file of toReparse){
|
|
122
110
|
// A doc only gets extract/store from features that apply to it (currently: embed, via
|
|
123
|
-
// FileStat.embed -- true iff
|
|
111
|
+
// FileStat.embed -- true iff the config names an embedding model).
|
|
124
112
|
const fileFeatures = features.filter((feature)=>!feature.enabledForFile || feature.enabledForFile(cfg, file));
|
|
125
113
|
const { doc, warnings: fileWarnings } = parseFile(file, fileFeatures);
|
|
126
114
|
report.tick(++parsedCount);
|
|
@@ -192,20 +180,15 @@ export function reconcile(db, cfg, baseDir) {
|
|
|
192
180
|
});
|
|
193
181
|
// Frontmatter upsert first: content's rowid lookup below depends on this row existing.
|
|
194
182
|
insert.run(...values);
|
|
195
|
-
//
|
|
196
|
-
// cold build (empty table, nothing to delete) is wasted work, and doing it
|
|
197
|
-
// unconditionally previously made the crawl quadratic when it scanned by column --
|
|
198
|
-
// measured 4x time per note-count doubling at 13k/26k notes.
|
|
183
|
+
// Only for docs that have rows: an unconditional delete made cold crawls quadratic.
|
|
199
184
|
if (existing.has(doc.relPath)) {
|
|
200
185
|
var _feature_remove1;
|
|
201
186
|
delBody.run(doc.relPath);
|
|
202
187
|
for (const feature of features)(_feature_remove1 = feature.remove) === null || _feature_remove1 === void 0 ? void 0 : _feature_remove1.call(feature, db, doc.relPath, delta);
|
|
203
188
|
}
|
|
204
189
|
insertBody.run(doc.relPath, doc.search.title, doc.search.summary, doc.search.text, doc.relPath);
|
|
205
|
-
//
|
|
206
|
-
//
|
|
207
|
-
// (see open()), so an unchanged doc's coverage is already correct on disk. New docs
|
|
208
|
-
// have no rows to clear -- skipping the delete keeps cold builds linear.
|
|
190
|
+
// A preset edit forces a full rebuild, so an unchanged doc's coverage is already
|
|
191
|
+
// correct; new docs have nothing to clear, which keeps cold builds linear.
|
|
209
192
|
if (existing.has(doc.relPath)) delPresetFiles.run(doc.relPath);
|
|
210
193
|
for (const presetName of doc.presets)insertPresetFile.run(doc.relPath, presetName);
|
|
211
194
|
for (const feature of features)(_feature_store = feature.store) === null || _feature_store === void 0 ? void 0 : _feature_store.call(feature, db, doc.relPath, doc.extracted[feature.name], delta);
|
|
@@ -258,10 +241,8 @@ export function open(cfg) {
|
|
|
258
241
|
const dbPath = join(stateDir, DB_FILENAME);
|
|
259
242
|
const db = new DatabaseSync(dbPath);
|
|
260
243
|
db.exec('PRAGMA journal_mode = WAL');
|
|
261
|
-
// Covers a concurrent watcher's bulk reconcile
|
|
262
|
-
//
|
|
263
|
-
// racing the watcher got SQLITE_BUSY. 30s bounds the wait at ~3x the largest measured
|
|
264
|
-
// reconcile; a query that outwaits it still fails loudly.
|
|
244
|
+
// Covers a concurrent watcher's bulk reconcile (~5s for 500 files at 26k notes). A query
|
|
245
|
+
// that outwaits it still fails loudly.
|
|
265
246
|
db.exec('PRAGMA busy_timeout = 30000');
|
|
266
247
|
registerFunctions(db);
|
|
267
248
|
db.exec('CREATE TABLE IF NOT EXISTS meta (key TEXT PRIMARY KEY, value TEXT)');
|
|
@@ -287,11 +268,8 @@ export function open(cfg) {
|
|
|
287
268
|
return open(cfg);
|
|
288
269
|
}
|
|
289
270
|
ensureSchema(db, cfg);
|
|
290
|
-
//
|
|
291
|
-
//
|
|
292
|
-
// one pathological build can't pin every later open to an unbounded wait. Installed
|
|
293
|
-
// before reconcile() below -- this open's own reconcile is exactly the operation that
|
|
294
|
-
// races a concurrent watcher's transaction and needs the derived wait.
|
|
271
|
+
// 3x the largest reconcile this cache has recorded, floored at 30s and capped at 10min.
|
|
272
|
+
// Installed before reconcile() -- that call is the one that races a watcher's transaction.
|
|
295
273
|
const recordedMaxMs = Number((_getMeta = getMeta(db, 'reconcile_max_ms')) !== null && _getMeta !== void 0 ? _getMeta : '0');
|
|
296
274
|
db.exec(`PRAGMA busy_timeout = ${Math.min(Math.max(30000, 3 * recordedMaxMs), 600000)}`);
|
|
297
275
|
const { parsed, warnings } = reconcile(db, cfg, cfg.baseDir);
|
package/dist/esm/db.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/db.ts"],"sourcesContent":["// The package's Node floor (>=22.16) is set here and nowhere else: node:sqlite arrived in\n// 22.5, but FTS5 -- which search.ts's whole lexical half is built on -- and\n// StatementSync.columns() both landed in 22.16. 22.15 fails with \"no such module: fts5\".\n// Nothing outside this file, features/, and commands.ts needs anything past Node 12, so\n// raise the floor only for a sqlite capability, and lower it for nothing.\nimport { mkdirSync, rmSync } from 'node:fs';\nimport { join } from 'node:path';\nimport { DatabaseSync } from 'node:sqlite';\nimport type { Config, ResolvedConfig } from './config.ts';\nimport { featureSignature, STATE_DIR } from './config.ts';\nimport { SenseError } from './errors.ts';\nimport { activeFeatures } from './features/index.ts';\nimport type { ReconcileDelta } from './features/types.ts';\nimport { progress } from './progress.ts';\nimport type { ParsedDoc } from './scan.ts';\nimport { listFiles, parseFile, RESERVED_COLUMNS } from './scan.ts';\n\n// path/_mtime/_size are core: every reparse legitimately rewrites them. Every other\n// RESERVED_COLUMNS entry that shows up as a real frontmatter column (currently only\n// rank's `_rank`) is feature-owned -- scan.ts already refuses to let frontmatter set it,\n// so it must never appear in the upsert below, or a reparse would blow its last computed\n// value away with NULL on every touch, not just the reconciles that recompute it.\nconst CORE_FRONTMATTER_COLUMNS = new Set(['path', '_mtime', '_size']);\n\n// Rows -> SQLite: core schema, reconcile loop, has(). Parsing lives in scan.ts;\n// everything beyond frontmatter + content lives in src/features/.\n\nexport const DB_FILENAME = 'cache.db';\n// Cache shape version, independent of the config's own `version`.\n// 7: presets replace layers -- frontmatter drops the `layer` column, a new\n// preset_files(preset, path) table tracks per-preset coverage for status/map (was: 6,\n// frontmatter gains the `layer` column).\nexport const SCHEMA_VERSION = '8';\n\n// SQLite's compile-time SQLITE_MAX_COLUMN, default 2000 (https://www.sqlite.org/limits.html).\nconst MAX_FRONTMATTER_COLUMNS = 2000;\n\nexport interface OpenResult {\n db: DatabaseSync;\n cfg: ResolvedConfig;\n dbPath: string;\n parsed: number;\n warnings: string[];\n}\n\nfunction quoteIdent(name: string): string {\n return `\"${name.split('\"').join('\"\"')}\"`;\n}\n\n// has(field, value): JSON-array field -> membership, string field -> substring, NULL -> false.\nfunction registerFunctions(db: DatabaseSync): void {\n db.function('has', { deterministic: true, varargs: false }, (field: unknown, value: unknown): number => {\n if (field === null || field === undefined) return 0;\n\n const needle = String(value);\n\n if (typeof field === 'string') {\n if (field.startsWith('[')) {\n try {\n const parsed = JSON.parse(field);\n if (Array.isArray(parsed)) {\n return parsed.some((item) => String(item) === needle) ? 1 : 0;\n }\n } catch {}\n }\n return field.includes(needle) ? 1 : 0;\n }\n\n return String(field).includes(needle) ? 1 : 0;\n });\n}\n\nfunction getColumns(db: DatabaseSync): Set<string> {\n const rows = db.prepare('PRAGMA table_info(frontmatter)').all() as Array<{ name: string }>;\n return new Set(rows.map((r) => r.name));\n}\n\n// Content is a separate table (not a column on frontmatter) so `SELECT * FROM frontmatter`\n// can't dump file text into context. Features add their own tables after the core ones.\nfunction ensureSchema(db: DatabaseSync, cfg: Config): void {\n db.exec(`CREATE TABLE IF NOT EXISTS frontmatter (\"path\" TEXT PRIMARY KEY, \"_mtime\" REAL, \"_size\" INTEGER)`);\n db.exec(`CREATE VIRTUAL TABLE IF NOT EXISTS content USING fts5(title, summary, text, path UNINDEXED, tokenize = 'porter unicode61')`);\n // Coverage, not ownership: a path can appear under several presets. Rebuilt per-file\n // alongside frontmatter/content at reconcile so status/map can report matched/embedded\n // counts per preset without recomputing globs at read time.\n // path leads the PK so the per-doc delete in reconcile is an index hit -- keyed the other\n // way it scans the whole table per doc, which made cold builds quadratic (measured 3x cost\n // per note-count doubling at 13k/26k). Coverage-by-preset reads get their own index.\n db.exec(`CREATE TABLE IF NOT EXISTS preset_files (\"path\" TEXT, preset TEXT, PRIMARY KEY (\"path\", preset))`);\n db.exec('CREATE INDEX IF NOT EXISTS preset_files_preset ON preset_files(preset)');\n for (const feature of activeFeatures(cfg)) feature.schema(db);\n if (getMeta(db, 'schema_version') === null) setMeta(db, 'schema_version', SCHEMA_VERSION);\n if (getMeta(db, 'features') === null) setMeta(db, 'features', featureSignature(cfg));\n}\n\nexport function getMeta(db: DatabaseSync, key: string): string | null {\n const row = db.prepare('SELECT value FROM meta WHERE key = ?').get(key) as { value: string } | undefined;\n return row ? row.value : null;\n}\n\nexport function setMeta(db: DatabaseSync, key: string, value: string | null): void {\n if (value === null) {\n db.prepare('DELETE FROM meta WHERE key = ?').run(key);\n return;\n }\n db.prepare('INSERT INTO meta (key, value) VALUES (?, ?) ON CONFLICT(key) DO UPDATE SET value = excluded.value').run(key, value);\n}\n\nexport function docCount(db: DatabaseSync): number {\n const row = db.prepare('SELECT COUNT(*) AS n FROM frontmatter').get() as { n: number };\n return row.n;\n}\n\nexport function reconcile(db: DatabaseSync, cfg: Config, baseDir: string): { parsed: number; warnings: string[] } {\n const files = listFiles(cfg, baseDir);\n const currentSet = new Set(files.map((f) => f.relPath));\n\n const existingRows = db.prepare(`SELECT \"path\", \"_mtime\", \"_size\" FROM frontmatter`).all() as Array<{\n path: string;\n _mtime: number;\n _size: number;\n }>;\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 = getColumns(db);\n const newColumns: string[] = [];\n const parsedDocs: ParsedDoc[] = [];\n const warnings: string[] = [];\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 let parsedCount = 0;\n for (const file of toReparse) {\n // A doc only gets extract/store from features that apply to it (currently: embed, via\n // FileStat.embed -- true iff a covering preset has semantic on).\n const fileFeatures = features.filter((feature) => !feature.enabledForFile || feature.enabledForFile(cfg, file));\n const { doc, warnings: fileWarnings } = parseFile(file, fileFeatures);\n report.tick(++parsedCount);\n warnings.push(...fileWarnings);\n for (const key of Object.keys(doc.data)) {\n if (!seenColumns.has(key)) {\n seenColumns.add(key);\n newColumns.push(key);\n }\n }\n parsedDocs.push(doc);\n }\n report.finish();\n\n const allColumns = [...seenColumns];\n // Fence before ALTERing: SQLite's own failure past this point is a raw\n // \"too many columns on sqlite_altertab_frontmatter\" with no indication of the boundary or the levers.\n if (allColumns.length > MAX_FRONTMATTER_COLUMNS) {\n throw new SenseError(\n 'COLUMN_LIMIT',\n `frontmatter would need ${allColumns.length} columns, crossing SQLite's compile-time SQLITE_MAX_COLUMN limit (default ${MAX_FRONTMATTER_COLUMNS}; see https://www.sqlite.org/limits.html). Narrow the presets' include globs so fewer/other files are indexed, or fix whatever is generating unbounded frontmatter keys.`\n );\n }\n // Columns the frontmatter upsert actually writes: core + parsed frontmatter keys, never a\n // feature-owned reserved column (see CORE_FRONTMATTER_COLUMNS above).\n const writableColumns = allColumns.filter((c) => CORE_FRONTMATTER_COLUMNS.has(c) || !RESERVED_COLUMNS.has(c));\n // ON CONFLICT UPDATE (not OR REPLACE) keeps the row's rowid stable across reparses --\n // content rows are coupled to that rowid below.\n const insertSql = `INSERT INTO frontmatter (${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\n const txStart = Date.now();\n db.exec('BEGIN');\n try {\n for (const col of newColumns) db.exec(`ALTER TABLE frontmatter ADD COLUMN ${quoteIdent(col)}`);\n // FTS5 has no upsert, so delete-before-insert into `content`; coupled to the\n // frontmatter rowid (indexed via its PRIMARY KEY) instead of the UNINDEXED `path`\n // column, which a per-row DELETE would otherwise scan the whole table to find.\n const delBody = db.prepare(`DELETE FROM content WHERE rowid = (SELECT rowid FROM frontmatter WHERE \"path\" = ?)`);\n const delPresetFiles = db.prepare(`DELETE FROM preset_files WHERE \"path\" = ?`);\n const insertPresetFile = db.prepare(`INSERT INTO preset_files (\"path\", preset) VALUES (?, ?)`);\n if (vanished.length > 0) {\n const del = db.prepare(`DELETE FROM frontmatter WHERE \"path\" = ?`);\n for (const path of vanished) {\n // content delete must run first: it looks up the frontmatter rowid by path, which\n // the frontmatter delete below would otherwise have already removed.\n delBody.run(path);\n del.run(path);\n delPresetFiles.run(path);\n for (const feature of features) feature.remove?.(db, path, delta);\n }\n }\n if (parsedDocs.length > 0) {\n const insert = db.prepare(insertSql);\n const insertBody = db.prepare(`INSERT INTO content (rowid, title, summary, text, \"path\") VALUES ((SELECT rowid FROM frontmatter WHERE \"path\" = ?), ?, ?, ?, ?)`);\n for (const doc of parsedDocs) {\n const values = writableColumns.map((col) => {\n if (col === 'path') return doc.relPath;\n if (col === '_mtime') return doc.mtimeMs;\n if (col === '_size') return doc.size;\n return doc.data[col] ?? null;\n });\n // Frontmatter upsert first: content's rowid lookup below depends on this row existing.\n insert.run(...values);\n // Delete-before-insert only for docs that have rows: an FTS5 DELETE by rowid on a\n // cold build (empty table, nothing to delete) is wasted work, and doing it\n // unconditionally previously made the crawl quadratic when it scanned by column --\n // measured 4x time per note-count doubling at 13k/26k notes.\n if (existing.has(doc.relPath)) {\n delBody.run(doc.relPath);\n for (const feature of features) feature.remove?.(db, doc.relPath, delta);\n }\n insertBody.run(doc.relPath, doc.search.title, doc.search.summary, doc.search.text, doc.relPath);\n // Coverage is glob-derived, not content-derived, but only reparsed/added docs are\n // touched here: a preset edit changes featureSignature and forces a full rebuild\n // (see open()), so an unchanged doc's coverage is already correct on disk. New docs\n // have no rows to clear -- skipping the delete keeps cold builds linear.\n if (existing.has(doc.relPath)) delPresetFiles.run(doc.relPath);\n for (const presetName of doc.presets) insertPresetFile.run(doc.relPath, presetName);\n for (const feature of features) feature.store?.(db, doc.relPath, doc.extracted[feature.name], delta);\n }\n }\n for (const feature of features) feature.afterReconcile?.(db, delta);\n db.exec('COMMIT');\n } catch (err) {\n db.exec('ROLLBACK');\n throw err;\n }\n\n // Reconcile's own write-transaction duration, for open()'s derived busy_timeout (F):\n // keep the observed max so a big watcher reconcile's lock hold is what the next open bounds its wait against.\n const durationMs = Date.now() - txStart;\n const prevRaw = getMeta(db, '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 (durationMs > prevMax) setMeta(db, 'reconcile_max_ms', String(durationMs));\n\n return { parsed: parsedDocs.length, warnings };\n}\n\n// Names what moved between two feature signatures (see config.featureSignature's format:\n// global features, embed provider, then one segment per preset) for the rebuild notice.\nfunction signatureDiff(before: string, after: string): string {\n // Segment keys: `features`, `embed`, `preset:<name>` (config.featureSignature's format).\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 const label = (key: string) => (key === 'embed' ? 'embed settings' : key.startsWith('preset:') ? `preset \"${key.slice(7)}\"` : 'features');\n return changed.size === 0 ? 'features' : [...changed].map(label).join(', ');\n}\n\nexport function open(cfg: ResolvedConfig): OpenResult {\n const stateDir = join(cfg.baseDir, STATE_DIR);\n mkdirSync(stateDir, { recursive: true });\n const dbPath = join(stateDir, DB_FILENAME);\n\n const db = new DatabaseSync(dbPath);\n db.exec('PRAGMA journal_mode = WAL');\n // Covers a concurrent watcher's bulk reconcile: the write transaction for 500 changed\n // files measures ~5s at 26k notes, so 5s expired exactly at the boundary and queries\n // racing the watcher got SQLITE_BUSY. 30s bounds the wait at ~3x the largest measured\n // reconcile; a query that outwaits it still fails loudly.\n db.exec('PRAGMA busy_timeout = 30000');\n registerFunctions(db);\n\n db.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 = getMeta(db, 'schema_version');\n const features = getMeta(db, 'features');\n const wantFeatures = featureSignature(cfg);\n if ((version !== null && version !== SCHEMA_VERSION) || (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 !== SCHEMA_VERSION) {\n console.error('sense: cache format changed (new sensemaking version); rebuilding the index');\n } else {\n const changed = signatureDiff(features ?? '', wantFeatures);\n console.error(`sense: config change (${changed}) rebuilds the index`);\n }\n db.close();\n rmSync(stateDir, { recursive: true, force: true });\n return open(cfg);\n }\n\n ensureSchema(db, cfg);\n\n // Derived from reconcile's own recorded max (F): 3x the largest reconcile this cache has\n // ever held its write transaction for, floored at the 30s default and capped at 10min so\n // one pathological build can't pin every later open to an unbounded wait. Installed\n // before reconcile() below -- this open's own reconcile is exactly the operation that\n // races a concurrent watcher's transaction and needs the derived wait.\n const recordedMaxMs = Number(getMeta(db, 'reconcile_max_ms') ?? '0');\n db.exec(`PRAGMA busy_timeout = ${Math.min(Math.max(30000, 3 * recordedMaxMs), 600_000)}`);\n\n const { parsed, warnings } = reconcile(db, cfg, cfg.baseDir);\n\n return { db, cfg, dbPath, parsed, warnings };\n}\n\n// Manual reset for a doubted cache.\nexport function rebuild(cfg: ResolvedConfig): OpenResult {\n rmSync(join(cfg.baseDir, STATE_DIR), { recursive: true, force: true });\n return open(cfg);\n}\n"],"names":["mkdirSync","rmSync","join","DatabaseSync","featureSignature","STATE_DIR","SenseError","activeFeatures","progress","listFiles","parseFile","RESERVED_COLUMNS","CORE_FRONTMATTER_COLUMNS","Set","DB_FILENAME","SCHEMA_VERSION","MAX_FRONTMATTER_COLUMNS","quoteIdent","name","split","registerFunctions","db","function","deterministic","varargs","field","value","undefined","needle","String","startsWith","parsed","JSON","parse","Array","isArray","some","item","includes","getColumns","rows","prepare","all","map","r","ensureSchema","cfg","exec","feature","schema","getMeta","setMeta","key","row","get","run","docCount","n","reconcile","baseDir","files","currentSet","f","relPath","existingRows","existing","Map","path","vanished","filter","has","toReparse","_mtime","mtimeMs","_size","size","length","warnings","features","seenColumns","newColumns","parsedDocs","report","parsedCount","file","fileFeatures","enabledForFile","doc","fileWarnings","tick","push","Object","keys","data","add","finish","allColumns","writableColumns","c","insertSql","added","delta","reparsed","d","txStart","Date","now","col","delBody","delPresetFiles","insertPresetFile","del","remove","insert","insertBody","values","search","title","summary","text","presetName","presets","store","extracted","afterReconcile","err","durationMs","prevRaw","prevMax","Number","signatureDiff","before","after","keyOf","part","slice","sig","a","b","changed","val","label","open","stateDir","recursive","dbPath","version","wantFeatures","console","error","close","force","recordedMaxMs","Math","min","max","rebuild"],"mappings":"AAAA,0FAA0F;AAC1F,4EAA4E;AAC5E,yFAAyF;AACzF,wFAAwF;AACxF,0EAA0E;AAC1E,SAASA,SAAS,EAAEC,MAAM,QAAQ,UAAU;AAC5C,SAASC,IAAI,QAAQ,YAAY;AACjC,SAASC,YAAY,QAAQ,cAAc;AAE3C,SAASC,gBAAgB,EAAEC,SAAS,QAAQ,cAAc;AAC1D,SAASC,UAAU,QAAQ,cAAc;AACzC,SAASC,cAAc,QAAQ,sBAAsB;AAErD,SAASC,QAAQ,QAAQ,gBAAgB;AAEzC,SAASC,SAAS,EAAEC,SAAS,EAAEC,gBAAgB,QAAQ,YAAY;AAEnE,oFAAoF;AACpF,oFAAoF;AACpF,yFAAyF;AACzF,yFAAyF;AACzF,kFAAkF;AAClF,MAAMC,2BAA2B,IAAIC,IAAI;IAAC;IAAQ;IAAU;CAAQ;AAEpE,gFAAgF;AAChF,kEAAkE;AAElE,OAAO,MAAMC,cAAc,WAAW;AACtC,kEAAkE;AAClE,2EAA2E;AAC3E,sFAAsF;AACtF,yCAAyC;AACzC,OAAO,MAAMC,iBAAiB,IAAI;AAElC,8FAA8F;AAC9F,MAAMC,0BAA0B;AAUhC,SAASC,WAAWC,IAAY;IAC9B,OAAO,CAAC,CAAC,EAAEA,KAAKC,KAAK,CAAC,KAAKjB,IAAI,CAAC,MAAM,CAAC,CAAC;AAC1C;AAEA,+FAA+F;AAC/F,SAASkB,kBAAkBC,EAAgB;IACzCA,GAAGC,QAAQ,CAAC,OAAO;QAAEC,eAAe;QAAMC,SAAS;IAAM,GAAG,CAACC,OAAgBC;QAC3E,IAAID,UAAU,QAAQA,UAAUE,WAAW,OAAO;QAElD,MAAMC,SAASC,OAAOH;QAEtB,IAAI,OAAOD,UAAU,UAAU;YAC7B,IAAIA,MAAMK,UAAU,CAAC,MAAM;gBACzB,IAAI;oBACF,MAAMC,SAASC,KAAKC,KAAK,CAACR;oBAC1B,IAAIS,MAAMC,OAAO,CAACJ,SAAS;wBACzB,OAAOA,OAAOK,IAAI,CAAC,CAACC,OAASR,OAAOQ,UAAUT,UAAU,IAAI;oBAC9D;gBACF,EAAE,OAAM,CAAC;YACX;YACA,OAAOH,MAAMa,QAAQ,CAACV,UAAU,IAAI;QACtC;QAEA,OAAOC,OAAOJ,OAAOa,QAAQ,CAACV,UAAU,IAAI;IAC9C;AACF;AAEA,SAASW,WAAWlB,EAAgB;IAClC,MAAMmB,OAAOnB,GAAGoB,OAAO,CAAC,kCAAkCC,GAAG;IAC7D,OAAO,IAAI7B,IAAI2B,KAAKG,GAAG,CAAC,CAACC,IAAMA,EAAE1B,IAAI;AACvC;AAEA,2FAA2F;AAC3F,wFAAwF;AACxF,SAAS2B,aAAaxB,EAAgB,EAAEyB,GAAW;IACjDzB,GAAG0B,IAAI,CAAC,CAAC,gGAAgG,CAAC;IAC1G1B,GAAG0B,IAAI,CAAC,CAAC,0HAA0H,CAAC;IACpI,qFAAqF;IACrF,uFAAuF;IACvF,4DAA4D;IAC5D,0FAA0F;IAC1F,2FAA2F;IAC3F,qFAAqF;IACrF1B,GAAG0B,IAAI,CAAC,CAAC,gGAAgG,CAAC;IAC1G1B,GAAG0B,IAAI,CAAC;IACR,KAAK,MAAMC,WAAWzC,eAAeuC,KAAME,QAAQC,MAAM,CAAC5B;IAC1D,IAAI6B,QAAQ7B,IAAI,sBAAsB,MAAM8B,QAAQ9B,IAAI,kBAAkBN;IAC1E,IAAImC,QAAQ7B,IAAI,gBAAgB,MAAM8B,QAAQ9B,IAAI,YAAYjB,iBAAiB0C;AACjF;AAEA,OAAO,SAASI,QAAQ7B,EAAgB,EAAE+B,GAAW;IACnD,MAAMC,MAAMhC,GAAGoB,OAAO,CAAC,wCAAwCa,GAAG,CAACF;IACnE,OAAOC,MAAMA,IAAI3B,KAAK,GAAG;AAC3B;AAEA,OAAO,SAASyB,QAAQ9B,EAAgB,EAAE+B,GAAW,EAAE1B,KAAoB;IACzE,IAAIA,UAAU,MAAM;QAClBL,GAAGoB,OAAO,CAAC,kCAAkCc,GAAG,CAACH;QACjD;IACF;IACA/B,GAAGoB,OAAO,CAAC,qGAAqGc,GAAG,CAACH,KAAK1B;AAC3H;AAEA,OAAO,SAAS8B,SAASnC,EAAgB;IACvC,MAAMgC,MAAMhC,GAAGoB,OAAO,CAAC,yCAAyCa,GAAG;IACnE,OAAOD,IAAII,CAAC;AACd;AAEA,OAAO,SAASC,UAAUrC,EAAgB,EAAEyB,GAAW,EAAEa,OAAe;IACtE,MAAMC,QAAQnD,UAAUqC,KAAKa;IAC7B,MAAME,aAAa,IAAIhD,IAAI+C,MAAMjB,GAAG,CAAC,CAACmB,IAAMA,EAAEC,OAAO;IAErD,MAAMC,eAAe3C,GAAGoB,OAAO,CAAC,CAAC,iDAAiD,CAAC,EAAEC,GAAG;IAKxF,MAAMuB,WAAW,IAAIC,IAAIF,aAAarB,GAAG,CAAC,CAACC,IAAM;YAACA,EAAEuB,IAAI;YAAEvB;SAAE;IAC5D,MAAMwB,WAAWJ,aAAaK,MAAM,CAAC,CAACzB,IAAM,CAACiB,WAAWS,GAAG,CAAC1B,EAAEuB,IAAI,GAAGxB,GAAG,CAAC,CAACC,IAAMA,EAAEuB,IAAI;IAEtF,MAAMI,YAAYX,MAAMS,MAAM,CAAC,CAACP;QAC9B,MAAMT,MAAMY,SAASX,GAAG,CAACQ,EAAEC,OAAO;QAClC,OAAO,CAACV,OAAOA,IAAImB,MAAM,KAAKV,EAAEW,OAAO,IAAIpB,IAAIqB,KAAK,KAAKZ,EAAEa,IAAI;IACjE;IAEA,IAAIP,SAASQ,MAAM,KAAK,KAAKL,UAAUK,MAAM,KAAK,GAAG,OAAO;QAAE7C,QAAQ;QAAG8C,UAAU,EAAE;IAAC;IAEtF,MAAMC,WAAWvE,eAAeuC;IAChC,MAAMiC,cAAcxC,WAAWlB;IAC/B,MAAM2D,aAAuB,EAAE;IAC/B,MAAMC,aAA0B,EAAE;IAClC,MAAMJ,WAAqB,EAAE;IAE7B,oFAAoF;IACpF,uDAAuD;IACvD,MAAMK,SAAS1E,SAAS,mBAAmB+D,UAAUK,MAAM;IAC3D,IAAIO,cAAc;IAClB,KAAK,MAAMC,QAAQb,UAAW;QAC5B,sFAAsF;QACtF,iEAAiE;QACjE,MAAMc,eAAeP,SAAST,MAAM,CAAC,CAACrB,UAAY,CAACA,QAAQsC,cAAc,IAAItC,QAAQsC,cAAc,CAACxC,KAAKsC;QACzG,MAAM,EAAEG,GAAG,EAAEV,UAAUW,YAAY,EAAE,GAAG9E,UAAU0E,MAAMC;QACxDH,OAAOO,IAAI,CAAC,EAAEN;QACdN,SAASa,IAAI,IAAIF;QACjB,KAAK,MAAMpC,OAAOuC,OAAOC,IAAI,CAACL,IAAIM,IAAI,EAAG;YACvC,IAAI,CAACd,YAAYT,GAAG,CAAClB,MAAM;gBACzB2B,YAAYe,GAAG,CAAC1C;gBAChB4B,WAAWU,IAAI,CAACtC;YAClB;QACF;QACA6B,WAAWS,IAAI,CAACH;IAClB;IACAL,OAAOa,MAAM;IAEb,MAAMC,aAAa;WAAIjB;KAAY;IACnC,uEAAuE;IACvE,sGAAsG;IACtG,IAAIiB,WAAWpB,MAAM,GAAG5D,yBAAyB;QAC/C,MAAM,IAAIV,WACR,gBACA,CAAC,uBAAuB,EAAE0F,WAAWpB,MAAM,CAAC,0EAA0E,EAAE5D,wBAAwB,wKAAwK,CAAC;IAE7T;IACA,0FAA0F;IAC1F,sEAAsE;IACtE,MAAMiF,kBAAkBD,WAAW3B,MAAM,CAAC,CAAC6B,IAAMtF,yBAAyB0D,GAAG,CAAC4B,MAAM,CAACvF,iBAAiB2D,GAAG,CAAC4B;IAC1G,sFAAsF;IACtF,gDAAgD;IAChD,MAAMC,YAAY,CAAC,yBAAyB,EAAEF,gBAAgBtD,GAAG,CAAC1B,YAAYf,IAAI,CAAC,MAAM,UAAU,EAAE+F,gBAAgBtD,GAAG,CAAC,IAAM,KAAKzC,IAAI,CAAC,MAAM,oCAAoC,EAAE+F,gBAClL5B,MAAM,CAAC,CAAC6B,IAAMA,MAAM,QACpBvD,GAAG,CAAC,CAACuD,IAAM,GAAGjF,WAAWiF,GAAG,YAAY,EAAEjF,WAAWiF,IAAI,EACzDhG,IAAI,CAAC,OAAO;IAEf,MAAMkG,QAAQ7B,UAAUF,MAAM,CAAC,CAACP,IAAM,CAACG,SAASK,GAAG,CAACR,EAAEC,OAAO,GAAGpB,GAAG,CAAC,CAACmB,IAAMA,EAAEC,OAAO;IACpF,MAAMsC,QAAwB;QAAEzC;QAAO0C,UAAUrB,WAAWtC,GAAG,CAAC,CAAC4D,IAAMA,EAAExC,OAAO;QAAGqC;QAAOhC;IAAS;IAEnG,MAAMoC,UAAUC,KAAKC,GAAG;IACxBrF,GAAG0B,IAAI,CAAC;IACR,IAAI;YAiD8BC;QAhDhC,KAAK,MAAM2D,OAAO3B,WAAY3D,GAAG0B,IAAI,CAAC,CAAC,mCAAmC,EAAE9B,WAAW0F,MAAM;QAC7F,6EAA6E;QAC7E,kFAAkF;QAClF,+EAA+E;QAC/E,MAAMC,UAAUvF,GAAGoB,OAAO,CAAC,CAAC,kFAAkF,CAAC;QAC/G,MAAMoE,iBAAiBxF,GAAGoB,OAAO,CAAC,CAAC,yCAAyC,CAAC;QAC7E,MAAMqE,mBAAmBzF,GAAGoB,OAAO,CAAC,CAAC,uDAAuD,CAAC;QAC7F,IAAI2B,SAASQ,MAAM,GAAG,GAAG;YACvB,MAAMmC,MAAM1F,GAAGoB,OAAO,CAAC,CAAC,wCAAwC,CAAC;YACjE,KAAK,MAAM0B,QAAQC,SAAU;oBAMKpB;gBALhC,kFAAkF;gBAClF,qEAAqE;gBACrE4D,QAAQrD,GAAG,CAACY;gBACZ4C,IAAIxD,GAAG,CAACY;gBACR0C,eAAetD,GAAG,CAACY;gBACnB,KAAK,MAAMnB,WAAW8B,UAAU9B,kBAAAA,QAAQgE,MAAM,cAAdhE,sCAAAA,qBAAAA,SAAiB3B,IAAI8C,MAAMkC;YAC7D;QACF;QACA,IAAIpB,WAAWL,MAAM,GAAG,GAAG;YACzB,MAAMqC,SAAS5F,GAAGoB,OAAO,CAAC0D;YAC1B,MAAMe,aAAa7F,GAAGoB,OAAO,CAAC,CAAC,+HAA+H,CAAC;YAC/J,KAAK,MAAM8C,OAAON,WAAY;oBAwBIjC;gBAvBhC,MAAMmE,SAASlB,gBAAgBtD,GAAG,CAAC,CAACgE;wBAI3BpB;oBAHP,IAAIoB,QAAQ,QAAQ,OAAOpB,IAAIxB,OAAO;oBACtC,IAAI4C,QAAQ,UAAU,OAAOpB,IAAId,OAAO;oBACxC,IAAIkC,QAAQ,SAAS,OAAOpB,IAAIZ,IAAI;oBACpC,QAAOY,gBAAAA,IAAIM,IAAI,CAACc,IAAI,cAAbpB,2BAAAA,gBAAiB;gBAC1B;gBACA,uFAAuF;gBACvF0B,OAAO1D,GAAG,IAAI4D;gBACd,kFAAkF;gBAClF,2EAA2E;gBAC3E,mFAAmF;gBACnF,6DAA6D;gBAC7D,IAAIlD,SAASK,GAAG,CAACiB,IAAIxB,OAAO,GAAG;wBAEGf;oBADhC4D,QAAQrD,GAAG,CAACgC,IAAIxB,OAAO;oBACvB,KAAK,MAAMf,WAAW8B,UAAU9B,mBAAAA,QAAQgE,MAAM,cAAdhE,uCAAAA,sBAAAA,SAAiB3B,IAAIkE,IAAIxB,OAAO,EAAEsC;gBACpE;gBACAa,WAAW3D,GAAG,CAACgC,IAAIxB,OAAO,EAAEwB,IAAI6B,MAAM,CAACC,KAAK,EAAE9B,IAAI6B,MAAM,CAACE,OAAO,EAAE/B,IAAI6B,MAAM,CAACG,IAAI,EAAEhC,IAAIxB,OAAO;gBAC9F,kFAAkF;gBAClF,iFAAiF;gBACjF,oFAAoF;gBACpF,yEAAyE;gBACzE,IAAIE,SAASK,GAAG,CAACiB,IAAIxB,OAAO,GAAG8C,eAAetD,GAAG,CAACgC,IAAIxB,OAAO;gBAC7D,KAAK,MAAMyD,cAAcjC,IAAIkC,OAAO,CAAEX,iBAAiBvD,GAAG,CAACgC,IAAIxB,OAAO,EAAEyD;gBACxE,KAAK,MAAMxE,WAAW8B,UAAU9B,iBAAAA,QAAQ0E,KAAK,cAAb1E,qCAAAA,oBAAAA,SAAgB3B,IAAIkE,IAAIxB,OAAO,EAAEwB,IAAIoC,SAAS,CAAC3E,QAAQ9B,IAAI,CAAC,EAAEmF;YAChG;QACF;QACA,KAAK,MAAMrD,WAAW8B,UAAU9B,0BAAAA,QAAQ4E,cAAc,cAAtB5E,8CAAAA,6BAAAA,SAAyB3B,IAAIgF;QAC7DhF,GAAG0B,IAAI,CAAC;IACV,EAAE,OAAO8E,KAAK;QACZxG,GAAG0B,IAAI,CAAC;QACR,MAAM8E;IACR;IAEA,qFAAqF;IACrF,8GAA8G;IAC9G,MAAMC,aAAarB,KAAKC,GAAG,KAAKF;IAChC,MAAMuB,UAAU7E,QAAQ7B,IAAI;IAC5B,yFAAyF;IACzF,+EAA+E;IAC/E,MAAM2G,UAAUD,YAAY,OAAO,CAAC,IAAIE,OAAOF;IAC/C,IAAID,aAAaE,SAAS7E,QAAQ9B,IAAI,oBAAoBQ,OAAOiG;IAEjE,OAAO;QAAE/F,QAAQkD,WAAWL,MAAM;QAAEC;IAAS;AAC/C;AAEA,yFAAyF;AACzF,wFAAwF;AACxF,SAASqD,cAAcC,MAAc,EAAEC,KAAa;IAClD,yFAAyF;IACzF,MAAMC,QAAQ,CAACC,OAAkBA,KAAKxG,UAAU,CAAC,aAAawG,KAAKnH,KAAK,CAAC,KAAKoH,KAAK,CAAC,GAAG,GAAGrI,IAAI,CAAC,OAAOoI,KAAKnH,KAAK,CAAC,IAAI,CAAC,EAAE;IACxH,MAAMc,QAAQ,CAACuG,MAAgB,IAAItE,IAAIsE,IAAIrH,KAAK,CAAC,KAAKwB,GAAG,CAAC,CAAC2F,OAAS;gBAACD,MAAMC;gBAAOA;aAAK;IACvF,MAAMG,IAAIxG,MAAMkG;IAChB,MAAMO,IAAIzG,MAAMmG;IAChB,MAAMO,UAAU,IAAI9H;IACpB,KAAK,MAAM,CAACuC,KAAKwF,IAAI,IAAIF,EAAG,IAAID,EAAEnF,GAAG,CAACF,SAASwF,KAAKD,QAAQ7C,GAAG,CAAC1C;IAChE,KAAK,MAAMA,OAAOqF,EAAE7C,IAAI,GAAI,IAAI,CAAC8C,EAAEpE,GAAG,CAAClB,MAAMuF,QAAQ7C,GAAG,CAAC1C;IACzD,MAAMyF,QAAQ,CAACzF,MAAiBA,QAAQ,UAAU,mBAAmBA,IAAItB,UAAU,CAAC,aAAa,CAAC,QAAQ,EAAEsB,IAAImF,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG;IAC9H,OAAOI,QAAQhE,IAAI,KAAK,IAAI,aAAa;WAAIgE;KAAQ,CAAChG,GAAG,CAACkG,OAAO3I,IAAI,CAAC;AACxE;AAEA,OAAO,SAAS4I,KAAKhG,GAAmB;QA0CTI;IAzC7B,MAAM6F,WAAW7I,KAAK4C,IAAIa,OAAO,EAAEtD;IACnCL,UAAU+I,UAAU;QAAEC,WAAW;IAAK;IACtC,MAAMC,SAAS/I,KAAK6I,UAAUjI;IAE9B,MAAMO,KAAK,IAAIlB,aAAa8I;IAC5B5H,GAAG0B,IAAI,CAAC;IACR,sFAAsF;IACtF,qFAAqF;IACrF,sFAAsF;IACtF,0DAA0D;IAC1D1B,GAAG0B,IAAI,CAAC;IACR3B,kBAAkBC;IAElBA,GAAG0B,IAAI,CAAC;IAER,uFAAuF;IACvF,qGAAqG;IACrG,MAAMmG,UAAUhG,QAAQ7B,IAAI;IAC5B,MAAMyD,WAAW5B,QAAQ7B,IAAI;IAC7B,MAAM8H,eAAe/I,iBAAiB0C;IACtC,IAAI,AAACoG,YAAY,QAAQA,YAAYnI,kBAAoB+D,aAAa,QAAQA,aAAaqE,cAAe;QACxG,uFAAuF;QACvF,wFAAwF;QACxF,IAAID,YAAY,QAAQA,YAAYnI,gBAAgB;YAClDqI,QAAQC,KAAK,CAAC;QAChB,OAAO;YACL,MAAMV,UAAUT,cAAcpD,qBAAAA,sBAAAA,WAAY,IAAIqE;YAC9CC,QAAQC,KAAK,CAAC,CAAC,sBAAsB,EAAEV,QAAQ,oBAAoB,CAAC;QACtE;QACAtH,GAAGiI,KAAK;QACRrJ,OAAO8I,UAAU;YAAEC,WAAW;YAAMO,OAAO;QAAK;QAChD,OAAOT,KAAKhG;IACd;IAEAD,aAAaxB,IAAIyB;IAEjB,yFAAyF;IACzF,yFAAyF;IACzF,oFAAoF;IACpF,sFAAsF;IACtF,uEAAuE;IACvE,MAAM0G,gBAAgBvB,QAAO/E,WAAAA,QAAQ7B,IAAI,iCAAZ6B,sBAAAA,WAAmC;IAChE7B,GAAG0B,IAAI,CAAC,CAAC,sBAAsB,EAAE0G,KAAKC,GAAG,CAACD,KAAKE,GAAG,CAAC,OAAO,IAAIH,gBAAgB,SAAU;IAExF,MAAM,EAAEzH,MAAM,EAAE8C,QAAQ,EAAE,GAAGnB,UAAUrC,IAAIyB,KAAKA,IAAIa,OAAO;IAE3D,OAAO;QAAEtC;QAAIyB;QAAKmG;QAAQlH;QAAQ8C;IAAS;AAC7C;AAEA,oCAAoC;AACpC,OAAO,SAAS+E,QAAQ9G,GAAmB;IACzC7C,OAAOC,KAAK4C,IAAIa,OAAO,EAAEtD,YAAY;QAAE2I,WAAW;QAAMO,OAAO;IAAK;IACpE,OAAOT,KAAKhG;AACd"}
|
|
1
|
+
{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/db.ts"],"sourcesContent":["// The Node floor (>=22.20) is explained here and nowhere else: 22.20 is the first release with\n// both FTS5 and row-returning INSERT ... RETURNING. Raise it only for a load-bearing capability.\nimport { mkdirSync, rmSync } from 'node:fs';\nimport { join } from 'node:path';\nimport { DatabaseSync } from 'node:sqlite';\nimport type { Config, ResolvedConfig } from './config.ts';\nimport { featureSignature, STATE_DIR } from './config.ts';\nimport { SenseError } from './errors.ts';\nimport { activeFeatures } from './features/index.ts';\nimport type { ReconcileDelta } from './features/types.ts';\nimport { progress } from './progress.ts';\nimport type { ParsedDoc } from './scan.ts';\nimport { listFiles, parseFile, RESERVED_COLUMNS } from './scan.ts';\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', '_size']);\n\n// Rows -> SQLite: core schema, reconcile loop, has(). Parsing lives in scan.ts;\n// everything beyond frontmatter + content lives in src/features/.\n\nexport const DB_FILENAME = 'cache.db';\n// Cache shape version, independent of the config's own `version`. Bumping it rebuilds\n// existing trees on first query.\nexport const SCHEMA_VERSION = '8';\n\n// SQLite's compile-time SQLITE_MAX_COLUMN, default 2000 (https://www.sqlite.org/limits.html).\nconst MAX_FRONTMATTER_COLUMNS = 2000;\n\nexport interface OpenResult {\n db: DatabaseSync;\n cfg: ResolvedConfig;\n dbPath: string;\n parsed: number;\n warnings: string[];\n}\n\nfunction quoteIdent(name: string): string {\n return `\"${name.split('\"').join('\"\"')}\"`;\n}\n\n// has(field, value): JSON-array field -> membership, string field -> substring, NULL -> false.\nfunction registerFunctions(db: DatabaseSync): void {\n db.function('has', { deterministic: true, varargs: false }, (field: unknown, value: unknown): number => {\n if (field === null || field === undefined) return 0;\n\n const needle = String(value);\n\n if (typeof field === 'string') {\n if (field.startsWith('[')) {\n try {\n const parsed = JSON.parse(field);\n if (Array.isArray(parsed)) {\n return parsed.some((item) => String(item) === needle) ? 1 : 0;\n }\n } catch {}\n }\n return field.includes(needle) ? 1 : 0;\n }\n\n return String(field).includes(needle) ? 1 : 0;\n });\n}\n\nfunction getColumns(db: DatabaseSync): Set<string> {\n const rows = db.prepare('PRAGMA table_info(frontmatter)').all() as Array<{ name: string }>;\n return new Set(rows.map((r) => r.name));\n}\n\n// Content is a separate table (not a column on frontmatter) so `SELECT * FROM frontmatter`\n// can't dump file text into context. Features add their own tables after the core ones.\nfunction ensureSchema(db: DatabaseSync, cfg: Config): void {\n db.exec(`CREATE TABLE IF NOT EXISTS frontmatter (\"path\" TEXT PRIMARY KEY, \"_mtime\" REAL, \"_size\" INTEGER)`);\n db.exec(`CREATE VIRTUAL TABLE IF NOT EXISTS content USING fts5(title, summary, text, path UNINDEXED, tokenize = 'porter unicode61')`);\n // Coverage, not ownership: a path can appear under several presets. path leads the PK so the\n // per-doc delete is an index hit -- keyed the other way, cold builds went quadratic.\n db.exec(`CREATE TABLE IF NOT EXISTS preset_files (\"path\" TEXT, preset TEXT, PRIMARY KEY (\"path\", preset))`);\n db.exec('CREATE INDEX IF NOT EXISTS preset_files_preset ON preset_files(preset)');\n for (const feature of activeFeatures(cfg)) feature.schema(db);\n if (getMeta(db, 'schema_version') === null) setMeta(db, 'schema_version', SCHEMA_VERSION);\n if (getMeta(db, 'features') === null) setMeta(db, 'features', featureSignature(cfg));\n}\n\nexport function getMeta(db: DatabaseSync, key: string): string | null {\n const row = db.prepare('SELECT value FROM meta WHERE key = ?').get(key) as { value: string } | undefined;\n return row ? row.value : null;\n}\n\nexport function setMeta(db: DatabaseSync, key: string, value: string | null): void {\n if (value === null) {\n db.prepare('DELETE FROM meta WHERE key = ?').run(key);\n return;\n }\n db.prepare('INSERT INTO meta (key, value) VALUES (?, ?) ON CONFLICT(key) DO UPDATE SET value = excluded.value').run(key, value);\n}\n\nexport function docCount(db: DatabaseSync): number {\n const row = db.prepare('SELECT COUNT(*) AS n FROM frontmatter').get() as { n: number };\n return row.n;\n}\n\nexport function reconcile(db: DatabaseSync, cfg: Config, baseDir: string): { parsed: number; warnings: string[] } {\n const files = listFiles(cfg, baseDir);\n const currentSet = new Set(files.map((f) => f.relPath));\n\n const existingRows = db.prepare(`SELECT \"path\", \"_mtime\", \"_size\" FROM frontmatter`).all() as Array<{\n path: string;\n _mtime: number;\n _size: number;\n }>;\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 = getColumns(db);\n const newColumns: string[] = [];\n const parsedDocs: ParsedDoc[] = [];\n const warnings: string[] = [];\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 let parsedCount = 0;\n for (const file of toReparse) {\n // A doc only gets extract/store from features that apply to it (currently: embed, via\n // FileStat.embed -- true iff the config names an embedding model).\n const fileFeatures = features.filter((feature) => !feature.enabledForFile || feature.enabledForFile(cfg, file));\n const { doc, warnings: fileWarnings } = parseFile(file, fileFeatures);\n report.tick(++parsedCount);\n warnings.push(...fileWarnings);\n for (const key of Object.keys(doc.data)) {\n if (!seenColumns.has(key)) {\n seenColumns.add(key);\n newColumns.push(key);\n }\n }\n parsedDocs.push(doc);\n }\n report.finish();\n\n const allColumns = [...seenColumns];\n // Fence before ALTERing: SQLite's own failure past this point is a raw\n // \"too many columns on sqlite_altertab_frontmatter\" with no indication of the boundary or the levers.\n if (allColumns.length > MAX_FRONTMATTER_COLUMNS) {\n throw new SenseError(\n 'COLUMN_LIMIT',\n `frontmatter would need ${allColumns.length} columns, crossing SQLite's compile-time SQLITE_MAX_COLUMN limit (default ${MAX_FRONTMATTER_COLUMNS}; see https://www.sqlite.org/limits.html). Narrow the presets' include globs so fewer/other files are indexed, or fix whatever is generating unbounded frontmatter keys.`\n );\n }\n // Columns the frontmatter upsert actually writes: core + parsed frontmatter keys, never a\n // feature-owned reserved column (see CORE_FRONTMATTER_COLUMNS above).\n const writableColumns = allColumns.filter((c) => CORE_FRONTMATTER_COLUMNS.has(c) || !RESERVED_COLUMNS.has(c));\n // ON CONFLICT UPDATE (not OR REPLACE) keeps the row's rowid stable across reparses --\n // content rows are coupled to that rowid below.\n const insertSql = `INSERT INTO frontmatter (${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\n const txStart = Date.now();\n db.exec('BEGIN');\n try {\n for (const col of newColumns) db.exec(`ALTER TABLE frontmatter ADD COLUMN ${quoteIdent(col)}`);\n // FTS5 has no upsert, so delete-before-insert into `content`; coupled to the\n // frontmatter rowid (indexed via its PRIMARY KEY) instead of the UNINDEXED `path`\n // column, which a per-row DELETE would otherwise scan the whole table to find.\n const delBody = db.prepare(`DELETE FROM content WHERE rowid = (SELECT rowid FROM frontmatter WHERE \"path\" = ?)`);\n const delPresetFiles = db.prepare(`DELETE FROM preset_files WHERE \"path\" = ?`);\n const insertPresetFile = db.prepare(`INSERT INTO preset_files (\"path\", preset) VALUES (?, ?)`);\n if (vanished.length > 0) {\n const del = db.prepare(`DELETE FROM frontmatter WHERE \"path\" = ?`);\n for (const path of vanished) {\n // content delete must run first: it looks up the frontmatter rowid by path, which\n // the frontmatter delete below would otherwise have already removed.\n delBody.run(path);\n del.run(path);\n delPresetFiles.run(path);\n for (const feature of features) feature.remove?.(db, path, delta);\n }\n }\n if (parsedDocs.length > 0) {\n const insert = db.prepare(insertSql);\n const insertBody = db.prepare(`INSERT INTO content (rowid, title, summary, text, \"path\") VALUES ((SELECT rowid FROM frontmatter WHERE \"path\" = ?), ?, ?, ?, ?)`);\n for (const doc of parsedDocs) {\n const values = writableColumns.map((col) => {\n if (col === 'path') return doc.relPath;\n if (col === '_mtime') return doc.mtimeMs;\n if (col === '_size') return doc.size;\n return doc.data[col] ?? null;\n });\n // Frontmatter upsert first: content's rowid lookup below depends on this row existing.\n insert.run(...values);\n // Only for docs that have rows: an unconditional delete made cold crawls quadratic.\n if (existing.has(doc.relPath)) {\n delBody.run(doc.relPath);\n for (const feature of features) feature.remove?.(db, doc.relPath, delta);\n }\n insertBody.run(doc.relPath, doc.search.title, doc.search.summary, doc.search.text, doc.relPath);\n // A preset edit forces a full rebuild, so an unchanged doc's coverage is already\n // correct; new docs have nothing to clear, which keeps cold builds linear.\n if (existing.has(doc.relPath)) delPresetFiles.run(doc.relPath);\n for (const presetName of doc.presets) insertPresetFile.run(doc.relPath, presetName);\n for (const feature of features) feature.store?.(db, doc.relPath, doc.extracted[feature.name], delta);\n }\n }\n for (const feature of features) feature.afterReconcile?.(db, delta);\n db.exec('COMMIT');\n } catch (err) {\n db.exec('ROLLBACK');\n throw err;\n }\n\n // Reconcile's own write-transaction duration, for open()'s derived busy_timeout (F):\n // keep the observed max so a big watcher reconcile's lock hold is what the next open bounds its wait against.\n const durationMs = Date.now() - txStart;\n const prevRaw = getMeta(db, '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 (durationMs > prevMax) setMeta(db, 'reconcile_max_ms', String(durationMs));\n\n return { parsed: parsedDocs.length, warnings };\n}\n\n// Names what moved between two feature signatures (see config.featureSignature's format:\n// global features, embed provider, then one segment per preset) for the rebuild notice.\nfunction signatureDiff(before: string, after: string): string {\n // Segment keys: `features`, `embed`, `preset:<name>` (config.featureSignature's format).\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 const label = (key: string) => (key === 'embed' ? 'embed settings' : key.startsWith('preset:') ? `preset \"${key.slice(7)}\"` : 'features');\n return changed.size === 0 ? 'features' : [...changed].map(label).join(', ');\n}\n\nexport function open(cfg: ResolvedConfig): OpenResult {\n const stateDir = join(cfg.baseDir, STATE_DIR);\n mkdirSync(stateDir, { recursive: true });\n const dbPath = join(stateDir, DB_FILENAME);\n\n const db = new DatabaseSync(dbPath);\n db.exec('PRAGMA journal_mode = WAL');\n // Covers a concurrent watcher's bulk reconcile (~5s for 500 files at 26k notes). A query\n // that outwaits it still fails loudly.\n db.exec('PRAGMA busy_timeout = 30000');\n registerFunctions(db);\n\n db.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 = getMeta(db, 'schema_version');\n const features = getMeta(db, 'features');\n const wantFeatures = featureSignature(cfg);\n if ((version !== null && version !== SCHEMA_VERSION) || (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 !== SCHEMA_VERSION) {\n console.error('sense: cache format changed (new sensemaking version); rebuilding the index');\n } else {\n const changed = signatureDiff(features ?? '', wantFeatures);\n console.error(`sense: config change (${changed}) rebuilds the index`);\n }\n db.close();\n rmSync(stateDir, { recursive: true, force: true });\n return open(cfg);\n }\n\n ensureSchema(db, 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(getMeta(db, 'reconcile_max_ms') ?? '0');\n db.exec(`PRAGMA busy_timeout = ${Math.min(Math.max(30000, 3 * recordedMaxMs), 600_000)}`);\n\n const { parsed, warnings } = reconcile(db, cfg, cfg.baseDir);\n\n return { db, cfg, dbPath, parsed, warnings };\n}\n\n// Manual reset for a doubted cache.\nexport function rebuild(cfg: ResolvedConfig): OpenResult {\n rmSync(join(cfg.baseDir, STATE_DIR), { recursive: true, force: true });\n return open(cfg);\n}\n"],"names":["mkdirSync","rmSync","join","DatabaseSync","featureSignature","STATE_DIR","SenseError","activeFeatures","progress","listFiles","parseFile","RESERVED_COLUMNS","CORE_FRONTMATTER_COLUMNS","Set","DB_FILENAME","SCHEMA_VERSION","MAX_FRONTMATTER_COLUMNS","quoteIdent","name","split","registerFunctions","db","function","deterministic","varargs","field","value","undefined","needle","String","startsWith","parsed","JSON","parse","Array","isArray","some","item","includes","getColumns","rows","prepare","all","map","r","ensureSchema","cfg","exec","feature","schema","getMeta","setMeta","key","row","get","run","docCount","n","reconcile","baseDir","files","currentSet","f","relPath","existingRows","existing","Map","path","vanished","filter","has","toReparse","_mtime","mtimeMs","_size","size","length","warnings","features","seenColumns","newColumns","parsedDocs","report","parsedCount","file","fileFeatures","enabledForFile","doc","fileWarnings","tick","push","Object","keys","data","add","finish","allColumns","writableColumns","c","insertSql","added","delta","reparsed","d","txStart","Date","now","col","delBody","delPresetFiles","insertPresetFile","del","remove","insert","insertBody","values","search","title","summary","text","presetName","presets","store","extracted","afterReconcile","err","durationMs","prevRaw","prevMax","Number","signatureDiff","before","after","keyOf","part","slice","sig","a","b","changed","val","label","open","stateDir","recursive","dbPath","version","wantFeatures","console","error","close","force","recordedMaxMs","Math","min","max","rebuild"],"mappings":"AAAA,+FAA+F;AAC/F,iGAAiG;AACjG,SAASA,SAAS,EAAEC,MAAM,QAAQ,UAAU;AAC5C,SAASC,IAAI,QAAQ,YAAY;AACjC,SAASC,YAAY,QAAQ,cAAc;AAE3C,SAASC,gBAAgB,EAAEC,SAAS,QAAQ,cAAc;AAC1D,SAASC,UAAU,QAAQ,cAAc;AACzC,SAASC,cAAc,QAAQ,sBAAsB;AAErD,SAASC,QAAQ,QAAQ,gBAAgB;AAEzC,SAASC,SAAS,EAAEC,SAAS,EAAEC,gBAAgB,QAAQ,YAAY;AAEnE,6FAA6F;AAC7F,4EAA4E;AAC5E,MAAMC,2BAA2B,IAAIC,IAAI;IAAC;IAAQ;IAAU;CAAQ;AAEpE,gFAAgF;AAChF,kEAAkE;AAElE,OAAO,MAAMC,cAAc,WAAW;AACtC,sFAAsF;AACtF,iCAAiC;AACjC,OAAO,MAAMC,iBAAiB,IAAI;AAElC,8FAA8F;AAC9F,MAAMC,0BAA0B;AAUhC,SAASC,WAAWC,IAAY;IAC9B,OAAO,CAAC,CAAC,EAAEA,KAAKC,KAAK,CAAC,KAAKjB,IAAI,CAAC,MAAM,CAAC,CAAC;AAC1C;AAEA,+FAA+F;AAC/F,SAASkB,kBAAkBC,EAAgB;IACzCA,GAAGC,QAAQ,CAAC,OAAO;QAAEC,eAAe;QAAMC,SAAS;IAAM,GAAG,CAACC,OAAgBC;QAC3E,IAAID,UAAU,QAAQA,UAAUE,WAAW,OAAO;QAElD,MAAMC,SAASC,OAAOH;QAEtB,IAAI,OAAOD,UAAU,UAAU;YAC7B,IAAIA,MAAMK,UAAU,CAAC,MAAM;gBACzB,IAAI;oBACF,MAAMC,SAASC,KAAKC,KAAK,CAACR;oBAC1B,IAAIS,MAAMC,OAAO,CAACJ,SAAS;wBACzB,OAAOA,OAAOK,IAAI,CAAC,CAACC,OAASR,OAAOQ,UAAUT,UAAU,IAAI;oBAC9D;gBACF,EAAE,OAAM,CAAC;YACX;YACA,OAAOH,MAAMa,QAAQ,CAACV,UAAU,IAAI;QACtC;QAEA,OAAOC,OAAOJ,OAAOa,QAAQ,CAACV,UAAU,IAAI;IAC9C;AACF;AAEA,SAASW,WAAWlB,EAAgB;IAClC,MAAMmB,OAAOnB,GAAGoB,OAAO,CAAC,kCAAkCC,GAAG;IAC7D,OAAO,IAAI7B,IAAI2B,KAAKG,GAAG,CAAC,CAACC,IAAMA,EAAE1B,IAAI;AACvC;AAEA,2FAA2F;AAC3F,wFAAwF;AACxF,SAAS2B,aAAaxB,EAAgB,EAAEyB,GAAW;IACjDzB,GAAG0B,IAAI,CAAC,CAAC,gGAAgG,CAAC;IAC1G1B,GAAG0B,IAAI,CAAC,CAAC,0HAA0H,CAAC;IACpI,6FAA6F;IAC7F,qFAAqF;IACrF1B,GAAG0B,IAAI,CAAC,CAAC,gGAAgG,CAAC;IAC1G1B,GAAG0B,IAAI,CAAC;IACR,KAAK,MAAMC,WAAWzC,eAAeuC,KAAME,QAAQC,MAAM,CAAC5B;IAC1D,IAAI6B,QAAQ7B,IAAI,sBAAsB,MAAM8B,QAAQ9B,IAAI,kBAAkBN;IAC1E,IAAImC,QAAQ7B,IAAI,gBAAgB,MAAM8B,QAAQ9B,IAAI,YAAYjB,iBAAiB0C;AACjF;AAEA,OAAO,SAASI,QAAQ7B,EAAgB,EAAE+B,GAAW;IACnD,MAAMC,MAAMhC,GAAGoB,OAAO,CAAC,wCAAwCa,GAAG,CAACF;IACnE,OAAOC,MAAMA,IAAI3B,KAAK,GAAG;AAC3B;AAEA,OAAO,SAASyB,QAAQ9B,EAAgB,EAAE+B,GAAW,EAAE1B,KAAoB;IACzE,IAAIA,UAAU,MAAM;QAClBL,GAAGoB,OAAO,CAAC,kCAAkCc,GAAG,CAACH;QACjD;IACF;IACA/B,GAAGoB,OAAO,CAAC,qGAAqGc,GAAG,CAACH,KAAK1B;AAC3H;AAEA,OAAO,SAAS8B,SAASnC,EAAgB;IACvC,MAAMgC,MAAMhC,GAAGoB,OAAO,CAAC,yCAAyCa,GAAG;IACnE,OAAOD,IAAII,CAAC;AACd;AAEA,OAAO,SAASC,UAAUrC,EAAgB,EAAEyB,GAAW,EAAEa,OAAe;IACtE,MAAMC,QAAQnD,UAAUqC,KAAKa;IAC7B,MAAME,aAAa,IAAIhD,IAAI+C,MAAMjB,GAAG,CAAC,CAACmB,IAAMA,EAAEC,OAAO;IAErD,MAAMC,eAAe3C,GAAGoB,OAAO,CAAC,CAAC,iDAAiD,CAAC,EAAEC,GAAG;IAKxF,MAAMuB,WAAW,IAAIC,IAAIF,aAAarB,GAAG,CAAC,CAACC,IAAM;YAACA,EAAEuB,IAAI;YAAEvB;SAAE;IAC5D,MAAMwB,WAAWJ,aAAaK,MAAM,CAAC,CAACzB,IAAM,CAACiB,WAAWS,GAAG,CAAC1B,EAAEuB,IAAI,GAAGxB,GAAG,CAAC,CAACC,IAAMA,EAAEuB,IAAI;IAEtF,MAAMI,YAAYX,MAAMS,MAAM,CAAC,CAACP;QAC9B,MAAMT,MAAMY,SAASX,GAAG,CAACQ,EAAEC,OAAO;QAClC,OAAO,CAACV,OAAOA,IAAImB,MAAM,KAAKV,EAAEW,OAAO,IAAIpB,IAAIqB,KAAK,KAAKZ,EAAEa,IAAI;IACjE;IAEA,IAAIP,SAASQ,MAAM,KAAK,KAAKL,UAAUK,MAAM,KAAK,GAAG,OAAO;QAAE7C,QAAQ;QAAG8C,UAAU,EAAE;IAAC;IAEtF,MAAMC,WAAWvE,eAAeuC;IAChC,MAAMiC,cAAcxC,WAAWlB;IAC/B,MAAM2D,aAAuB,EAAE;IAC/B,MAAMC,aAA0B,EAAE;IAClC,MAAMJ,WAAqB,EAAE;IAE7B,oFAAoF;IACpF,uDAAuD;IACvD,MAAMK,SAAS1E,SAAS,mBAAmB+D,UAAUK,MAAM;IAC3D,IAAIO,cAAc;IAClB,KAAK,MAAMC,QAAQb,UAAW;QAC5B,sFAAsF;QACtF,mEAAmE;QACnE,MAAMc,eAAeP,SAAST,MAAM,CAAC,CAACrB,UAAY,CAACA,QAAQsC,cAAc,IAAItC,QAAQsC,cAAc,CAACxC,KAAKsC;QACzG,MAAM,EAAEG,GAAG,EAAEV,UAAUW,YAAY,EAAE,GAAG9E,UAAU0E,MAAMC;QACxDH,OAAOO,IAAI,CAAC,EAAEN;QACdN,SAASa,IAAI,IAAIF;QACjB,KAAK,MAAMpC,OAAOuC,OAAOC,IAAI,CAACL,IAAIM,IAAI,EAAG;YACvC,IAAI,CAACd,YAAYT,GAAG,CAAClB,MAAM;gBACzB2B,YAAYe,GAAG,CAAC1C;gBAChB4B,WAAWU,IAAI,CAACtC;YAClB;QACF;QACA6B,WAAWS,IAAI,CAACH;IAClB;IACAL,OAAOa,MAAM;IAEb,MAAMC,aAAa;WAAIjB;KAAY;IACnC,uEAAuE;IACvE,sGAAsG;IACtG,IAAIiB,WAAWpB,MAAM,GAAG5D,yBAAyB;QAC/C,MAAM,IAAIV,WACR,gBACA,CAAC,uBAAuB,EAAE0F,WAAWpB,MAAM,CAAC,0EAA0E,EAAE5D,wBAAwB,wKAAwK,CAAC;IAE7T;IACA,0FAA0F;IAC1F,sEAAsE;IACtE,MAAMiF,kBAAkBD,WAAW3B,MAAM,CAAC,CAAC6B,IAAMtF,yBAAyB0D,GAAG,CAAC4B,MAAM,CAACvF,iBAAiB2D,GAAG,CAAC4B;IAC1G,sFAAsF;IACtF,gDAAgD;IAChD,MAAMC,YAAY,CAAC,yBAAyB,EAAEF,gBAAgBtD,GAAG,CAAC1B,YAAYf,IAAI,CAAC,MAAM,UAAU,EAAE+F,gBAAgBtD,GAAG,CAAC,IAAM,KAAKzC,IAAI,CAAC,MAAM,oCAAoC,EAAE+F,gBAClL5B,MAAM,CAAC,CAAC6B,IAAMA,MAAM,QACpBvD,GAAG,CAAC,CAACuD,IAAM,GAAGjF,WAAWiF,GAAG,YAAY,EAAEjF,WAAWiF,IAAI,EACzDhG,IAAI,CAAC,OAAO;IAEf,MAAMkG,QAAQ7B,UAAUF,MAAM,CAAC,CAACP,IAAM,CAACG,SAASK,GAAG,CAACR,EAAEC,OAAO,GAAGpB,GAAG,CAAC,CAACmB,IAAMA,EAAEC,OAAO;IACpF,MAAMsC,QAAwB;QAAEzC;QAAO0C,UAAUrB,WAAWtC,GAAG,CAAC,CAAC4D,IAAMA,EAAExC,OAAO;QAAGqC;QAAOhC;IAAS;IAEnG,MAAMoC,UAAUC,KAAKC,GAAG;IACxBrF,GAAG0B,IAAI,CAAC;IACR,IAAI;YA4C8BC;QA3ChC,KAAK,MAAM2D,OAAO3B,WAAY3D,GAAG0B,IAAI,CAAC,CAAC,mCAAmC,EAAE9B,WAAW0F,MAAM;QAC7F,6EAA6E;QAC7E,kFAAkF;QAClF,+EAA+E;QAC/E,MAAMC,UAAUvF,GAAGoB,OAAO,CAAC,CAAC,kFAAkF,CAAC;QAC/G,MAAMoE,iBAAiBxF,GAAGoB,OAAO,CAAC,CAAC,yCAAyC,CAAC;QAC7E,MAAMqE,mBAAmBzF,GAAGoB,OAAO,CAAC,CAAC,uDAAuD,CAAC;QAC7F,IAAI2B,SAASQ,MAAM,GAAG,GAAG;YACvB,MAAMmC,MAAM1F,GAAGoB,OAAO,CAAC,CAAC,wCAAwC,CAAC;YACjE,KAAK,MAAM0B,QAAQC,SAAU;oBAMKpB;gBALhC,kFAAkF;gBAClF,qEAAqE;gBACrE4D,QAAQrD,GAAG,CAACY;gBACZ4C,IAAIxD,GAAG,CAACY;gBACR0C,eAAetD,GAAG,CAACY;gBACnB,KAAK,MAAMnB,WAAW8B,UAAU9B,kBAAAA,QAAQgE,MAAM,cAAdhE,sCAAAA,qBAAAA,SAAiB3B,IAAI8C,MAAMkC;YAC7D;QACF;QACA,IAAIpB,WAAWL,MAAM,GAAG,GAAG;YACzB,MAAMqC,SAAS5F,GAAGoB,OAAO,CAAC0D;YAC1B,MAAMe,aAAa7F,GAAGoB,OAAO,CAAC,CAAC,+HAA+H,CAAC;YAC/J,KAAK,MAAM8C,OAAON,WAAY;oBAmBIjC;gBAlBhC,MAAMmE,SAASlB,gBAAgBtD,GAAG,CAAC,CAACgE;wBAI3BpB;oBAHP,IAAIoB,QAAQ,QAAQ,OAAOpB,IAAIxB,OAAO;oBACtC,IAAI4C,QAAQ,UAAU,OAAOpB,IAAId,OAAO;oBACxC,IAAIkC,QAAQ,SAAS,OAAOpB,IAAIZ,IAAI;oBACpC,QAAOY,gBAAAA,IAAIM,IAAI,CAACc,IAAI,cAAbpB,2BAAAA,gBAAiB;gBAC1B;gBACA,uFAAuF;gBACvF0B,OAAO1D,GAAG,IAAI4D;gBACd,oFAAoF;gBACpF,IAAIlD,SAASK,GAAG,CAACiB,IAAIxB,OAAO,GAAG;wBAEGf;oBADhC4D,QAAQrD,GAAG,CAACgC,IAAIxB,OAAO;oBACvB,KAAK,MAAMf,WAAW8B,UAAU9B,mBAAAA,QAAQgE,MAAM,cAAdhE,uCAAAA,sBAAAA,SAAiB3B,IAAIkE,IAAIxB,OAAO,EAAEsC;gBACpE;gBACAa,WAAW3D,GAAG,CAACgC,IAAIxB,OAAO,EAAEwB,IAAI6B,MAAM,CAACC,KAAK,EAAE9B,IAAI6B,MAAM,CAACE,OAAO,EAAE/B,IAAI6B,MAAM,CAACG,IAAI,EAAEhC,IAAIxB,OAAO;gBAC9F,iFAAiF;gBACjF,2EAA2E;gBAC3E,IAAIE,SAASK,GAAG,CAACiB,IAAIxB,OAAO,GAAG8C,eAAetD,GAAG,CAACgC,IAAIxB,OAAO;gBAC7D,KAAK,MAAMyD,cAAcjC,IAAIkC,OAAO,CAAEX,iBAAiBvD,GAAG,CAACgC,IAAIxB,OAAO,EAAEyD;gBACxE,KAAK,MAAMxE,WAAW8B,UAAU9B,iBAAAA,QAAQ0E,KAAK,cAAb1E,qCAAAA,oBAAAA,SAAgB3B,IAAIkE,IAAIxB,OAAO,EAAEwB,IAAIoC,SAAS,CAAC3E,QAAQ9B,IAAI,CAAC,EAAEmF;YAChG;QACF;QACA,KAAK,MAAMrD,WAAW8B,UAAU9B,0BAAAA,QAAQ4E,cAAc,cAAtB5E,8CAAAA,6BAAAA,SAAyB3B,IAAIgF;QAC7DhF,GAAG0B,IAAI,CAAC;IACV,EAAE,OAAO8E,KAAK;QACZxG,GAAG0B,IAAI,CAAC;QACR,MAAM8E;IACR;IAEA,qFAAqF;IACrF,8GAA8G;IAC9G,MAAMC,aAAarB,KAAKC,GAAG,KAAKF;IAChC,MAAMuB,UAAU7E,QAAQ7B,IAAI;IAC5B,yFAAyF;IACzF,+EAA+E;IAC/E,MAAM2G,UAAUD,YAAY,OAAO,CAAC,IAAIE,OAAOF;IAC/C,IAAID,aAAaE,SAAS7E,QAAQ9B,IAAI,oBAAoBQ,OAAOiG;IAEjE,OAAO;QAAE/F,QAAQkD,WAAWL,MAAM;QAAEC;IAAS;AAC/C;AAEA,yFAAyF;AACzF,wFAAwF;AACxF,SAASqD,cAAcC,MAAc,EAAEC,KAAa;IAClD,yFAAyF;IACzF,MAAMC,QAAQ,CAACC,OAAkBA,KAAKxG,UAAU,CAAC,aAAawG,KAAKnH,KAAK,CAAC,KAAKoH,KAAK,CAAC,GAAG,GAAGrI,IAAI,CAAC,OAAOoI,KAAKnH,KAAK,CAAC,IAAI,CAAC,EAAE;IACxH,MAAMc,QAAQ,CAACuG,MAAgB,IAAItE,IAAIsE,IAAIrH,KAAK,CAAC,KAAKwB,GAAG,CAAC,CAAC2F,OAAS;gBAACD,MAAMC;gBAAOA;aAAK;IACvF,MAAMG,IAAIxG,MAAMkG;IAChB,MAAMO,IAAIzG,MAAMmG;IAChB,MAAMO,UAAU,IAAI9H;IACpB,KAAK,MAAM,CAACuC,KAAKwF,IAAI,IAAIF,EAAG,IAAID,EAAEnF,GAAG,CAACF,SAASwF,KAAKD,QAAQ7C,GAAG,CAAC1C;IAChE,KAAK,MAAMA,OAAOqF,EAAE7C,IAAI,GAAI,IAAI,CAAC8C,EAAEpE,GAAG,CAAClB,MAAMuF,QAAQ7C,GAAG,CAAC1C;IACzD,MAAMyF,QAAQ,CAACzF,MAAiBA,QAAQ,UAAU,mBAAmBA,IAAItB,UAAU,CAAC,aAAa,CAAC,QAAQ,EAAEsB,IAAImF,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG;IAC9H,OAAOI,QAAQhE,IAAI,KAAK,IAAI,aAAa;WAAIgE;KAAQ,CAAChG,GAAG,CAACkG,OAAO3I,IAAI,CAAC;AACxE;AAEA,OAAO,SAAS4I,KAAKhG,GAAmB;QAqCTI;IApC7B,MAAM6F,WAAW7I,KAAK4C,IAAIa,OAAO,EAAEtD;IACnCL,UAAU+I,UAAU;QAAEC,WAAW;IAAK;IACtC,MAAMC,SAAS/I,KAAK6I,UAAUjI;IAE9B,MAAMO,KAAK,IAAIlB,aAAa8I;IAC5B5H,GAAG0B,IAAI,CAAC;IACR,yFAAyF;IACzF,uCAAuC;IACvC1B,GAAG0B,IAAI,CAAC;IACR3B,kBAAkBC;IAElBA,GAAG0B,IAAI,CAAC;IAER,uFAAuF;IACvF,qGAAqG;IACrG,MAAMmG,UAAUhG,QAAQ7B,IAAI;IAC5B,MAAMyD,WAAW5B,QAAQ7B,IAAI;IAC7B,MAAM8H,eAAe/I,iBAAiB0C;IACtC,IAAI,AAACoG,YAAY,QAAQA,YAAYnI,kBAAoB+D,aAAa,QAAQA,aAAaqE,cAAe;QACxG,uFAAuF;QACvF,wFAAwF;QACxF,IAAID,YAAY,QAAQA,YAAYnI,gBAAgB;YAClDqI,QAAQC,KAAK,CAAC;QAChB,OAAO;YACL,MAAMV,UAAUT,cAAcpD,qBAAAA,sBAAAA,WAAY,IAAIqE;YAC9CC,QAAQC,KAAK,CAAC,CAAC,sBAAsB,EAAEV,QAAQ,oBAAoB,CAAC;QACtE;QACAtH,GAAGiI,KAAK;QACRrJ,OAAO8I,UAAU;YAAEC,WAAW;YAAMO,OAAO;QAAK;QAChD,OAAOT,KAAKhG;IACd;IAEAD,aAAaxB,IAAIyB;IAEjB,wFAAwF;IACxF,2FAA2F;IAC3F,MAAM0G,gBAAgBvB,QAAO/E,WAAAA,QAAQ7B,IAAI,iCAAZ6B,sBAAAA,WAAmC;IAChE7B,GAAG0B,IAAI,CAAC,CAAC,sBAAsB,EAAE0G,KAAKC,GAAG,CAACD,KAAKE,GAAG,CAAC,OAAO,IAAIH,gBAAgB,SAAU;IAExF,MAAM,EAAEzH,MAAM,EAAE8C,QAAQ,EAAE,GAAGnB,UAAUrC,IAAIyB,KAAKA,IAAIa,OAAO;IAE3D,OAAO;QAAEtC;QAAIyB;QAAKmG;QAAQlH;QAAQ8C;IAAS;AAC7C;AAEA,oCAAoC;AACpC,OAAO,SAAS+E,QAAQ9G,GAAmB;IACzC7C,OAAOC,KAAK4C,IAAIa,OAAO,EAAEtD,YAAY;QAAE2I,WAAW;QAAMO,OAAO;IAAK;IACpE,OAAOT,KAAKhG;AACd"}
|
package/dist/esm/errors.d.ts
CHANGED
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
export type SenseErrorCode = 'CONFIG_NOT_FOUND' | 'CONFIG_EXISTS' | 'CONFIG_INVALID' | 'CONFIG_VERSION_UNSUPPORTED' | 'WATCH_ACTIVE' | 'NOTE_NOT_FOUND' | 'NOTE_AMBIGUOUS' | 'EMBED_DISABLED' | 'EMBED_MODEL' | 'SEARCH_SYNTAX' | 'COLUMN_LIMIT' | 'PRESET_UNKNOWN';
|
|
1
|
+
export type SenseErrorCode = 'CONFIG_NOT_FOUND' | 'CONFIG_EXISTS' | 'CONFIG_INVALID' | 'CONFIG_VERSION_UNSUPPORTED' | 'WATCH_ACTIVE' | 'NOTE_NOT_FOUND' | 'NOTE_AMBIGUOUS' | 'EMBED_DISABLED' | 'EMBED_MODEL' | 'EMBED_MODEL_MISSING' | 'SEARCH_SYNTAX' | 'COLUMN_LIMIT' | 'PRESET_UNKNOWN';
|
|
2
2
|
export declare class SenseError extends Error {
|
|
3
3
|
code: SenseErrorCode;
|
|
4
4
|
constructor(code: SenseErrorCode, message: string);
|
package/dist/esm/errors.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/errors.ts"],"sourcesContent":["// Library code throws; only cli.ts prints and exits.\nexport type SenseErrorCode = 'CONFIG_NOT_FOUND' | 'CONFIG_EXISTS' | 'CONFIG_INVALID' | 'CONFIG_VERSION_UNSUPPORTED' | 'WATCH_ACTIVE' | 'NOTE_NOT_FOUND' | 'NOTE_AMBIGUOUS' | 'EMBED_DISABLED' | 'EMBED_MODEL' | 'SEARCH_SYNTAX' | 'COLUMN_LIMIT' | 'PRESET_UNKNOWN';\n\nexport class SenseError extends Error {\n code: SenseErrorCode;\n\n constructor(code: SenseErrorCode, message: string) {\n super(message);\n this.name = 'SenseError';\n this.code = code;\n }\n}\n"],"names":["SenseError","Error","code","message","name"],"mappings":"AAAA,qDAAqD;AAGrD,OAAO,MAAMA,mBAAmBC;IAG9B,YAAYC,IAAoB,EAAEC,OAAe,CAAE;QACjD,KAAK,CAACA;QACN,IAAI,CAACC,IAAI,GAAG;QACZ,IAAI,CAACF,IAAI,GAAGA;IACd;AACF"}
|
|
1
|
+
{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/errors.ts"],"sourcesContent":["// Library code throws; only cli.ts prints and exits.\nexport type SenseErrorCode = 'CONFIG_NOT_FOUND' | 'CONFIG_EXISTS' | 'CONFIG_INVALID' | 'CONFIG_VERSION_UNSUPPORTED' | 'WATCH_ACTIVE' | 'NOTE_NOT_FOUND' | 'NOTE_AMBIGUOUS' | 'EMBED_DISABLED' | 'EMBED_MODEL' | 'EMBED_MODEL_MISSING' | 'SEARCH_SYNTAX' | 'COLUMN_LIMIT' | 'PRESET_UNKNOWN';\n\nexport class SenseError extends Error {\n code: SenseErrorCode;\n\n constructor(code: SenseErrorCode, message: string) {\n super(message);\n this.name = 'SenseError';\n this.code = code;\n }\n}\n"],"names":["SenseError","Error","code","message","name"],"mappings":"AAAA,qDAAqD;AAGrD,OAAO,MAAMA,mBAAmBC;IAG9B,YAAYC,IAAoB,EAAEC,OAAe,CAAE;QACjD,KAAK,CAACA;QACN,IAAI,CAACC,IAAI,GAAG;QACZ,IAAI,CAACF,IAAI,GAAGA;IACd;AACF"}
|
|
@@ -7,9 +7,23 @@ export interface EmbedProvider {
|
|
|
7
7
|
embed(texts: string[]): Promise<Float32Array[]>;
|
|
8
8
|
}
|
|
9
9
|
export declare const embed: Feature;
|
|
10
|
+
export declare const MODEL_FILENAMES: string;
|
|
11
|
+
export declare function isDownloadable(model: string): boolean;
|
|
12
|
+
export declare function modelDir(model: string): string;
|
|
13
|
+
export declare function hasModelFiles(model: string): boolean;
|
|
14
|
+
export declare function modelPresent(cfg: Config): boolean;
|
|
15
|
+
export declare function downloadModel(model: string, onFile?: (file: string, dir: string) => void): Promise<string>;
|
|
10
16
|
export declare function embedPending(db: DatabaseSync, cfg: Config, baseDir: string): Promise<void>;
|
|
11
17
|
export declare function semanticCandidates(db: DatabaseSync, cfg: Config, terms: string, fetch: number): Promise<Array<{
|
|
12
18
|
path: string;
|
|
13
19
|
lines: string;
|
|
14
20
|
similarity: number;
|
|
15
21
|
}>>;
|
|
22
|
+
export declare function similarNotes(db: DatabaseSync, _cfg: Config, path: string, opts: {
|
|
23
|
+
exclude: Set<string>;
|
|
24
|
+
allowed?: Set<string>;
|
|
25
|
+
k: number;
|
|
26
|
+
}): Array<{
|
|
27
|
+
path: string;
|
|
28
|
+
similarity: number;
|
|
29
|
+
}>;
|
|
@@ -5,18 +5,16 @@ import { embedConfig } from '../config.js';
|
|
|
5
5
|
import { SenseError } from '../errors.js';
|
|
6
6
|
import { progress } from '../progress.js';
|
|
7
7
|
import { parseFile } from '../scan.js';
|
|
8
|
-
//
|
|
9
|
-
//
|
|
10
|
-
// Reconcile stores dirty rows inside its transaction; embedding tops up on the next
|
|
11
|
-
// semantic query (embedPending), so staleness costs recall, never correctness.
|
|
8
|
+
// Heading-based chunks, int8 vectors with a per-vector scale, NULL vector = not yet embedded.
|
|
9
|
+
// Reconcile writes dirty rows; embedding tops up on the next search, so staleness costs recall.
|
|
12
10
|
// Storage lever fixed by the bake-off (BENCHMARKING.md): int8 at 256 dims is
|
|
13
11
|
// quality-free vs f32-512 when fused. Queries stay f32 at the same dims.
|
|
14
12
|
const STORE_DIMS = 256;
|
|
13
|
+
// Seed chunks that participate in a `related` scan; see similarNotes for the measurement.
|
|
14
|
+
const TARGET_CHUNK_CAP = 16;
|
|
15
15
|
const BATCH = 64;
|
|
16
|
-
// Deterministic
|
|
17
|
-
//
|
|
18
|
-
// preamble kept; whole file when no headings. The title/summary prefix mirrors the
|
|
19
|
-
// bm25 column weighting.
|
|
16
|
+
// Deterministic, so embed time can re-derive text from stored line ranges. Heading-delimited,
|
|
17
|
+
// preamble kept, whole file when no headings; the title/summary prefix mirrors bm25 weighting.
|
|
20
18
|
function chunksOf(raw, search) {
|
|
21
19
|
const lines = raw.split('\n');
|
|
22
20
|
const starts = [];
|
|
@@ -61,8 +59,8 @@ export const embed = {
|
|
|
61
59
|
remove (db, path) {
|
|
62
60
|
db.prepare('DELETE FROM embeddings WHERE "path" = ?').run(path);
|
|
63
61
|
},
|
|
64
|
-
// A
|
|
65
|
-
//
|
|
62
|
+
// A tree with no embedding model never had extract() run for the doc (db.ts's per-file
|
|
63
|
+
// filter skips it), so extracted is undefined here -- store nothing, i.e. no rows.
|
|
66
64
|
store (db, path, extracted) {
|
|
67
65
|
if (!extracted) return;
|
|
68
66
|
const insert = db.prepare('INSERT INTO embeddings ("path", chunk, start_line, end_line, scale, vector) VALUES (?, ?, ?, ?, NULL, NULL)');
|
|
@@ -75,6 +73,45 @@ export const embed = {
|
|
|
75
73
|
// --- static type: Model2Vec safetensors + pure-JS tokenizer, model cached in ~/.cache ---
|
|
76
74
|
// Encode convention from model2vec/model.py: no special tokens, drop unk ids, mean-pool,
|
|
77
75
|
// L2-normalize.
|
|
76
|
+
const MODEL_FILES = [
|
|
77
|
+
'model.safetensors',
|
|
78
|
+
'tokenizer.json'
|
|
79
|
+
];
|
|
80
|
+
// The pair, for messages that name what a local model directory must contain.
|
|
81
|
+
export const MODEL_FILENAMES = MODEL_FILES.join(' and ');
|
|
82
|
+
// A Hugging Face repo id: one slash, HF's charset. Anything else is a path, used as one rather
|
|
83
|
+
// than flattened into a cache key Windows would reject.
|
|
84
|
+
const HF_MODEL_ID = /^[A-Za-z0-9][\w.-]*\/[A-Za-z0-9][\w.-]*$/;
|
|
85
|
+
// Downloadable iff it names a Hugging Face repo; a local path is the caller's to populate.
|
|
86
|
+
export function isDownloadable(model) {
|
|
87
|
+
return HF_MODEL_ID.test(model);
|
|
88
|
+
}
|
|
89
|
+
// Machine-wide, not per-tree: `.sense/` is the index a rebuild throws away, while a 124 MB
|
|
90
|
+
// model is shared by every tree on the machine. Honors XDG_CACHE_HOME where it is set;
|
|
91
|
+
// elsewhere ~/.cache, which is also where Hugging Face's own libraries keep theirs.
|
|
92
|
+
function cacheRoot() {
|
|
93
|
+
const xdg = process.env.XDG_CACHE_HOME;
|
|
94
|
+
return join(xdg && xdg.length > 0 ? xdg : join(homedir(), '.cache'), 'sensemaking', 'models');
|
|
95
|
+
}
|
|
96
|
+
// A model is either a local directory the caller pointed at, or the cache `sense download`
|
|
97
|
+
// fills. Nothing here fetches: an absent model degrades search to its other signals rather
|
|
98
|
+
// than pulling 124 MB out of a command that reads like a query.
|
|
99
|
+
export function modelDir(model) {
|
|
100
|
+
if (!HF_MODEL_ID.test(model)) return model;
|
|
101
|
+
return join(cacheRoot(), model.replace(/\//g, '--'));
|
|
102
|
+
}
|
|
103
|
+
// Both files, so an interrupted download reads as absent and the next one resumes it.
|
|
104
|
+
export function hasModelFiles(model) {
|
|
105
|
+
const dir = modelDir(model);
|
|
106
|
+
return MODEL_FILES.every((file)=>existsSync(join(dir, file)));
|
|
107
|
+
}
|
|
108
|
+
// api providers have nothing to download, so they are never "missing" here; an unreachable
|
|
109
|
+
// endpoint surfaces as an EMBED_MODEL error at call time instead.
|
|
110
|
+
export function modelPresent(cfg) {
|
|
111
|
+
const e = embedConfig(cfg);
|
|
112
|
+
if (!e) return false;
|
|
113
|
+
return e.type === 'api' || hasModelFiles(e.model);
|
|
114
|
+
}
|
|
78
115
|
function fetchToFile(url, dest) {
|
|
79
116
|
return fetch(url).then(async (res)=>{
|
|
80
117
|
if (!res.ok) throw new SenseError('EMBED_MODEL', `model download failed: ${url} -> HTTP ${res.status}`);
|
|
@@ -82,23 +119,33 @@ function fetchToFile(url, dest) {
|
|
|
82
119
|
renameSync(`${dest}.part`, dest);
|
|
83
120
|
});
|
|
84
121
|
}
|
|
122
|
+
// The only code path that touches the network for weights. `sense download` calls it; no
|
|
123
|
+
// query ever does. Idempotent: a file already on disk is left alone.
|
|
124
|
+
export async function downloadModel(model, onFile) {
|
|
125
|
+
if (!isDownloadable(model)) {
|
|
126
|
+
throw new SenseError('EMBED_MODEL', `embed.model "${model}" is a local path, not a Hugging Face model id, so there is nothing to download; put model.safetensors and tokenizer.json in that directory, or name a model id like "minishlab/potion-retrieval-32M"`);
|
|
127
|
+
}
|
|
128
|
+
const dir = modelDir(model);
|
|
129
|
+
mkdirSync(dir, {
|
|
130
|
+
recursive: true
|
|
131
|
+
});
|
|
132
|
+
for (const file of MODEL_FILES){
|
|
133
|
+
if (existsSync(join(dir, file))) continue;
|
|
134
|
+
onFile === null || onFile === void 0 ? void 0 : onFile(file, dir);
|
|
135
|
+
await fetchToFile(`https://huggingface.co/${model}/resolve/main/${file}`, join(dir, file));
|
|
136
|
+
}
|
|
137
|
+
return dir;
|
|
138
|
+
}
|
|
85
139
|
async function staticProvider(model) {
|
|
86
140
|
var _ref;
|
|
87
141
|
var _tokenizerJson_model, _tokenizerJson_model_vocab, _tokenizerJson_model1;
|
|
88
|
-
|
|
89
|
-
|
|
90
|
-
|
|
91
|
-
|
|
92
|
-
|
|
93
|
-
|
|
94
|
-
|
|
95
|
-
'model.safetensors',
|
|
96
|
-
'tokenizer.json'
|
|
97
|
-
]){
|
|
98
|
-
if (!existsSync(join(dir, file))) {
|
|
99
|
-
console.error(`fetching ${model}/${file} into ${dir} (once; delete to refetch)`);
|
|
100
|
-
await fetchToFile(`https://huggingface.co/${model}/resolve/main/${file}`, join(dir, file));
|
|
101
|
-
}
|
|
142
|
+
const dir = modelDir(model);
|
|
143
|
+
for (const file of MODEL_FILES){
|
|
144
|
+
if (!existsSync(join(dir, file))) {
|
|
145
|
+
// A local path the caller controls gets told what is missing where; only a repo id can
|
|
146
|
+
// be fixed by downloading.
|
|
147
|
+
const fix = isDownloadable(model) ? 'run `sense download`' : `expected ${MODEL_FILENAMES} in that directory`;
|
|
148
|
+
throw new SenseError('EMBED_MODEL_MISSING', `embed model ${model} is not available (looked in ${dir}); ${fix}`);
|
|
102
149
|
}
|
|
103
150
|
}
|
|
104
151
|
const raw = readFileSync(join(dir, 'model.safetensors'));
|
|
@@ -178,7 +225,7 @@ const providers = new Map();
|
|
|
178
225
|
function getProvider(cfg) {
|
|
179
226
|
var _e_url;
|
|
180
227
|
const e = embedConfig(cfg);
|
|
181
|
-
if (!e) throw new SenseError('EMBED_DISABLED', '
|
|
228
|
+
if (!e) throw new SenseError('EMBED_DISABLED', 'this tree has no embedding model: add an "embed" block naming one to sense.config.json, then run `sense download`');
|
|
182
229
|
const sig = `${e.type}:${e.model}:${(_e_url = e.url) !== null && _e_url !== void 0 ? _e_url : ''}`;
|
|
183
230
|
let p = providers.get(sig);
|
|
184
231
|
if (!p) {
|
|
@@ -224,7 +271,7 @@ export async function embedPending(db, cfg, baseDir) {
|
|
|
224
271
|
let chunks;
|
|
225
272
|
try {
|
|
226
273
|
// presets/embed are irrelevant here -- re-deriving chunk text for a doc that already
|
|
227
|
-
// has embeddings rows means
|
|
274
|
+
// has embeddings rows means the tree had an embedding model at reconcile time.
|
|
228
275
|
chunks = parseFile({
|
|
229
276
|
relPath: path,
|
|
230
277
|
absPath: join(baseDir, path),
|
|
@@ -245,7 +292,7 @@ export async function embedPending(db, cfg, baseDir) {
|
|
|
245
292
|
});
|
|
246
293
|
}
|
|
247
294
|
const update = db.prepare('UPDATE embeddings SET scale = ?, vector = ? WHERE "path" = ? AND chunk = ?');
|
|
248
|
-
// The lazy build is the one long silence a first
|
|
295
|
+
// The lazy build is the one long silence a first search hits (measured 23s at
|
|
249
296
|
// 26k notes); progress makes it distinguishable from a hang.
|
|
250
297
|
const report = progress('embedding chunks', jobs.length);
|
|
251
298
|
for(let i = 0; i < jobs.length; i += BATCH){
|
|
@@ -268,14 +315,9 @@ export async function embedPending(db, cfg, baseDir) {
|
|
|
268
315
|
}
|
|
269
316
|
report.finish();
|
|
270
317
|
}
|
|
271
|
-
//
|
|
272
|
-
//
|
|
273
|
-
//
|
|
274
|
-
// Returns the cosine similarity alongside each candidate: stored and query vectors are both
|
|
275
|
-
// L2-normalised before quantisation, so the dot product is cosine in [-1, 1]. `find` surfaces
|
|
276
|
-
// it because the fused rank score cannot express match quality -- nearest-neighbour search
|
|
277
|
-
// always returns a nearest neighbour, so without a magnitude an agent cannot tell a real hit
|
|
278
|
-
// from the best of a bad lot.
|
|
318
|
+
// Best chunk per file by cosine, its line range riding along; FTS5 operators are stripped as
|
|
319
|
+
// lexical syntax. Similarity comes back because the fused score cannot express match quality,
|
|
320
|
+
// and nearest-neighbour search always returns a neighbour however far away.
|
|
279
321
|
export async function semanticCandidates(db, cfg, terms, fetch1) {
|
|
280
322
|
var _terms_match;
|
|
281
323
|
const baseDir = cfg.baseDir;
|
|
@@ -313,3 +355,36 @@ export async function semanticCandidates(db, cfg, terms, fetch1) {
|
|
|
313
355
|
similarity: Math.round(b.score * 1000) / 1000
|
|
314
356
|
}));
|
|
315
357
|
}
|
|
358
|
+
// Note-to-note similarity is the max cosine over (target chunk, other chunk) pairs, one linear
|
|
359
|
+
// scan of stored vectors. Reads only what is stored, so it stays sync.
|
|
360
|
+
export function similarNotes(db, _cfg, path, opts) {
|
|
361
|
+
// Cost is target_chunks x stored_chunks, so a heading-dense seed multiplies a full-corpus
|
|
362
|
+
// scan (12.7s at 201 chunks/note). Sample evenly, so late sections still get a vote.
|
|
363
|
+
const targetRows = db.prepare('SELECT scale, vector FROM embeddings WHERE "path" = ? AND vector IS NOT NULL ORDER BY chunk').all(path);
|
|
364
|
+
if (targetRows.length === 0) return [];
|
|
365
|
+
const step = Math.max(1, Math.ceil(targetRows.length / TARGET_CHUNK_CAP));
|
|
366
|
+
const target = targetRows.filter((_, i)=>i % step === 0).map((row)=>({
|
|
367
|
+
v: new Int8Array(row.vector.buffer, row.vector.byteOffset, row.vector.byteLength),
|
|
368
|
+
scale: row.scale
|
|
369
|
+
}));
|
|
370
|
+
const rows = db.prepare('SELECT "path", scale, vector FROM embeddings WHERE vector IS NOT NULL').all();
|
|
371
|
+
const best = new Map();
|
|
372
|
+
for (const row of rows){
|
|
373
|
+
if (row.path === path || opts.exclude.has(row.path) || opts.allowed && !opts.allowed.has(row.path)) continue;
|
|
374
|
+
const other = new Int8Array(row.vector.buffer, row.vector.byteOffset, row.vector.byteLength);
|
|
375
|
+
for (const t of target){
|
|
376
|
+
let dot = 0;
|
|
377
|
+
const len = Math.min(t.v.length, other.length);
|
|
378
|
+
for(let d = 0; d < len; d++)dot += t.v[d] * other[d];
|
|
379
|
+
const score = dot * t.scale * row.scale;
|
|
380
|
+
const existing = best.get(row.path);
|
|
381
|
+
if (existing === undefined || score > existing) best.set(row.path, score);
|
|
382
|
+
}
|
|
383
|
+
}
|
|
384
|
+
return [
|
|
385
|
+
...best.entries()
|
|
386
|
+
].sort((a, b)=>b[1] - a[1]).slice(0, opts.k).map(([p, score])=>({
|
|
387
|
+
path: p,
|
|
388
|
+
similarity: Math.round(score * 1000) / 1000
|
|
389
|
+
}));
|
|
390
|
+
}
|