sensemaking 0.27.0 → 0.27.1
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/dist/cjs/chunk/extract.js.map +1 -1
- package/dist/cjs/chunk/group.js.map +1 -1
- package/dist/cjs/chunk/index.js.map +1 -1
- package/dist/cjs/chunk/parse.js.map +1 -1
- package/dist/cjs/chunk/parser.js.map +1 -1
- package/dist/cjs/chunk/tokens.js.map +1 -1
- package/dist/cjs/chunk/version.js.map +1 -1
- package/dist/cjs/cli/build.js.map +1 -1
- package/dist/cjs/cli/download.js.map +1 -1
- package/dist/cjs/cli/index.js.map +1 -1
- package/dist/cjs/cli/init.js.map +1 -1
- package/dist/cjs/cli/map.js.map +1 -1
- package/dist/cjs/cli/named.js.map +1 -1
- package/dist/cjs/cli/path.js.map +1 -1
- package/dist/cjs/cli/peek.js.map +1 -1
- package/dist/cjs/cli/related.js.map +1 -1
- package/dist/cjs/cli/search.js.map +1 -1
- package/dist/cjs/cli/shared.js.map +1 -1
- package/dist/cjs/cli/sql.js.map +1 -1
- package/dist/cjs/cli/status.js.map +1 -1
- package/dist/cjs/cli/watch.js.map +1 -1
- package/dist/cjs/cli.js.map +1 -1
- package/dist/cjs/commands/index.js.map +1 -1
- package/dist/cjs/commands/map.js.map +1 -1
- package/dist/cjs/commands/path.js.map +1 -1
- package/dist/cjs/commands/peek.js.map +1 -1
- package/dist/cjs/commands/related.js.map +1 -1
- package/dist/cjs/commands/scope.js.map +1 -1
- package/dist/cjs/commands/search.js.map +1 -1
- package/dist/cjs/commands/signals.js.map +1 -1
- package/dist/cjs/commands/snapshot.js.map +1 -1
- package/dist/cjs/commands/status.js.map +1 -1
- package/dist/cjs/config/access.js.map +1 -1
- package/dist/cjs/config/index.js.map +1 -1
- package/dist/cjs/config/load.js.map +1 -1
- package/dist/cjs/config/resolve.js.map +1 -1
- package/dist/cjs/config/signals.js.map +1 -1
- package/dist/cjs/config/types.js.map +1 -1
- package/dist/cjs/config/validate.js.map +1 -1
- package/dist/cjs/embed/cohere.js.map +1 -1
- package/dist/cjs/embed/distribution.js.map +1 -1
- package/dist/cjs/embed/endpoint.js.map +1 -1
- package/dist/cjs/embed/handoff.js.map +1 -1
- package/dist/cjs/embed/http.js.map +1 -1
- package/dist/cjs/embed/identity.js.map +1 -1
- package/dist/cjs/embed/langfit.js.map +1 -1
- package/dist/cjs/embed/languages.js.map +1 -1
- package/dist/cjs/embed/openai.js.map +1 -1
- package/dist/cjs/embed/query.js.map +1 -1
- package/dist/cjs/embed/registry.js.map +1 -1
- package/dist/cjs/embed/static.js.map +1 -1
- package/dist/cjs/embed/store.js.map +1 -1
- package/dist/cjs/embed/types.js.map +1 -1
- package/dist/cjs/errors.js.map +1 -1
- package/dist/cjs/features/embed.js.map +1 -1
- package/dist/cjs/features/fences.js.map +1 -1
- package/dist/cjs/features/index.js.map +1 -1
- package/dist/cjs/features/links.js.map +1 -1
- package/dist/cjs/features/rank.js.map +1 -1
- package/dist/cjs/features/sections.js.map +1 -1
- package/dist/cjs/features/tags.js.map +1 -1
- package/dist/cjs/graph/graph.js.map +1 -1
- package/dist/cjs/graph/traverse.js.map +1 -1
- package/dist/cjs/index.js.map +1 -1
- package/dist/cjs/lib/atomic-write.js.map +1 -1
- package/dist/cjs/lib/guarded-tick.js.map +1 -1
- package/dist/cjs/lib/serial-query.js.map +1 -1
- package/dist/cjs/lib/watch-notification.js.map +1 -1
- package/dist/cjs/lib/worker-file.js.map +1 -1
- package/dist/cjs/output/column-hint.js.map +1 -1
- package/dist/cjs/output/output.js.map +1 -1
- package/dist/cjs/output/progress.js.map +1 -1
- package/dist/cjs/output/search-error.js.map +1 -1
- package/dist/cjs/scan/frontmatter.js.map +1 -1
- package/dist/cjs/scan/index.js.map +1 -1
- package/dist/cjs/scan/list.js.map +1 -1
- package/dist/cjs/scan/pool.js.map +1 -1
- package/dist/cjs/scan/reparse.js.map +1 -1
- package/dist/cjs/scan/worker-error.js.map +1 -1
- package/dist/cjs/store/batch.js.map +1 -1
- package/dist/cjs/store/builder.js.map +1 -1
- package/dist/cjs/store/cache.js.map +1 -1
- package/dist/cjs/store/duckdb/batch.js.map +1 -1
- package/dist/cjs/store/duckdb/connection.js.map +1 -1
- package/dist/cjs/store/duckdb/fieldStats.js.map +1 -1
- package/dist/cjs/store/duckdb/lexical.d.cts +1 -6
- package/dist/cjs/store/duckdb/lexical.d.ts +1 -6
- package/dist/cjs/store/duckdb/lexical.js +28 -39
- package/dist/cjs/store/duckdb/lexical.js.map +1 -1
- package/dist/cjs/store/duckdb/native.d.cts +1 -2
- package/dist/cjs/store/duckdb/native.d.ts +1 -2
- package/dist/cjs/store/duckdb/native.js +1 -6
- package/dist/cjs/store/duckdb/native.js.map +1 -1
- package/dist/cjs/store/duckdb/open.js.map +1 -1
- package/dist/cjs/store/duckdb/ordered-bm25.d.cts +3 -16
- package/dist/cjs/store/duckdb/ordered-bm25.d.ts +3 -16
- package/dist/cjs/store/duckdb/ordered-bm25.js +27 -79
- package/dist/cjs/store/duckdb/ordered-bm25.js.map +1 -1
- package/dist/cjs/store/duckdb/reconcile.js.map +1 -1
- package/dist/cjs/store/duckdb/sql-functions.js.map +1 -1
- package/dist/cjs/store/duckdb/store.js.map +1 -1
- package/dist/cjs/store/duckdb/vectors.js.map +1 -1
- package/dist/cjs/store/embed-scope.js.map +1 -1
- package/dist/cjs/store/feature-scope.js.map +1 -1
- package/dist/cjs/store/index.js.map +1 -1
- package/dist/cjs/store/lock-wait.js.map +1 -1
- package/dist/cjs/store/native.js.map +1 -1
- package/dist/cjs/store/open.js.map +1 -1
- package/dist/cjs/store/preset-scope.js.map +1 -1
- package/dist/cjs/store/readiness.js.map +1 -1
- package/dist/cjs/store/reconcile.js.map +1 -1
- package/dist/cjs/store/shared.js.map +1 -1
- package/dist/cjs/store/signature.js.map +1 -1
- package/dist/cjs/store/sql-functions.js.map +1 -1
- package/dist/cjs/store/sqlite/connection.js.map +1 -1
- package/dist/cjs/store/sqlite/fieldStats.js.map +1 -1
- package/dist/cjs/store/sqlite/lexical.js.map +1 -1
- package/dist/cjs/store/sqlite/open.js.map +1 -1
- package/dist/cjs/store/sqlite/reconcile.js.map +1 -1
- package/dist/cjs/store/sqlite/sql-functions.js.map +1 -1
- package/dist/cjs/store/sqlite/store.js.map +1 -1
- package/dist/cjs/store/sqlite/vectors.js.map +1 -1
- package/dist/cjs/store/stages.js.map +1 -1
- package/dist/cjs/store/transaction.js.map +1 -1
- package/dist/cjs/store/turso/connection.js.map +1 -1
- package/dist/cjs/store/turso/fieldStats.js.map +1 -1
- package/dist/cjs/store/turso/lexical-text.js.map +1 -1
- package/dist/cjs/store/turso/lexical.js.map +1 -1
- package/dist/cjs/store/turso/native.js.map +1 -1
- package/dist/cjs/store/turso/open.js.map +1 -1
- package/dist/cjs/store/turso/reconcile.js.map +1 -1
- package/dist/cjs/store/turso/sql-functions.js.map +1 -1
- package/dist/cjs/store/turso/store.js.map +1 -1
- package/dist/cjs/store/turso/vectors.js.map +1 -1
- package/dist/cjs/store/types.js.map +1 -1
- package/dist/cjs/store/vectors.js.map +1 -1
- package/dist/cjs/text/segment.js.map +1 -1
- package/dist/cjs/text/strip.js.map +1 -1
- package/dist/cjs/watch-claim.js.map +1 -1
- package/dist/cjs/watch.js.map +1 -1
- package/dist/cjs/workers/parse.js.map +1 -1
- package/dist/cjs/workers/watch-heartbeat.js.map +1 -1
- package/dist/esm/chunk/extract.js.map +1 -1
- package/dist/esm/chunk/group.js.map +1 -1
- package/dist/esm/chunk/index.js.map +1 -1
- package/dist/esm/chunk/parse.js.map +1 -1
- package/dist/esm/chunk/parser.js.map +1 -1
- package/dist/esm/chunk/tokens.js.map +1 -1
- package/dist/esm/chunk/types.js.map +1 -1
- package/dist/esm/chunk/version.js.map +1 -1
- package/dist/esm/cli/build.js.map +1 -1
- package/dist/esm/cli/download.js.map +1 -1
- package/dist/esm/cli/index.js.map +1 -1
- package/dist/esm/cli/init.js.map +1 -1
- package/dist/esm/cli/map.js.map +1 -1
- package/dist/esm/cli/named.js.map +1 -1
- package/dist/esm/cli/path.js.map +1 -1
- package/dist/esm/cli/peek.js.map +1 -1
- package/dist/esm/cli/related.js.map +1 -1
- package/dist/esm/cli/search.js.map +1 -1
- package/dist/esm/cli/shared.js.map +1 -1
- package/dist/esm/cli/sql.js.map +1 -1
- package/dist/esm/cli/status.js.map +1 -1
- package/dist/esm/cli/types.js.map +1 -1
- package/dist/esm/cli/watch.js.map +1 -1
- package/dist/esm/cli.js.map +1 -1
- package/dist/esm/commands/index.js.map +1 -1
- package/dist/esm/commands/map.js.map +1 -1
- package/dist/esm/commands/path.js.map +1 -1
- package/dist/esm/commands/peek.js.map +1 -1
- package/dist/esm/commands/related.js.map +1 -1
- package/dist/esm/commands/scope.js.map +1 -1
- package/dist/esm/commands/search.js.map +1 -1
- package/dist/esm/commands/signals.js.map +1 -1
- package/dist/esm/commands/snapshot.js.map +1 -1
- package/dist/esm/commands/status.js.map +1 -1
- package/dist/esm/config/access.js.map +1 -1
- package/dist/esm/config/index.js.map +1 -1
- package/dist/esm/config/load.js.map +1 -1
- package/dist/esm/config/resolve.js.map +1 -1
- package/dist/esm/config/signals.js.map +1 -1
- package/dist/esm/config/types.js.map +1 -1
- package/dist/esm/config/validate.js.map +1 -1
- package/dist/esm/embed/cohere.js.map +1 -1
- package/dist/esm/embed/distribution.js.map +1 -1
- package/dist/esm/embed/endpoint.js.map +1 -1
- package/dist/esm/embed/handoff.js.map +1 -1
- package/dist/esm/embed/http.js.map +1 -1
- package/dist/esm/embed/identity.js.map +1 -1
- package/dist/esm/embed/langfit.js.map +1 -1
- package/dist/esm/embed/languages.js.map +1 -1
- package/dist/esm/embed/openai.js.map +1 -1
- package/dist/esm/embed/query.js.map +1 -1
- package/dist/esm/embed/registry.js.map +1 -1
- package/dist/esm/embed/static.js.map +1 -1
- package/dist/esm/embed/store.js.map +1 -1
- package/dist/esm/embed/types.js.map +1 -1
- package/dist/esm/errors.js.map +1 -1
- package/dist/esm/features/embed.js.map +1 -1
- package/dist/esm/features/fences.js.map +1 -1
- package/dist/esm/features/index.js.map +1 -1
- package/dist/esm/features/links.js.map +1 -1
- package/dist/esm/features/rank.js.map +1 -1
- package/dist/esm/features/sections.js.map +1 -1
- package/dist/esm/features/tags.js.map +1 -1
- package/dist/esm/features/types.js.map +1 -1
- package/dist/esm/graph/graph.js.map +1 -1
- package/dist/esm/graph/traverse.js.map +1 -1
- package/dist/esm/index.js.map +1 -1
- package/dist/esm/lib/atomic-write.js.map +1 -1
- package/dist/esm/lib/guarded-tick.js.map +1 -1
- package/dist/esm/lib/serial-query.js.map +1 -1
- package/dist/esm/lib/watch-notification.js.map +1 -1
- package/dist/esm/lib/worker-file.js.map +1 -1
- package/dist/esm/output/column-hint.js.map +1 -1
- package/dist/esm/output/output.js.map +1 -1
- package/dist/esm/output/progress.js.map +1 -1
- package/dist/esm/output/search-error.js.map +1 -1
- package/dist/esm/scan/frontmatter.js.map +1 -1
- package/dist/esm/scan/index.js.map +1 -1
- package/dist/esm/scan/list.js.map +1 -1
- package/dist/esm/scan/pool.js.map +1 -1
- package/dist/esm/scan/reparse.js.map +1 -1
- package/dist/esm/scan/worker-error.js.map +1 -1
- package/dist/esm/store/batch.js.map +1 -1
- package/dist/esm/store/builder.js.map +1 -1
- package/dist/esm/store/cache.js.map +1 -1
- package/dist/esm/store/duckdb/batch.js.map +1 -1
- package/dist/esm/store/duckdb/connection.js.map +1 -1
- package/dist/esm/store/duckdb/fieldStats.js.map +1 -1
- package/dist/esm/store/duckdb/lexical.d.ts +1 -6
- package/dist/esm/store/duckdb/lexical.js +13 -17
- package/dist/esm/store/duckdb/lexical.js.map +1 -1
- package/dist/esm/store/duckdb/native.d.ts +1 -2
- package/dist/esm/store/duckdb/native.js +1 -3
- package/dist/esm/store/duckdb/native.js.map +1 -1
- package/dist/esm/store/duckdb/open.js.map +1 -1
- package/dist/esm/store/duckdb/ordered-bm25.d.ts +3 -16
- package/dist/esm/store/duckdb/ordered-bm25.js +18 -54
- package/dist/esm/store/duckdb/ordered-bm25.js.map +1 -1
- package/dist/esm/store/duckdb/reconcile.js.map +1 -1
- package/dist/esm/store/duckdb/sql-functions.js.map +1 -1
- package/dist/esm/store/duckdb/store.js.map +1 -1
- package/dist/esm/store/duckdb/vectors.js.map +1 -1
- package/dist/esm/store/embed-scope.js.map +1 -1
- package/dist/esm/store/feature-scope.js.map +1 -1
- package/dist/esm/store/index.js.map +1 -1
- package/dist/esm/store/lock-wait.js.map +1 -1
- package/dist/esm/store/native.js.map +1 -1
- package/dist/esm/store/open.js.map +1 -1
- package/dist/esm/store/preset-scope.js.map +1 -1
- package/dist/esm/store/readiness.js.map +1 -1
- package/dist/esm/store/reconcile.js.map +1 -1
- package/dist/esm/store/shared.js.map +1 -1
- package/dist/esm/store/signature.js.map +1 -1
- package/dist/esm/store/sql-functions.js.map +1 -1
- package/dist/esm/store/sqlite/connection.js.map +1 -1
- package/dist/esm/store/sqlite/fieldStats.js.map +1 -1
- package/dist/esm/store/sqlite/lexical.js.map +1 -1
- package/dist/esm/store/sqlite/open.js.map +1 -1
- package/dist/esm/store/sqlite/reconcile.js.map +1 -1
- package/dist/esm/store/sqlite/sql-functions.js.map +1 -1
- package/dist/esm/store/sqlite/store.js.map +1 -1
- package/dist/esm/store/sqlite/vectors.js.map +1 -1
- package/dist/esm/store/stages.js.map +1 -1
- package/dist/esm/store/transaction.js.map +1 -1
- package/dist/esm/store/turso/connection.js.map +1 -1
- package/dist/esm/store/turso/fieldStats.js.map +1 -1
- package/dist/esm/store/turso/lexical-text.js.map +1 -1
- package/dist/esm/store/turso/lexical.js.map +1 -1
- package/dist/esm/store/turso/native.js.map +1 -1
- package/dist/esm/store/turso/open.js.map +1 -1
- package/dist/esm/store/turso/reconcile.js.map +1 -1
- package/dist/esm/store/turso/sql-functions.js.map +1 -1
- package/dist/esm/store/turso/store.js.map +1 -1
- package/dist/esm/store/turso/vectors.js.map +1 -1
- package/dist/esm/store/types.js.map +1 -1
- package/dist/esm/store/vectors.js.map +1 -1
- package/dist/esm/text/segment.js.map +1 -1
- package/dist/esm/text/strip.js.map +1 -1
- package/dist/esm/watch-claim.js.map +1 -1
- package/dist/esm/watch.js.map +1 -1
- package/dist/esm/workers/parse.js.map +1 -1
- package/dist/esm/workers/watch-heartbeat.js.map +1 -1
- package/package.json +2 -2
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"sources":["/Users/kevin/Dev/
|
|
1
|
+
{"version":3,"sources":["/Users/kevin/Dev/open-source/ai/sensemaking/src/store/turso/fieldStats.ts"],"sourcesContent":["import { quoteIdent } from '../shared.ts';\nimport type { Connection, FieldStat } from '../types.ts';\n\n// Fork of sqlite/fieldStats.ts (house precedent: each store owns its engine-specific SQL):\n// Turso is SQLite dialect (spike-verified: typeof(), GROUP_CONCAT, FILTER behave identically to node:sqlite), so the SQL text is identical -- only the Connection differs.\nexport async function fieldStats(conn: Connection, columns: string[], scopeWhere: string): Promise<FieldStat[]> {\n const exprs = columns.map((name, i) => {\n const quoted = quoteIdent(name);\n return `COUNT(${quoted}) AS n${i}, GROUP_CONCAT(DISTINCT typeof(${quoted})) FILTER (WHERE ${quoted} IS NOT NULL) AS t${i}`;\n });\n const stmt = await conn.prepare(`SELECT ${exprs.join(', ')} FROM frontmatter ${scopeWhere}`);\n const row = (await stmt.get()) as Record<string, number | string | null>;\n return columns.map((name, i) => ({\n field: name,\n coverage: Number(row[`n${i}`] ?? 0),\n type: [...new Set(((row[`t${i}`] as string) ?? '').split(',').filter(Boolean))].sort().join(','),\n }));\n}\n"],"names":["fieldStats","conn","columns","scopeWhere","exprs","stmt","row","map","name","i","quoted","quoteIdent","prepare","join","get","field","coverage","Number","type","Set","split","filter","Boolean","sort"],"mappings":";;;;+BAKsBA;;;eAAAA;;;wBALK;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAKpB,SAAeA,WAAWC,IAAgB,EAAEC,OAAiB,EAAEC,UAAkB;;YAChFC,OAIAC,MACAC;;;;oBALAF,QAAQF,QAAQK,GAAG,CAAC,SAACC,MAAMC;wBAC/B,IAAMC,SAASC,IAAAA,oBAAU,EAACH;wBAC1B,OAAO,AAAC,SAAuBC,OAAfC,QAAO,UAA2CA,OAAnCD,GAAE,mCAA2DC,OAA1BA,QAAO,qBAA8CD,OAA3BC,QAAO,sBAAsB,OAAFD;oBACzH;oBACa;;wBAAMR,KAAKW,OAAO,CAAC,AAAC,UAA8CT,OAArCC,MAAMS,IAAI,CAAC,OAAM,sBAA+B,OAAXV;;;oBAAzEE,OAAO;oBACA;;wBAAMA,KAAKS,GAAG;;;oBAArBR,MAAO;oBACb;;wBAAOJ,QAAQK,GAAG,CAAC,SAACC,MAAMC;gCAEPH,OACGA;mCAHW;gCAC/BS,OAAOP;gCACPQ,UAAUC,QAAOX,QAAAA,GAAG,CAAC,AAAC,IAAK,OAAFG,GAAI,cAAZH,mBAAAA,QAAgB;gCACjCY,MAAM,AAAC,qBAAG,IAAIC,IAAI,EAAEb,SAAAA,GAAG,CAAC,AAAC,IAAK,OAAFG,GAAI,cAAZH,oBAAAA,SAA2B,IAAIc,KAAK,CAAC,KAAKC,MAAM,CAACC,WAAWC,IAAI,GAAGV,IAAI,CAAC;4BAC9F;;;;;IACF"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"sources":["/Users/kevin/Dev/
|
|
1
|
+
{"version":3,"sources":["/Users/kevin/Dev/open-source/ai/sensemaking/src/store/turso/lexical-text.ts"],"sourcesContent":["import { stemmer } from 'stemmer';\nimport { foldForSearch, searchTokens } from '../../text/segment.ts';\n\nexport function stemFolded(text: string): string {\n let out = '';\n let end = 0;\n for (const token of searchTokens(text)) {\n out += text.slice(end, token.start).replace(/[\\u0027\\u2019]/gu, ' ');\n out += token.unspaced ? token.text : stemmer(foldForSearch(token.text));\n end = token.end;\n }\n return out + text.slice(end).replace(/[\\u0027\\u2019]/gu, ' ');\n}\n"],"names":["stemFolded","text","out","end","searchTokens","token","slice","start","replace","unspaced","stemmer","foldForSearch"],"mappings":";;;;+BAGgBA;;;eAAAA;;;uBAHQ;yBACoB;AAErC,SAASA,WAAWC,IAAY;IACrC,IAAIC,MAAM;IACV,IAAIC,MAAM;QACL,kCAAA,2BAAA;;QAAL,QAAK,YAAeC,IAAAA,uBAAY,EAACH,0BAA5B,SAAA,6BAAA,QAAA,yBAAA,iCAAmC;YAAnC,IAAMI,QAAN;YACHH,OAAOD,KAAKK,KAAK,CAACH,KAAKE,MAAME,KAAK,EAAEC,OAAO,CAAC,kCAAoB;YAChEN,OAAOG,MAAMI,QAAQ,GAAGJ,MAAMJ,IAAI,GAAGS,IAAAA,gBAAO,EAACC,IAAAA,wBAAa,EAACN,MAAMJ,IAAI;YACrEE,MAAME,MAAMF,GAAG;QACjB;;QAJK;QAAA;;;iBAAA,6BAAA;gBAAA;;;gBAAA;sBAAA;;;;IAKL,OAAOD,MAAMD,KAAKK,KAAK,CAACH,KAAKK,OAAO,CAAC,kCAAoB;AAC3D"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/turso/lexical.ts"],"sourcesContent":["// T1: Tantivy FTS via fts_match/fts_score on the default-tokenizer index for bare words and\n// quoted phrases, plus a second ngram-tokenized index over \"_ngram\" sidecars for unspaced-script runs.\nimport { SenseError } from '../../errors.ts';\nimport { matchesSearchPhrase, type SearchToken, searchTokens } from '../../text/segment.ts';\nimport type { Connection, LexicalHit, LexicalQueryOptions } from '../types.ts';\nimport { stemFolded } from './lexical-text.ts';\n\nconst FIELDS = ['title_stem', 'summary_stem', 'text_stem'] as const;\nconst NGRAM_FIELDS = ['title_ngram', 'summary_ngram', 'text_ngram'] as const;\nconst AUTHORED_FIELDS = ['title', 'summary', 'text'] as const;\ntype AuthoredField = (typeof AUTHORED_FIELDS)[number];\n\n// FTS5 operator syntax a bare query would otherwise treat as literal text (PRINCIPLES:\n// no-silent-modes). `^` is rejected either way: boost to Tantivy, initial-token match to FTS5.\nconst FTS5_OPERATORS: Array<{ label: string; re: RegExp }> = [\n { label: 'prefix query', re: /[\\p{L}\\p{N}_]+\\*(?=\\s|$)/u },\n { label: 'boolean operator', re: /(?:^|\\s)(?:AND|OR|NOT)(?=\\s|$)/ },\n { label: 'NEAR operator', re: /(?:^|\\s)NEAR\\b/ },\n { label: 'boost/initial-token operator', re: /\\S*\\^\\S*/ },\n { label: 'column filter', re: /(?:^|\\s)[\\p{L}_]\\w*\\s*:/u },\n { label: 'grouping operator', re: /[()]/ },\n];\n\nfunction unsupportedOperator(terms: string): { label: string; token: string } | null {\n const withoutPhrases = terms.replace(/\"[^\"]*\"/g, 'phrase');\n for (const { label, re } of FTS5_OPERATORS) {\n const m = withoutPhrases.match(re);\n if (m) return { label, token: m[0].trim() };\n }\n return null;\n}\n\n// A run whose script marks no word boundaries indexes as one opaque token under every tokenizer\n// but ngram, so such runs -- bare or quoted -- go to the ngram sidecar index instead.\nfunction splitTerms(terms: string): { words: string[]; phrases: SearchToken[][]; unspaced: string[]; emptyPhrase: boolean } {\n const words: string[] = [];\n const phrases: SearchToken[][] = [];\n const unspaced: string[] = [];\n let emptyPhrase = false;\n const withoutPhrases = terms.replace(/\"([^\"]*)\"/g, (_m, inner: string) => {\n const phrase = inner.trim();\n const tokens = searchTokens(phrase);\n if (tokens.length === 0) {\n emptyPhrase = true;\n return ' ';\n }\n unspaced.push(...tokens.filter((token) => token.unspaced).map((token) => token.text));\n phrases.push(tokens);\n return ' ';\n });\n for (const tok of withoutPhrases.split(/\\s+/)) {\n if (tok.length === 0) continue;\n for (const token of searchTokens(tok)) {\n if (token.unspaced) unspaced.push(token.text);\n else words.push(token.text);\n }\n }\n return { words, phrases, unspaced, emptyPhrase };\n}\n\n// fts_match/fts_score's query argument must be a SQL literal: bound as `?`, the row set stays\n// correct but fts_score silently returns 0 for every row (PRINCIPLES: no-silent-modes).\nfunction escapeFtsLiteral(query: string): string {\n return query.replace(/'/g, \"''\");\n}\n\n// Terms are quoted (\"O'Brien's\" is a parse error bare) and AND-joined: Tantivy's default for\n// bare terms is disjunctive, so space-joining would loosen \"apple banana\" into \"apple OR banana\".\nfunction buildQueries(words: string[], phrases: SearchToken[][], unspaced: string[]): { defaultQuery: string; ngramQuery: string } {\n const phraseParts = phrases.flatMap((phrase) => {\n const latin = phrase.filter((token) => !token.unspaced).map((token) => stemFolded(token.text));\n if (latin.length === 0) return [];\n return phrase.some((token) => token.unspaced) ? latin : [latin.join(' ')];\n });\n const defaultParts = [...words.map(stemFolded), ...phraseParts].map((t) => `\"${t}\"`);\n const unspacedParts = unspaced.map((t) => `\"${t}\"`);\n return { defaultQuery: defaultParts.join(' AND '), ngramQuery: unspacedParts.join(' AND ') };\n}\n\n// fts_score() is real only as a SELECT's bare projected expression, and only for the index that\n// SELECT's fts_match() chose; every other form silently returns 0. Hence one SELECT per index.\nfunction ftsBranch(cols: readonly string[], query: string, opts: LexicalQueryOptions): string {\n const colList = cols.map((c) => `content.${c}`).join(', ');\n const lit = escapeFtsLiteral(query);\n return `SELECT content.\"path\" AS path, fts_score(${colList}, '${lit}') AS score\n FROM content ${opts.whereJoin} WHERE fts_match(${colList}, '${lit}') ${opts.whereCond} ${opts.scopeCond}`;\n}\n\n// Ranked query, scoped by the caller-built SQL fragments (same shape sqlite's/duckdb's\n// queryLexical take). `snippets` is always NULL: fts_highlight returns the full column text, not a bounded snippet, so every row goes through the caller's JS snippet fallback (commands/search.ts).\nexport async function queryLexical(conn: Connection, terms: string, opts: LexicalQueryOptions): Promise<LexicalHit[]> {\n const unsupported = unsupportedOperator(terms);\n if (unsupported !== null) {\n throw new SenseError('STORE_CAPABILITY_MISSING', `store \"turso\" does not implement FTS5's ${unsupported.label} (\"${unsupported.token}\") in this build; rephrase \"${terms.trim()}\" without it, or set \"store\" to \"sqlite\" in this tree's config to search it as written`);\n }\n const { words, phrases, unspaced, emptyPhrase } = splitTerms(terms);\n if (emptyPhrase) return [];\n const { defaultQuery, ngramQuery } = buildQueries(words, phrases, unspaced);\n if (defaultQuery === '' && ngramQuery === '') return [];\n\n const branches: string[] = [];\n if (defaultQuery !== '') branches.push(ftsBranch(FIELDS, defaultQuery, opts));\n if (ngramQuery !== '') branches.push(ftsBranch(NGRAM_FIELDS, ngramQuery, opts));\n\n // HAVING COUNT(*) = branches.length is the AND across indexes; with one branch it is a no-op,\n // `path` being content's primary key. fts_score is higher-is-better BM25, so DESC. Phrase\n // candidates carry authored fields through the join below so the shared verifier can reject\n // cross-field, reordered, and non-adjacent matches before applying the caller's limit.\n const ranked = `SELECT path, SUM(score) AS score FROM (\n ${branches.join(' UNION ALL ')}\n ) parts GROUP BY path HAVING COUNT(*) = ${branches.length}`;\n const phraseSelect = phrases.length > 0 ? ', content.title, content.summary, content.text' : '';\n const phraseJoin = phrases.length > 0 ? ' JOIN content ON content.path = sq.path' : '';\n const sql = `SELECT sq.path, sq.score${phraseSelect} FROM (\n ${ranked}\n ) sq${phraseJoin} ORDER BY sq.score DESC, sq.path${phrases.length > 0 ? '' : ' LIMIT ?'}`;\n const stmt = await conn.prepare(sql);\n const rows = (await stmt.all(...(phrases.length > 0 ? [] : [opts.limit]))) as Array<LexicalHit & Partial<Record<AuthoredField, string>>>;\n // A concurrent FTS read during an open write transaction can yield a {path: null} row. Raised,\n // not filtered: filtering returns a silently short result (PRINCIPLES: no-silent-modes).\n if (rows.some((r) => r.path === null)) {\n throw new SenseError('LEXICAL_NULL_PATH', 'store \"turso\" returned a search hit with no path (a known engine anomaly under concurrent FTS reads); retry the query');\n }\n const matches = phrases.length === 0 ? rows : rows.filter((row) => phrases.every((phrase) => AUTHORED_FIELDS.some((field) => matchesSearchPhrase(row[field] ?? '', phrase))));\n return matches.slice(0, opts.limit).map(({ path }) => ({ path }));\n}\n"],"names":["queryLexical","FIELDS","NGRAM_FIELDS","AUTHORED_FIELDS","FTS5_OPERATORS","label","re","unsupportedOperator","terms","withoutPhrases","replace","m","match","token","trim","splitTerms","words","phrases","unspaced","emptyPhrase","_m","inner","phrase","tokens","searchTokens","length","push","filter","map","text","split","tok","escapeFtsLiteral","query","buildQueries","phraseParts","flatMap","latin","stemFolded","some","join","defaultParts","t","unspacedParts","defaultQuery","ngramQuery","ftsBranch","cols","opts","colList","c","lit","whereJoin","whereCond","scopeCond","conn","stmt","unsupported","branches","ranked","phraseSelect","phraseJoin","sql","rows","matches","SenseError","prepare","all","limit","r","path","row","every","field","matchesSearchPhrase","slice"],"mappings":"AAAA,4FAA4F;AAC5F,uGAAuG;;;;;+BAyFjFA;;;eAAAA;;;wBAxFK;yBACyC;6BAEzC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAE3B,IAAMC,SAAS;IAAC;IAAc;IAAgB;CAAY;AAC1D,IAAMC,eAAe;IAAC;IAAe;IAAiB;CAAa;AACnE,IAAMC,kBAAkB;IAAC;IAAS;IAAW;CAAO;AAGpD,uFAAuF;AACvF,+FAA+F;AAC/F,IAAMC,iBAAuD;IAC3D;QAAEC,OAAO;QAAgBC,IAAI;IAA4B;IACzD;QAAED,OAAO;QAAoBC,IAAI;IAAiC;IAClE;QAAED,OAAO;QAAiBC,IAAI;IAAiB;IAC/C;QAAED,OAAO;QAAgCC,IAAI;IAAW;IACxD;QAAED,OAAO;QAAiBC,IAAI;IAA2B;IACzD;QAAED,OAAO;QAAqBC,IAAI;IAAO;CAC1C;AAED,SAASC,oBAAoBC,KAAa;IACxC,IAAMC,iBAAiBD,MAAME,OAAO,CAAC,YAAY;QAC5C,kCAAA,2BAAA;;QAAL,QAAK,YAAuBN,mCAAvB,SAAA,6BAAA,QAAA,yBAAA,iCAAuC;YAAvC,kBAAA,aAAQC,oBAAAA,OAAOC,iBAAAA;YAClB,IAAMK,IAAIF,eAAeG,KAAK,CAACN;YAC/B,IAAIK,GAAG,OAAO;gBAAEN,OAAAA;gBAAOQ,OAAOF,CAAC,CAAC,EAAE,CAACG,IAAI;YAAG;QAC5C;;QAHK;QAAA;;;iBAAA,6BAAA;gBAAA;;;gBAAA;sBAAA;;;;IAIL,OAAO;AACT;AAEA,gGAAgG;AAChG,sFAAsF;AACtF,SAASC,WAAWP,KAAa;IAC/B,IAAMQ,QAAkB,EAAE;IAC1B,IAAMC,UAA2B,EAAE;IACnC,IAAMC,WAAqB,EAAE;IAC7B,IAAIC,cAAc;IAClB,IAAMV,iBAAiBD,MAAME,OAAO,CAAC,cAAc,SAACU,IAAIC;YAOtDH;QANA,IAAMI,SAASD,MAAMP,IAAI;QACzB,IAAMS,SAASC,IAAAA,uBAAY,EAACF;QAC5B,IAAIC,OAAOE,MAAM,KAAK,GAAG;YACvBN,cAAc;YACd,OAAO;QACT;QACAD,CAAAA,YAAAA,UAASQ,IAAI,OAAbR,WAAc,qBAAGK,OAAOI,MAAM,CAAC,SAACd;mBAAUA,MAAMK,QAAQ;WAAEU,GAAG,CAAC,SAACf;mBAAUA,MAAMgB,IAAI;;QACnFZ,QAAQS,IAAI,CAACH;QACb,OAAO;IACT;QACK,kCAAA,2BAAA;;QAAL,QAAK,YAAad,eAAeqB,KAAK,CAAC,2BAAlC,SAAA,6BAAA,QAAA,yBAAA,iCAA0C;YAA1C,IAAMC,MAAN;YACH,IAAIA,IAAIN,MAAM,KAAK,GAAG;gBACjB,mCAAA,4BAAA;;gBAAL,QAAK,aAAeD,IAAAA,uBAAY,EAACO,yBAA5B,UAAA,8BAAA,SAAA,0BAAA,kCAAkC;oBAAlC,IAAMlB,QAAN;oBACH,IAAIA,MAAMK,QAAQ,EAAEA,SAASQ,IAAI,CAACb,MAAMgB,IAAI;yBACvCb,MAAMU,IAAI,CAACb,MAAMgB,IAAI;gBAC5B;;gBAHK;gBAAA;;;yBAAA,8BAAA;wBAAA;;;wBAAA;8BAAA;;;;QAIP;;QANK;QAAA;;;iBAAA,6BAAA;gBAAA;;;gBAAA;sBAAA;;;;IAOL,OAAO;QAAEb,OAAAA;QAAOC,SAAAA;QAASC,UAAAA;QAAUC,aAAAA;IAAY;AACjD;AAEA,8FAA8F;AAC9F,wFAAwF;AACxF,SAASa,iBAAiBC,KAAa;IACrC,OAAOA,MAAMvB,OAAO,CAAC,MAAM;AAC7B;AAEA,6FAA6F;AAC7F,kGAAkG;AAClG,SAASwB,aAAalB,KAAe,EAAEC,OAAwB,EAAEC,QAAkB;IACjF,IAAMiB,cAAclB,QAAQmB,OAAO,CAAC,SAACd;QACnC,IAAMe,QAAQf,OAAOK,MAAM,CAAC,SAACd;mBAAU,CAACA,MAAMK,QAAQ;WAAEU,GAAG,CAAC,SAACf;mBAAUyB,IAAAA,yBAAU,EAACzB,MAAMgB,IAAI;;QAC5F,IAAIQ,MAAMZ,MAAM,KAAK,GAAG,OAAO,EAAE;QACjC,OAAOH,OAAOiB,IAAI,CAAC,SAAC1B;mBAAUA,MAAMK,QAAQ;aAAImB,QAAQ;YAACA,MAAMG,IAAI,CAAC;SAAK;IAC3E;IACA,IAAMC,eAAe,AAAC,qBAAGzB,MAAMY,GAAG,CAACU,yBAAU,UAAG,qBAAGH,cAAaP,GAAG,CAAC,SAACc;eAAM,AAAC,IAAK,OAAFA,GAAE;;IACjF,IAAMC,gBAAgBzB,SAASU,GAAG,CAAC,SAACc;eAAM,AAAC,IAAK,OAAFA,GAAE;;IAChD,OAAO;QAAEE,cAAcH,aAAaD,IAAI,CAAC;QAAUK,YAAYF,cAAcH,IAAI,CAAC;IAAS;AAC7F;AAEA,gGAAgG;AAChG,+FAA+F;AAC/F,SAASM,UAAUC,IAAuB,EAAEd,KAAa,EAAEe,IAAyB;IAClF,IAAMC,UAAUF,KAAKnB,GAAG,CAAC,SAACsB;eAAM,AAAC,WAAY,OAAFA;OAAKV,IAAI,CAAC;IACrD,IAAMW,MAAMnB,iBAAiBC;IAC7B,OAAO,AAAC,4CAAwDkB,OAAbF,SAAQ,OAC1CD,OAD+CG,KAAI,kCACjBF,OAAlCD,KAAKI,SAAS,EAAC,qBAAgCD,OAAbF,SAAQ,OAAcD,OAATG,KAAI,OAAuBH,OAAlBA,KAAKK,SAAS,EAAC,KAAkB,OAAfL,KAAKM,SAAS;AAC3G;AAIO,SAAetD,aAAauD,IAAgB,EAAE/C,KAAa,EAAEwC,IAAyB;;YA2BvEQ,OA1BdC,aAI4C1C,aAA1CC,OAAOC,SAASC,UAAUC,aAEGe,eAA7BU,cAAcC,YAGhBa,UAQAC,QAGAC,cACAC,YACAC,KAGAN,MACAO,MAMAC;;;;oBAhCAP,cAAclD,oBAAoBC;oBACxC,IAAIiD,gBAAgB,MAAM;wBACxB,MAAM,IAAIQ,oBAAU,CAAC,4BAA4B,AAAC,4CAAiER,OAAvBA,YAAYpD,KAAK,EAAC,OAAqDG,OAAhDiD,YAAY5C,KAAK,EAAC,gCAA2C,OAAbL,MAAMM,IAAI,IAAG;oBAClL;oBACkDC,cAAAA,WAAWP,QAArDQ,QAA0CD,YAA1CC,OAAOC,UAAmCF,YAAnCE,SAASC,WAA0BH,YAA1BG,UAAUC,cAAgBJ,YAAhBI;oBAClC,IAAIA,aAAa;;;;oBACoBe,gBAAAA,aAAalB,OAAOC,SAASC,WAA1D0B,eAA6BV,cAA7BU,cAAcC,aAAeX,cAAfW;oBACtB,IAAID,iBAAiB,MAAMC,eAAe,IAAI;;;;oBAExCa;oBACN,IAAId,iBAAiB,IAAIc,SAAShC,IAAI,CAACoB,UAAU7C,QAAQ2C,cAAcI;oBACvE,IAAIH,eAAe,IAAIa,SAAShC,IAAI,CAACoB,UAAU5C,cAAc2C,YAAYG;oBAEzE,8FAA8F;oBAC9F,0FAA0F;oBAC1F,4FAA4F;oBAC5F,uFAAuF;oBACjFW,SAAS,AAAC,kDAE4BD,OADtCA,SAASlB,IAAI,CAAC,gBAAe,kDACyB,OAAhBkB,SAASjC,MAAM;oBACrDmC,eAAe3C,QAAQQ,MAAM,GAAG,IAAI,mDAAmD;oBACvFoC,aAAa5C,QAAQQ,MAAM,GAAG,IAAI,4CAA4C;oBAC9EqC,MAAM,AAAC,2BACTH,OADmCC,cAAa,iBAE9CC,OADFF,QAAO,YACwC1C,OAA7C4C,YAAW,oCAAuE,OAArC5C,QAAQQ,MAAM,GAAG,IAAI,KAAK;oBAChE;;wBAAM8B,KAAKW,OAAO,CAACJ;;;oBAA1BN,OAAO;oBACC;;wBAAMA,CAAAA,QAAAA,MAAKW,GAAG,OAARX,OAAS,qBAAIvC,QAAQQ,MAAM,GAAG;4BAAUuB,KAAKoB,KAAK;;;;oBAAhEL,OAAQ;oBACd,+FAA+F;oBAC/F,yFAAyF;oBACzF,IAAIA,KAAKxB,IAAI,CAAC,SAAC8B;+BAAMA,EAAEC,IAAI,KAAK;wBAAO;wBACrC,MAAM,IAAIL,oBAAU,CAAC,qBAAqB;oBAC5C;oBACMD,UAAU/C,QAAQQ,MAAM,KAAK,IAAIsC,OAAOA,KAAKpC,MAAM,CAAC,SAAC4C;+BAAQtD,QAAQuD,KAAK,CAAC,SAAClD;mCAAWnB,gBAAgBoC,IAAI,CAAC,SAACkC;oCAA8BF;uCAApBG,IAAAA,8BAAmB,GAACH,aAAAA,GAAG,CAACE,MAAM,cAAVF,wBAAAA,aAAc,IAAIjD;;;;oBACnK;;wBAAO0C,QAAQW,KAAK,CAAC,GAAG3B,KAAKoB,KAAK,EAAExC,GAAG,CAAC;gCAAG0C,aAAAA;mCAAY;gCAAEA,MAAAA;4BAAK;;;;;IAChE"}
|
|
1
|
+
{"version":3,"sources":["/Users/kevin/Dev/open-source/ai/sensemaking/src/store/turso/lexical.ts"],"sourcesContent":["// T1: Tantivy FTS via fts_match/fts_score on the default-tokenizer index for bare words and\n// quoted phrases, plus a second ngram-tokenized index over \"_ngram\" sidecars for unspaced-script runs.\nimport { SenseError } from '../../errors.ts';\nimport { matchesSearchPhrase, type SearchToken, searchTokens } from '../../text/segment.ts';\nimport type { Connection, LexicalHit, LexicalQueryOptions } from '../types.ts';\nimport { stemFolded } from './lexical-text.ts';\n\nconst FIELDS = ['title_stem', 'summary_stem', 'text_stem'] as const;\nconst NGRAM_FIELDS = ['title_ngram', 'summary_ngram', 'text_ngram'] as const;\nconst AUTHORED_FIELDS = ['title', 'summary', 'text'] as const;\ntype AuthoredField = (typeof AUTHORED_FIELDS)[number];\n\n// FTS5 operator syntax a bare query would otherwise treat as literal text (PRINCIPLES:\n// no-silent-modes). `^` is rejected either way: boost to Tantivy, initial-token match to FTS5.\nconst FTS5_OPERATORS: Array<{ label: string; re: RegExp }> = [\n { label: 'prefix query', re: /[\\p{L}\\p{N}_]+\\*(?=\\s|$)/u },\n { label: 'boolean operator', re: /(?:^|\\s)(?:AND|OR|NOT)(?=\\s|$)/ },\n { label: 'NEAR operator', re: /(?:^|\\s)NEAR\\b/ },\n { label: 'boost/initial-token operator', re: /\\S*\\^\\S*/ },\n { label: 'column filter', re: /(?:^|\\s)[\\p{L}_]\\w*\\s*:/u },\n { label: 'grouping operator', re: /[()]/ },\n];\n\nfunction unsupportedOperator(terms: string): { label: string; token: string } | null {\n const withoutPhrases = terms.replace(/\"[^\"]*\"/g, 'phrase');\n for (const { label, re } of FTS5_OPERATORS) {\n const m = withoutPhrases.match(re);\n if (m) return { label, token: m[0].trim() };\n }\n return null;\n}\n\n// A run whose script marks no word boundaries indexes as one opaque token under every tokenizer\n// but ngram, so such runs -- bare or quoted -- go to the ngram sidecar index instead.\nfunction splitTerms(terms: string): { words: string[]; phrases: SearchToken[][]; unspaced: string[]; emptyPhrase: boolean } {\n const words: string[] = [];\n const phrases: SearchToken[][] = [];\n const unspaced: string[] = [];\n let emptyPhrase = false;\n const withoutPhrases = terms.replace(/\"([^\"]*)\"/g, (_m, inner: string) => {\n const phrase = inner.trim();\n const tokens = searchTokens(phrase);\n if (tokens.length === 0) {\n emptyPhrase = true;\n return ' ';\n }\n unspaced.push(...tokens.filter((token) => token.unspaced).map((token) => token.text));\n phrases.push(tokens);\n return ' ';\n });\n for (const tok of withoutPhrases.split(/\\s+/)) {\n if (tok.length === 0) continue;\n for (const token of searchTokens(tok)) {\n if (token.unspaced) unspaced.push(token.text);\n else words.push(token.text);\n }\n }\n return { words, phrases, unspaced, emptyPhrase };\n}\n\n// fts_match/fts_score's query argument must be a SQL literal: bound as `?`, the row set stays\n// correct but fts_score silently returns 0 for every row (PRINCIPLES: no-silent-modes).\nfunction escapeFtsLiteral(query: string): string {\n return query.replace(/'/g, \"''\");\n}\n\n// Terms are quoted (\"O'Brien's\" is a parse error bare) and AND-joined: Tantivy's default for\n// bare terms is disjunctive, so space-joining would loosen \"apple banana\" into \"apple OR banana\".\nfunction buildQueries(words: string[], phrases: SearchToken[][], unspaced: string[]): { defaultQuery: string; ngramQuery: string } {\n const phraseParts = phrases.flatMap((phrase) => {\n const latin = phrase.filter((token) => !token.unspaced).map((token) => stemFolded(token.text));\n if (latin.length === 0) return [];\n return phrase.some((token) => token.unspaced) ? latin : [latin.join(' ')];\n });\n const defaultParts = [...words.map(stemFolded), ...phraseParts].map((t) => `\"${t}\"`);\n const unspacedParts = unspaced.map((t) => `\"${t}\"`);\n return { defaultQuery: defaultParts.join(' AND '), ngramQuery: unspacedParts.join(' AND ') };\n}\n\n// fts_score() is real only as a SELECT's bare projected expression, and only for the index that\n// SELECT's fts_match() chose; every other form silently returns 0. Hence one SELECT per index.\nfunction ftsBranch(cols: readonly string[], query: string, opts: LexicalQueryOptions): string {\n const colList = cols.map((c) => `content.${c}`).join(', ');\n const lit = escapeFtsLiteral(query);\n return `SELECT content.\"path\" AS path, fts_score(${colList}, '${lit}') AS score\n FROM content ${opts.whereJoin} WHERE fts_match(${colList}, '${lit}') ${opts.whereCond} ${opts.scopeCond}`;\n}\n\n// Ranked query, scoped by the caller-built SQL fragments (same shape sqlite's/duckdb's\n// queryLexical take). `snippets` is always NULL: fts_highlight returns the full column text, not a bounded snippet, so every row goes through the caller's JS snippet fallback (commands/search.ts).\nexport async function queryLexical(conn: Connection, terms: string, opts: LexicalQueryOptions): Promise<LexicalHit[]> {\n const unsupported = unsupportedOperator(terms);\n if (unsupported !== null) {\n throw new SenseError('STORE_CAPABILITY_MISSING', `store \"turso\" does not implement FTS5's ${unsupported.label} (\"${unsupported.token}\") in this build; rephrase \"${terms.trim()}\" without it, or set \"store\" to \"sqlite\" in this tree's config to search it as written`);\n }\n const { words, phrases, unspaced, emptyPhrase } = splitTerms(terms);\n if (emptyPhrase) return [];\n const { defaultQuery, ngramQuery } = buildQueries(words, phrases, unspaced);\n if (defaultQuery === '' && ngramQuery === '') return [];\n\n const branches: string[] = [];\n if (defaultQuery !== '') branches.push(ftsBranch(FIELDS, defaultQuery, opts));\n if (ngramQuery !== '') branches.push(ftsBranch(NGRAM_FIELDS, ngramQuery, opts));\n\n // HAVING COUNT(*) = branches.length is the AND across indexes; with one branch it is a no-op,\n // `path` being content's primary key. fts_score is higher-is-better BM25, so DESC. Phrase\n // candidates carry authored fields through the join below so the shared verifier can reject\n // cross-field, reordered, and non-adjacent matches before applying the caller's limit.\n const ranked = `SELECT path, SUM(score) AS score FROM (\n ${branches.join(' UNION ALL ')}\n ) parts GROUP BY path HAVING COUNT(*) = ${branches.length}`;\n const phraseSelect = phrases.length > 0 ? ', content.title, content.summary, content.text' : '';\n const phraseJoin = phrases.length > 0 ? ' JOIN content ON content.path = sq.path' : '';\n const sql = `SELECT sq.path, sq.score${phraseSelect} FROM (\n ${ranked}\n ) sq${phraseJoin} ORDER BY sq.score DESC, sq.path${phrases.length > 0 ? '' : ' LIMIT ?'}`;\n const stmt = await conn.prepare(sql);\n const rows = (await stmt.all(...(phrases.length > 0 ? [] : [opts.limit]))) as Array<LexicalHit & Partial<Record<AuthoredField, string>>>;\n // A concurrent FTS read during an open write transaction can yield a {path: null} row. Raised,\n // not filtered: filtering returns a silently short result (PRINCIPLES: no-silent-modes).\n if (rows.some((r) => r.path === null)) {\n throw new SenseError('LEXICAL_NULL_PATH', 'store \"turso\" returned a search hit with no path (a known engine anomaly under concurrent FTS reads); retry the query');\n }\n const matches = phrases.length === 0 ? rows : rows.filter((row) => phrases.every((phrase) => AUTHORED_FIELDS.some((field) => matchesSearchPhrase(row[field] ?? '', phrase))));\n return matches.slice(0, opts.limit).map(({ path }) => ({ path }));\n}\n"],"names":["queryLexical","FIELDS","NGRAM_FIELDS","AUTHORED_FIELDS","FTS5_OPERATORS","label","re","unsupportedOperator","terms","withoutPhrases","replace","m","match","token","trim","splitTerms","words","phrases","unspaced","emptyPhrase","_m","inner","phrase","tokens","searchTokens","length","push","filter","map","text","split","tok","escapeFtsLiteral","query","buildQueries","phraseParts","flatMap","latin","stemFolded","some","join","defaultParts","t","unspacedParts","defaultQuery","ngramQuery","ftsBranch","cols","opts","colList","c","lit","whereJoin","whereCond","scopeCond","conn","stmt","unsupported","branches","ranked","phraseSelect","phraseJoin","sql","rows","matches","SenseError","prepare","all","limit","r","path","row","every","field","matchesSearchPhrase","slice"],"mappings":"AAAA,4FAA4F;AAC5F,uGAAuG;;;;;+BAyFjFA;;;eAAAA;;;wBAxFK;yBACyC;6BAEzC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAE3B,IAAMC,SAAS;IAAC;IAAc;IAAgB;CAAY;AAC1D,IAAMC,eAAe;IAAC;IAAe;IAAiB;CAAa;AACnE,IAAMC,kBAAkB;IAAC;IAAS;IAAW;CAAO;AAGpD,uFAAuF;AACvF,+FAA+F;AAC/F,IAAMC,iBAAuD;IAC3D;QAAEC,OAAO;QAAgBC,IAAI;IAA4B;IACzD;QAAED,OAAO;QAAoBC,IAAI;IAAiC;IAClE;QAAED,OAAO;QAAiBC,IAAI;IAAiB;IAC/C;QAAED,OAAO;QAAgCC,IAAI;IAAW;IACxD;QAAED,OAAO;QAAiBC,IAAI;IAA2B;IACzD;QAAED,OAAO;QAAqBC,IAAI;IAAO;CAC1C;AAED,SAASC,oBAAoBC,KAAa;IACxC,IAAMC,iBAAiBD,MAAME,OAAO,CAAC,YAAY;QAC5C,kCAAA,2BAAA;;QAAL,QAAK,YAAuBN,mCAAvB,SAAA,6BAAA,QAAA,yBAAA,iCAAuC;YAAvC,kBAAA,aAAQC,oBAAAA,OAAOC,iBAAAA;YAClB,IAAMK,IAAIF,eAAeG,KAAK,CAACN;YAC/B,IAAIK,GAAG,OAAO;gBAAEN,OAAAA;gBAAOQ,OAAOF,CAAC,CAAC,EAAE,CAACG,IAAI;YAAG;QAC5C;;QAHK;QAAA;;;iBAAA,6BAAA;gBAAA;;;gBAAA;sBAAA;;;;IAIL,OAAO;AACT;AAEA,gGAAgG;AAChG,sFAAsF;AACtF,SAASC,WAAWP,KAAa;IAC/B,IAAMQ,QAAkB,EAAE;IAC1B,IAAMC,UAA2B,EAAE;IACnC,IAAMC,WAAqB,EAAE;IAC7B,IAAIC,cAAc;IAClB,IAAMV,iBAAiBD,MAAME,OAAO,CAAC,cAAc,SAACU,IAAIC;YAOtDH;QANA,IAAMI,SAASD,MAAMP,IAAI;QACzB,IAAMS,SAASC,IAAAA,uBAAY,EAACF;QAC5B,IAAIC,OAAOE,MAAM,KAAK,GAAG;YACvBN,cAAc;YACd,OAAO;QACT;QACAD,CAAAA,YAAAA,UAASQ,IAAI,OAAbR,WAAc,qBAAGK,OAAOI,MAAM,CAAC,SAACd;mBAAUA,MAAMK,QAAQ;WAAEU,GAAG,CAAC,SAACf;mBAAUA,MAAMgB,IAAI;;QACnFZ,QAAQS,IAAI,CAACH;QACb,OAAO;IACT;QACK,kCAAA,2BAAA;;QAAL,QAAK,YAAad,eAAeqB,KAAK,CAAC,2BAAlC,SAAA,6BAAA,QAAA,yBAAA,iCAA0C;YAA1C,IAAMC,MAAN;YACH,IAAIA,IAAIN,MAAM,KAAK,GAAG;gBACjB,mCAAA,4BAAA;;gBAAL,QAAK,aAAeD,IAAAA,uBAAY,EAACO,yBAA5B,UAAA,8BAAA,SAAA,0BAAA,kCAAkC;oBAAlC,IAAMlB,QAAN;oBACH,IAAIA,MAAMK,QAAQ,EAAEA,SAASQ,IAAI,CAACb,MAAMgB,IAAI;yBACvCb,MAAMU,IAAI,CAACb,MAAMgB,IAAI;gBAC5B;;gBAHK;gBAAA;;;yBAAA,8BAAA;wBAAA;;;wBAAA;8BAAA;;;;QAIP;;QANK;QAAA;;;iBAAA,6BAAA;gBAAA;;;gBAAA;sBAAA;;;;IAOL,OAAO;QAAEb,OAAAA;QAAOC,SAAAA;QAASC,UAAAA;QAAUC,aAAAA;IAAY;AACjD;AAEA,8FAA8F;AAC9F,wFAAwF;AACxF,SAASa,iBAAiBC,KAAa;IACrC,OAAOA,MAAMvB,OAAO,CAAC,MAAM;AAC7B;AAEA,6FAA6F;AAC7F,kGAAkG;AAClG,SAASwB,aAAalB,KAAe,EAAEC,OAAwB,EAAEC,QAAkB;IACjF,IAAMiB,cAAclB,QAAQmB,OAAO,CAAC,SAACd;QACnC,IAAMe,QAAQf,OAAOK,MAAM,CAAC,SAACd;mBAAU,CAACA,MAAMK,QAAQ;WAAEU,GAAG,CAAC,SAACf;mBAAUyB,IAAAA,yBAAU,EAACzB,MAAMgB,IAAI;;QAC5F,IAAIQ,MAAMZ,MAAM,KAAK,GAAG,OAAO,EAAE;QACjC,OAAOH,OAAOiB,IAAI,CAAC,SAAC1B;mBAAUA,MAAMK,QAAQ;aAAImB,QAAQ;YAACA,MAAMG,IAAI,CAAC;SAAK;IAC3E;IACA,IAAMC,eAAe,AAAC,qBAAGzB,MAAMY,GAAG,CAACU,yBAAU,UAAG,qBAAGH,cAAaP,GAAG,CAAC,SAACc;eAAM,AAAC,IAAK,OAAFA,GAAE;;IACjF,IAAMC,gBAAgBzB,SAASU,GAAG,CAAC,SAACc;eAAM,AAAC,IAAK,OAAFA,GAAE;;IAChD,OAAO;QAAEE,cAAcH,aAAaD,IAAI,CAAC;QAAUK,YAAYF,cAAcH,IAAI,CAAC;IAAS;AAC7F;AAEA,gGAAgG;AAChG,+FAA+F;AAC/F,SAASM,UAAUC,IAAuB,EAAEd,KAAa,EAAEe,IAAyB;IAClF,IAAMC,UAAUF,KAAKnB,GAAG,CAAC,SAACsB;eAAM,AAAC,WAAY,OAAFA;OAAKV,IAAI,CAAC;IACrD,IAAMW,MAAMnB,iBAAiBC;IAC7B,OAAO,AAAC,4CAAwDkB,OAAbF,SAAQ,OAC1CD,OAD+CG,KAAI,kCACjBF,OAAlCD,KAAKI,SAAS,EAAC,qBAAgCD,OAAbF,SAAQ,OAAcD,OAATG,KAAI,OAAuBH,OAAlBA,KAAKK,SAAS,EAAC,KAAkB,OAAfL,KAAKM,SAAS;AAC3G;AAIO,SAAetD,aAAauD,IAAgB,EAAE/C,KAAa,EAAEwC,IAAyB;;YA2BvEQ,OA1BdC,aAI4C1C,aAA1CC,OAAOC,SAASC,UAAUC,aAEGe,eAA7BU,cAAcC,YAGhBa,UAQAC,QAGAC,cACAC,YACAC,KAGAN,MACAO,MAMAC;;;;oBAhCAP,cAAclD,oBAAoBC;oBACxC,IAAIiD,gBAAgB,MAAM;wBACxB,MAAM,IAAIQ,oBAAU,CAAC,4BAA4B,AAAC,4CAAiER,OAAvBA,YAAYpD,KAAK,EAAC,OAAqDG,OAAhDiD,YAAY5C,KAAK,EAAC,gCAA2C,OAAbL,MAAMM,IAAI,IAAG;oBAClL;oBACkDC,cAAAA,WAAWP,QAArDQ,QAA0CD,YAA1CC,OAAOC,UAAmCF,YAAnCE,SAASC,WAA0BH,YAA1BG,UAAUC,cAAgBJ,YAAhBI;oBAClC,IAAIA,aAAa;;;;oBACoBe,gBAAAA,aAAalB,OAAOC,SAASC,WAA1D0B,eAA6BV,cAA7BU,cAAcC,aAAeX,cAAfW;oBACtB,IAAID,iBAAiB,MAAMC,eAAe,IAAI;;;;oBAExCa;oBACN,IAAId,iBAAiB,IAAIc,SAAShC,IAAI,CAACoB,UAAU7C,QAAQ2C,cAAcI;oBACvE,IAAIH,eAAe,IAAIa,SAAShC,IAAI,CAACoB,UAAU5C,cAAc2C,YAAYG;oBAEzE,8FAA8F;oBAC9F,0FAA0F;oBAC1F,4FAA4F;oBAC5F,uFAAuF;oBACjFW,SAAS,AAAC,kDAE4BD,OADtCA,SAASlB,IAAI,CAAC,gBAAe,kDACyB,OAAhBkB,SAASjC,MAAM;oBACrDmC,eAAe3C,QAAQQ,MAAM,GAAG,IAAI,mDAAmD;oBACvFoC,aAAa5C,QAAQQ,MAAM,GAAG,IAAI,4CAA4C;oBAC9EqC,MAAM,AAAC,2BACTH,OADmCC,cAAa,iBAE9CC,OADFF,QAAO,YACwC1C,OAA7C4C,YAAW,oCAAuE,OAArC5C,QAAQQ,MAAM,GAAG,IAAI,KAAK;oBAChE;;wBAAM8B,KAAKW,OAAO,CAACJ;;;oBAA1BN,OAAO;oBACC;;wBAAMA,CAAAA,QAAAA,MAAKW,GAAG,OAARX,OAAS,qBAAIvC,QAAQQ,MAAM,GAAG;4BAAUuB,KAAKoB,KAAK;;;;oBAAhEL,OAAQ;oBACd,+FAA+F;oBAC/F,yFAAyF;oBACzF,IAAIA,KAAKxB,IAAI,CAAC,SAAC8B;+BAAMA,EAAEC,IAAI,KAAK;wBAAO;wBACrC,MAAM,IAAIL,oBAAU,CAAC,qBAAqB;oBAC5C;oBACMD,UAAU/C,QAAQQ,MAAM,KAAK,IAAIsC,OAAOA,KAAKpC,MAAM,CAAC,SAAC4C;+BAAQtD,QAAQuD,KAAK,CAAC,SAAClD;mCAAWnB,gBAAgBoC,IAAI,CAAC,SAACkC;oCAA8BF;uCAApBG,IAAAA,8BAAmB,GAACH,aAAAA,GAAG,CAACE,MAAM,cAAVF,wBAAAA,aAAc,IAAIjD;;;;oBACnK;;wBAAO0C,QAAQW,KAAK,CAAC,GAAG3B,KAAKoB,KAAK,EAAExC,GAAG,CAAC;gCAAG0C,aAAAA;mCAAY;gCAAEA,MAAAA;4BAAK;;;;;IAChE"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"sources":["/Users/kevin/Dev/
|
|
1
|
+
{"version":3,"sources":["/Users/kevin/Dev/open-source/ai/sensemaking/src/store/turso/native.ts"],"sourcesContent":["import { loadOrInstall, type NativeDescriptor, packageNodeModules } from '../native.ts';\n\nexport const TURSO_PACKAGE = '@tursodatabase/database';\n\nconst TURSO: NativeDescriptor = { store: 'turso', pkg: TURSO_PACKAGE, sizeHint: '~16MB' };\n\n// Every connect() in this store goes through these: the same 30s floor sqlite opens with, and the\n// FTS index method ensureSchema() needs. reclaimSpace adds 'vacuum'; nothing else may (PLAN 3.54).\nexport const CONNECT_OPTS = { timeout: 30_000, experimental: ['index_method'] } as const;\n\nlet tursoApiPromise: Promise<typeof import('@tursodatabase/database')> | undefined;\n\n// One accessor every consumer must go through, not each its own bare-specifier import: Node\n// keeps returning a pre-install \"not found\" miss to any new lookup of the same path afterward.\nexport function tursoApi(): Promise<typeof import('@tursodatabase/database')> {\n if (!tursoApiPromise) tursoApiPromise = loadOrInstall<typeof import('@tursodatabase/database')>(TURSO, packageNodeModules());\n return tursoApiPromise;\n}\n"],"names":["CONNECT_OPTS","TURSO_PACKAGE","tursoApi","TURSO","store","pkg","sizeHint","timeout","experimental","tursoApiPromise","loadOrInstall","packageNodeModules"],"mappings":";;;;;;;;;;;QAQaA;eAAAA;;QANAC;eAAAA;;QAYGC;eAAAA;;;wBAdyD;AAElE,IAAMD,gBAAgB;AAE7B,IAAME,QAA0B;IAAEC,OAAO;IAASC,KAAKJ;IAAeK,UAAU;AAAQ;AAIjF,IAAMN,eAAe;IAAEO,SAAS;IAAQC,cAAc;QAAC;KAAe;AAAC;AAE9E,IAAIC;AAIG,SAASP;IACd,IAAI,CAACO,iBAAiBA,kBAAkBC,IAAAA,uBAAa,EAA2CP,OAAOQ,IAAAA,4BAAkB;IACzH,OAAOF;AACT"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"sources":["/Users/kevin/Dev/
|
|
1
|
+
{"version":3,"sources":["/Users/kevin/Dev/open-source/ai/sensemaking/src/store/turso/open.ts"],"sourcesContent":["import type { Database } from '@tursodatabase/database';\nimport type { Config, ResolvedConfig } from '../../config/index.ts';\nimport { STORE_DIMS } from '../../embed/types.ts';\nimport { SenseError } from '../../errors.ts';\nimport { activeFeatures } from '../../features/index.ts';\nimport type { InternalOpenOptions, OpenResult } from '../open.ts';\nimport { openWithDialect } from '../open.ts';\nimport { getMeta, setMeta } from '../shared.ts';\nimport type { Connection, OpenDialect } from '../types.ts';\nimport { checkpointWal, createConnection } from './connection.ts';\nimport { CONNECT_OPTS, TURSO_PACKAGE, tursoApi } from './native.ts';\nimport { CONTENT_FTS_DDL, tursoDialect } from './reconcile.ts';\nimport { createStore } from './store.ts';\n\nexport const DB_FILENAME = 'cache.turso.db';\n// Independent of sqlite's and duckdb's SCHEMA_VERSION: each store's cache shape evolves\n// separately. Covers the FTS indexes, the \"_ngram\" sidecar columns, and embeddings.vector's width.\nexport const SCHEMA_VERSION = '10';\n\nexport type { OpenResult };\n\n// This store's Handle (types.ts's OpenDialect<Handle>) is the connected Database itself: no\n// extra native state to thread, unlike duckdb's separate instance/connection pair.\nasync function ensureCoreSchema(_handle: Database, conn: Connection): Promise<void> {\n await conn.exec(`CREATE TABLE IF NOT EXISTS frontmatter (\"path\" TEXT PRIMARY KEY, \"_mtime\" REAL, \"_ctime\" REAL, \"_size\" INTEGER, \"_parse_error\" TEXT)`);\n await conn.exec(`CREATE TABLE IF NOT EXISTS content (\"path\" TEXT PRIMARY KEY, title TEXT, summary TEXT, text TEXT, title_stem TEXT, summary_stem TEXT, text_stem TEXT, title_ngram TEXT, summary_ngram TEXT, text_ngram TEXT)`);\n await conn.exec(`CREATE TABLE IF NOT EXISTS indexed_sources (\"path\" TEXT PRIMARY KEY, text TEXT NOT NULL)`);\n for (const ddl of CONTENT_FTS_DDL) await conn.exec(ddl);\n await conn.exec(`CREATE TABLE IF NOT EXISTS preset_files (\"path\" TEXT, preset TEXT, PRIMARY KEY (\"path\", preset))`);\n await conn.exec('CREATE INDEX IF NOT EXISTS preset_files_preset ON preset_files(preset)');\n if ((await getMeta(conn, 'schema_version')) === null) await setMeta(conn, 'schema_version', SCHEMA_VERSION);\n}\n\nasync function ensureFeatureSchema(_handle: Database, conn: Connection, cfg: Config): Promise<void> {\n for (const feature of activeFeatures(cfg)) {\n // Native F32_BLOB(STORE_DIMS) instead of the embed feature's engine-neutral BLOB DDL. `scale`\n // is kept unused, so the shared reconcile-time INSERT/DELETE names a column both stores have.\n if (feature.name === 'embed') {\n await conn.exec(`CREATE TABLE IF NOT EXISTS embeddings (\"path\" TEXT, chunk INTEGER, start_line INTEGER, end_line INTEGER, content_identity TEXT NOT NULL, scale REAL, vector F32_BLOB(${STORE_DIMS}), PRIMARY KEY (\"path\", chunk))`);\n continue;\n }\n await feature.schema(conn);\n }\n}\n\nasync function close(handle: Database, options: { observational?: boolean } = {}): Promise<void> {\n if (!options.observational) await checkpointWal(handle);\n await handle.close();\n}\n\nasync function setDerivedBusyTimeout(_handle: Database, conn: Connection, ms: number): Promise<void> {\n await conn.exec(`PRAGMA busy_timeout = ${ms}`);\n}\n\nasync function connect(dbPath: string, _cfg: ResolvedConfig, options: { existingOnly: boolean; observational: boolean }): Promise<{ handle: Database; conn: Connection }> {\n // Dynamic, not a top-level import: a sqlite or duckdb tree must never attempt to resolve this\n // optional dependency until a turso tree is actually opened. Installed on first use if missing.\n let turso: Awaited<ReturnType<typeof tursoApi>>;\n try {\n turso = await tursoApi();\n } catch (err) {\n if (err instanceof SenseError) throw err;\n throw new SenseError('STORE_DEPENDENCY_MISSING', `store \"turso\" needs the ${TURSO_PACKAGE} package (${(err as Error).message})`);\n }\n\n let db: Database;\n try {\n // Turso's readonly mode rejects native FTS SELECTs and TEMP query tables. Existing-only opens\n // stay query-capable; shared no-build orchestration and observational close prevent upkeep writes.\n db = await turso.connect(dbPath, { ...CONNECT_OPTS, experimental: [...CONNECT_OPTS.experimental], ...(options.existingOnly ? { fileMustExist: true } : {}) });\n } catch (err) {\n throw new SenseError('STORE_DEPENDENCY_MISSING', `store \"turso\" failed to open ${dbPath}: ${(err as Error).message}`);\n }\n return { handle: db, conn: createConnection(db) };\n}\n\n// This store's dialect (types.ts's OpenDialect) for the shared orchestration in store/open.ts.\nexport const tursoOpenDialect: OpenDialect<Database> = {\n filename: DB_FILENAME,\n schemaVersion: SCHEMA_VERSION,\n reconcileDialect: tursoDialect,\n connect,\n close,\n // \"Locking error: Failed locking file ...\", worded per platform: posix \"File is locked by another\n // process\", Windows \"another process has locked a portion of the file (os error 33)\". Distinct from\n // the write-time \"database is locked\" its connect-time `timeout` covers; that one never reaches here.\n isLocked: (err) => /File is locked by another process|locked a portion of the file/.test(err.message),\n ensureCoreSchema,\n ensureFeatureSchema,\n setDerivedBusyTimeout,\n createStore: (handle, conn, _cfg, options) => createStore(handle, conn, options),\n};\n\nexport async function openTurso(cfg: ResolvedConfig, options?: InternalOpenOptions): Promise<OpenResult> {\n return openWithDialect(cfg, tursoOpenDialect, options);\n}\n"],"names":["DB_FILENAME","SCHEMA_VERSION","openTurso","tursoOpenDialect","ensureCoreSchema","_handle","conn","ddl","exec","CONTENT_FTS_DDL","getMeta","setMeta","ensureFeatureSchema","cfg","feature","activeFeatures","name","STORE_DIMS","schema","close","handle","options","observational","checkpointWal","setDerivedBusyTimeout","ms","connect","dbPath","_cfg","turso","err","db","tursoApi","SenseError","TURSO_PACKAGE","message","CONNECT_OPTS","experimental","existingOnly","fileMustExist","createConnection","filename","schemaVersion","reconcileDialect","tursoDialect","isLocked","test","createStore","openWithDialect"],"mappings":";;;;;;;;;;;QAcaA;eAAAA;;QAGAC;eAAAA;;QA4ESC;eAAAA;;QAhBTC;eAAAA;;;uBA3Ec;wBACA;uBACI;sBAEC;wBACC;4BAEe;wBACM;2BACR;uBAClB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAErB,IAAMH,cAAc;AAGpB,IAAMC,iBAAiB;AAI9B,4FAA4F;AAC5F,mFAAmF;AACnF,SAAeG,iBAAiBC,OAAiB,EAAEC,IAAgB;;YAI5D,2BAAA,mBAAA,gBAAA,WAAA,OAAMC;;;;oBAHX;;wBAAMD,KAAKE,IAAI,CAAC;;;oBAAhB;oBACA;;wBAAMF,KAAKE,IAAI,CAAC;;;oBAAhB;oBACA;;wBAAMF,KAAKE,IAAI,CAAC;;;oBAAhB;oBACK,kCAAA,2BAAA;;;;;;;;;oBAAA,YAAaC,4BAAe;;;2BAA5B,6BAAA,QAAA;;;;oBAAMF,MAAN;oBAA8B;;wBAAMD,KAAKE,IAAI,CAACD;;;oBAAhB;;;oBAA9B;;;;;;;;;;;;oBAAA;oBAAA;;;;;;;6BAAA,6BAAA;4BAAA;;;4BAAA;kCAAA;;;;;;;oBACL;;wBAAMD,KAAKE,IAAI,CAAC;;;oBAAhB;oBACA;;wBAAMF,KAAKE,IAAI,CAAC;;;oBAAhB;oBACK;;wBAAME,IAAAA,iBAAO,EAACJ,MAAM;;;yBAArB,CAAA,AAAC,kBAA2C,IAAG,GAA/C;;;;oBAAkD;;wBAAMK,IAAAA,iBAAO,EAACL,MAAM,kBAAkBL;;;oBAAtC;;;;;;;;IACxD;;AAEA,SAAeW,oBAAoBP,OAAiB,EAAEC,IAAgB,EAAEO,GAAW;;YAC5E,2BAAA,mBAAA,gBAAA,WAAA,OAAMC;;;;oBAAN,kCAAA,2BAAA;;;;;;;;;oBAAA,YAAiBC,IAAAA,uBAAc,EAACF;;;2BAAhC,6BAAA,QAAA;;;;oBAAMC,UAAN;yBAGCA,CAAAA,QAAQE,IAAI,KAAK,OAAM,GAAvBF;;;;oBACF;;wBAAMR,KAAKE,IAAI,CAAC,AAAC,wKAAkL,OAAXS,mBAAU,EAAC;;;oBAAnM;oBACA;;;;;oBAEF;;wBAAMH,QAAQI,MAAM,CAACZ;;;oBAArB;;;oBAPG;;;;;;;;;;;;oBAAA;oBAAA;;;;;;;6BAAA,6BAAA;4BAAA;;;4BAAA;kCAAA;;;;;;;;;;;;IASP;;AAEA,SAAea;wCAAMC,MAAgB;YAAEC;;;;;oBAAAA,UAAAA,oEAAuC,CAAC;yBACzE,CAACA,QAAQC,aAAa,EAAtB;;;;oBAAwB;;wBAAMC,IAAAA,2BAAa,EAACH;;;oBAApB;;;oBAC5B;;wBAAMA,OAAOD,KAAK;;;oBAAlB;;;;;;IACF;;AAEA,SAAeK,sBAAsBnB,OAAiB,EAAEC,IAAgB,EAAEmB,EAAU;;;;;oBAClF;;wBAAMnB,KAAKE,IAAI,CAAC,AAAC,yBAA2B,OAAHiB;;;oBAAzC;;;;;;IACF;;AAEA,SAAeC,QAAQC,MAAc,EAAEC,IAAoB,EAAEP,OAA0D;;YAGjHQ,OAGKC,KAKLC,IAKKD;;;;;;;;;;oBAXC;;wBAAME,IAAAA,kBAAQ;;;oBAAtBH,QAAQ;;;;;;oBACDC;oBACP,IAAIA,AAAG,YAAHA,KAAeG,oBAAU,GAAE,MAAMH;oBACrC,MAAM,IAAIG,oBAAU,CAAC,4BAA4B,AAAC,2BAAoD,OAA1BC,uBAAa,EAAC,cAAmC,OAAvB,AAACJ,IAAcK,OAAO,EAAC;;;;;;;;oBAOxH;;wBAAMN,MAAMH,OAAO,CAACC,QAAQ,uDAAKS,sBAAY;4BAAEC,cAAe,qBAAGD,sBAAY,CAACC,YAAY;4BAAOhB,QAAQiB,YAAY,GAAG;4BAAEC,eAAe;wBAAK,IAAI,CAAC;;;oBAFxJ,8FAA8F;oBAC9F,mGAAmG;oBACnGR,KAAK;;;;;;oBACED;oBACP,MAAM,IAAIG,oBAAU,CAAC,4BAA4B,AAAC,gCAA0C,OAAXN,QAAO,MAA2B,OAAvB,AAACG,KAAcK,OAAO;;oBAEpH;;wBAAO;4BAAEf,QAAQW;4BAAIzB,MAAMkC,IAAAA,8BAAgB,EAACT;wBAAI;;;;IAClD;;AAGO,IAAM5B,mBAA0C;IACrDsC,UAAUzC;IACV0C,eAAezC;IACf0C,kBAAkBC,yBAAY;IAC9BlB,SAAAA;IACAP,OAAAA;IACA,kGAAkG;IAClG,oGAAoG;IACpG,sGAAsG;IACtG0B,UAAU,SAAVA,SAAWf;eAAQ,iEAAiEgB,IAAI,CAAChB,IAAIK,OAAO;;IACpG/B,kBAAAA;IACAQ,qBAAAA;IACAY,uBAAAA;IACAuB,aAAa,SAAbA,YAAc3B,QAAQd,MAAMsB,MAAMP;eAAY0B,IAAAA,oBAAW,EAAC3B,QAAQd,MAAMe;;AAC1E;AAEO,SAAenB,UAAUW,GAAmB,EAAEQ,OAA6B;;;YAChF;;gBAAO2B,IAAAA,uBAAe,EAACnC,KAAKV,kBAAkBkB;;;IAChD"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"sources":["/Users/kevin/Dev/
|
|
1
|
+
{"version":3,"sources":["/Users/kevin/Dev/open-source/ai/sensemaking/src/store/turso/reconcile.ts"],"sourcesContent":["import type { Config } from '../../config/index.ts';\nimport { SenseError } from '../../errors.ts';\nimport type { ReconcileDelta } from '../../features/types.ts';\nimport type { ParsedDoc } from '../../scan/index.ts';\nimport { hasUnspacedRun } from '../../text/segment.ts';\nimport { quoteIdent, recordReconcileDuration } from '../shared.ts';\nimport { BEGIN_WRITE } from '../transaction.ts';\nimport type { Connection, ReconcileDialect } from '../types.ts';\nimport { stemFolded } from './lexical-text.ts';\n\n// This store's dialect (types.ts's ReconcileDialect) for the shared orchestration in\n// store/reconcile.ts. content is a plain table with \"_ngram\" sidecars for lexical.ts.\n\n// The ngram index is scoped to disjoint \"_ngram\" sidecar columns: a second index over the same\n// columns makes a bare substring match a whole word, defeating the prefix-query rejection.\n// The two FTS indexes, named here (not open.ts) because reconcile drops and rebuilds them around\n// a bulk load: Tantivy maintains them per inserted row, which is quadratic in what is already indexed.\nexport const CONTENT_FTS_DDL = [\n `CREATE INDEX IF NOT EXISTS content_fts ON content USING fts (title_stem, summary_stem, text_stem) WITH (weights = 'title_stem=10.0,summary_stem=5.0,text_stem=1.0')`,\n `CREATE INDEX IF NOT EXISTS content_fts_ngram ON content USING fts (title_ngram, summary_ngram, text_ngram) WITH (tokenizer='ngram', weights='title_ngram=10.0,summary_ngram=5.0,text_ngram=1.0')`,\n] as const;\nexport const CONTENT_FTS_NAMES = ['content_fts', 'content_fts_ngram'] as const;\n\n// '' when the field has no unspaced-script run (the common case), so the ngram index carries\n// nothing for it -- same \"pay for nothing when absent\" shape as sqlite's segmentField sidecars.\nfunction ngramSidecar(text: string): string {\n return hasUnspacedRun(text) ? text : '';\n}\n\n// Not a compile-time cap on ALTER TABLE ADD COLUMN (spike-measured: turso accepts 10,000 with\n// no error); the real fence is a SELECT projecting more than this many result columns, which fails to prepare.\nconst MAX_FRONTMATTER_COLUMNS = 2000;\n\n// Retained policy boundary from 2026-08-30 corpus-scale measurements; the optimum after later\n// update-path changes remains pending matched corpus-scale crossover evidence.\nexport const FTS_REBUILD_THRESHOLD = 250;\n\nexport type TursoFtsStrategy = 'incremental' | 'rebuild';\n\n// No rowid coupling (unlike sqlite's FTS5 content): `path` is content's own primary key.\nconst INSERT_CONTENT_SQL = `INSERT INTO content (\"path\", title, summary, text, title_stem, summary_stem, text_stem, title_ngram, summary_ngram, text_ngram) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`;\n\nfunction contentRow(doc: ParsedDoc): unknown[] {\n const { title, summary, text } = doc.search;\n return [doc.relPath, title, summary, text, stemFolded(title), stemFolded(summary), stemFolded(text), ngramSidecar(title), ngramSidecar(summary), ngramSidecar(text)];\n}\n\nexport function tursoFtsStrategy(delta: ReconcileDelta): TursoFtsStrategy {\n const churn = delta.reparsed.length + delta.vanished.length;\n return delta.files.length === 0 || churn > FTS_REBUILD_THRESHOLD ? 'rebuild' : 'incremental';\n}\n\n// Kept transaction-free because the shared reconcile owns the one write transaction. The explicit\n// strategy is internal composition for real-engine diagnostics; production selects it from delta.\nexport async function reconcileTursoContentWithStrategy(conn: Connection, touched: string[], docs: ParsedDoc[], strategy: TursoFtsStrategy): Promise<void> {\n // Tantivy indexes per inserted row at a cost that grows with the batch, so a large insert is\n // superlinear and a rebuild wins past the threshold.\n const bulk = strategy === 'rebuild';\n if (bulk) for (const name of CONTENT_FTS_NAMES) await conn.exec(`DROP INDEX IF EXISTS ${name}`);\n\n // content is a plain table keyed by its own path (no rowid subquery, unlike sqlite's FTS5\n // content), so vanished and reparsed docs delete in one pass.\n if (touched.length > 0) {\n const placeholders = touched.map(() => '?').join(', ');\n await conn.runBatch(`DELETE FROM content WHERE \"path\" IN (${placeholders})`, [touched]);\n }\n if (docs.length > 0) await conn.runBatch(INSERT_CONTENT_SQL, docs.map(contentRow));\n if (bulk) for (const ddl of CONTENT_FTS_DDL) await conn.exec(ddl);\n}\n\nasync function reconcileTursoContent(conn: Connection, touched: string[], docs: ParsedDoc[], delta: ReconcileDelta, _cfg: Config): Promise<void> {\n await reconcileTursoContentWithStrategy(conn, touched, docs, tursoFtsStrategy(delta));\n}\n\n// turso's ADD COLUMN is metadata-only, so a loop costs nothing extra over one statement.\nasync function addColumns(conn: Connection, names: string[]): Promise<void> {\n for (const name of names) await conn.exec(`ALTER TABLE frontmatter ADD COLUMN ${quoteIdent(name)}`);\n}\n\nexport const tursoDialect: ReconcileDialect = {\n beginMode: () => BEGIN_WRITE,\n checkColumnLimit(count) {\n if (count > MAX_FRONTMATTER_COLUMNS) {\n throw new SenseError(\n 'COLUMN_LIMIT',\n `frontmatter would need ${count} columns, crossing turso's SELECT result-set column limit (${MAX_FRONTMATTER_COLUMNS}; ALTER TABLE ADD COLUMN itself accepts far more, but a query projecting past this many columns fails to prepare). Narrow the presets' include globs so fewer/other files are indexed, or fix whatever is generating unbounded frontmatter keys.`\n );\n }\n },\n addColumns,\n reconcileContent: reconcileTursoContent,\n recordDuration: recordReconcileDuration,\n};\n"],"names":["CONTENT_FTS_DDL","CONTENT_FTS_NAMES","FTS_REBUILD_THRESHOLD","reconcileTursoContentWithStrategy","tursoDialect","tursoFtsStrategy","ngramSidecar","text","hasUnspacedRun","MAX_FRONTMATTER_COLUMNS","INSERT_CONTENT_SQL","contentRow","doc","search","title","summary","relPath","stemFolded","delta","churn","reparsed","length","vanished","files","conn","touched","docs","strategy","bulk","name","placeholders","ddl","exec","map","join","runBatch","reconcileTursoContent","_cfg","addColumns","names","quoteIdent","beginMode","BEGIN_WRITE","checkColumnLimit","count","SenseError","reconcileContent","recordDuration","recordReconcileDuration"],"mappings":";;;;;;;;;;;QAiBaA;eAAAA;;QAIAC;eAAAA;;QAcAC;eAAAA;;QAmBSC;eAAAA;;QAyBTC;eAAAA;;QAhCGC;eAAAA;;;wBA9CW;yBAGI;wBACqB;6BACxB;6BAED;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AASpB,IAAML,kBAAkB;IAC7B;IACA;CACD;AACM,IAAMC,oBAAoB;IAAC;IAAe;CAAoB;AAErE,6FAA6F;AAC7F,gGAAgG;AAChG,SAASK,aAAaC,IAAY;IAChC,OAAOC,IAAAA,yBAAc,EAACD,QAAQA,OAAO;AACvC;AAEA,8FAA8F;AAC9F,+GAA+G;AAC/G,IAAME,0BAA0B;AAIzB,IAAMP,wBAAwB;AAIrC,yFAAyF;AACzF,IAAMQ,qBAAqB;AAE3B,SAASC,WAAWC,GAAc;IAChC,IAAiCA,cAAAA,IAAIC,MAAM,EAAnCC,QAAyBF,YAAzBE,OAAOC,UAAkBH,YAAlBG,SAASR,OAASK,YAATL;IACxB,OAAO;QAACK,IAAII,OAAO;QAAEF;QAAOC;QAASR;QAAMU,IAAAA,yBAAU,EAACH;QAAQG,IAAAA,yBAAU,EAACF;QAAUE,IAAAA,yBAAU,EAACV;QAAOD,aAAaQ;QAAQR,aAAaS;QAAUT,aAAaC;KAAM;AACtK;AAEO,SAASF,iBAAiBa,KAAqB;IACpD,IAAMC,QAAQD,MAAME,QAAQ,CAACC,MAAM,GAAGH,MAAMI,QAAQ,CAACD,MAAM;IAC3D,OAAOH,MAAMK,KAAK,CAACF,MAAM,KAAK,KAAKF,QAAQjB,wBAAwB,YAAY;AACjF;AAIO,SAAeC,kCAAkCqB,IAAgB,EAAEC,OAAiB,EAAEC,IAAiB,EAAEC,QAA0B;;YAGlIC,MACS,2BAAA,mBAAA,gBAAA,WAAA,OAAMC,WAKbC,cAIO,4BAAA,oBAAA,iBAAA,YAAA,QAAMC;;;;oBAZrB,6FAA6F;oBAC7F,qDAAqD;oBAC/CH,OAAOD,aAAa;oBACX,kCAAA,2BAAA;yBAAXC,MAAAA;;;;;;;;;;;;oBAAW,YAAc3B;;;2BAAd,6BAAA,QAAA;;;;oBAAM4B,OAAN;oBAAiC;;wBAAML,KAAKQ,IAAI,CAAC,AAAC,wBAA4B,OAALH;;;oBAAxC;;;oBAAjC;;;;;;;;;;;;oBAAA;oBAAA;;;;;;;6BAAA,6BAAA;4BAAA;;;4BAAA;kCAAA;;;;;;;yBAIXJ,CAAAA,QAAQJ,MAAM,GAAG,CAAA,GAAjBI;;;;oBACIK,eAAeL,QAAQQ,GAAG,CAAC;+BAAM;uBAAKC,IAAI,CAAC;oBACjD;;wBAAMV,KAAKW,QAAQ,CAAC,AAAC,wCAAoD,OAAbL,cAAa;4BAAKL;;;;oBAA9E;;;yBAEEC,CAAAA,KAAKL,MAAM,GAAG,CAAA,GAAdK;;;;oBAAiB;;wBAAMF,KAAKW,QAAQ,CAACzB,oBAAoBgB,KAAKO,GAAG,CAACtB;;;oBAAjD;;;oBACN,mCAAA,4BAAA;yBAAXiB,MAAAA;;;;;;;;;;;;oBAAW,aAAa5B;;;2BAAb,8BAAA,SAAA;;;;oBAAM+B,MAAN;oBAA8B;;wBAAMP,KAAKQ,IAAI,CAACD;;;oBAAhB;;;oBAA9B;;;;;;;;;;;;oBAAA;oBAAA;;;;;;;6BAAA,8BAAA;4BAAA;;;4BAAA;kCAAA;;;;;;;;;;;;IACjB;;AAEA,SAAeK,sBAAsBZ,IAAgB,EAAEC,OAAiB,EAAEC,IAAiB,EAAER,KAAqB,EAAEmB,IAAY;;;;;oBAC9H;;wBAAMlC,kCAAkCqB,MAAMC,SAASC,MAAMrB,iBAAiBa;;;oBAA9E;;;;;;IACF;;AAEA,yFAAyF;AACzF,SAAeoB,WAAWd,IAAgB,EAAEe,KAAe;;YACpD,2BAAA,mBAAA,gBAAA,WAAA,OAAMV;;;;oBAAN,kCAAA,2BAAA;;;;;;;;;oBAAA,YAAcU;;;2BAAd,6BAAA,QAAA;;;;oBAAMV,OAAN;oBAAqB;;wBAAML,KAAKQ,IAAI,CAAC,AAAC,sCAAsD,OAAjBQ,IAAAA,oBAAU,EAACX;;;oBAAjE;;;oBAArB;;;;;;;;;;;;oBAAA;oBAAA;;;;;;;6BAAA,6BAAA;4BAAA;;;4BAAA;kCAAA;;;;;;;;;;;;IACP;;AAEO,IAAMzB,eAAiC;IAC5CqC,WAAW,SAAXA;eAAiBC,0BAAW;;IAC5BC,kBAAAA,SAAAA,iBAAiBC,KAAK;QACpB,IAAIA,QAAQnC,yBAAyB;YACnC,MAAM,IAAIoC,oBAAU,CAClB,gBACA,AAAC,0BAA4FpC,OAAnEmC,OAAM,+DAAqF,OAAxBnC,yBAAwB;QAEzH;IACF;IACA6B,YAAAA;IACAQ,kBAAkBV;IAClBW,gBAAgBC,iCAAuB;AACzC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/turso/sql-functions.ts"],"sourcesContent":["import { SenseError } from '../../errors.ts';\n\n// Turso can't register UDFs: has()/basename() are rewritten into portable SQL below; segment()\n// (Intl.Segmenter grapheme clustering, types.ts's 'segment' capability) has none, so it's rejected like an unsupported FTS5 operator.\n\n// Non-global: recursion (readArgs' contents rewritten via rewriteFunctions) reenters this module's\n// regexes before the outer scan resumes, and a global regex's lastIndex does not nest across that.\nconst CALL_ANYWHERE = /\\b(has|basename|segment)\\s*\\(/i;\nconst CALL_AT_START = /^(has|basename|segment)\\s*\\(/i;\n\nfunction isWordChar(ch: string | undefined): boolean {\n return ch !== undefined && /[A-Za-z0-9_]/.test(ch);\n}\n\n// The index just past the quoted literal or SQL comment starting at i, or null if i starts\n// neither. Shared by the scan and readArgs so both leave quoted/comment text opaque to a call match.\nfunction skipOpaque(sql: string, i: number): number | null {\n const ch = sql[i];\n if (ch === \"'\" || ch === '\"') {\n let j = i + 1;\n while (j < sql.length) {\n if (sql[j] === ch) {\n if (sql[j + 1] === ch) {\n j += 2;\n continue;\n }\n return j + 1;\n }\n j++;\n }\n return j; // unterminated literal: consume to end rather than loop forever\n }\n if (ch === '-' && sql[i + 1] === '-') {\n const nl = sql.indexOf('\\n', i);\n return nl === -1 ? sql.length : nl + 1;\n }\n if (ch === '/' && sql[i + 1] === '*') {\n const close = sql.indexOf('*/', i + 2);\n return close === -1 ? sql.length : close + 2;\n }\n return null;\n}\n\n// `?N`/named placeholders bind by number/name, so repeating one is safe; an anonymous `?` takes\n// the next slot per occurrence, so it's numbered here first -- sqlHas/sqlBasename duplicate their arguments.\nfunction numberPlaceholders(sql: string): string {\n let out = '';\n let i = 0;\n let n = 0;\n while (i < sql.length) {\n const skip = skipOpaque(sql, i);\n if (skip !== null) {\n out += sql.slice(i, skip);\n i = skip;\n continue;\n }\n if (sql[i] === '?' && !/[0-9]/.test(sql[i + 1] ?? '')) {\n n++;\n out += `?${n}`;\n i++;\n continue;\n }\n out += sql[i];\n i++;\n }\n return out;\n}\n\n// The un-rewritten argument texts of the call whose '(' sits at parenStart, and the index just\n// past its matching ')'. Paren depth is tracked so a nested call's commas/parens aren't mistaken for ours.\nfunction readArgs(sql: string, parenStart: number): { args: string[]; end: number } {\n const args: string[] = [];\n let depth = 1;\n let cur = '';\n let i = parenStart + 1;\n while (i < sql.length) {\n const skip = skipOpaque(sql, i);\n if (skip !== null) {\n cur += sql.slice(i, skip);\n i = skip;\n continue;\n }\n const ch = sql[i];\n if (ch === '(') {\n depth++;\n cur += ch;\n i++;\n continue;\n }\n if (ch === ')') {\n depth--;\n i++;\n if (depth === 0) break;\n cur += ch;\n continue;\n }\n if (ch === ',' && depth === 1) {\n args.push(cur.trim());\n cur = '';\n i++;\n continue;\n }\n cur += ch;\n i++;\n }\n if (depth !== 0) throw new Error(`turso: unterminated argument list starting at \"${sql.slice(parenStart, parenStart + 30)}...\"`);\n if (cur.trim() !== '') args.push(cur.trim());\n return { args, end: i };\n}\n\n// field.startsWith('[') && JSON.parse(field) is an array, matched by membership; every other\n// field, stringified, matched by substring (sql-functions.ts's hasImpl, mirrored branch for branch).\nfunction sqlHas(fieldExpr: string, valueExpr: string): string {\n const isArray = `substr(${fieldExpr},1,1) = '[' AND json_valid(${fieldExpr})`;\n // json_each's argument is re-guarded by its own CASE (falling back to '[]'): turso evaluates a\n // CASE branch's table function even when untaken, and a non-array argument there raises rather than yielding 0 rows (spike-verified).\n return `(CASE\n WHEN ${fieldExpr} IS NULL THEN 0\n WHEN ${isArray} THEN\n (SELECT COUNT(*) FROM json_each(CASE WHEN ${isArray} THEN ${fieldExpr} ELSE '[]' END) je WHERE CAST(je.value AS TEXT) = CAST(${valueExpr} AS TEXT)) > 0\n ELSE instr(CAST(${fieldExpr} AS TEXT), CAST(${valueExpr} AS TEXT)) > 0\n END)`;\n}\n\n// posix.basename(String(path)), suffix stripped Unix-style (sql-functions.ts's basenameImpl). No\n// last-'/' primitive exists, so reverse() finds the first '/' from the end, then reverse() restores order.\nfunction sqlBasename(pathExpr: string, suffixExpr: string): string {\n const trimmed = `rtrim(CAST(${pathExpr} AS TEXT), '/')`;\n const name = `CASE\n WHEN ${trimmed} = '' THEN ''\n WHEN instr(${trimmed}, '/') = 0 THEN ${trimmed}\n ELSE reverse(substr(reverse(${trimmed}), 1, instr(reverse(${trimmed}), '/') - 1))\n END`;\n return `(CASE WHEN ${pathExpr} IS NULL THEN NULL ELSE\n (CASE WHEN ${suffixExpr} IS NOT NULL AND CAST(${suffixExpr} AS TEXT) != '' AND CAST(${suffixExpr} AS TEXT) != (${name})\n AND substr((${name}), -length(CAST(${suffixExpr} AS TEXT))) = CAST(${suffixExpr} AS TEXT)\n THEN substr((${name}), 1, length((${name})) - length(CAST(${suffixExpr} AS TEXT)))\n ELSE (${name}) END)\n END)`;\n}\n\n// Rewrites every has()/basename() call into the portable SQL above, recursing into each call's own\n// arguments first so a nested `instr(basename(f.path), 'x')` resolves inside out; segment() instead raises STORE_CAPABILITY_MISSING, named, like an unsupported FTS5 operator.\nexport function rewriteFunctions(sql: string): string {\n if (!CALL_ANYWHERE.test(sql)) return sql;\n // Numbered once, over the whole statement, before any argument is duplicated below -- rewrite()\n // recurses over already-numbered substrings, never renumbering them a second time.\n return rewrite(numberPlaceholders(sql));\n}\n\nfunction rewrite(sql: string): string {\n let out = '';\n let i = 0;\n while (i < sql.length) {\n const skip = skipOpaque(sql, i);\n if (skip !== null) {\n out += sql.slice(i, skip);\n i = skip;\n continue;\n }\n const m = !isWordChar(sql[i - 1]) ? CALL_AT_START.exec(sql.slice(i)) : null;\n if (m) {\n const name = m[1].toLowerCase();\n const parenStart = i + m[0].length - 1;\n if (name === 'segment') {\n throw new SenseError('STORE_CAPABILITY_MISSING', `store \"turso\" does not implement segment() as a SQL function in this build (its client cannot register UDFs); rephrase the query without it, or set \"store\" to \"sqlite\" in this tree's config to run it as written`);\n }\n const { args, end } = readArgs(sql, parenStart);\n const rewrittenArgs = args.map(rewrite);\n if (name === 'has') {\n if (rewrittenArgs.length !== 2) throw new Error(`turso: has() takes exactly 2 arguments (field, value), got ${rewrittenArgs.length}`);\n out += sqlHas(rewrittenArgs[0], rewrittenArgs[1]);\n } else {\n if (rewrittenArgs.length < 1 || rewrittenArgs.length > 2) throw new Error(`turso: basename() takes 1 or 2 arguments (path, suffix?), got ${rewrittenArgs.length}`);\n out += sqlBasename(rewrittenArgs[0], rewrittenArgs[1] ?? 'NULL');\n }\n i = end;\n continue;\n }\n out += sql[i];\n i++;\n }\n return out;\n}\n"],"names":["rewriteFunctions","CALL_ANYWHERE","CALL_AT_START","isWordChar","ch","undefined","test","skipOpaque","sql","i","j","length","nl","indexOf","close","numberPlaceholders","out","n","skip","slice","readArgs","parenStart","args","depth","cur","push","trim","Error","end","sqlHas","fieldExpr","valueExpr","isArray","sqlBasename","pathExpr","suffixExpr","trimmed","name","rewrite","m","exec","toLowerCase","SenseError","rewrittenArgs","map"],"mappings":";;;;+BA+IgBA;;;eAAAA;;;wBA/IW;AAE3B,+FAA+F;AAC/F,sIAAsI;AAEtI,mGAAmG;AACnG,mGAAmG;AACnG,IAAMC,gBAAgB;AACtB,IAAMC,gBAAgB;AAEtB,SAASC,WAAWC,EAAsB;IACxC,OAAOA,OAAOC,aAAa,eAAeC,IAAI,CAACF;AACjD;AAEA,2FAA2F;AAC3F,qGAAqG;AACrG,SAASG,WAAWC,GAAW,EAAEC,CAAS;IACxC,IAAML,KAAKI,GAAG,CAACC,EAAE;IACjB,IAAIL,OAAO,OAAOA,OAAO,KAAK;QAC5B,IAAIM,IAAID,IAAI;QACZ,MAAOC,IAAIF,IAAIG,MAAM,CAAE;YACrB,IAAIH,GAAG,CAACE,EAAE,KAAKN,IAAI;gBACjB,IAAII,GAAG,CAACE,IAAI,EAAE,KAAKN,IAAI;oBACrBM,KAAK;oBACL;gBACF;gBACA,OAAOA,IAAI;YACb;YACAA;QACF;QACA,OAAOA,GAAG,gEAAgE;IAC5E;IACA,IAAIN,OAAO,OAAOI,GAAG,CAACC,IAAI,EAAE,KAAK,KAAK;QACpC,IAAMG,KAAKJ,IAAIK,OAAO,CAAC,MAAMJ;QAC7B,OAAOG,OAAO,CAAC,IAAIJ,IAAIG,MAAM,GAAGC,KAAK;IACvC;IACA,IAAIR,OAAO,OAAOI,GAAG,CAACC,IAAI,EAAE,KAAK,KAAK;QACpC,IAAMK,QAAQN,IAAIK,OAAO,CAAC,MAAMJ,IAAI;QACpC,OAAOK,UAAU,CAAC,IAAIN,IAAIG,MAAM,GAAGG,QAAQ;IAC7C;IACA,OAAO;AACT;AAEA,gGAAgG;AAChG,6GAA6G;AAC7G,SAASC,mBAAmBP,GAAW;IACrC,IAAIQ,MAAM;IACV,IAAIP,IAAI;IACR,IAAIQ,IAAI;IACR,MAAOR,IAAID,IAAIG,MAAM,CAAE;YAOeH;QANpC,IAAMU,OAAOX,WAAWC,KAAKC;QAC7B,IAAIS,SAAS,MAAM;YACjBF,OAAOR,IAAIW,KAAK,CAACV,GAAGS;YACpBT,IAAIS;YACJ;QACF;QACA,IAAIV,GAAG,CAACC,EAAE,KAAK,OAAO,CAAC,QAAQH,IAAI,EAACE,QAAAA,GAAG,CAACC,IAAI,EAAE,cAAVD,mBAAAA,QAAc,KAAK;YACrDS;YACAD,OAAO,AAAC,IAAK,OAAFC;YACXR;YACA;QACF;QACAO,OAAOR,GAAG,CAACC,EAAE;QACbA;IACF;IACA,OAAOO;AACT;AAEA,+FAA+F;AAC/F,2GAA2G;AAC3G,SAASI,SAASZ,GAAW,EAAEa,UAAkB;IAC/C,IAAMC,OAAiB,EAAE;IACzB,IAAIC,QAAQ;IACZ,IAAIC,MAAM;IACV,IAAIf,IAAIY,aAAa;IACrB,MAAOZ,IAAID,IAAIG,MAAM,CAAE;QACrB,IAAMO,OAAOX,WAAWC,KAAKC;QAC7B,IAAIS,SAAS,MAAM;YACjBM,OAAOhB,IAAIW,KAAK,CAACV,GAAGS;YACpBT,IAAIS;YACJ;QACF;QACA,IAAMd,KAAKI,GAAG,CAACC,EAAE;QACjB,IAAIL,OAAO,KAAK;YACdmB;YACAC,OAAOpB;YACPK;YACA;QACF;QACA,IAAIL,OAAO,KAAK;YACdmB;YACAd;YACA,IAAIc,UAAU,GAAG;YACjBC,OAAOpB;YACP;QACF;QACA,IAAIA,OAAO,OAAOmB,UAAU,GAAG;YAC7BD,KAAKG,IAAI,CAACD,IAAIE,IAAI;YAClBF,MAAM;YACNf;YACA;QACF;QACAe,OAAOpB;QACPK;IACF;IACA,IAAIc,UAAU,GAAG,MAAM,IAAII,MAAM,AAAC,kDAAwF,OAAvCnB,IAAIW,KAAK,CAACE,YAAYA,aAAa,KAAI;IAC1H,IAAIG,IAAIE,IAAI,OAAO,IAAIJ,KAAKG,IAAI,CAACD,IAAIE,IAAI;IACzC,OAAO;QAAEJ,MAAAA;QAAMM,KAAKnB;IAAE;AACxB;AAEA,6FAA6F;AAC7F,qGAAqG;AACrG,SAASoB,OAAOC,SAAiB,EAAEC,SAAiB;IAClD,IAAMC,UAAU,AAAC,UAAgDF,OAAvCA,WAAU,+BAAuC,OAAVA,WAAU;IAC3E,+FAA+F;IAC/F,sIAAsI;IACtI,OAAO,AAAC,mBAECE,OADAF,WAAU,8BAE6BE,OADvCA,SAAQ,2DAC+CF,OAAhBE,SAAQ,UAA2ED,OAAnED,WAAU,2DACtDA,OAD+GC,WAAU,wCAC7FA,OAA5BD,WAAU,oBAA4B,OAAVC,WAAU;AAE5D;AAEA,iGAAiG;AACjG,2GAA2G;AAC3G,SAASE,YAAYC,QAAgB,EAAEC,UAAkB;IACvD,IAAMC,UAAU,AAAC,cAAsB,OAATF,UAAS;IACvC,IAAMG,OAAO,AAAC,oBAEGD,OADNA,SAAQ,oCACwBA,OAA1BA,SAAQ,oBACSA,OADSA,SAAQ,wCACaA,OAA9BA,SAAQ,wBAA8B,OAARA,SAAQ;IAExE,OAAO,AAAC,cACOD,OADMD,UAAS,4CACoBC,OAAnCA,YAAW,0BAA8DA,OAAtCA,YAAW,6BAAsDE,OAA3BF,YAAW,kBAC7EE,OAD6FA,MAAK,6BAC3EF,OAAvBE,MAAK,oBAAkDF,OAAhCA,YAAW,uBACtCE,OAD2DF,YAAW,iCACjDE,OAArBA,MAAK,kBAAwCF,OAAxBE,MAAK,qBACjCA,OADoDF,YAAW,4BAC1D,OAALE,MAAK;AAElB;AAIO,SAASrC,iBAAiBQ,GAAW;IAC1C,IAAI,CAACP,cAAcK,IAAI,CAACE,MAAM,OAAOA;IACrC,gGAAgG;IAChG,mFAAmF;IACnF,OAAO8B,QAAQvB,mBAAmBP;AACpC;AAEA,SAAS8B,QAAQ9B,GAAW;IAC1B,IAAIQ,MAAM;IACV,IAAIP,IAAI;IACR,MAAOA,IAAID,IAAIG,MAAM,CAAE;QACrB,IAAMO,OAAOX,WAAWC,KAAKC;QAC7B,IAAIS,SAAS,MAAM;YACjBF,OAAOR,IAAIW,KAAK,CAACV,GAAGS;YACpBT,IAAIS;YACJ;QACF;QACA,IAAMqB,IAAI,CAACpC,WAAWK,GAAG,CAACC,IAAI,EAAE,IAAIP,cAAcsC,IAAI,CAAChC,IAAIW,KAAK,CAACV,MAAM;QACvE,IAAI8B,GAAG;YACL,IAAMF,OAAOE,CAAC,CAAC,EAAE,CAACE,WAAW;YAC7B,IAAMpB,aAAaZ,IAAI8B,CAAC,CAAC,EAAE,CAAC5B,MAAM,GAAG;YACrC,IAAI0B,SAAS,WAAW;gBACtB,MAAM,IAAIK,oBAAU,CAAC,4BAA4B;YACnD;YACA,IAAsBtB,YAAAA,SAASZ,KAAKa,aAA5BC,OAAcF,UAAdE,MAAMM,MAAQR,UAARQ;YACd,IAAMe,gBAAgBrB,KAAKsB,GAAG,CAACN;YAC/B,IAAID,SAAS,OAAO;gBAClB,IAAIM,cAAchC,MAAM,KAAK,GAAG,MAAM,IAAIgB,MAAM,AAAC,8DAAkF,OAArBgB,cAAchC,MAAM;gBAClIK,OAAOa,OAAOc,aAAa,CAAC,EAAE,EAAEA,aAAa,CAAC,EAAE;YAClD,OAAO;oBAEgCA;gBADrC,IAAIA,cAAchC,MAAM,GAAG,KAAKgC,cAAchC,MAAM,GAAG,GAAG,MAAM,IAAIgB,MAAM,AAAC,iEAAqF,OAArBgB,cAAchC,MAAM;gBAC/JK,OAAOiB,YAAYU,aAAa,CAAC,EAAE,GAAEA,kBAAAA,aAAa,CAAC,EAAE,cAAhBA,6BAAAA,kBAAoB;YAC3D;YACAlC,IAAImB;YACJ;QACF;QACAZ,OAAOR,GAAG,CAACC,EAAE;QACbA;IACF;IACA,OAAOO;AACT"}
|
|
1
|
+
{"version":3,"sources":["/Users/kevin/Dev/open-source/ai/sensemaking/src/store/turso/sql-functions.ts"],"sourcesContent":["import { SenseError } from '../../errors.ts';\n\n// Turso can't register UDFs: has()/basename() are rewritten into portable SQL below; segment()\n// (Intl.Segmenter grapheme clustering, types.ts's 'segment' capability) has none, so it's rejected like an unsupported FTS5 operator.\n\n// Non-global: recursion (readArgs' contents rewritten via rewriteFunctions) reenters this module's\n// regexes before the outer scan resumes, and a global regex's lastIndex does not nest across that.\nconst CALL_ANYWHERE = /\\b(has|basename|segment)\\s*\\(/i;\nconst CALL_AT_START = /^(has|basename|segment)\\s*\\(/i;\n\nfunction isWordChar(ch: string | undefined): boolean {\n return ch !== undefined && /[A-Za-z0-9_]/.test(ch);\n}\n\n// The index just past the quoted literal or SQL comment starting at i, or null if i starts\n// neither. Shared by the scan and readArgs so both leave quoted/comment text opaque to a call match.\nfunction skipOpaque(sql: string, i: number): number | null {\n const ch = sql[i];\n if (ch === \"'\" || ch === '\"') {\n let j = i + 1;\n while (j < sql.length) {\n if (sql[j] === ch) {\n if (sql[j + 1] === ch) {\n j += 2;\n continue;\n }\n return j + 1;\n }\n j++;\n }\n return j; // unterminated literal: consume to end rather than loop forever\n }\n if (ch === '-' && sql[i + 1] === '-') {\n const nl = sql.indexOf('\\n', i);\n return nl === -1 ? sql.length : nl + 1;\n }\n if (ch === '/' && sql[i + 1] === '*') {\n const close = sql.indexOf('*/', i + 2);\n return close === -1 ? sql.length : close + 2;\n }\n return null;\n}\n\n// `?N`/named placeholders bind by number/name, so repeating one is safe; an anonymous `?` takes\n// the next slot per occurrence, so it's numbered here first -- sqlHas/sqlBasename duplicate their arguments.\nfunction numberPlaceholders(sql: string): string {\n let out = '';\n let i = 0;\n let n = 0;\n while (i < sql.length) {\n const skip = skipOpaque(sql, i);\n if (skip !== null) {\n out += sql.slice(i, skip);\n i = skip;\n continue;\n }\n if (sql[i] === '?' && !/[0-9]/.test(sql[i + 1] ?? '')) {\n n++;\n out += `?${n}`;\n i++;\n continue;\n }\n out += sql[i];\n i++;\n }\n return out;\n}\n\n// The un-rewritten argument texts of the call whose '(' sits at parenStart, and the index just\n// past its matching ')'. Paren depth is tracked so a nested call's commas/parens aren't mistaken for ours.\nfunction readArgs(sql: string, parenStart: number): { args: string[]; end: number } {\n const args: string[] = [];\n let depth = 1;\n let cur = '';\n let i = parenStart + 1;\n while (i < sql.length) {\n const skip = skipOpaque(sql, i);\n if (skip !== null) {\n cur += sql.slice(i, skip);\n i = skip;\n continue;\n }\n const ch = sql[i];\n if (ch === '(') {\n depth++;\n cur += ch;\n i++;\n continue;\n }\n if (ch === ')') {\n depth--;\n i++;\n if (depth === 0) break;\n cur += ch;\n continue;\n }\n if (ch === ',' && depth === 1) {\n args.push(cur.trim());\n cur = '';\n i++;\n continue;\n }\n cur += ch;\n i++;\n }\n if (depth !== 0) throw new Error(`turso: unterminated argument list starting at \"${sql.slice(parenStart, parenStart + 30)}...\"`);\n if (cur.trim() !== '') args.push(cur.trim());\n return { args, end: i };\n}\n\n// field.startsWith('[') && JSON.parse(field) is an array, matched by membership; every other\n// field, stringified, matched by substring (sql-functions.ts's hasImpl, mirrored branch for branch).\nfunction sqlHas(fieldExpr: string, valueExpr: string): string {\n const isArray = `substr(${fieldExpr},1,1) = '[' AND json_valid(${fieldExpr})`;\n // json_each's argument is re-guarded by its own CASE (falling back to '[]'): turso evaluates a\n // CASE branch's table function even when untaken, and a non-array argument there raises rather than yielding 0 rows (spike-verified).\n return `(CASE\n WHEN ${fieldExpr} IS NULL THEN 0\n WHEN ${isArray} THEN\n (SELECT COUNT(*) FROM json_each(CASE WHEN ${isArray} THEN ${fieldExpr} ELSE '[]' END) je WHERE CAST(je.value AS TEXT) = CAST(${valueExpr} AS TEXT)) > 0\n ELSE instr(CAST(${fieldExpr} AS TEXT), CAST(${valueExpr} AS TEXT)) > 0\n END)`;\n}\n\n// posix.basename(String(path)), suffix stripped Unix-style (sql-functions.ts's basenameImpl). No\n// last-'/' primitive exists, so reverse() finds the first '/' from the end, then reverse() restores order.\nfunction sqlBasename(pathExpr: string, suffixExpr: string): string {\n const trimmed = `rtrim(CAST(${pathExpr} AS TEXT), '/')`;\n const name = `CASE\n WHEN ${trimmed} = '' THEN ''\n WHEN instr(${trimmed}, '/') = 0 THEN ${trimmed}\n ELSE reverse(substr(reverse(${trimmed}), 1, instr(reverse(${trimmed}), '/') - 1))\n END`;\n return `(CASE WHEN ${pathExpr} IS NULL THEN NULL ELSE\n (CASE WHEN ${suffixExpr} IS NOT NULL AND CAST(${suffixExpr} AS TEXT) != '' AND CAST(${suffixExpr} AS TEXT) != (${name})\n AND substr((${name}), -length(CAST(${suffixExpr} AS TEXT))) = CAST(${suffixExpr} AS TEXT)\n THEN substr((${name}), 1, length((${name})) - length(CAST(${suffixExpr} AS TEXT)))\n ELSE (${name}) END)\n END)`;\n}\n\n// Rewrites every has()/basename() call into the portable SQL above, recursing into each call's own\n// arguments first so a nested `instr(basename(f.path), 'x')` resolves inside out; segment() instead raises STORE_CAPABILITY_MISSING, named, like an unsupported FTS5 operator.\nexport function rewriteFunctions(sql: string): string {\n if (!CALL_ANYWHERE.test(sql)) return sql;\n // Numbered once, over the whole statement, before any argument is duplicated below -- rewrite()\n // recurses over already-numbered substrings, never renumbering them a second time.\n return rewrite(numberPlaceholders(sql));\n}\n\nfunction rewrite(sql: string): string {\n let out = '';\n let i = 0;\n while (i < sql.length) {\n const skip = skipOpaque(sql, i);\n if (skip !== null) {\n out += sql.slice(i, skip);\n i = skip;\n continue;\n }\n const m = !isWordChar(sql[i - 1]) ? CALL_AT_START.exec(sql.slice(i)) : null;\n if (m) {\n const name = m[1].toLowerCase();\n const parenStart = i + m[0].length - 1;\n if (name === 'segment') {\n throw new SenseError('STORE_CAPABILITY_MISSING', `store \"turso\" does not implement segment() as a SQL function in this build (its client cannot register UDFs); rephrase the query without it, or set \"store\" to \"sqlite\" in this tree's config to run it as written`);\n }\n const { args, end } = readArgs(sql, parenStart);\n const rewrittenArgs = args.map(rewrite);\n if (name === 'has') {\n if (rewrittenArgs.length !== 2) throw new Error(`turso: has() takes exactly 2 arguments (field, value), got ${rewrittenArgs.length}`);\n out += sqlHas(rewrittenArgs[0], rewrittenArgs[1]);\n } else {\n if (rewrittenArgs.length < 1 || rewrittenArgs.length > 2) throw new Error(`turso: basename() takes 1 or 2 arguments (path, suffix?), got ${rewrittenArgs.length}`);\n out += sqlBasename(rewrittenArgs[0], rewrittenArgs[1] ?? 'NULL');\n }\n i = end;\n continue;\n }\n out += sql[i];\n i++;\n }\n return out;\n}\n"],"names":["rewriteFunctions","CALL_ANYWHERE","CALL_AT_START","isWordChar","ch","undefined","test","skipOpaque","sql","i","j","length","nl","indexOf","close","numberPlaceholders","out","n","skip","slice","readArgs","parenStart","args","depth","cur","push","trim","Error","end","sqlHas","fieldExpr","valueExpr","isArray","sqlBasename","pathExpr","suffixExpr","trimmed","name","rewrite","m","exec","toLowerCase","SenseError","rewrittenArgs","map"],"mappings":";;;;+BA+IgBA;;;eAAAA;;;wBA/IW;AAE3B,+FAA+F;AAC/F,sIAAsI;AAEtI,mGAAmG;AACnG,mGAAmG;AACnG,IAAMC,gBAAgB;AACtB,IAAMC,gBAAgB;AAEtB,SAASC,WAAWC,EAAsB;IACxC,OAAOA,OAAOC,aAAa,eAAeC,IAAI,CAACF;AACjD;AAEA,2FAA2F;AAC3F,qGAAqG;AACrG,SAASG,WAAWC,GAAW,EAAEC,CAAS;IACxC,IAAML,KAAKI,GAAG,CAACC,EAAE;IACjB,IAAIL,OAAO,OAAOA,OAAO,KAAK;QAC5B,IAAIM,IAAID,IAAI;QACZ,MAAOC,IAAIF,IAAIG,MAAM,CAAE;YACrB,IAAIH,GAAG,CAACE,EAAE,KAAKN,IAAI;gBACjB,IAAII,GAAG,CAACE,IAAI,EAAE,KAAKN,IAAI;oBACrBM,KAAK;oBACL;gBACF;gBACA,OAAOA,IAAI;YACb;YACAA;QACF;QACA,OAAOA,GAAG,gEAAgE;IAC5E;IACA,IAAIN,OAAO,OAAOI,GAAG,CAACC,IAAI,EAAE,KAAK,KAAK;QACpC,IAAMG,KAAKJ,IAAIK,OAAO,CAAC,MAAMJ;QAC7B,OAAOG,OAAO,CAAC,IAAIJ,IAAIG,MAAM,GAAGC,KAAK;IACvC;IACA,IAAIR,OAAO,OAAOI,GAAG,CAACC,IAAI,EAAE,KAAK,KAAK;QACpC,IAAMK,QAAQN,IAAIK,OAAO,CAAC,MAAMJ,IAAI;QACpC,OAAOK,UAAU,CAAC,IAAIN,IAAIG,MAAM,GAAGG,QAAQ;IAC7C;IACA,OAAO;AACT;AAEA,gGAAgG;AAChG,6GAA6G;AAC7G,SAASC,mBAAmBP,GAAW;IACrC,IAAIQ,MAAM;IACV,IAAIP,IAAI;IACR,IAAIQ,IAAI;IACR,MAAOR,IAAID,IAAIG,MAAM,CAAE;YAOeH;QANpC,IAAMU,OAAOX,WAAWC,KAAKC;QAC7B,IAAIS,SAAS,MAAM;YACjBF,OAAOR,IAAIW,KAAK,CAACV,GAAGS;YACpBT,IAAIS;YACJ;QACF;QACA,IAAIV,GAAG,CAACC,EAAE,KAAK,OAAO,CAAC,QAAQH,IAAI,EAACE,QAAAA,GAAG,CAACC,IAAI,EAAE,cAAVD,mBAAAA,QAAc,KAAK;YACrDS;YACAD,OAAO,AAAC,IAAK,OAAFC;YACXR;YACA;QACF;QACAO,OAAOR,GAAG,CAACC,EAAE;QACbA;IACF;IACA,OAAOO;AACT;AAEA,+FAA+F;AAC/F,2GAA2G;AAC3G,SAASI,SAASZ,GAAW,EAAEa,UAAkB;IAC/C,IAAMC,OAAiB,EAAE;IACzB,IAAIC,QAAQ;IACZ,IAAIC,MAAM;IACV,IAAIf,IAAIY,aAAa;IACrB,MAAOZ,IAAID,IAAIG,MAAM,CAAE;QACrB,IAAMO,OAAOX,WAAWC,KAAKC;QAC7B,IAAIS,SAAS,MAAM;YACjBM,OAAOhB,IAAIW,KAAK,CAACV,GAAGS;YACpBT,IAAIS;YACJ;QACF;QACA,IAAMd,KAAKI,GAAG,CAACC,EAAE;QACjB,IAAIL,OAAO,KAAK;YACdmB;YACAC,OAAOpB;YACPK;YACA;QACF;QACA,IAAIL,OAAO,KAAK;YACdmB;YACAd;YACA,IAAIc,UAAU,GAAG;YACjBC,OAAOpB;YACP;QACF;QACA,IAAIA,OAAO,OAAOmB,UAAU,GAAG;YAC7BD,KAAKG,IAAI,CAACD,IAAIE,IAAI;YAClBF,MAAM;YACNf;YACA;QACF;QACAe,OAAOpB;QACPK;IACF;IACA,IAAIc,UAAU,GAAG,MAAM,IAAII,MAAM,AAAC,kDAAwF,OAAvCnB,IAAIW,KAAK,CAACE,YAAYA,aAAa,KAAI;IAC1H,IAAIG,IAAIE,IAAI,OAAO,IAAIJ,KAAKG,IAAI,CAACD,IAAIE,IAAI;IACzC,OAAO;QAAEJ,MAAAA;QAAMM,KAAKnB;IAAE;AACxB;AAEA,6FAA6F;AAC7F,qGAAqG;AACrG,SAASoB,OAAOC,SAAiB,EAAEC,SAAiB;IAClD,IAAMC,UAAU,AAAC,UAAgDF,OAAvCA,WAAU,+BAAuC,OAAVA,WAAU;IAC3E,+FAA+F;IAC/F,sIAAsI;IACtI,OAAO,AAAC,mBAECE,OADAF,WAAU,8BAE6BE,OADvCA,SAAQ,2DAC+CF,OAAhBE,SAAQ,UAA2ED,OAAnED,WAAU,2DACtDA,OAD+GC,WAAU,wCAC7FA,OAA5BD,WAAU,oBAA4B,OAAVC,WAAU;AAE5D;AAEA,iGAAiG;AACjG,2GAA2G;AAC3G,SAASE,YAAYC,QAAgB,EAAEC,UAAkB;IACvD,IAAMC,UAAU,AAAC,cAAsB,OAATF,UAAS;IACvC,IAAMG,OAAO,AAAC,oBAEGD,OADNA,SAAQ,oCACwBA,OAA1BA,SAAQ,oBACSA,OADSA,SAAQ,wCACaA,OAA9BA,SAAQ,wBAA8B,OAARA,SAAQ;IAExE,OAAO,AAAC,cACOD,OADMD,UAAS,4CACoBC,OAAnCA,YAAW,0BAA8DA,OAAtCA,YAAW,6BAAsDE,OAA3BF,YAAW,kBAC7EE,OAD6FA,MAAK,6BAC3EF,OAAvBE,MAAK,oBAAkDF,OAAhCA,YAAW,uBACtCE,OAD2DF,YAAW,iCACjDE,OAArBA,MAAK,kBAAwCF,OAAxBE,MAAK,qBACjCA,OADoDF,YAAW,4BAC1D,OAALE,MAAK;AAElB;AAIO,SAASrC,iBAAiBQ,GAAW;IAC1C,IAAI,CAACP,cAAcK,IAAI,CAACE,MAAM,OAAOA;IACrC,gGAAgG;IAChG,mFAAmF;IACnF,OAAO8B,QAAQvB,mBAAmBP;AACpC;AAEA,SAAS8B,QAAQ9B,GAAW;IAC1B,IAAIQ,MAAM;IACV,IAAIP,IAAI;IACR,MAAOA,IAAID,IAAIG,MAAM,CAAE;QACrB,IAAMO,OAAOX,WAAWC,KAAKC;QAC7B,IAAIS,SAAS,MAAM;YACjBF,OAAOR,IAAIW,KAAK,CAACV,GAAGS;YACpBT,IAAIS;YACJ;QACF;QACA,IAAMqB,IAAI,CAACpC,WAAWK,GAAG,CAACC,IAAI,EAAE,IAAIP,cAAcsC,IAAI,CAAChC,IAAIW,KAAK,CAACV,MAAM;QACvE,IAAI8B,GAAG;YACL,IAAMF,OAAOE,CAAC,CAAC,EAAE,CAACE,WAAW;YAC7B,IAAMpB,aAAaZ,IAAI8B,CAAC,CAAC,EAAE,CAAC5B,MAAM,GAAG;YACrC,IAAI0B,SAAS,WAAW;gBACtB,MAAM,IAAIK,oBAAU,CAAC,4BAA4B;YACnD;YACA,IAAsBtB,YAAAA,SAASZ,KAAKa,aAA5BC,OAAcF,UAAdE,MAAMM,MAAQR,UAARQ;YACd,IAAMe,gBAAgBrB,KAAKsB,GAAG,CAACN;YAC/B,IAAID,SAAS,OAAO;gBAClB,IAAIM,cAAchC,MAAM,KAAK,GAAG,MAAM,IAAIgB,MAAM,AAAC,8DAAkF,OAArBgB,cAAchC,MAAM;gBAClIK,OAAOa,OAAOc,aAAa,CAAC,EAAE,EAAEA,aAAa,CAAC,EAAE;YAClD,OAAO;oBAEgCA;gBADrC,IAAIA,cAAchC,MAAM,GAAG,KAAKgC,cAAchC,MAAM,GAAG,GAAG,MAAM,IAAIgB,MAAM,AAAC,iEAAqF,OAArBgB,cAAchC,MAAM;gBAC/JK,OAAOiB,YAAYU,aAAa,CAAC,EAAE,GAAEA,kBAAAA,aAAa,CAAC,EAAE,cAAhBA,6BAAAA,kBAAoB;YAC3D;YACAlC,IAAImB;YACJ;QACF;QACAZ,OAAOR,GAAG,CAACC,EAAE;QACbA;IACF;IACA,OAAOO;AACT"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"sources":["/Users/kevin/Dev/
|
|
1
|
+
{"version":3,"sources":["/Users/kevin/Dev/open-source/ai/sensemaking/src/store/turso/store.ts"],"sourcesContent":["import type { Database } from '@tursodatabase/database';\nimport { STORE_DIMS } from '../../embed/types.ts';\nimport { getColumns, getMeta, setMeta } from '../shared.ts';\nimport { withTransaction } from '../transaction.ts';\nimport type { Capability, Connection, Statement, Store, VectorStore } from '../types.ts';\nimport { createNativeVectorScopeOwner, hasVectorRow, pendingRows } from '../vectors.ts';\nimport { cacheFilePath, checkpointWal, fileSize, reclaimSpace } from './connection.ts';\nimport { fieldStats } from './fieldStats.ts';\nimport { queryLexical } from './lexical.ts';\nimport { scanCandidates, scanSimilar, writeVectorBatch } from './vectors.ts';\n\n// No 'segment': has()/basename() resolve via SQL rewrite below (sql-functions.ts), but segment() has no SQL form and the client cannot register it as a UDF.\nexport const CAPABILITIES: ReadonlySet<Capability> = new Set(['vectors']);\n\n// close() reclaims once the cache file outgrows its compact size by this factor: the disk\n// overhead a user should accept (PLAN 3.52), in postgres autovacuum scale-factor shape.\nconst BLOAT_FACTOR = 1.5;\n\n// Shares one Connection instance (conn) with the builder's own reconcile call so transaction depth\n// (see transaction.ts) is tracked against the same object everywhere.\nexport function createStore(db: Database, conn: Connection, options: { observational?: boolean } = {}): Store {\n const vectorScope = createNativeVectorScopeOwner();\n // The column's fixed DDL width (STORE_DIMS) is what every scan binds against, not the\n // interface's per-call storeDims -- see vectors.ts's padded() for why a shorter vector is still correct against a wider column.\n const vectors: VectorStore = {\n pending: () => pendingRows(conn),\n writeVectors: (rows) => writeVectorBatch(conn, STORE_DIMS, rows),\n candidates: (qv, _storeDims, fetch, allowed) => scanCandidates(conn, qv, STORE_DIMS, fetch, allowed, vectorScope.activeFor(allowed)),\n similar: (path, opts) => scanSimilar(conn, STORE_DIMS, path, opts, vectorScope.activeFor(opts.allowed)),\n hasVector: (path) => hasVectorRow(conn, path),\n };\n vectorScope.bind(vectors);\n\n return {\n name: 'turso',\n capabilities: CAPABILITIES,\n async exec(sql: string): Promise<void> {\n await conn.exec(sql);\n },\n async prepare(sql: string): Promise<Statement> {\n return conn.prepare(sql);\n },\n async runBatch(sql: string, paramRows: unknown[][]): Promise<void> {\n await conn.runBatch(sql, paramRows);\n },\n async transaction<T>(fn: () => Promise<T>): Promise<T> {\n return withTransaction(conn, fn);\n },\n docs: {\n async columns() {\n return [...(await getColumns(conn))];\n },\n fieldStats: (columns, scopeWhere) => fieldStats(conn, columns, scopeWhere),\n },\n lexical: {\n query: (terms, opts) => queryLexical(conn, terms, opts),\n },\n vectors,\n async engineStatus() {\n // Read back rather than recomputed: this is what open() actually set (3x the largest\n // recorded reconcile, floored at 30s, capped at 10min). Turso's PRAGMA busy_timeout names its column \"busy_timeout\" (spike-verified), not \"timeout\" like real SQLite.\n const row = (await (await conn.prepare('PRAGMA busy_timeout')).get()) as { busy_timeout: number };\n return { busy_timeout: `${row.busy_timeout}ms (derived: 3x the largest reconcile this cache has recorded, floored at 30000ms)` };\n },\n raw: {\n async prepare(sql: string) {\n const stmt = await conn.prepare(sql);\n stmt.setReadBigInts(true); // int64 past 2^53 arrives as BigInt instead of losing precision\n return {\n columns: () => stmt.columns(),\n // sense sql streams through the client's own async generator.\n iterate: async function* (...params: unknown[]) {\n yield* stmt.iterate(...params);\n },\n };\n },\n },\n async close() {\n if (options.observational) {\n await db.close();\n return;\n }\n // Checkpoint before measuring: the last reconcile's pages sit in the WAL, not the file yet.\n await checkpointWal(db);\n let reclaimPath: string | null = null;\n try {\n const path = await cacheFilePath(conn);\n const base = Number((await getMeta(conn, 'compact_size')) ?? '0');\n const size = path === null ? null : await fileSize(path);\n if (size !== null) {\n // No baseline yet (cold build or an upgraded cache): record it, never reclaim here, so a\n // cold build pays no VACUUM (PLAN 3.49 Track A) and an upgraded cache heals on a crossing.\n if (base === 0) {\n await setMeta(conn, 'compact_size', String(size));\n // The meta write landed past the checkpoint; leave no WAL for the next opener.\n await checkpointWal(db);\n } else if (size > BLOAT_FACTOR * base) {\n reclaimPath = path;\n }\n }\n } catch (err) {\n console.error(`sense: turso cache bloat check failed, the cache will keep its space until one succeeds: ${(err as Error).message}`);\n }\n await db.close();\n // After this store's connection is closed, never before: turso hands the file to one\n // connection at a time, and reclaimSpace opens its own to carry the `vacuum` flag (PLAN 3.54).\n if (reclaimPath !== null) await reclaimSpace(reclaimPath);\n },\n };\n}\n"],"names":["CAPABILITIES","createStore","Set","BLOAT_FACTOR","db","conn","options","vectorScope","createNativeVectorScopeOwner","vectors","pending","pendingRows","writeVectors","rows","writeVectorBatch","STORE_DIMS","candidates","qv","_storeDims","fetch","allowed","scanCandidates","activeFor","similar","path","opts","scanSimilar","hasVector","hasVectorRow","bind","name","capabilities","exec","sql","prepare","runBatch","paramRows","transaction","fn","withTransaction","docs","columns","getColumns","fieldStats","scopeWhere","lexical","query","terms","queryLexical","engineStatus","row","get","busy_timeout","raw","stmt","setReadBigInts","iterate","params","close","reclaimPath","base","size","err","observational","checkpointWal","cacheFilePath","getMeta","Number","fileSize","setMeta","String","console","error","message","reclaimSpace"],"mappings":";;;;;;;;;;;QAYaA;eAAAA;;QAQGC;eAAAA;;;uBAnBW;wBACkB;6BACb;yBAEwC;4BACH;4BAC1C;yBACE;0BACiC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAGvD,IAAMD,eAAwC,IAAIE,IAAI;IAAC;CAAU;AAExE,0FAA0F;AAC1F,wFAAwF;AACxF,IAAMC,eAAe;AAId,SAASF,YAAYG,EAAY,EAAEC,IAAgB;QAAEC,UAAAA,iEAAuC,CAAC;IAClG,IAAMC,cAAcC,IAAAA,uCAA4B;IAChD,sFAAsF;IACtF,gIAAgI;IAChI,IAAMC,UAAuB;QAC3BC,SAAS,SAATA;mBAAeC,IAAAA,sBAAW,EAACN;;QAC3BO,cAAc,SAAdA,aAAeC;mBAASC,IAAAA,4BAAgB,EAACT,MAAMU,mBAAU,EAAEF;;QAC3DG,YAAY,SAAZA,WAAaC,IAAIC,YAAYC,OAAOC;mBAAYC,IAAAA,0BAAc,EAAChB,MAAMY,IAAIF,mBAAU,EAAEI,OAAOC,SAASb,YAAYe,SAAS,CAACF;;QAC3HG,SAAS,SAATA,QAAUC,MAAMC;mBAASC,IAAAA,uBAAW,EAACrB,MAAMU,mBAAU,EAAES,MAAMC,MAAMlB,YAAYe,SAAS,CAACG,KAAKL,OAAO;;QACrGO,WAAW,SAAXA,UAAYH;mBAASI,IAAAA,uBAAY,EAACvB,MAAMmB;;IAC1C;IACAjB,YAAYsB,IAAI,CAACpB;IAEjB,OAAO;QACLqB,MAAM;QACNC,cAAc/B;QACRgC,MAAN,SAAMA,KAAKC,GAAW;;;;;4BACpB;;gCAAM5B,KAAK2B,IAAI,CAACC;;;4BAAhB;;;;;;YACF;;QACMC,SAAN,SAAMA,QAAQD,GAAW;;;oBACvB;;wBAAO5B,KAAK6B,OAAO,CAACD;;;YACtB;;QACME,UAAN,SAAMA,SAASF,GAAW,EAAEG,SAAsB;;;;;4BAChD;;gCAAM/B,KAAK8B,QAAQ,CAACF,KAAKG;;;4BAAzB;;;;;;YACF;;QACMC,aAAN,SAAMA,YAAeC,EAAoB;;;oBACvC;;wBAAOC,IAAAA,8BAAe,EAAClC,MAAMiC;;;YAC/B;;QACAE,MAAM;YACEC,SAAN,SAAMA;;;;;gCACQ;;oCAAMC,IAAAA,oBAAU,EAACrC;;;gCAA7B;;oCAAQ;wCAAI;;;;;gBACd;;YACAsC,YAAY,SAAZA,WAAaF,SAASG;uBAAeD,IAAAA,wBAAU,EAACtC,MAAMoC,SAASG;;QACjE;QACAC,SAAS;YACPC,OAAO,SAAPA,MAAQC,OAAOtB;uBAASuB,IAAAA,uBAAY,EAAC3C,MAAM0C,OAAOtB;;QACpD;QACAhB,SAAAA;QACMwC,cAAN,SAAMA;;oBAGEC;;;;4BAAc;;gCAAM7C,KAAK6B,OAAO,CAAC;;;4BAA1B;;gCAAO,cAA2CiB,GAAG;;;4BAA5DD,MAAO;4BACb;;gCAAO;oCAAEE,cAAc,AAAC,GAAmB,OAAjBF,IAAIE,YAAY,EAAC;gCAAoF;;;;YACjI;;QACAC,KAAK;YACGnB,SAAN,SAAMA,QAAQD,GAAW;;wBACjBqB;;;;gCAAO;;oCAAMjD,KAAK6B,OAAO,CAACD;;;gCAA1BqB,OAAO;gCACbA,KAAKC,cAAc,CAAC,OAAO,gEAAgE;gCAC3F;;oCAAO;wCACLd,SAAS,SAATA;mDAAea,KAAKb,OAAO;;wCAC3B,8DAA8D;wCAC9De,SAAS,SAATA;4CAA0B,IAAA,IAAA,OAAA,UAAA,QAAA,AAAGC,SAAH,UAAA,OAAA,OAAA,GAAA,OAAA,MAAA;gDAAGA,OAAH,QAAA,SAAA,CAAA,KAAoB;;;oDACrCH;;;;4DAAP;;2EAAA,0CAAOA,CAAAA,QAAAA,MAAKE,OAAO,OAAZF,OAAa,qBAAGG;;;4DAAvB;;;;;;4CACF;;oCACF;;;;gBACF;;QACF;QACMC,OAAN,SAAMA;;oBAOAC,aAGmB,MADfnC,MACAoC,MACAC,YAYCC;;;;iCAtBLxD,QAAQyD,aAAa,EAArBzD;;;;4BACF;;gCAAMF,GAAGsD,KAAK;;;4BAAd;4BACA;;;;4BAEF,4FAA4F;4BAC5F;;gCAAMM,IAAAA,2BAAa,EAAC5D;;;4BAApB;4BACIuD,cAA6B;;;;;;;;;4BAElB;;gCAAMM,IAAAA,2BAAa,EAAC5D;;;4BAA3BmB,OAAO;4BACQ;;gCAAM0C,IAAAA,iBAAO,EAAC7D,MAAM;;;4BAAnCuD,OAAOO;iCAAQ,OAAA,2BAAA,kBAAA,OAAwC;;iCAChD3C,CAAAA,SAAS,IAAG,GAAZA;;;;mCAAgB;;;;;;4BAAO;;gCAAM4C,IAAAA,sBAAQ,EAAC5C;;;mCAAf;;;4BAA9BqC;iCACFA,CAAAA,SAAS,IAAG,GAAZA;;;;iCAGED,CAAAA,SAAS,CAAA,GAATA;;;;4BACF;;gCAAMS,IAAAA,iBAAO,EAAChE,MAAM,gBAAgBiE,OAAOT;;;4BAA3C;4BACA,+EAA+E;4BAC/E;;gCAAMG,IAAAA,2BAAa,EAAC5D;;;4BAApB;;;;;;4BACK,IAAIyD,OAAO1D,eAAeyD,MAAM;gCACrCD,cAAcnC;4BAChB;;;;;;;;4BAEKsC;4BACPS,QAAQC,KAAK,CAAC,AAAC,4FAAkH,OAAvB,AAACV,IAAcW,OAAO;;;;;;4BAElI;;gCAAMrE,GAAGsD,KAAK;;;4BAAd;iCAGIC,CAAAA,gBAAgB,IAAG,GAAnBA;;;;4BAAsB;;gCAAMe,IAAAA,0BAAY,EAACf;;;4BAAnB;;;;;;;;YAC5B;;IACF;AACF"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/turso/vectors.ts"],"sourcesContent":["import type { Connection, VectorCandidate, VectorSimilar, VectorWriteRow } from '../types.ts';\nimport { asCosine, compareVectorScores, NATIVE_VECTOR_SCOPE_TABLE, sampleEvenly } from '../vectors.ts';\n\n// Native F32_BLOB(dims) columns, dims fixed at DDL time (open.ts, STORE_DIMS). Both scans\n// score in JS: a per-row vector_distance_cos measured 3.5-8x slower on this row engine.\n\n// A model narrower than the column zero-pads: the added dimensions contribute nothing to\n// either dot product or norm, so cosine is unchanged.\nfunction padded(values: ArrayLike<number>, dims: number): Float32Array {\n const out = new Float32Array(dims);\n for (let d = 0; d < Math.min(values.length, dims); d++) out[d] = values[d];\n return out;\n}\n\n// A bound Float32Array stores truncated bytes with no error; a Buffer view over the same\n// bytes round-trips exact (measured 0.7.2).\nfunction toBlob(v: Float32Array): Buffer {\n return Buffer.from(v.buffer, v.byteOffset, v.byteLength);\n}\n\nfunction decode(vector: Buffer, dims: number): Float32Array {\n return new Float32Array(vector.buffer, vector.byteOffset, dims);\n}\n\n// Rows arrive int8-quantized with a per-vector scale (embed/query.ts's toStore); this store\n// dequantizes to floats, as duckdb does.\nfunction dequantize(row: VectorWriteRow, dims: number): Float32Array {\n const q = new Int8Array(row.vector.buffer, row.vector.byteOffset, row.vector.byteLength);\n const out = new Float32Array(dims);\n for (let d = 0; d < Math.min(q.length, dims); d++) out[d] = q[d] * row.scale;\n return out;\n}\n\nfunction vectorNorm(vector: Float32Array): number {\n let norm = 0;\n for (let d = 0; d < vector.length; d++) norm += vector[d] * vector[d];\n return Math.sqrt(norm);\n}\n\nfunction cosineSimilarity(a: Float32Array, b: Float32Array, aNorm: number, bNorm: number): number {\n let dot = 0;\n for (let d = 0; d < a.length; d++) dot += a[d] * b[d];\n return dot / (aNorm * bNorm + 1e-32);\n}\n\n// `scale` stays NULL: the column exists so the row shape matches the other stores, and\n// cosine is scale-invariant once the value is dequantized here.\nexport async function writeVectorBatch(conn: Connection, dims: number, rows: VectorWriteRow[]): Promise<void> {\n if (rows.length === 0) return;\n await conn.runBatch(\n 'UPDATE embeddings SET vector = ? WHERE \"path\" = ? AND chunk = ?',\n rows.map((row) => [toBlob(dequantize(row, dims)), row.path, row.chunk])\n );\n}\n\n// Best chunk per file by cosine. GROUP BY with MIN() cannot do this on 0.7.2: the aggregate\n// is right but the bare start_line/end_line come from an arbitrary row in the group.\nexport async function scanCandidates(conn: Connection, qv: Float32Array, dims: number, fetch: number, allowed?: Set<string>, nativeScope = false): Promise<VectorCandidate[]> {\n const q = padded(qv, dims);\n const queryNorm = vectorNorm(q);\n const scope = nativeScope ? ` AND \"path\" IN (SELECT \"path\" FROM ${NATIVE_VECTOR_SCOPE_TABLE})` : '';\n const stmt = await conn.prepare(`SELECT \"path\", chunk, start_line, end_line, vector FROM embeddings WHERE vector IS NOT NULL${scope} ORDER BY \"path\", chunk`);\n const rows = (await stmt.all()) as Array<{ path: string; chunk: number; start_line: number; end_line: number; vector: Buffer }>;\n\n const best = new Map<string, { score: number; chunk: number; lines: string }>();\n for (const row of rows) {\n if (allowed && !allowed.has(row.path)) continue;\n const vector = decode(row.vector, dims);\n const score = cosineSimilarity(q, vector, queryNorm, vectorNorm(vector));\n const existing = best.get(row.path);\n if (!existing || score > existing.score || (score === existing.score && row.chunk < existing.chunk)) {\n best.set(row.path, { score, chunk: row.chunk, lines: `L${row.start_line}-${row.end_line}` });\n }\n }\n return [...best.entries()]\n .map(([path, value]) => ({ path, ...value }))\n .sort(compareVectorScores)\n .slice(0, fetch)\n .map(({ path, lines, score }) => ({ path, lines, similarity: asCosine(score) }));\n}\n\n// Max cosine over (target chunk, other chunk) pairs. The SQL cross-join duckdb uses measured\n// 6-8x slower here at 26k chunks (2026-08-30).\nexport async function scanSimilar(conn: Connection, dims: number, path: string, opts: { exclude: Set<string>; allowed?: Set<string>; k: number }, nativeScope = false): Promise<VectorSimilar[]> {\n if (opts.allowed && opts.allowed.size === 0) return [];\n const targetStmt = await conn.prepare('SELECT vector FROM embeddings WHERE \"path\" = ? AND vector IS NOT NULL ORDER BY chunk');\n const targetRows = (await targetStmt.all(path)) as Array<{ vector: Buffer }>;\n if (targetRows.length === 0) return [];\n const targets = sampleEvenly(targetRows).map((row) => {\n const vector = decode(row.vector, dims);\n return { vector, norm: vectorNorm(vector) };\n });\n\n const scope = nativeScope ? ` AND \"path\" IN (SELECT \"path\" FROM ${NATIVE_VECTOR_SCOPE_TABLE})` : '';\n const stmt = await conn.prepare(`SELECT \"path\", vector FROM embeddings WHERE vector IS NOT NULL${scope}`);\n const rows = (await stmt.all()) as Array<{ path: string; vector: Buffer }>;\n const best = new Map<string, number>();\n for (const row of rows) {\n if (row.path === path || opts.exclude.has(row.path) || (opts.allowed && !opts.allowed.has(row.path))) continue;\n const other = decode(row.vector, dims);\n const otherNorm = vectorNorm(other);\n for (const t of targets) {\n const score = cosineSimilarity(t.vector, other, t.norm, otherNorm);\n const existing = best.get(row.path);\n if (existing === undefined || score > existing) best.set(row.path, score);\n }\n }\n\n return [...best.entries()]\n .map(([p, score]) => ({ path: p, score }))\n .sort(compareVectorScores)\n .slice(0, opts.k)\n .map(({ path: p, score }) => ({ path: p, similarity: asCosine(score) }));\n}\n"],"names":["scanCandidates","scanSimilar","writeVectorBatch","padded","values","dims","out","Float32Array","d","Math","min","length","toBlob","v","Buffer","from","buffer","byteOffset","byteLength","decode","vector","dequantize","row","q","Int8Array","scale","vectorNorm","norm","sqrt","cosineSimilarity","a","b","aNorm","bNorm","dot","conn","rows","runBatch","map","path","chunk","qv","fetch","allowed","nativeScope","queryNorm","scope","stmt","best","score","existing","NATIVE_VECTOR_SCOPE_TABLE","prepare","all","Map","has","get","set","lines","start_line","end_line","entries","value","sort","compareVectorScores","slice","similarity","asCosine","opts","targetStmt","targetRows","targets","other","otherNorm","t","size","sampleEvenly","exclude","undefined","p","k"],"mappings":";;;;;;;;;;;QAyDsBA;eAAAA;;QA0BAC;eAAAA;;QApCAC;eAAAA;;;yBA9CiE;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAEvF,0FAA0F;AAC1F,wFAAwF;AAExF,yFAAyF;AACzF,sDAAsD;AACtD,SAASC,OAAOC,MAAyB,EAAEC,IAAY;IACrD,IAAMC,MAAM,IAAIC,aAAaF;IAC7B,IAAK,IAAIG,IAAI,GAAGA,IAAIC,KAAKC,GAAG,CAACN,OAAOO,MAAM,EAAEN,OAAOG,IAAKF,GAAG,CAACE,EAAE,GAAGJ,MAAM,CAACI,EAAE;IAC1E,OAAOF;AACT;AAEA,yFAAyF;AACzF,4CAA4C;AAC5C,SAASM,OAAOC,CAAe;IAC7B,OAAOC,OAAOC,IAAI,CAACF,EAAEG,MAAM,EAAEH,EAAEI,UAAU,EAAEJ,EAAEK,UAAU;AACzD;AAEA,SAASC,OAAOC,MAAc,EAAEf,IAAY;IAC1C,OAAO,IAAIE,aAAaa,OAAOJ,MAAM,EAAEI,OAAOH,UAAU,EAAEZ;AAC5D;AAEA,4FAA4F;AAC5F,yCAAyC;AACzC,SAASgB,WAAWC,GAAmB,EAAEjB,IAAY;IACnD,IAAMkB,IAAI,IAAIC,UAAUF,IAAIF,MAAM,CAACJ,MAAM,EAAEM,IAAIF,MAAM,CAACH,UAAU,EAAEK,IAAIF,MAAM,CAACF,UAAU;IACvF,IAAMZ,MAAM,IAAIC,aAAaF;IAC7B,IAAK,IAAIG,IAAI,GAAGA,IAAIC,KAAKC,GAAG,CAACa,EAAEZ,MAAM,EAAEN,OAAOG,IAAKF,GAAG,CAACE,EAAE,GAAGe,CAAC,CAACf,EAAE,GAAGc,IAAIG,KAAK;IAC5E,OAAOnB;AACT;AAEA,SAASoB,WAAWN,MAAoB;IACtC,IAAIO,OAAO;IACX,IAAK,IAAInB,IAAI,GAAGA,IAAIY,OAAOT,MAAM,EAAEH,IAAKmB,QAAQP,MAAM,CAACZ,EAAE,GAAGY,MAAM,CAACZ,EAAE;IACrE,OAAOC,KAAKmB,IAAI,CAACD;AACnB;AAEA,SAASE,iBAAiBC,CAAe,EAAEC,CAAe,EAAEC,KAAa,EAAEC,KAAa;IACtF,IAAIC,MAAM;IACV,IAAK,IAAI1B,IAAI,GAAGA,IAAIsB,EAAEnB,MAAM,EAAEH,IAAK0B,OAAOJ,CAAC,CAACtB,EAAE,GAAGuB,CAAC,CAACvB,EAAE;IACrD,OAAO0B,MAAOF,CAAAA,QAAQC,QAAQ,KAAI;AACpC;AAIO,SAAe/B,iBAAiBiC,IAAgB,EAAE9B,IAAY,EAAE+B,IAAsB;;;;;oBAC3F,IAAIA,KAAKzB,MAAM,KAAK,GAAG;;;oBACvB;;wBAAMwB,KAAKE,QAAQ,CACjB,mEACAD,KAAKE,GAAG,CAAC,SAAChB;mCAAQ;gCAACV,OAAOS,WAAWC,KAAKjB;gCAAQiB,IAAIiB,IAAI;gCAAEjB,IAAIkB,KAAK;6BAAC;;;;oBAFxE;;;;;;IAIF;;AAIO,SAAexC,eAAemC,IAAgB,EAAEM,EAAgB,EAAEpC,IAAY,EAAEqC,KAAa,EAAEC,OAAqB;QAAEC,cAAAA,iEAAc;;YACnIrB,GACAsB,WACAC,OACAC,MACAX,MAEAY,MACD,2BAAA,mBAAA,gBAAA,WAAA,OAAM1B,KAEHF,QACA6B,OACAC;;;;oBAXF3B,IAAIpB,OAAOsC,IAAIpC;oBACfwC,YAAYnB,WAAWH;oBACvBuB,QAAQF,cAAc,AAAC,sCAA+D,OAA1BO,oCAAyB,EAAC,OAAK;oBACpF;;wBAAMhB,KAAKiB,OAAO,CAAC,AAAC,8FAAmG,OAANN,OAAM;;;oBAA9HC,OAAO;oBACC;;wBAAMA,KAAKM,GAAG;;;oBAAtBjB,OAAQ;oBAERY,OAAO,IAAIM;oBACZ,kCAAA,2BAAA;;wBAAL,IAAK,YAAalB,2BAAb,6BAAA,QAAA,yBAAA,iCAAmB;4BAAbd,MAAN;4BACH,IAAIqB,WAAW,CAACA,QAAQY,GAAG,CAACjC,IAAIiB,IAAI,GAAG;4BACjCnB,SAASD,OAAOG,IAAIF,MAAM,EAAEf;4BAC5B4C,QAAQpB,iBAAiBN,GAAGH,QAAQyB,WAAWnB,WAAWN;4BAC1D8B,WAAWF,KAAKQ,GAAG,CAAClC,IAAIiB,IAAI;4BAClC,IAAI,CAACW,YAAYD,QAAQC,SAASD,KAAK,IAAKA,UAAUC,SAASD,KAAK,IAAI3B,IAAIkB,KAAK,GAAGU,SAASV,KAAK,EAAG;gCACnGQ,KAAKS,GAAG,CAACnC,IAAIiB,IAAI,EAAE;oCAAEU,OAAAA;oCAAOT,OAAOlB,IAAIkB,KAAK;oCAAEkB,OAAO,AAAC,IAAqBpC,OAAlBA,IAAIqC,UAAU,EAAC,KAAgB,OAAbrC,IAAIsC,QAAQ;gCAAG;4BAC5F;wBACF;;wBARK;wBAAA;;;iCAAA,6BAAA;gCAAA;;;gCAAA;sCAAA;;;;oBASL;;wBAAQ,qBAAGZ,KAAKa,OAAO,IACpBvB,GAAG,CAAC;qEAAEC,kBAAMuB;mCAAY;gCAAEvB,MAAAA;+BAASuB;2BACnCC,IAAI,CAACC,8BAAmB,EACxBC,KAAK,CAAC,GAAGvB,OACTJ,GAAG,CAAC;gCAAGC,aAAAA,MAAMmB,cAAAA,OAAOT,cAAAA;mCAAa;gCAAEV,MAAAA;gCAAMmB,OAAAA;gCAAOQ,YAAYC,IAAAA,mBAAQ,EAAClB;4BAAO;;;;;IACjF;;AAIO,SAAehD,YAAYkC,IAAgB,EAAE9B,IAAY,EAAEkC,IAAY,EAAE6B,IAAgE;QAAExB,cAAAA,iEAAc;;YAExJyB,YACAC,YAEAC,SAKAzB,OACAC,MACAX,MACAY,MACD,2BAAA,mBAAA,gBAAA,WAAA,OAAM1B,KAEHkD,OACAC,WACD,4BAAA,oBAAA,iBAAA,YAAA,QAAMC,GACHzB,OACAC;;;;oBAnBV,IAAIkB,KAAKzB,OAAO,IAAIyB,KAAKzB,OAAO,CAACgC,IAAI,KAAK,GAAG;;;;oBAC1B;;wBAAMxC,KAAKiB,OAAO,CAAC;;;oBAAhCiB,aAAa;oBACC;;wBAAMA,WAAWhB,GAAG,CAACd;;;oBAAnC+B,aAAc;oBACpB,IAAIA,WAAW3D,MAAM,KAAK,GAAG;;;;oBACvB4D,UAAUK,IAAAA,uBAAY,EAACN,YAAYhC,GAAG,CAAC,SAAChB;wBAC5C,IAAMF,SAASD,OAAOG,IAAIF,MAAM,EAAEf;wBAClC,OAAO;4BAAEe,QAAAA;4BAAQO,MAAMD,WAAWN;wBAAQ;oBAC5C;oBAEM0B,QAAQF,cAAc,AAAC,sCAA+D,OAA1BO,oCAAyB,EAAC,OAAK;oBACpF;;wBAAMhB,KAAKiB,OAAO,CAAC,AAAC,iEAAsE,OAANN;;;oBAA3FC,OAAO;oBACC;;wBAAMA,KAAKM,GAAG;;;oBAAtBjB,OAAQ;oBACRY,OAAO,IAAIM;oBACZ,kCAAA,2BAAA;;wBAAL,IAAK,YAAalB,2BAAb,6BAAA,QAAA,yBAAA,iCAAmB;4BAAbd,MAAN;4BACH,IAAIA,IAAIiB,IAAI,KAAKA,QAAQ6B,KAAKS,OAAO,CAACtB,GAAG,CAACjC,IAAIiB,IAAI,KAAM6B,KAAKzB,OAAO,IAAI,CAACyB,KAAKzB,OAAO,CAACY,GAAG,CAACjC,IAAIiB,IAAI,GAAI;4BAChGiC,QAAQrD,OAAOG,IAAIF,MAAM,EAAEf;4BAC3BoE,YAAY/C,WAAW8C;4BACxB,mCAAA,4BAAA;;gCAAL,IAAK,aAAWD,8BAAX,8BAAA,SAAA,0BAAA,kCAAoB;oCAAdG,IAAN;oCACGzB,QAAQpB,iBAAiB6C,EAAEtD,MAAM,EAAEoD,OAAOE,EAAE/C,IAAI,EAAE8C;oCAClDvB,WAAWF,KAAKQ,GAAG,CAAClC,IAAIiB,IAAI;oCAClC,IAAIW,aAAa4B,aAAa7B,QAAQC,UAAUF,KAAKS,GAAG,CAACnC,IAAIiB,IAAI,EAAEU;gCACrE;;gCAJK;gCAAA;;;yCAAA,8BAAA;wCAAA;;;wCAAA;8CAAA;;;;wBAKP;;wBATK;wBAAA;;;iCAAA,6BAAA;gCAAA;;;gCAAA;sCAAA;;;;oBAWL;;wBAAQ,qBAAGD,KAAKa,OAAO,IACpBvB,GAAG,CAAC;qEAAEyC,eAAG9B;mCAAY;gCAAEV,MAAMwC;gCAAG9B,OAAAA;4BAAM;2BACtCc,IAAI,CAACC,8BAAmB,EACxBC,KAAK,CAAC,GAAGG,KAAKY,CAAC,EACf1C,GAAG,CAAC;gCAAGC,AAAMwC,UAANxC,MAASU,cAAAA;mCAAa;gCAAEV,MAAMwC;gCAAGb,YAAYC,IAAAA,mBAAQ,EAAClB;4BAAO;;;;;IACzE"}
|
|
1
|
+
{"version":3,"sources":["/Users/kevin/Dev/open-source/ai/sensemaking/src/store/turso/vectors.ts"],"sourcesContent":["import type { Connection, VectorCandidate, VectorSimilar, VectorWriteRow } from '../types.ts';\nimport { asCosine, compareVectorScores, NATIVE_VECTOR_SCOPE_TABLE, sampleEvenly } from '../vectors.ts';\n\n// Native F32_BLOB(dims) columns, dims fixed at DDL time (open.ts, STORE_DIMS). Both scans\n// score in JS: a per-row vector_distance_cos measured 3.5-8x slower on this row engine.\n\n// A model narrower than the column zero-pads: the added dimensions contribute nothing to\n// either dot product or norm, so cosine is unchanged.\nfunction padded(values: ArrayLike<number>, dims: number): Float32Array {\n const out = new Float32Array(dims);\n for (let d = 0; d < Math.min(values.length, dims); d++) out[d] = values[d];\n return out;\n}\n\n// A bound Float32Array stores truncated bytes with no error; a Buffer view over the same\n// bytes round-trips exact (measured 0.7.2).\nfunction toBlob(v: Float32Array): Buffer {\n return Buffer.from(v.buffer, v.byteOffset, v.byteLength);\n}\n\nfunction decode(vector: Buffer, dims: number): Float32Array {\n return new Float32Array(vector.buffer, vector.byteOffset, dims);\n}\n\n// Rows arrive int8-quantized with a per-vector scale (embed/query.ts's toStore); this store\n// dequantizes to floats, as duckdb does.\nfunction dequantize(row: VectorWriteRow, dims: number): Float32Array {\n const q = new Int8Array(row.vector.buffer, row.vector.byteOffset, row.vector.byteLength);\n const out = new Float32Array(dims);\n for (let d = 0; d < Math.min(q.length, dims); d++) out[d] = q[d] * row.scale;\n return out;\n}\n\nfunction vectorNorm(vector: Float32Array): number {\n let norm = 0;\n for (let d = 0; d < vector.length; d++) norm += vector[d] * vector[d];\n return Math.sqrt(norm);\n}\n\nfunction cosineSimilarity(a: Float32Array, b: Float32Array, aNorm: number, bNorm: number): number {\n let dot = 0;\n for (let d = 0; d < a.length; d++) dot += a[d] * b[d];\n return dot / (aNorm * bNorm + 1e-32);\n}\n\n// `scale` stays NULL: the column exists so the row shape matches the other stores, and\n// cosine is scale-invariant once the value is dequantized here.\nexport async function writeVectorBatch(conn: Connection, dims: number, rows: VectorWriteRow[]): Promise<void> {\n if (rows.length === 0) return;\n await conn.runBatch(\n 'UPDATE embeddings SET vector = ? WHERE \"path\" = ? AND chunk = ?',\n rows.map((row) => [toBlob(dequantize(row, dims)), row.path, row.chunk])\n );\n}\n\n// Best chunk per file by cosine. GROUP BY with MIN() cannot do this on 0.7.2: the aggregate\n// is right but the bare start_line/end_line come from an arbitrary row in the group.\nexport async function scanCandidates(conn: Connection, qv: Float32Array, dims: number, fetch: number, allowed?: Set<string>, nativeScope = false): Promise<VectorCandidate[]> {\n const q = padded(qv, dims);\n const queryNorm = vectorNorm(q);\n const scope = nativeScope ? ` AND \"path\" IN (SELECT \"path\" FROM ${NATIVE_VECTOR_SCOPE_TABLE})` : '';\n const stmt = await conn.prepare(`SELECT \"path\", chunk, start_line, end_line, vector FROM embeddings WHERE vector IS NOT NULL${scope} ORDER BY \"path\", chunk`);\n const rows = (await stmt.all()) as Array<{ path: string; chunk: number; start_line: number; end_line: number; vector: Buffer }>;\n\n const best = new Map<string, { score: number; chunk: number; lines: string }>();\n for (const row of rows) {\n if (allowed && !allowed.has(row.path)) continue;\n const vector = decode(row.vector, dims);\n const score = cosineSimilarity(q, vector, queryNorm, vectorNorm(vector));\n const existing = best.get(row.path);\n if (!existing || score > existing.score || (score === existing.score && row.chunk < existing.chunk)) {\n best.set(row.path, { score, chunk: row.chunk, lines: `L${row.start_line}-${row.end_line}` });\n }\n }\n return [...best.entries()]\n .map(([path, value]) => ({ path, ...value }))\n .sort(compareVectorScores)\n .slice(0, fetch)\n .map(({ path, lines, score }) => ({ path, lines, similarity: asCosine(score) }));\n}\n\n// Max cosine over (target chunk, other chunk) pairs. The SQL cross-join duckdb uses measured\n// 6-8x slower here at 26k chunks (2026-08-30).\nexport async function scanSimilar(conn: Connection, dims: number, path: string, opts: { exclude: Set<string>; allowed?: Set<string>; k: number }, nativeScope = false): Promise<VectorSimilar[]> {\n if (opts.allowed && opts.allowed.size === 0) return [];\n const targetStmt = await conn.prepare('SELECT vector FROM embeddings WHERE \"path\" = ? AND vector IS NOT NULL ORDER BY chunk');\n const targetRows = (await targetStmt.all(path)) as Array<{ vector: Buffer }>;\n if (targetRows.length === 0) return [];\n const targets = sampleEvenly(targetRows).map((row) => {\n const vector = decode(row.vector, dims);\n return { vector, norm: vectorNorm(vector) };\n });\n\n const scope = nativeScope ? ` AND \"path\" IN (SELECT \"path\" FROM ${NATIVE_VECTOR_SCOPE_TABLE})` : '';\n const stmt = await conn.prepare(`SELECT \"path\", vector FROM embeddings WHERE vector IS NOT NULL${scope}`);\n const rows = (await stmt.all()) as Array<{ path: string; vector: Buffer }>;\n const best = new Map<string, number>();\n for (const row of rows) {\n if (row.path === path || opts.exclude.has(row.path) || (opts.allowed && !opts.allowed.has(row.path))) continue;\n const other = decode(row.vector, dims);\n const otherNorm = vectorNorm(other);\n for (const t of targets) {\n const score = cosineSimilarity(t.vector, other, t.norm, otherNorm);\n const existing = best.get(row.path);\n if (existing === undefined || score > existing) best.set(row.path, score);\n }\n }\n\n return [...best.entries()]\n .map(([p, score]) => ({ path: p, score }))\n .sort(compareVectorScores)\n .slice(0, opts.k)\n .map(({ path: p, score }) => ({ path: p, similarity: asCosine(score) }));\n}\n"],"names":["scanCandidates","scanSimilar","writeVectorBatch","padded","values","dims","out","Float32Array","d","Math","min","length","toBlob","v","Buffer","from","buffer","byteOffset","byteLength","decode","vector","dequantize","row","q","Int8Array","scale","vectorNorm","norm","sqrt","cosineSimilarity","a","b","aNorm","bNorm","dot","conn","rows","runBatch","map","path","chunk","qv","fetch","allowed","nativeScope","queryNorm","scope","stmt","best","score","existing","NATIVE_VECTOR_SCOPE_TABLE","prepare","all","Map","has","get","set","lines","start_line","end_line","entries","value","sort","compareVectorScores","slice","similarity","asCosine","opts","targetStmt","targetRows","targets","other","otherNorm","t","size","sampleEvenly","exclude","undefined","p","k"],"mappings":";;;;;;;;;;;QAyDsBA;eAAAA;;QA0BAC;eAAAA;;QApCAC;eAAAA;;;yBA9CiE;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAEvF,0FAA0F;AAC1F,wFAAwF;AAExF,yFAAyF;AACzF,sDAAsD;AACtD,SAASC,OAAOC,MAAyB,EAAEC,IAAY;IACrD,IAAMC,MAAM,IAAIC,aAAaF;IAC7B,IAAK,IAAIG,IAAI,GAAGA,IAAIC,KAAKC,GAAG,CAACN,OAAOO,MAAM,EAAEN,OAAOG,IAAKF,GAAG,CAACE,EAAE,GAAGJ,MAAM,CAACI,EAAE;IAC1E,OAAOF;AACT;AAEA,yFAAyF;AACzF,4CAA4C;AAC5C,SAASM,OAAOC,CAAe;IAC7B,OAAOC,OAAOC,IAAI,CAACF,EAAEG,MAAM,EAAEH,EAAEI,UAAU,EAAEJ,EAAEK,UAAU;AACzD;AAEA,SAASC,OAAOC,MAAc,EAAEf,IAAY;IAC1C,OAAO,IAAIE,aAAaa,OAAOJ,MAAM,EAAEI,OAAOH,UAAU,EAAEZ;AAC5D;AAEA,4FAA4F;AAC5F,yCAAyC;AACzC,SAASgB,WAAWC,GAAmB,EAAEjB,IAAY;IACnD,IAAMkB,IAAI,IAAIC,UAAUF,IAAIF,MAAM,CAACJ,MAAM,EAAEM,IAAIF,MAAM,CAACH,UAAU,EAAEK,IAAIF,MAAM,CAACF,UAAU;IACvF,IAAMZ,MAAM,IAAIC,aAAaF;IAC7B,IAAK,IAAIG,IAAI,GAAGA,IAAIC,KAAKC,GAAG,CAACa,EAAEZ,MAAM,EAAEN,OAAOG,IAAKF,GAAG,CAACE,EAAE,GAAGe,CAAC,CAACf,EAAE,GAAGc,IAAIG,KAAK;IAC5E,OAAOnB;AACT;AAEA,SAASoB,WAAWN,MAAoB;IACtC,IAAIO,OAAO;IACX,IAAK,IAAInB,IAAI,GAAGA,IAAIY,OAAOT,MAAM,EAAEH,IAAKmB,QAAQP,MAAM,CAACZ,EAAE,GAAGY,MAAM,CAACZ,EAAE;IACrE,OAAOC,KAAKmB,IAAI,CAACD;AACnB;AAEA,SAASE,iBAAiBC,CAAe,EAAEC,CAAe,EAAEC,KAAa,EAAEC,KAAa;IACtF,IAAIC,MAAM;IACV,IAAK,IAAI1B,IAAI,GAAGA,IAAIsB,EAAEnB,MAAM,EAAEH,IAAK0B,OAAOJ,CAAC,CAACtB,EAAE,GAAGuB,CAAC,CAACvB,EAAE;IACrD,OAAO0B,MAAOF,CAAAA,QAAQC,QAAQ,KAAI;AACpC;AAIO,SAAe/B,iBAAiBiC,IAAgB,EAAE9B,IAAY,EAAE+B,IAAsB;;;;;oBAC3F,IAAIA,KAAKzB,MAAM,KAAK,GAAG;;;oBACvB;;wBAAMwB,KAAKE,QAAQ,CACjB,mEACAD,KAAKE,GAAG,CAAC,SAAChB;mCAAQ;gCAACV,OAAOS,WAAWC,KAAKjB;gCAAQiB,IAAIiB,IAAI;gCAAEjB,IAAIkB,KAAK;6BAAC;;;;oBAFxE;;;;;;IAIF;;AAIO,SAAexC,eAAemC,IAAgB,EAAEM,EAAgB,EAAEpC,IAAY,EAAEqC,KAAa,EAAEC,OAAqB;QAAEC,cAAAA,iEAAc;;YACnIrB,GACAsB,WACAC,OACAC,MACAX,MAEAY,MACD,2BAAA,mBAAA,gBAAA,WAAA,OAAM1B,KAEHF,QACA6B,OACAC;;;;oBAXF3B,IAAIpB,OAAOsC,IAAIpC;oBACfwC,YAAYnB,WAAWH;oBACvBuB,QAAQF,cAAc,AAAC,sCAA+D,OAA1BO,oCAAyB,EAAC,OAAK;oBACpF;;wBAAMhB,KAAKiB,OAAO,CAAC,AAAC,8FAAmG,OAANN,OAAM;;;oBAA9HC,OAAO;oBACC;;wBAAMA,KAAKM,GAAG;;;oBAAtBjB,OAAQ;oBAERY,OAAO,IAAIM;oBACZ,kCAAA,2BAAA;;wBAAL,IAAK,YAAalB,2BAAb,6BAAA,QAAA,yBAAA,iCAAmB;4BAAbd,MAAN;4BACH,IAAIqB,WAAW,CAACA,QAAQY,GAAG,CAACjC,IAAIiB,IAAI,GAAG;4BACjCnB,SAASD,OAAOG,IAAIF,MAAM,EAAEf;4BAC5B4C,QAAQpB,iBAAiBN,GAAGH,QAAQyB,WAAWnB,WAAWN;4BAC1D8B,WAAWF,KAAKQ,GAAG,CAAClC,IAAIiB,IAAI;4BAClC,IAAI,CAACW,YAAYD,QAAQC,SAASD,KAAK,IAAKA,UAAUC,SAASD,KAAK,IAAI3B,IAAIkB,KAAK,GAAGU,SAASV,KAAK,EAAG;gCACnGQ,KAAKS,GAAG,CAACnC,IAAIiB,IAAI,EAAE;oCAAEU,OAAAA;oCAAOT,OAAOlB,IAAIkB,KAAK;oCAAEkB,OAAO,AAAC,IAAqBpC,OAAlBA,IAAIqC,UAAU,EAAC,KAAgB,OAAbrC,IAAIsC,QAAQ;gCAAG;4BAC5F;wBACF;;wBARK;wBAAA;;;iCAAA,6BAAA;gCAAA;;;gCAAA;sCAAA;;;;oBASL;;wBAAQ,qBAAGZ,KAAKa,OAAO,IACpBvB,GAAG,CAAC;qEAAEC,kBAAMuB;mCAAY;gCAAEvB,MAAAA;+BAASuB;2BACnCC,IAAI,CAACC,8BAAmB,EACxBC,KAAK,CAAC,GAAGvB,OACTJ,GAAG,CAAC;gCAAGC,aAAAA,MAAMmB,cAAAA,OAAOT,cAAAA;mCAAa;gCAAEV,MAAAA;gCAAMmB,OAAAA;gCAAOQ,YAAYC,IAAAA,mBAAQ,EAAClB;4BAAO;;;;;IACjF;;AAIO,SAAehD,YAAYkC,IAAgB,EAAE9B,IAAY,EAAEkC,IAAY,EAAE6B,IAAgE;QAAExB,cAAAA,iEAAc;;YAExJyB,YACAC,YAEAC,SAKAzB,OACAC,MACAX,MACAY,MACD,2BAAA,mBAAA,gBAAA,WAAA,OAAM1B,KAEHkD,OACAC,WACD,4BAAA,oBAAA,iBAAA,YAAA,QAAMC,GACHzB,OACAC;;;;oBAnBV,IAAIkB,KAAKzB,OAAO,IAAIyB,KAAKzB,OAAO,CAACgC,IAAI,KAAK,GAAG;;;;oBAC1B;;wBAAMxC,KAAKiB,OAAO,CAAC;;;oBAAhCiB,aAAa;oBACC;;wBAAMA,WAAWhB,GAAG,CAACd;;;oBAAnC+B,aAAc;oBACpB,IAAIA,WAAW3D,MAAM,KAAK,GAAG;;;;oBACvB4D,UAAUK,IAAAA,uBAAY,EAACN,YAAYhC,GAAG,CAAC,SAAChB;wBAC5C,IAAMF,SAASD,OAAOG,IAAIF,MAAM,EAAEf;wBAClC,OAAO;4BAAEe,QAAAA;4BAAQO,MAAMD,WAAWN;wBAAQ;oBAC5C;oBAEM0B,QAAQF,cAAc,AAAC,sCAA+D,OAA1BO,oCAAyB,EAAC,OAAK;oBACpF;;wBAAMhB,KAAKiB,OAAO,CAAC,AAAC,iEAAsE,OAANN;;;oBAA3FC,OAAO;oBACC;;wBAAMA,KAAKM,GAAG;;;oBAAtBjB,OAAQ;oBACRY,OAAO,IAAIM;oBACZ,kCAAA,2BAAA;;wBAAL,IAAK,YAAalB,2BAAb,6BAAA,QAAA,yBAAA,iCAAmB;4BAAbd,MAAN;4BACH,IAAIA,IAAIiB,IAAI,KAAKA,QAAQ6B,KAAKS,OAAO,CAACtB,GAAG,CAACjC,IAAIiB,IAAI,KAAM6B,KAAKzB,OAAO,IAAI,CAACyB,KAAKzB,OAAO,CAACY,GAAG,CAACjC,IAAIiB,IAAI,GAAI;4BAChGiC,QAAQrD,OAAOG,IAAIF,MAAM,EAAEf;4BAC3BoE,YAAY/C,WAAW8C;4BACxB,mCAAA,4BAAA;;gCAAL,IAAK,aAAWD,8BAAX,8BAAA,SAAA,0BAAA,kCAAoB;oCAAdG,IAAN;oCACGzB,QAAQpB,iBAAiB6C,EAAEtD,MAAM,EAAEoD,OAAOE,EAAE/C,IAAI,EAAE8C;oCAClDvB,WAAWF,KAAKQ,GAAG,CAAClC,IAAIiB,IAAI;oCAClC,IAAIW,aAAa4B,aAAa7B,QAAQC,UAAUF,KAAKS,GAAG,CAACnC,IAAIiB,IAAI,EAAEU;gCACrE;;gCAJK;gCAAA;;;yCAAA,8BAAA;wCAAA;;;wCAAA;8CAAA;;;;wBAKP;;wBATK;wBAAA;;;iCAAA,6BAAA;gCAAA;;;gCAAA;sCAAA;;;;oBAWL;;wBAAQ,qBAAGD,KAAKa,OAAO,IACpBvB,GAAG,CAAC;qEAAEyC,eAAG9B;mCAAY;gCAAEV,MAAMwC;gCAAG9B,OAAAA;4BAAM;2BACtCc,IAAI,CAACC,8BAAmB,EACxBC,KAAK,CAAC,GAAGG,KAAKY,CAAC,EACf1C,GAAG,CAAC;gCAAGC,AAAMwC,UAANxC,MAASU,cAAAA;mCAAa;gCAAEV,MAAMwC;gCAAGb,YAAYC,IAAAA,mBAAQ,EAAClB;4BAAO;;;;;IACzE"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/types.ts"],"sourcesContent":["// The backing-store interface: a minimal portable statement surface (exec/prepare), plus\n// dedicated interfaces exactly where engines diverge (lexical index, vector scan, raw sql).\n\nimport type { Config, ResolvedConfig, StoreName } from '../config/index.ts';\nimport type { ReconcileDelta } from '../features/types.ts';\nimport type { ParsedDoc } from '../scan/index.ts';\n\n// 'vectors': the store's VectorStore is functionally implemented, not present-but-inert; checked\n// at open (index.ts) before a config with an \"embed\" block is allowed onto a store lacking it.\n// 'segment': the engine can run segment() (Intl.Segmenter grapheme clustering) as a SQL function;\n// turso's client cannot register it and has no rewrite for it, unlike has()/basename(), which now\n// work everywhere. An array, not a bare union, so a runtime check reads the same list the type does.\nexport const CAPABILITY_NAMES = ['vectors', 'segment'] as const;\nexport type Capability = (typeof CAPABILITY_NAMES)[number];\n\nexport interface RunResult {\n changes: number | bigint;\n lastInsertRowid: number | bigint;\n}\n\n// A prepared statement's async surface: every supported engine crosses a real async boundary\n// for a query (DuckDB has no synchronous client), so run/get/all return Promises.\nexport interface Statement {\n run(...params: unknown[]): Promise<RunResult>;\n get(...params: unknown[]): Promise<unknown>;\n all(...params: unknown[]): Promise<unknown[]>;\n iterate(...params: unknown[]): AsyncIterable<unknown>;\n columns(): Array<{ name: string }>;\n setReadBigInts(enabled: boolean): void;\n}\n\n// Connection surface feature-owned SQL runs against inside a store's hot loops (schema,\n// reconcile), portable so a feature never imports an engine-specific client type.\nexport interface Connection {\n exec(sql: string): Promise<void>;\n prepare(sql: string): Promise<Statement>;\n runBatch(sql: string, paramRows: unknown[][]): Promise<void>;\n // Bulk-inserts rows that cannot conflict, through a path that binds no per-value parameters.\n // Optional: a store that has nothing faster than its own INSERT omits it and callers fall back\n // (appendRows in shared.ts). `columns` names the values each row carries; a table column the\n // caller does not write takes its default.\n appendRows?(table: string, columns: string[], rows: unknown[][]): Promise<void>;\n}\n\n// One reconcile algorithm (src/store/reconcile.ts), parameterised per engine. reconcileContent is\n// the load-bearing member: it lands every content-table change and owns its own multi-step\n// strategy (sqlite FTS5 incremental, duckdb combined delete/insert, turso incremental-or-rebuild).\nexport interface ReconcileDialect {\n // BEGIN mode for the whole reconcile: sqlite/turso 'BEGIN IMMEDIATE', duckdb 'BEGIN'.\n beginMode(): string;\n // Throws SenseError('COLUMN_LIMIT', ...) past this store's own column ceiling and reasoning.\n checkColumnLimit(count: number): void;\n // Adds `names` to frontmatter, already filtered to columns this connection doesn't have yet.\n // sqlite/turso loop (their ADD COLUMN is metadata-only); duckdb issues one statement per call.\n addColumns(conn: Connection, names: string[]): Promise<void>;\n // Replaces this engine's existing-row frontmatter upsert, inside the caller's\n // publication transaction. Preserve non-writable columns and repair missing rows.\n updateFrontmatter?(conn: Connection, writableColumns: string[], rows: unknown[][]): Promise<void>;\n // Reconciles searchable content for `touched` paths and parsed `docs`; `delta` carries the tree\n // state a strategy may need. Must not open its own transaction, and must not return before its own\n // multi-step strategy (e.g. turso's DROP/rebuild) completes.\n reconcileContent(conn: Connection, touched: string[], docs: ParsedDoc[], delta: ReconcileDelta, cfg: Config): Promise<void>;\n // Records this reconcile's write-transaction duration. sqlite/turso use it for open()'s derived\n // busy_timeout; duckdb has no such PRAGMA and omits it.\n recordDuration?(conn: Connection, ms: number): Promise<void>;\n}\n\n// One open algorithm (src/store/open.ts), parameterised per engine. `Handle` is whatever this\n// store needs to close the connection and construct its Store (sqlite: {db}; duckdb:\n// {instance, duckdb}; turso: db) -- opaque to the shared orchestration, threaded through unchanged.\nexport interface OpenDialect<Handle> {\n // Cache filename under STATE_DIR, e.g. 'cache.db'.\n filename: string;\n // Cache shape version, independent of the config's own `version`; bumping it rebuilds an existing tree.\n schemaVersion: string;\n reconcileDialect: ReconcileDialect;\n // Opens the physical connection and applies pragmas due before any SQL runs (sqlite: busy_timeout\n // + WAL; turso: connect-time timeout; duckdb: none).\n connect(dbPath: string, cfg: ResolvedConfig, options: { existingOnly: boolean; observational: boolean }): Promise<{ handle: Handle; conn: Connection }>;\n // Releases the handle, for both the rebuild-and-reopen branch and error cleanup on this attempt.\n close(handle: Handle, options?: { observational?: boolean }): Promise<void>;\n // True when connect() failed because another process holds the cache file, which the orchestration\n // retries. Each engine words it differently, so the match is the dialect's own (`native-not-emulated`).\n // Absent for sqlite, whose file lock is shared and whose concurrent-open failure is a different defect.\n // Matching on message text is forced, not chosen: measured 2026-09-02, duckdb throws a plain Error\n // whose only own properties are stack and message, and turso sets code to the constant\n // 'GenericFailure' on every failure alike with rawCode undefined. Neither exposes anything a\n // predicate could switch on, so each dialect pins its engine's wordings and unit-tests them.\n isLocked?(err: Error): boolean;\n // Minimal, configuration-independent schema needed before the builder can capture a planning\n // snapshot. This bootstrap may commit independently because no feature generation is published.\n ensureCoreSchema(handle: Handle, conn: Connection): Promise<void>;\n // Configuration-dependent persistent DDL. The builder invokes this only inside its fenced write\n // transaction, so a failed feature enable cannot leave catalog changes under the old signature.\n ensureFeatureSchema(handle: Handle, conn: Connection, cfg: Config): Promise<void>;\n // Connection-local preparation with no durable cache effect (DuckDB's TEMP vector stage).\n prepareConnection?(handle: Handle, conn: Connection, cfg: Config): Promise<void>;\n // Installs the derived busy_timeout PRAGMA right before reconcile (sqlite/turso); absent for\n // duckdb, which has no such PRAGMA.\n setDerivedBusyTimeout?(handle: Handle, conn: Connection, ms: number): Promise<void>;\n // Persistent lexical preparation belongs to build, while no-build calls only the readiness\n // check. SQLite and Turso maintain lexical state during reconcile and omit both hooks.\n prepareLexical?(conn: Connection): Promise<void>;\n assertLexicalReady?(conn: Connection): Promise<void>;\n createStore(handle: Handle, conn: Connection, cfg: ResolvedConfig, options: { observational: boolean }): Store;\n}\n\nexport interface FieldStat {\n field: string;\n coverage: number;\n type: string;\n}\n\nexport interface DocumentStore {\n // Frontmatter column names (including internal ones; callers filter).\n columns(): Promise<string[]>;\n // Per-column coverage (non-null count) and observed type set, aggregated in one SQL query,\n // never one row per note. `scopeWhere` is a caller-built WHERE fragment, per LexicalQueryOptions.\n fieldStats(columns: string[], scopeWhere: string): Promise<FieldStat[]>;\n}\n\n// A ranked match, nothing more. Snippets are cut from the document above the store\n// (commands/search.ts), identically for every engine, so no store carries them.\nexport interface LexicalHit {\n path: string;\n}\n\nexport interface LexicalQueryOptions {\n whereJoin: string;\n whereCond: string;\n scopeCond: string;\n limit: number;\n}\n\nexport interface LexicalIndex {\n // Ranked word-match query with snippet, scoped by the caller-built SQL fragments (the same\n // fragments narrowByWhere/materializeScope produce elsewhere).\n query(terms: string, opts: LexicalQueryOptions): Promise<LexicalHit[]>;\n}\n\nexport interface VectorCandidate {\n path: string;\n lines: string;\n similarity: number;\n}\n\nexport interface VectorSimilar {\n path: string;\n similarity: number;\n}\n\nexport interface VectorWriteRow {\n path: string;\n chunk: number;\n scale: number;\n vector: Buffer;\n}\n\nexport interface VectorStore {\n // Rows whose vector is NULL: never embedded, or added since.\n pending(): Promise<Array<{ path: string; chunk: number }>>;\n // One batch write per call (never per row) so a provider's embedding batch stays inside a\n // single store method.\n writeVectors(rows: VectorWriteRow[]): Promise<void>;\n candidates(queryVector: Float32Array, storeDims: number, fetch: number, allowed?: Set<string>): Promise<VectorCandidate[]>;\n similar(path: string, opts: { exclude: Set<string>; allowed?: Set<string>; k: number }): Promise<VectorSimilar[]>;\n hasVector(path: string): Promise<boolean>;\n}\n\n// The `sense sql` passthrough: string in, streamed rows out. Each store registers the same\n// sense-supplied functions and applies its own read-bigints/error-translation behavior.\nexport interface RawStatement {\n columns(): Array<{ name: string }>;\n iterate(...params: unknown[]): AsyncIterable<unknown>;\n}\n\nexport interface SqlSession {\n prepare(sql: string): Promise<RawStatement>;\n}\n\nexport interface Store {\n readonly name: StoreName;\n readonly capabilities: ReadonlySet<Capability>;\n exec(sql: string): Promise<void>;\n prepare(sql: string): Promise<Statement>;\n // One crossing, N rows: every external write loop (search's candidate insert, graph ring's\n // temp-table writes) goes through this instead of looping over `run()`. Same contract as Connection.runBatch.\n runBatch(sql: string, paramRows: unknown[][]): Promise<void>;\n // Pins one snapshot across multi-statement reads; `map` and `peek` use it. A network-bound\n // write (search's embedding top-up) must stay outside it. Nesting joins the enclosing transaction.\n transaction<T>(fn: () => Promise<T>): Promise<T>;\n docs: DocumentStore;\n lexical: LexicalIndex;\n vectors: VectorStore;\n raw: SqlSession;\n // Engine-level facts for `sense status` (e.g. a derived busy_timeout PRAGMA reading); each\n // store owns what it reports and how it is worded. Empty when there is nothing to report.\n engineStatus(): Promise<Record<string, string>>;\n close(): Promise<void>;\n}\n"],"names":["CAPABILITY_NAMES"],"mappings":"AAAA,yFAAyF;AACzF,4FAA4F;;;;;+BAW/EA;;;eAAAA;;;AAAN,IAAMA,mBAAmB;IAAC;IAAW;CAAU"}
|
|
1
|
+
{"version":3,"sources":["/Users/kevin/Dev/open-source/ai/sensemaking/src/store/types.ts"],"sourcesContent":["// The backing-store interface: a minimal portable statement surface (exec/prepare), plus\n// dedicated interfaces exactly where engines diverge (lexical index, vector scan, raw sql).\n\nimport type { Config, ResolvedConfig, StoreName } from '../config/index.ts';\nimport type { ReconcileDelta } from '../features/types.ts';\nimport type { ParsedDoc } from '../scan/index.ts';\n\n// 'vectors': the store's VectorStore is functionally implemented, not present-but-inert; checked\n// at open (index.ts) before a config with an \"embed\" block is allowed onto a store lacking it.\n// 'segment': the engine can run segment() (Intl.Segmenter grapheme clustering) as a SQL function;\n// turso's client cannot register it and has no rewrite for it, unlike has()/basename(), which now\n// work everywhere. An array, not a bare union, so a runtime check reads the same list the type does.\nexport const CAPABILITY_NAMES = ['vectors', 'segment'] as const;\nexport type Capability = (typeof CAPABILITY_NAMES)[number];\n\nexport interface RunResult {\n changes: number | bigint;\n lastInsertRowid: number | bigint;\n}\n\n// A prepared statement's async surface: every supported engine crosses a real async boundary\n// for a query (DuckDB has no synchronous client), so run/get/all return Promises.\nexport interface Statement {\n run(...params: unknown[]): Promise<RunResult>;\n get(...params: unknown[]): Promise<unknown>;\n all(...params: unknown[]): Promise<unknown[]>;\n iterate(...params: unknown[]): AsyncIterable<unknown>;\n columns(): Array<{ name: string }>;\n setReadBigInts(enabled: boolean): void;\n}\n\n// Connection surface feature-owned SQL runs against inside a store's hot loops (schema,\n// reconcile), portable so a feature never imports an engine-specific client type.\nexport interface Connection {\n exec(sql: string): Promise<void>;\n prepare(sql: string): Promise<Statement>;\n runBatch(sql: string, paramRows: unknown[][]): Promise<void>;\n // Bulk-inserts rows that cannot conflict, through a path that binds no per-value parameters.\n // Optional: a store that has nothing faster than its own INSERT omits it and callers fall back\n // (appendRows in shared.ts). `columns` names the values each row carries; a table column the\n // caller does not write takes its default.\n appendRows?(table: string, columns: string[], rows: unknown[][]): Promise<void>;\n}\n\n// One reconcile algorithm (src/store/reconcile.ts), parameterised per engine. reconcileContent is\n// the load-bearing member: it lands every content-table change and owns its own multi-step\n// strategy (sqlite FTS5 incremental, duckdb combined delete/insert, turso incremental-or-rebuild).\nexport interface ReconcileDialect {\n // BEGIN mode for the whole reconcile: sqlite/turso 'BEGIN IMMEDIATE', duckdb 'BEGIN'.\n beginMode(): string;\n // Throws SenseError('COLUMN_LIMIT', ...) past this store's own column ceiling and reasoning.\n checkColumnLimit(count: number): void;\n // Adds `names` to frontmatter, already filtered to columns this connection doesn't have yet.\n // sqlite/turso loop (their ADD COLUMN is metadata-only); duckdb issues one statement per call.\n addColumns(conn: Connection, names: string[]): Promise<void>;\n // Replaces this engine's existing-row frontmatter upsert, inside the caller's\n // publication transaction. Preserve non-writable columns and repair missing rows.\n updateFrontmatter?(conn: Connection, writableColumns: string[], rows: unknown[][]): Promise<void>;\n // Reconciles searchable content for `touched` paths and parsed `docs`; `delta` carries the tree\n // state a strategy may need. Must not open its own transaction, and must not return before its own\n // multi-step strategy (e.g. turso's DROP/rebuild) completes.\n reconcileContent(conn: Connection, touched: string[], docs: ParsedDoc[], delta: ReconcileDelta, cfg: Config): Promise<void>;\n // Records this reconcile's write-transaction duration. sqlite/turso use it for open()'s derived\n // busy_timeout; duckdb has no such PRAGMA and omits it.\n recordDuration?(conn: Connection, ms: number): Promise<void>;\n}\n\n// One open algorithm (src/store/open.ts), parameterised per engine. `Handle` is whatever this\n// store needs to close the connection and construct its Store (sqlite: {db}; duckdb:\n// {instance, duckdb}; turso: db) -- opaque to the shared orchestration, threaded through unchanged.\nexport interface OpenDialect<Handle> {\n // Cache filename under STATE_DIR, e.g. 'cache.db'.\n filename: string;\n // Cache shape version, independent of the config's own `version`; bumping it rebuilds an existing tree.\n schemaVersion: string;\n reconcileDialect: ReconcileDialect;\n // Opens the physical connection and applies pragmas due before any SQL runs (sqlite: busy_timeout\n // + WAL; turso: connect-time timeout; duckdb: none).\n connect(dbPath: string, cfg: ResolvedConfig, options: { existingOnly: boolean; observational: boolean }): Promise<{ handle: Handle; conn: Connection }>;\n // Releases the handle, for both the rebuild-and-reopen branch and error cleanup on this attempt.\n close(handle: Handle, options?: { observational?: boolean }): Promise<void>;\n // True when connect() failed because another process holds the cache file, which the orchestration\n // retries. Each engine words it differently, so the match is the dialect's own (`native-not-emulated`).\n // Absent for sqlite, whose file lock is shared and whose concurrent-open failure is a different defect.\n // Matching on message text is forced, not chosen: measured 2026-09-02, duckdb throws a plain Error\n // whose only own properties are stack and message, and turso sets code to the constant\n // 'GenericFailure' on every failure alike with rawCode undefined. Neither exposes anything a\n // predicate could switch on, so each dialect pins its engine's wordings and unit-tests them.\n isLocked?(err: Error): boolean;\n // Minimal, configuration-independent schema needed before the builder can capture a planning\n // snapshot. This bootstrap may commit independently because no feature generation is published.\n ensureCoreSchema(handle: Handle, conn: Connection): Promise<void>;\n // Configuration-dependent persistent DDL. The builder invokes this only inside its fenced write\n // transaction, so a failed feature enable cannot leave catalog changes under the old signature.\n ensureFeatureSchema(handle: Handle, conn: Connection, cfg: Config): Promise<void>;\n // Connection-local preparation with no durable cache effect (DuckDB's TEMP vector stage).\n prepareConnection?(handle: Handle, conn: Connection, cfg: Config): Promise<void>;\n // Installs the derived busy_timeout PRAGMA right before reconcile (sqlite/turso); absent for\n // duckdb, which has no such PRAGMA.\n setDerivedBusyTimeout?(handle: Handle, conn: Connection, ms: number): Promise<void>;\n // Persistent lexical preparation belongs to build, while no-build calls only the readiness\n // check. SQLite and Turso maintain lexical state during reconcile and omit both hooks.\n prepareLexical?(conn: Connection): Promise<void>;\n assertLexicalReady?(conn: Connection): Promise<void>;\n createStore(handle: Handle, conn: Connection, cfg: ResolvedConfig, options: { observational: boolean }): Store;\n}\n\nexport interface FieldStat {\n field: string;\n coverage: number;\n type: string;\n}\n\nexport interface DocumentStore {\n // Frontmatter column names (including internal ones; callers filter).\n columns(): Promise<string[]>;\n // Per-column coverage (non-null count) and observed type set, aggregated in one SQL query,\n // never one row per note. `scopeWhere` is a caller-built WHERE fragment, per LexicalQueryOptions.\n fieldStats(columns: string[], scopeWhere: string): Promise<FieldStat[]>;\n}\n\n// A ranked match, nothing more. Snippets are cut from the document above the store\n// (commands/search.ts), identically for every engine, so no store carries them.\nexport interface LexicalHit {\n path: string;\n}\n\nexport interface LexicalQueryOptions {\n whereJoin: string;\n whereCond: string;\n scopeCond: string;\n limit: number;\n}\n\nexport interface LexicalIndex {\n // Ranked word-match query with snippet, scoped by the caller-built SQL fragments (the same\n // fragments narrowByWhere/materializeScope produce elsewhere).\n query(terms: string, opts: LexicalQueryOptions): Promise<LexicalHit[]>;\n}\n\nexport interface VectorCandidate {\n path: string;\n lines: string;\n similarity: number;\n}\n\nexport interface VectorSimilar {\n path: string;\n similarity: number;\n}\n\nexport interface VectorWriteRow {\n path: string;\n chunk: number;\n scale: number;\n vector: Buffer;\n}\n\nexport interface VectorStore {\n // Rows whose vector is NULL: never embedded, or added since.\n pending(): Promise<Array<{ path: string; chunk: number }>>;\n // One batch write per call (never per row) so a provider's embedding batch stays inside a\n // single store method.\n writeVectors(rows: VectorWriteRow[]): Promise<void>;\n candidates(queryVector: Float32Array, storeDims: number, fetch: number, allowed?: Set<string>): Promise<VectorCandidate[]>;\n similar(path: string, opts: { exclude: Set<string>; allowed?: Set<string>; k: number }): Promise<VectorSimilar[]>;\n hasVector(path: string): Promise<boolean>;\n}\n\n// The `sense sql` passthrough: string in, streamed rows out. Each store registers the same\n// sense-supplied functions and applies its own read-bigints/error-translation behavior.\nexport interface RawStatement {\n columns(): Array<{ name: string }>;\n iterate(...params: unknown[]): AsyncIterable<unknown>;\n}\n\nexport interface SqlSession {\n prepare(sql: string): Promise<RawStatement>;\n}\n\nexport interface Store {\n readonly name: StoreName;\n readonly capabilities: ReadonlySet<Capability>;\n exec(sql: string): Promise<void>;\n prepare(sql: string): Promise<Statement>;\n // One crossing, N rows: every external write loop (search's candidate insert, graph ring's\n // temp-table writes) goes through this instead of looping over `run()`. Same contract as Connection.runBatch.\n runBatch(sql: string, paramRows: unknown[][]): Promise<void>;\n // Pins one snapshot across multi-statement reads; `map` and `peek` use it. A network-bound\n // write (search's embedding top-up) must stay outside it. Nesting joins the enclosing transaction.\n transaction<T>(fn: () => Promise<T>): Promise<T>;\n docs: DocumentStore;\n lexical: LexicalIndex;\n vectors: VectorStore;\n raw: SqlSession;\n // Engine-level facts for `sense status` (e.g. a derived busy_timeout PRAGMA reading); each\n // store owns what it reports and how it is worded. Empty when there is nothing to report.\n engineStatus(): Promise<Record<string, string>>;\n close(): Promise<void>;\n}\n"],"names":["CAPABILITY_NAMES"],"mappings":"AAAA,yFAAyF;AACzF,4FAA4F;;;;;+BAW/EA;;;eAAAA;;;AAAN,IAAMA,mBAAmB;IAAC;IAAW;CAAU"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"sources":["/Users/kevin/Dev/
|
|
1
|
+
{"version":3,"sources":["/Users/kevin/Dev/open-source/ai/sensemaking/src/store/vectors.ts"],"sourcesContent":["import type { Connection, VectorStore } from './types.ts';\n\nexport const NATIVE_VECTOR_SCOPE_TABLE = '_vector_scope';\n\ntype NativeVectorScopePhase = 'materializing' | 'active' | 'cleaning';\n\ninterface NativeVectorScopeState {\n token: object;\n allowed: Set<string>;\n phase: NativeVectorScopePhase;\n}\n\ninterface NativeVectorScopeOwnerState {\n vectors?: VectorStore;\n scope?: NativeVectorScopeState;\n}\n\ninterface NativeVectorScopeOwner {\n bind(vectors: VectorStore): void;\n activeFor(allowed: Set<string> | undefined): boolean;\n}\n\ninterface NativeVectorScopeReservation {\n activate(): void;\n deactivate(): void;\n invalidate(): void;\n}\n\nconst nativeVectorScopeOwners = new WeakMap<VectorStore, NativeVectorScopeOwnerState>();\n\n// One owner is constructed beside one native store connection, then bound to that connection's\n// VectorStore wrapper. The adapter asks its own owner for a private boolean; no Set or reservation\n// can activate another Store, even when a caller reuses the same Set object across both.\nexport function createNativeVectorScopeOwner(): NativeVectorScopeOwner {\n const owner: NativeVectorScopeOwnerState = {};\n return {\n bind(vectors) {\n if (owner.vectors) throw new Error('native vector scope owner is already bound');\n if (nativeVectorScopeOwners.has(vectors)) throw new Error('VectorStore is already bound to a native vector scope owner');\n owner.vectors = vectors;\n nativeVectorScopeOwners.set(vectors, owner);\n },\n activeFor(allowed) {\n return allowed !== undefined && owner.scope?.phase === 'active' && owner.scope.allowed === allowed;\n },\n };\n}\n\nexport function reserveNativeVectorScope(vectors: VectorStore, allowed: Set<string>): NativeVectorScopeReservation | null {\n const owner = nativeVectorScopeOwners.get(vectors);\n if (!owner) return null;\n if (owner.scope) throw new Error('native vector scope operation is already active for this Store');\n\n const token = {};\n owner.scope = { token, allowed, phase: 'materializing' };\n let valid = true;\n\n const current = (): NativeVectorScopeState => {\n if (!valid || owner.scope?.token !== token) throw new Error('native vector scope reservation is no longer valid');\n return owner.scope;\n };\n\n return {\n activate() {\n const scope = current();\n if (scope.phase !== 'materializing') throw new Error(`native vector scope cannot activate from ${scope.phase}`);\n scope.phase = 'active';\n },\n deactivate() {\n const scope = current();\n if (scope.phase === 'cleaning') throw new Error('native vector scope cleanup already started');\n scope.phase = 'cleaning';\n },\n invalidate() {\n if (!valid) return;\n if (owner.scope?.token === token) owner.scope = undefined;\n valid = false;\n },\n };\n}\n\n// Seed chunks that participate in a `related` scan. Cost is target_chunks x stored_chunks, so a\n// heading-dense seed multiplies a full-corpus scan (12.7s at 201 chunks/note unsampled).\nexport const TARGET_CHUNK_CAP = 16;\n\n// Evenly samples down to at most `cap` rows, so late sections of a long note still get a vote\n// instead of being cut off by a fixed prefix.\nexport function sampleEvenly<T>(rows: T[], cap: number = TARGET_CHUNK_CAP): T[] {\n const step = Math.max(1, Math.ceil(rows.length / cap));\n return rows.filter((_, i) => i % step === 0);\n}\n\n// Every store computes a real cosine; clamp float error at the ends and round for the public score.\nexport function asCosine(score: number): number {\n return Math.round(Math.min(1, Math.max(-1, score)) * 1000) / 1000;\n}\n\n// Rank by the computed cosine before rounding for display. Bytewise path order resolves only an\n// exact score tie, so two distinguishable similarities never cross a caller's result boundary.\nexport function compareVectorScores(a: { path: string; score: number }, b: { path: string; score: number }): number {\n const byScore = b.score - a.score;\n if (byScore !== 0) return byScore;\n return Buffer.compare(Buffer.from(a.path), Buffer.from(b.path));\n}\n\nexport async function pendingRows(conn: Connection): Promise<Array<{ path: string; chunk: number }>> {\n const stmt = await conn.prepare('SELECT \"path\", chunk FROM embeddings WHERE vector IS NULL ORDER BY \"path\", chunk');\n return (await stmt.all()) as Array<{ path: string; chunk: number }>;\n}\n\n// Whether a note has any chunk with a vector. Distinguishes \"nothing is near this note\" from\n// \"this note has no text\", which look the same in an empty result.\nexport async function hasVectorRow(conn: Connection, path: string): Promise<boolean> {\n const stmt = await conn.prepare('SELECT 1 AS ok FROM embeddings WHERE \"path\" = ? AND vector IS NOT NULL LIMIT 1');\n const row = (await stmt.get(path)) as { ok: number } | undefined;\n return row !== undefined;\n}\n"],"names":["NATIVE_VECTOR_SCOPE_TABLE","TARGET_CHUNK_CAP","asCosine","compareVectorScores","createNativeVectorScopeOwner","hasVectorRow","pendingRows","reserveNativeVectorScope","sampleEvenly","nativeVectorScopeOwners","WeakMap","owner","bind","vectors","Error","has","set","activeFor","allowed","undefined","scope","phase","get","token","valid","current","activate","deactivate","invalidate","rows","cap","step","Math","max","ceil","length","filter","_","i","score","round","min","a","b","byScore","Buffer","compare","from","path","conn","stmt","prepare","all","row"],"mappings":";;;;;;;;;;;QAEaA;eAAAA;;QAiFAC;eAAAA;;QAUGC;eAAAA;;QAMAC;eAAAA;;QAlEAC;eAAAA;;QA+EMC;eAAAA;;QAPAC;eAAAA;;QAzDNC;eAAAA;;QAuCAC;eAAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AArFT,IAAMR,4BAA4B;AA0BzC,IAAMS,0BAA0B,IAAIC;AAK7B,SAASN;IACd,IAAMO,QAAqC,CAAC;IAC5C,OAAO;QACLC,MAAAA,SAAAA,KAAKC,OAAO;YACV,IAAIF,MAAME,OAAO,EAAE,MAAM,IAAIC,MAAM;YACnC,IAAIL,wBAAwBM,GAAG,CAACF,UAAU,MAAM,IAAIC,MAAM;YAC1DH,MAAME,OAAO,GAAGA;YAChBJ,wBAAwBO,GAAG,CAACH,SAASF;QACvC;QACAM,WAAAA,SAAAA,UAAUC,OAAO;gBACiBP;YAAhC,OAAOO,YAAYC,aAAaR,EAAAA,eAAAA,MAAMS,KAAK,cAAXT,mCAAAA,aAAaU,KAAK,MAAK,YAAYV,MAAMS,KAAK,CAACF,OAAO,KAAKA;QAC7F;IACF;AACF;AAEO,SAASX,yBAAyBM,OAAoB,EAAEK,OAAoB;IACjF,IAAMP,QAAQF,wBAAwBa,GAAG,CAACT;IAC1C,IAAI,CAACF,OAAO,OAAO;IACnB,IAAIA,MAAMS,KAAK,EAAE,MAAM,IAAIN,MAAM;IAEjC,IAAMS,QAAQ,CAAC;IACfZ,MAAMS,KAAK,GAAG;QAAEG,OAAAA;QAAOL,SAAAA;QAASG,OAAO;IAAgB;IACvD,IAAIG,QAAQ;IAEZ,IAAMC,UAAU;YACAd;QAAd,IAAI,CAACa,SAASb,EAAAA,eAAAA,MAAMS,KAAK,cAAXT,mCAAAA,aAAaY,KAAK,MAAKA,OAAO,MAAM,IAAIT,MAAM;QAC5D,OAAOH,MAAMS,KAAK;IACpB;IAEA,OAAO;QACLM,UAAAA,SAAAA;YACE,IAAMN,QAAQK;YACd,IAAIL,MAAMC,KAAK,KAAK,iBAAiB,MAAM,IAAIP,MAAM,AAAC,4CAAuD,OAAZM,MAAMC,KAAK;YAC5GD,MAAMC,KAAK,GAAG;QAChB;QACAM,YAAAA,SAAAA;YACE,IAAMP,QAAQK;YACd,IAAIL,MAAMC,KAAK,KAAK,YAAY,MAAM,IAAIP,MAAM;YAChDM,MAAMC,KAAK,GAAG;QAChB;QACAO,YAAAA,SAAAA;gBAEMjB;YADJ,IAAI,CAACa,OAAO;YACZ,IAAIb,EAAAA,eAAAA,MAAMS,KAAK,cAAXT,mCAAAA,aAAaY,KAAK,MAAKA,OAAOZ,MAAMS,KAAK,GAAGD;YAChDK,QAAQ;QACV;IACF;AACF;AAIO,IAAMvB,mBAAmB;AAIzB,SAASO,aAAgBqB,IAAS;QAAEC,MAAAA,iEAAc7B;IACvD,IAAM8B,OAAOC,KAAKC,GAAG,CAAC,GAAGD,KAAKE,IAAI,CAACL,KAAKM,MAAM,GAAGL;IACjD,OAAOD,KAAKO,MAAM,CAAC,SAACC,GAAGC;eAAMA,IAAIP,SAAS;;AAC5C;AAGO,SAAS7B,SAASqC,KAAa;IACpC,OAAOP,KAAKQ,KAAK,CAACR,KAAKS,GAAG,CAAC,GAAGT,KAAKC,GAAG,CAAC,CAAC,GAAGM,UAAU,QAAQ;AAC/D;AAIO,SAASpC,oBAAoBuC,CAAkC,EAAEC,CAAkC;IACxG,IAAMC,UAAUD,EAAEJ,KAAK,GAAGG,EAAEH,KAAK;IACjC,IAAIK,YAAY,GAAG,OAAOA;IAC1B,OAAOC,OAAOC,OAAO,CAACD,OAAOE,IAAI,CAACL,EAAEM,IAAI,GAAGH,OAAOE,IAAI,CAACJ,EAAEK,IAAI;AAC/D;AAEO,SAAe1C,YAAY2C,IAAgB;;YAC1CC;;;;oBAAO;;wBAAMD,KAAKE,OAAO,CAAC;;;oBAA1BD,OAAO;oBACL;;wBAAMA,KAAKE,GAAG;;;oBAAtB;;wBAAQ;;;;IACV;;AAIO,SAAe/C,aAAa4C,IAAgB,EAAED,IAAY;;YACzDE,MACAG;;;;oBADO;;wBAAMJ,KAAKE,OAAO,CAAC;;;oBAA1BD,OAAO;oBACA;;wBAAMA,KAAK5B,GAAG,CAAC0B;;;oBAAtBK,MAAO;oBACb;;wBAAOA,QAAQlC;;;;IACjB"}
|