sensemaking 0.18.4 → 0.19.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (368) hide show
  1. package/README.md +22 -4
  2. package/dist/cjs/chunk/group.js +10 -23
  3. package/dist/cjs/chunk/group.js.map +1 -1
  4. package/dist/cjs/chunk/parse.js +2 -5
  5. package/dist/cjs/chunk/parse.js.map +1 -1
  6. package/dist/cjs/cli/index.js.map +1 -1
  7. package/dist/cjs/cli/shared.js +6 -10
  8. package/dist/cjs/cli/shared.js.map +1 -1
  9. package/dist/cjs/cli/status.js +44 -10
  10. package/dist/cjs/cli/status.js.map +1 -1
  11. package/dist/cjs/cli.js +2 -8
  12. package/dist/cjs/cli.js.map +1 -1
  13. package/dist/cjs/commands/map.js +1 -3
  14. package/dist/cjs/commands/map.js.map +1 -1
  15. package/dist/cjs/commands/peek.js.map +1 -1
  16. package/dist/cjs/commands/related.js +3 -5
  17. package/dist/cjs/commands/related.js.map +1 -1
  18. package/dist/cjs/commands/scope.js.map +1 -1
  19. package/dist/cjs/commands/search.js +6 -8
  20. package/dist/cjs/commands/search.js.map +1 -1
  21. package/dist/cjs/commands/signals.js.map +1 -1
  22. package/dist/cjs/config/access.js.map +1 -1
  23. package/dist/cjs/config/index.d.cts +1 -1
  24. package/dist/cjs/config/index.d.ts +1 -1
  25. package/dist/cjs/config/index.js +3 -0
  26. package/dist/cjs/config/index.js.map +1 -1
  27. package/dist/cjs/config/load.js +10 -24
  28. package/dist/cjs/config/load.js.map +1 -1
  29. package/dist/cjs/config/resolve.js.map +1 -1
  30. package/dist/cjs/config/signals.js +2 -3
  31. package/dist/cjs/config/signals.js.map +1 -1
  32. package/dist/cjs/config/types.d.cts +2 -1
  33. package/dist/cjs/config/types.d.ts +2 -1
  34. package/dist/cjs/config/types.js +8 -0
  35. package/dist/cjs/config/types.js.map +1 -1
  36. package/dist/cjs/config/validate.d.cts +1 -1
  37. package/dist/cjs/config/validate.d.ts +1 -1
  38. package/dist/cjs/config/validate.js +9 -10
  39. package/dist/cjs/config/validate.js.map +1 -1
  40. package/dist/cjs/embed/cohere.js +2 -2
  41. package/dist/cjs/embed/cohere.js.map +1 -1
  42. package/dist/cjs/embed/distribution.js +3 -3
  43. package/dist/cjs/embed/distribution.js.map +1 -1
  44. package/dist/cjs/embed/handoff.d.cts +3 -0
  45. package/dist/cjs/embed/handoff.d.ts +3 -0
  46. package/dist/cjs/embed/handoff.js +39 -0
  47. package/dist/cjs/embed/handoff.js.map +1 -0
  48. package/dist/cjs/embed/identity.d.cts +11 -10
  49. package/dist/cjs/embed/identity.d.ts +11 -10
  50. package/dist/cjs/embed/identity.js +30 -15
  51. package/dist/cjs/embed/identity.js.map +1 -1
  52. package/dist/cjs/embed/langfit.js +2 -3
  53. package/dist/cjs/embed/langfit.js.map +1 -1
  54. package/dist/cjs/embed/languages.js +2 -4
  55. package/dist/cjs/embed/languages.js.map +1 -1
  56. package/dist/cjs/embed/openai.js +2 -2
  57. package/dist/cjs/embed/openai.js.map +1 -1
  58. package/dist/cjs/embed/query.js +58 -20
  59. package/dist/cjs/embed/query.js.map +1 -1
  60. package/dist/cjs/embed/registry.js +1 -2
  61. package/dist/cjs/embed/registry.js.map +1 -1
  62. package/dist/cjs/embed/static.d.cts +1 -1
  63. package/dist/cjs/embed/static.d.ts +1 -1
  64. package/dist/cjs/embed/static.js +8 -10
  65. package/dist/cjs/embed/static.js.map +1 -1
  66. package/dist/cjs/embed/store.d.cts +2 -2
  67. package/dist/cjs/embed/store.d.ts +2 -2
  68. package/dist/cjs/embed/store.js +48 -34
  69. package/dist/cjs/embed/store.js.map +1 -1
  70. package/dist/cjs/embed/types.js +1 -3
  71. package/dist/cjs/embed/types.js.map +1 -1
  72. package/dist/cjs/errors.d.cts +1 -1
  73. package/dist/cjs/errors.d.ts +1 -1
  74. package/dist/cjs/errors.js.map +1 -1
  75. package/dist/cjs/features/embed.js +19 -17
  76. package/dist/cjs/features/embed.js.map +1 -1
  77. package/dist/cjs/features/fences.js +5 -10
  78. package/dist/cjs/features/fences.js.map +1 -1
  79. package/dist/cjs/features/links.js +20 -37
  80. package/dist/cjs/features/links.js.map +1 -1
  81. package/dist/cjs/features/sections.js +21 -19
  82. package/dist/cjs/features/sections.js.map +1 -1
  83. package/dist/cjs/features/tags.js +5 -8
  84. package/dist/cjs/features/tags.js.map +1 -1
  85. package/dist/cjs/graph/traverse.js +1 -2
  86. package/dist/cjs/graph/traverse.js.map +1 -1
  87. package/dist/cjs/index.d.cts +1 -1
  88. package/dist/cjs/index.d.ts +1 -1
  89. package/dist/cjs/index.js +3 -0
  90. package/dist/cjs/index.js.map +1 -1
  91. package/dist/cjs/output/column-hint.js +3 -6
  92. package/dist/cjs/output/column-hint.js.map +1 -1
  93. package/dist/cjs/output/output.js +7 -14
  94. package/dist/cjs/output/output.js.map +1 -1
  95. package/dist/cjs/output/search-error.js +2 -5
  96. package/dist/cjs/output/search-error.js.map +1 -1
  97. package/dist/cjs/scan/frontmatter.d.cts +1 -0
  98. package/dist/cjs/scan/frontmatter.d.ts +1 -0
  99. package/dist/cjs/scan/frontmatter.js +14 -18
  100. package/dist/cjs/scan/frontmatter.js.map +1 -1
  101. package/dist/cjs/scan/list.js +2 -3
  102. package/dist/cjs/scan/list.js.map +1 -1
  103. package/dist/cjs/scan/reparse.js +12 -26
  104. package/dist/cjs/scan/reparse.js.map +1 -1
  105. package/dist/cjs/scan/worker-error.js.map +1 -1
  106. package/dist/cjs/store/cache.d.cts +2 -0
  107. package/dist/cjs/store/cache.d.ts +2 -0
  108. package/dist/cjs/store/cache.js +20 -0
  109. package/dist/cjs/store/cache.js.map +1 -0
  110. package/dist/cjs/store/duckdb/batch.js +4 -16
  111. package/dist/cjs/store/duckdb/batch.js.map +1 -1
  112. package/dist/cjs/store/duckdb/connection.js +8 -20
  113. package/dist/cjs/store/duckdb/connection.js.map +1 -1
  114. package/dist/cjs/store/duckdb/fieldStats.js +2 -8
  115. package/dist/cjs/store/duckdb/fieldStats.js.map +1 -1
  116. package/dist/cjs/store/duckdb/lexical.js +14 -28
  117. package/dist/cjs/store/duckdb/lexical.js.map +1 -1
  118. package/dist/cjs/store/duckdb/native.js.map +1 -1
  119. package/dist/cjs/store/duckdb/open.js +12 -13
  120. package/dist/cjs/store/duckdb/open.js.map +1 -1
  121. package/dist/cjs/store/duckdb/reconcile.js +4 -11
  122. package/dist/cjs/store/duckdb/reconcile.js.map +1 -1
  123. package/dist/cjs/store/duckdb/sql-functions.js +2 -4
  124. package/dist/cjs/store/duckdb/sql-functions.js.map +1 -1
  125. package/dist/cjs/store/duckdb/store.js +2 -3
  126. package/dist/cjs/store/duckdb/store.js.map +1 -1
  127. package/dist/cjs/store/duckdb/vectors.js +8 -18
  128. package/dist/cjs/store/duckdb/vectors.js.map +1 -1
  129. package/dist/cjs/store/index.d.cts +3 -2
  130. package/dist/cjs/store/index.d.ts +3 -2
  131. package/dist/cjs/store/index.js +7 -8
  132. package/dist/cjs/store/index.js.map +1 -1
  133. package/dist/cjs/store/native.js +6 -16
  134. package/dist/cjs/store/native.js.map +1 -1
  135. package/dist/cjs/store/sql-functions.js.map +1 -1
  136. package/dist/cjs/store/sqlite/connection.js +4 -8
  137. package/dist/cjs/store/sqlite/connection.js.map +1 -1
  138. package/dist/cjs/store/sqlite/fieldStats.js.map +1 -1
  139. package/dist/cjs/store/sqlite/lexical.js +3 -5
  140. package/dist/cjs/store/sqlite/lexical.js.map +1 -1
  141. package/dist/cjs/store/sqlite/open.d.cts +0 -1
  142. package/dist/cjs/store/sqlite/open.d.ts +0 -1
  143. package/dist/cjs/store/sqlite/open.js +22 -40
  144. package/dist/cjs/store/sqlite/open.js.map +1 -1
  145. package/dist/cjs/store/sqlite/reconcile.js +2 -3
  146. package/dist/cjs/store/sqlite/reconcile.js.map +1 -1
  147. package/dist/cjs/store/sqlite/sql-functions.js +2 -4
  148. package/dist/cjs/store/sqlite/sql-functions.js.map +1 -1
  149. package/dist/cjs/store/sqlite/store.js.map +1 -1
  150. package/dist/cjs/store/transaction.js +1 -3
  151. package/dist/cjs/store/transaction.js.map +1 -1
  152. package/dist/cjs/store/turso/connection.js +2 -6
  153. package/dist/cjs/store/turso/connection.js.map +1 -1
  154. package/dist/cjs/store/turso/fieldStats.js.map +1 -1
  155. package/dist/cjs/store/turso/lexical.d.cts +2 -0
  156. package/dist/cjs/store/turso/lexical.d.ts +2 -0
  157. package/dist/cjs/store/turso/lexical.js +347 -0
  158. package/dist/cjs/store/turso/lexical.js.map +1 -0
  159. package/dist/cjs/store/turso/native.js.map +1 -1
  160. package/dist/cjs/store/turso/open.d.cts +1 -1
  161. package/dist/cjs/store/turso/open.d.ts +1 -1
  162. package/dist/cjs/store/turso/open.js +81 -54
  163. package/dist/cjs/store/turso/open.js.map +1 -1
  164. package/dist/cjs/store/turso/reconcile.js +22 -22
  165. package/dist/cjs/store/turso/reconcile.js.map +1 -1
  166. package/dist/cjs/store/turso/store.js +18 -41
  167. package/dist/cjs/store/turso/store.js.map +1 -1
  168. package/dist/cjs/store/turso/vectors.d.cts +8 -0
  169. package/dist/cjs/store/turso/vectors.d.ts +8 -0
  170. package/dist/cjs/store/turso/vectors.js +435 -0
  171. package/dist/cjs/store/turso/vectors.js.map +1 -0
  172. package/dist/cjs/store/types.js +2 -3
  173. package/dist/cjs/store/types.js.map +1 -1
  174. package/dist/cjs/store/vectors.js.map +1 -1
  175. package/dist/cjs/text/segment.d.cts +1 -0
  176. package/dist/cjs/text/segment.d.ts +1 -0
  177. package/dist/cjs/text/segment.js +10 -14
  178. package/dist/cjs/text/segment.js.map +1 -1
  179. package/dist/cjs/watch.js +2 -5
  180. package/dist/cjs/watch.js.map +1 -1
  181. package/dist/cjs/workers/parse.js.map +1 -1
  182. package/dist/esm/chunk/group.js +10 -23
  183. package/dist/esm/chunk/group.js.map +1 -1
  184. package/dist/esm/chunk/parse.js +4 -8
  185. package/dist/esm/chunk/parse.js.map +1 -1
  186. package/dist/esm/cli/index.js +3 -5
  187. package/dist/esm/cli/index.js.map +1 -1
  188. package/dist/esm/cli/shared.js +12 -22
  189. package/dist/esm/cli/shared.js.map +1 -1
  190. package/dist/esm/cli/status.js +43 -9
  191. package/dist/esm/cli/status.js.map +1 -1
  192. package/dist/esm/cli.js +2 -8
  193. package/dist/esm/cli.js.map +1 -1
  194. package/dist/esm/commands/map.js +1 -3
  195. package/dist/esm/commands/map.js.map +1 -1
  196. package/dist/esm/commands/peek.js.map +1 -1
  197. package/dist/esm/commands/related.js +3 -5
  198. package/dist/esm/commands/related.js.map +1 -1
  199. package/dist/esm/commands/scope.js +3 -5
  200. package/dist/esm/commands/scope.js.map +1 -1
  201. package/dist/esm/commands/search.js +10 -16
  202. package/dist/esm/commands/search.js.map +1 -1
  203. package/dist/esm/commands/signals.js +3 -7
  204. package/dist/esm/commands/signals.js.map +1 -1
  205. package/dist/esm/config/access.js +8 -13
  206. package/dist/esm/config/access.js.map +1 -1
  207. package/dist/esm/config/index.d.ts +1 -1
  208. package/dist/esm/config/index.js +1 -1
  209. package/dist/esm/config/index.js.map +1 -1
  210. package/dist/esm/config/load.js +10 -24
  211. package/dist/esm/config/load.js.map +1 -1
  212. package/dist/esm/config/resolve.js +2 -3
  213. package/dist/esm/config/resolve.js.map +1 -1
  214. package/dist/esm/config/signals.js +3 -5
  215. package/dist/esm/config/signals.js.map +1 -1
  216. package/dist/esm/config/types.d.ts +2 -1
  217. package/dist/esm/config/types.js +8 -2
  218. package/dist/esm/config/types.js.map +1 -1
  219. package/dist/esm/config/validate.d.ts +1 -1
  220. package/dist/esm/config/validate.js +7 -10
  221. package/dist/esm/config/validate.js.map +1 -1
  222. package/dist/esm/embed/cohere.js +2 -2
  223. package/dist/esm/embed/cohere.js.map +1 -1
  224. package/dist/esm/embed/distribution.js +1 -1
  225. package/dist/esm/embed/distribution.js.map +1 -1
  226. package/dist/esm/embed/handoff.d.ts +3 -0
  227. package/dist/esm/embed/handoff.js +21 -0
  228. package/dist/esm/embed/handoff.js.map +1 -0
  229. package/dist/esm/embed/identity.d.ts +11 -10
  230. package/dist/esm/embed/identity.js +29 -26
  231. package/dist/esm/embed/identity.js.map +1 -1
  232. package/dist/esm/embed/langfit.js +2 -3
  233. package/dist/esm/embed/langfit.js.map +1 -1
  234. package/dist/esm/embed/languages.js +2 -4
  235. package/dist/esm/embed/languages.js.map +1 -1
  236. package/dist/esm/embed/openai.js +2 -2
  237. package/dist/esm/embed/openai.js.map +1 -1
  238. package/dist/esm/embed/query.js +21 -6
  239. package/dist/esm/embed/query.js.map +1 -1
  240. package/dist/esm/embed/registry.js +1 -2
  241. package/dist/esm/embed/registry.js.map +1 -1
  242. package/dist/esm/embed/static.d.ts +1 -1
  243. package/dist/esm/embed/static.js +8 -10
  244. package/dist/esm/embed/static.js.map +1 -1
  245. package/dist/esm/embed/store.d.ts +2 -2
  246. package/dist/esm/embed/store.js +28 -23
  247. package/dist/esm/embed/store.js.map +1 -1
  248. package/dist/esm/embed/types.js +1 -3
  249. package/dist/esm/embed/types.js.map +1 -1
  250. package/dist/esm/errors.d.ts +1 -1
  251. package/dist/esm/errors.js.map +1 -1
  252. package/dist/esm/features/embed.js +16 -16
  253. package/dist/esm/features/embed.js.map +1 -1
  254. package/dist/esm/features/fences.js +9 -19
  255. package/dist/esm/features/fences.js.map +1 -1
  256. package/dist/esm/features/links.js +24 -49
  257. package/dist/esm/features/links.js.map +1 -1
  258. package/dist/esm/features/sections.js +20 -20
  259. package/dist/esm/features/sections.js.map +1 -1
  260. package/dist/esm/features/tags.js +5 -8
  261. package/dist/esm/features/tags.js.map +1 -1
  262. package/dist/esm/features/types.js.map +1 -1
  263. package/dist/esm/graph/traverse.js +1 -2
  264. package/dist/esm/graph/traverse.js.map +1 -1
  265. package/dist/esm/index.d.ts +1 -1
  266. package/dist/esm/index.js +1 -1
  267. package/dist/esm/index.js.map +1 -1
  268. package/dist/esm/output/column-hint.js +3 -6
  269. package/dist/esm/output/column-hint.js.map +1 -1
  270. package/dist/esm/output/output.js +9 -20
  271. package/dist/esm/output/output.js.map +1 -1
  272. package/dist/esm/output/search-error.js +2 -5
  273. package/dist/esm/output/search-error.js.map +1 -1
  274. package/dist/esm/scan/frontmatter.d.ts +1 -0
  275. package/dist/esm/scan/frontmatter.js +13 -18
  276. package/dist/esm/scan/frontmatter.js.map +1 -1
  277. package/dist/esm/scan/list.js +2 -3
  278. package/dist/esm/scan/list.js.map +1 -1
  279. package/dist/esm/scan/reparse.js +16 -34
  280. package/dist/esm/scan/reparse.js.map +1 -1
  281. package/dist/esm/scan/worker-error.js.map +1 -1
  282. package/dist/esm/store/cache.d.ts +2 -0
  283. package/dist/esm/store/cache.js +11 -0
  284. package/dist/esm/store/cache.js.map +1 -0
  285. package/dist/esm/store/duckdb/batch.js +4 -16
  286. package/dist/esm/store/duckdb/batch.js.map +1 -1
  287. package/dist/esm/store/duckdb/connection.js +8 -20
  288. package/dist/esm/store/duckdb/connection.js.map +1 -1
  289. package/dist/esm/store/duckdb/fieldStats.js +4 -12
  290. package/dist/esm/store/duckdb/fieldStats.js.map +1 -1
  291. package/dist/esm/store/duckdb/lexical.js +19 -35
  292. package/dist/esm/store/duckdb/lexical.js.map +1 -1
  293. package/dist/esm/store/duckdb/native.js +2 -7
  294. package/dist/esm/store/duckdb/native.js.map +1 -1
  295. package/dist/esm/store/duckdb/open.js +16 -23
  296. package/dist/esm/store/duckdb/open.js.map +1 -1
  297. package/dist/esm/store/duckdb/reconcile.js +4 -11
  298. package/dist/esm/store/duckdb/reconcile.js.map +1 -1
  299. package/dist/esm/store/duckdb/sql-functions.js +4 -9
  300. package/dist/esm/store/duckdb/sql-functions.js.map +1 -1
  301. package/dist/esm/store/duckdb/store.js +6 -14
  302. package/dist/esm/store/duckdb/store.js.map +1 -1
  303. package/dist/esm/store/duckdb/vectors.js +14 -31
  304. package/dist/esm/store/duckdb/vectors.js.map +1 -1
  305. package/dist/esm/store/index.d.ts +3 -2
  306. package/dist/esm/store/index.js +5 -6
  307. package/dist/esm/store/index.js.map +1 -1
  308. package/dist/esm/store/native.js +10 -25
  309. package/dist/esm/store/native.js.map +1 -1
  310. package/dist/esm/store/sql-functions.js +3 -5
  311. package/dist/esm/store/sql-functions.js.map +1 -1
  312. package/dist/esm/store/sqlite/connection.js +4 -8
  313. package/dist/esm/store/sqlite/connection.js.map +1 -1
  314. package/dist/esm/store/sqlite/fieldStats.js +2 -6
  315. package/dist/esm/store/sqlite/fieldStats.js.map +1 -1
  316. package/dist/esm/store/sqlite/lexical.js +5 -9
  317. package/dist/esm/store/sqlite/lexical.js.map +1 -1
  318. package/dist/esm/store/sqlite/open.d.ts +0 -1
  319. package/dist/esm/store/sqlite/open.js +21 -41
  320. package/dist/esm/store/sqlite/open.js.map +1 -1
  321. package/dist/esm/store/sqlite/reconcile.js +6 -13
  322. package/dist/esm/store/sqlite/reconcile.js.map +1 -1
  323. package/dist/esm/store/sqlite/sql-functions.js +2 -4
  324. package/dist/esm/store/sqlite/sql-functions.js.map +1 -1
  325. package/dist/esm/store/sqlite/store.js +1 -2
  326. package/dist/esm/store/sqlite/store.js.map +1 -1
  327. package/dist/esm/store/transaction.js +2 -5
  328. package/dist/esm/store/transaction.js.map +1 -1
  329. package/dist/esm/store/turso/connection.js +2 -6
  330. package/dist/esm/store/turso/connection.js.map +1 -1
  331. package/dist/esm/store/turso/fieldStats.js +2 -5
  332. package/dist/esm/store/turso/fieldStats.js.map +1 -1
  333. package/dist/esm/store/turso/lexical.d.ts +2 -0
  334. package/dist/esm/store/turso/lexical.js +128 -0
  335. package/dist/esm/store/turso/lexical.js.map +1 -0
  336. package/dist/esm/store/turso/native.js +2 -7
  337. package/dist/esm/store/turso/native.js.map +1 -1
  338. package/dist/esm/store/turso/open.d.ts +1 -1
  339. package/dist/esm/store/turso/open.js +36 -29
  340. package/dist/esm/store/turso/open.js.map +1 -1
  341. package/dist/esm/store/turso/reconcile.js +22 -22
  342. package/dist/esm/store/turso/reconcile.js.map +1 -1
  343. package/dist/esm/store/turso/store.js +17 -28
  344. package/dist/esm/store/turso/store.js.map +1 -1
  345. package/dist/esm/store/turso/vectors.d.ts +8 -0
  346. package/dist/esm/store/turso/vectors.js +98 -0
  347. package/dist/esm/store/turso/vectors.js.map +1 -0
  348. package/dist/esm/store/types.js +2 -3
  349. package/dist/esm/store/types.js.map +1 -1
  350. package/dist/esm/store/vectors.js +2 -3
  351. package/dist/esm/store/vectors.js.map +1 -1
  352. package/dist/esm/text/segment.d.ts +1 -0
  353. package/dist/esm/text/segment.js +13 -21
  354. package/dist/esm/text/segment.js.map +1 -1
  355. package/dist/esm/watch.js +2 -5
  356. package/dist/esm/watch.js.map +1 -1
  357. package/dist/esm/workers/parse.js.map +1 -1
  358. package/package.json +3 -1
  359. package/schema.json +2 -2
  360. package/skills/sense/SKILL.md +2 -2
  361. package/skills/sense-setup/SKILL.md +3 -3
  362. package/dist/cjs/store/meta.d.cts +0 -3
  363. package/dist/cjs/store/meta.d.ts +0 -3
  364. package/dist/cjs/store/meta.js +0 -215
  365. package/dist/cjs/store/meta.js.map +0 -1
  366. package/dist/esm/store/meta.d.ts +0 -3
  367. package/dist/esm/store/meta.js +0 -17
  368. package/dist/esm/store/meta.js.map +0 -1
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/duckdb/lexical.ts"],"sourcesContent":["// D1: fts BM25 for ranking, contains() scans for exact substring / phrase verification /\n// unspaced scripts, JS excerpts for every doc (no snippet()).\nimport { SenseError } from '../../errors.ts';\nimport { UNSPACED_SCRIPTS } from '../../text/segment.ts';\nimport { withTransaction } from '../transaction.ts';\nimport type { Connection, LexicalHit, LexicalQueryOptions } from '../types.ts';\n\nconst FIELDS = ['title', 'summary', 'text'] as const;\ntype Field = (typeof FIELDS)[number];\n// Mirrors sqlite's bm25(content, 10.0, 5.0, 1.0, ...) column-weight intent; DuckDB's\n// match_bm25 has no field_weights argument (verified 1.5.5), so each field is scored\n// separately and combined here.\nconst FIELD_WEIGHT: Record<Field, number> = { title: 10, summary: 5, text: 1 };\n\nconst UNSPACED_RUN = new RegExp(`[${UNSPACED_SCRIPTS}]`, 'u');\n\n// FTS5 operator syntax (sqlite.org/fts5.html sec. 3) that words/substrings below would\n// otherwise silently treat as literal terms instead of honoring (PRINCIPLES: no-silent-modes).\n// Checked against terms with quoted spans blanked -- those go through contains() and are\n// supported.\nconst FTS5_OPERATORS: Array<{ label: string; re: RegExp }> = [\n { label: 'prefix query', re: /[\\p{L}\\p{N}_]+\\*(?=\\s|$)/u },\n { label: 'boolean operator', re: /(?:^|\\s)(?:AND|OR|NOT)(?=\\s|$)/ },\n { label: 'NEAR operator', re: /(?:^|\\s)NEAR\\b/ },\n { label: 'initial-token operator', re: /(?:^|\\s)\\^\\S+/ },\n { label: 'column filter', re: /(?:^|\\s)[\\p{L}_]\\w*\\s*:/u },\n];\n\nfunction unsupportedOperator(terms: string): { label: string; token: string } | null {\n const withoutPhrases = terms.replace(/\"[^\"]*\"/g, ' ');\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\ninterface FtsIndexState {\n stale: boolean;\n}\n\n// A run whose script marks no word boundaries makes match_bm25's whitespace tokenizer index\n// the whole run as one token (same gap FTS5 has without the `_seg` sidecar), so such runs --\n// and any author-quoted phrase, any script -- go through contains() instead of the fts index.\nfunction splitTerms(terms: string): { words: string[]; substrings: string[] } {\n const words: string[] = [];\n const substrings: string[] = [];\n const withoutPhrases = terms.replace(/\"([^\"]*)\"/g, (_m, inner: string) => {\n const phrase = inner.trim();\n if (phrase.length === 0) return ' ';\n substrings.push(phrase);\n if (!UNSPACED_RUN.test(phrase)) for (const w of phrase.split(/\\s+/)) if (w.length > 0) words.push(w);\n return ' ';\n });\n for (const tok of withoutPhrases.split(/\\s+/)) {\n if (tok.length === 0) continue;\n if (UNSPACED_RUN.test(tok)) substrings.push(tok);\n else words.push(tok);\n }\n return { words, substrings };\n}\n\n// No incremental update (verified 1.5.5: PRAGMA create_fts_index is rebuild-only), so this\n// pays the full rebuild -- but only once per store instance's first lexical query, and only\n// when content actually changed since (see FtsIndexState / duckdb/store.ts's markStale()).\nasync function ensureFtsFresh(conn: Connection, state: FtsIndexState): Promise<void> {\n if (!state.stale) return;\n await conn.exec('INSTALL fts; LOAD fts;');\n await withTransaction(conn, async () => {\n // stopwords='none': sqlite's porter/unicode61 tokenizer never removes stopwords either\n // (verified 1.5.5), and the fts extension's default 571-word English list would otherwise\n // silently drop common query words (e.g. \"and\") from the index but not from match_bm25's\n // conjunctive gate, making a bare multi-word query that contains one match nothing.\n await conn.exec(`PRAGMA create_fts_index('content', 'path', 'title', 'summary', 'text', stopwords='none', overwrite=1)`);\n });\n state.stale = false;\n}\n\n// Per-word-term and per-substring-term SQL fragments, field-weighted the same way for both\n// (title 10 / summary 5 / text 1), plus the params in the exact left-to-right order they are\n// emitted -- positional `?` binding requires that order to match the assembled SQL text. Score\n// and gate params are collected apart and joined at the end because the assembled SQL puts\n// every score part ahead of every gate part: appending to one array term by term matches that\n// text only while the query has words or substrings, never both, and silently shifts each `?`\n// by one as soon as it has both (PRINCIPLES: no-silent-modes -- a shifted bind answers a\n// different question rather than failing).\nfunction buildScoreAndGate(words: string[], substrings: string[]): { scoreSql: string; gateSql: string; params: unknown[] } {\n const scoreParts: string[] = [];\n const gateParts: string[] = [];\n const scoreParams: unknown[] = [];\n const gateParams: unknown[] = [];\n\n if (words.length > 0) {\n const wordQuery = words.join(' ');\n for (const field of FIELDS) {\n scoreParts.push(`${FIELD_WEIGHT[field]}.0 * COALESCE(fts_main_content.match_bm25(content.\"path\", ?, fields := '${field}', conjunctive := false), 0)`);\n scoreParams.push(wordQuery);\n }\n gateParts.push(`fts_main_content.match_bm25(content.\"path\", ?, conjunctive := true) IS NOT NULL`);\n gateParams.push(wordQuery);\n }\n\n for (const raw of substrings) {\n const needle = raw.toLowerCase();\n for (const field of FIELDS) {\n scoreParts.push(`${FIELD_WEIGHT[field]}.0 * CAST(contains(lower(content.${field}), ?) AS INTEGER)`);\n scoreParams.push(needle);\n }\n gateParts.push(`contains(lower(content.title) || ' ' || lower(content.summary) || ' ' || lower(content.text), ?)`);\n gateParams.push(needle);\n }\n\n return { scoreSql: scoreParts.join(' + '), gateSql: gateParts.join(' AND '), params: [...scoreParams, ...gateParams] };\n}\n\n// Ranked word-match query, scoped by the caller-built SQL fragments (same fragments sqlite's\n// queryLexical takes). `hit` is always NULL: no snippet() equivalent exists, so every row goes\n// through the caller's JS excerpt fallback (commands/search.ts) rather than a second excerpt path.\nexport async function queryLexical(conn: Connection, terms: string, opts: LexicalQueryOptions, state: FtsIndexState): Promise<LexicalHit[]> {\n const unsupported = unsupportedOperator(terms);\n if (unsupported !== null) {\n throw new SenseError('STORE_CAPABILITY_MISSING', `store \"duckdb\" does not implement FTS5's ${unsupported.label} (\"${unsupported.token}\") in this build; rephrase \"${terms.trim()}\" without it, or set \"store\" to \"sqlite\" in this tree's config to search it as written`);\n }\n const { whereJoin, whereCond, scopeCond, limit } = opts;\n const { words, substrings } = splitTerms(terms);\n if (words.length === 0 && substrings.length === 0) return [];\n\n if (words.length > 0) await ensureFtsFresh(conn, state);\n\n const { scoreSql, gateSql, params } = buildScoreAndGate(words, substrings);\n const sql = `SELECT path, NULL AS hit FROM (\n SELECT content.\"path\" AS path, (${scoreSql}) AS score\n FROM content\n ${whereJoin}\n WHERE ${gateSql}\n ${whereCond} ${scopeCond}\n ) sq ORDER BY score DESC, path LIMIT ?`;\n const stmt = await conn.prepare(sql);\n return (await stmt.all(...params, limit)) as unknown as LexicalHit[];\n}\n\n// One instance per store: `stale` starts true (a fresh connection cannot know whether an\n// on-disk fts schema still matches `content`), and store.ts's reconcile wrapper flips it back\n// to true whenever content changes, so the next query rebuilds before ranking against it.\nexport function createLexicalIndex(conn: Connection): { query: (terms: string, opts: LexicalQueryOptions) => Promise<LexicalHit[]>; markStale: () => void } {\n const state: FtsIndexState = { stale: true };\n return {\n query: (terms, opts) => queryLexical(conn, terms, opts, state),\n markStale: () => {\n state.stale = true;\n },\n };\n}\n"],"names":["SenseError","UNSPACED_SCRIPTS","withTransaction","FIELDS","FIELD_WEIGHT","title","summary","text","UNSPACED_RUN","RegExp","FTS5_OPERATORS","label","re","unsupportedOperator","terms","withoutPhrases","replace","m","match","token","trim","splitTerms","words","substrings","_m","inner","phrase","length","push","test","w","split","tok","ensureFtsFresh","conn","state","stale","exec","buildScoreAndGate","scoreParts","gateParts","scoreParams","gateParams","wordQuery","join","field","raw","needle","toLowerCase","scoreSql","gateSql","params","queryLexical","opts","unsupported","whereJoin","whereCond","scopeCond","limit","sql","stmt","prepare","all","createLexicalIndex","query","markStale"],"mappings":"AAAA,yFAAyF;AACzF,8DAA8D;AAC9D,SAASA,UAAU,QAAQ,kBAAkB;AAC7C,SAASC,gBAAgB,QAAQ,wBAAwB;AACzD,SAASC,eAAe,QAAQ,oBAAoB;AAGpD,MAAMC,SAAS;IAAC;IAAS;IAAW;CAAO;AAE3C,qFAAqF;AACrF,qFAAqF;AACrF,gCAAgC;AAChC,MAAMC,eAAsC;IAAEC,OAAO;IAAIC,SAAS;IAAGC,MAAM;AAAE;AAE7E,MAAMC,eAAe,IAAIC,OAAO,CAAC,CAAC,EAAER,iBAAiB,CAAC,CAAC,EAAE;AAEzD,uFAAuF;AACvF,+FAA+F;AAC/F,yFAAyF;AACzF,aAAa;AACb,MAAMS,iBAAuD;IAC3D;QAAEC,OAAO;QAAgBC,IAAI;IAA4B;IACzD;QAAED,OAAO;QAAoBC,IAAI;IAAiC;IAClE;QAAED,OAAO;QAAiBC,IAAI;IAAiB;IAC/C;QAAED,OAAO;QAA0BC,IAAI;IAAgB;IACvD;QAAED,OAAO;QAAiBC,IAAI;IAA2B;CAC1D;AAED,SAASC,oBAAoBC,KAAa;IACxC,MAAMC,iBAAiBD,MAAME,OAAO,CAAC,YAAY;IACjD,KAAK,MAAM,EAAEL,KAAK,EAAEC,EAAE,EAAE,IAAIF,eAAgB;QAC1C,MAAMO,IAAIF,eAAeG,KAAK,CAACN;QAC/B,IAAIK,GAAG,OAAO;YAAEN;YAAOQ,OAAOF,CAAC,CAAC,EAAE,CAACG,IAAI;QAAG;IAC5C;IACA,OAAO;AACT;AAMA,4FAA4F;AAC5F,6FAA6F;AAC7F,8FAA8F;AAC9F,SAASC,WAAWP,KAAa;IAC/B,MAAMQ,QAAkB,EAAE;IAC1B,MAAMC,aAAuB,EAAE;IAC/B,MAAMR,iBAAiBD,MAAME,OAAO,CAAC,cAAc,CAACQ,IAAIC;QACtD,MAAMC,SAASD,MAAML,IAAI;QACzB,IAAIM,OAAOC,MAAM,KAAK,GAAG,OAAO;QAChCJ,WAAWK,IAAI,CAACF;QAChB,IAAI,CAAClB,aAAaqB,IAAI,CAACH,SAAS;YAAA,KAAK,MAAMI,KAAKJ,OAAOK,KAAK,CAAC,OAAQ,IAAID,EAAEH,MAAM,GAAG,GAAGL,MAAMM,IAAI,CAACE;QAAE;QACpG,OAAO;IACT;IACA,KAAK,MAAME,OAAOjB,eAAegB,KAAK,CAAC,OAAQ;QAC7C,IAAIC,IAAIL,MAAM,KAAK,GAAG;QACtB,IAAInB,aAAaqB,IAAI,CAACG,MAAMT,WAAWK,IAAI,CAACI;aACvCV,MAAMM,IAAI,CAACI;IAClB;IACA,OAAO;QAAEV;QAAOC;IAAW;AAC7B;AAEA,2FAA2F;AAC3F,4FAA4F;AAC5F,2FAA2F;AAC3F,eAAeU,eAAeC,IAAgB,EAAEC,KAAoB;IAClE,IAAI,CAACA,MAAMC,KAAK,EAAE;IAClB,MAAMF,KAAKG,IAAI,CAAC;IAChB,MAAMnC,gBAAgBgC,MAAM;QAC1B,uFAAuF;QACvF,0FAA0F;QAC1F,yFAAyF;QACzF,oFAAoF;QACpF,MAAMA,KAAKG,IAAI,CAAC,CAAC,qGAAqG,CAAC;IACzH;IACAF,MAAMC,KAAK,GAAG;AAChB;AAEA,2FAA2F;AAC3F,6FAA6F;AAC7F,+FAA+F;AAC/F,2FAA2F;AAC3F,8FAA8F;AAC9F,8FAA8F;AAC9F,yFAAyF;AACzF,2CAA2C;AAC3C,SAASE,kBAAkBhB,KAAe,EAAEC,UAAoB;IAC9D,MAAMgB,aAAuB,EAAE;IAC/B,MAAMC,YAAsB,EAAE;IAC9B,MAAMC,cAAyB,EAAE;IACjC,MAAMC,aAAwB,EAAE;IAEhC,IAAIpB,MAAMK,MAAM,GAAG,GAAG;QACpB,MAAMgB,YAAYrB,MAAMsB,IAAI,CAAC;QAC7B,KAAK,MAAMC,SAAS1C,OAAQ;YAC1BoC,WAAWX,IAAI,CAAC,GAAGxB,YAAY,CAACyC,MAAM,CAAC,wEAAwE,EAAEA,MAAM,4BAA4B,CAAC;YACpJJ,YAAYb,IAAI,CAACe;QACnB;QACAH,UAAUZ,IAAI,CAAC,CAAC,+EAA+E,CAAC;QAChGc,WAAWd,IAAI,CAACe;IAClB;IAEA,KAAK,MAAMG,OAAOvB,WAAY;QAC5B,MAAMwB,SAASD,IAAIE,WAAW;QAC9B,KAAK,MAAMH,SAAS1C,OAAQ;YAC1BoC,WAAWX,IAAI,CAAC,GAAGxB,YAAY,CAACyC,MAAM,CAAC,iCAAiC,EAAEA,MAAM,iBAAiB,CAAC;YAClGJ,YAAYb,IAAI,CAACmB;QACnB;QACAP,UAAUZ,IAAI,CAAC,CAAC,gGAAgG,CAAC;QACjHc,WAAWd,IAAI,CAACmB;IAClB;IAEA,OAAO;QAAEE,UAAUV,WAAWK,IAAI,CAAC;QAAQM,SAASV,UAAUI,IAAI,CAAC;QAAUO,QAAQ;eAAIV;eAAgBC;SAAW;IAAC;AACvH;AAEA,6FAA6F;AAC7F,+FAA+F;AAC/F,mGAAmG;AACnG,OAAO,eAAeU,aAAalB,IAAgB,EAAEpB,KAAa,EAAEuC,IAAyB,EAAElB,KAAoB;IACjH,MAAMmB,cAAczC,oBAAoBC;IACxC,IAAIwC,gBAAgB,MAAM;QACxB,MAAM,IAAItD,WAAW,4BAA4B,CAAC,yCAAyC,EAAEsD,YAAY3C,KAAK,CAAC,GAAG,EAAE2C,YAAYnC,KAAK,CAAC,4BAA4B,EAAEL,MAAMM,IAAI,GAAG,sFAAsF,CAAC;IAC1Q;IACA,MAAM,EAAEmC,SAAS,EAAEC,SAAS,EAAEC,SAAS,EAAEC,KAAK,EAAE,GAAGL;IACnD,MAAM,EAAE/B,KAAK,EAAEC,UAAU,EAAE,GAAGF,WAAWP;IACzC,IAAIQ,MAAMK,MAAM,KAAK,KAAKJ,WAAWI,MAAM,KAAK,GAAG,OAAO,EAAE;IAE5D,IAAIL,MAAMK,MAAM,GAAG,GAAG,MAAMM,eAAeC,MAAMC;IAEjD,MAAM,EAAEc,QAAQ,EAAEC,OAAO,EAAEC,MAAM,EAAE,GAAGb,kBAAkBhB,OAAOC;IAC/D,MAAMoC,MAAM,CAAC;oCACqB,EAAEV,SAAS;;IAE3C,EAAEM,UAAU;UACN,EAAEL,QAAQ;IAChB,EAAEM,UAAU,CAAC,EAAEC,UAAU;wCACW,CAAC;IACvC,MAAMG,OAAO,MAAM1B,KAAK2B,OAAO,CAACF;IAChC,OAAQ,MAAMC,KAAKE,GAAG,IAAIX,QAAQO;AACpC;AAEA,yFAAyF;AACzF,8FAA8F;AAC9F,0FAA0F;AAC1F,OAAO,SAASK,mBAAmB7B,IAAgB;IACjD,MAAMC,QAAuB;QAAEC,OAAO;IAAK;IAC3C,OAAO;QACL4B,OAAO,CAAClD,OAAOuC,OAASD,aAAalB,MAAMpB,OAAOuC,MAAMlB;QACxD8B,WAAW;YACT9B,MAAMC,KAAK,GAAG;QAChB;IACF;AACF"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/duckdb/lexical.ts"],"sourcesContent":["// D1: fts BM25 for ranking, contains() scans for exact substring / phrase verification /\n// unspaced scripts, JS excerpts for every doc (no snippet()).\nimport { SenseError } from '../../errors.ts';\nimport { hasUnspacedRun } from '../../text/segment.ts';\nimport { withTransaction } from '../transaction.ts';\nimport type { Connection, LexicalHit, LexicalQueryOptions } from '../types.ts';\n\nconst FIELDS = ['title', 'summary', 'text'] as const;\ntype Field = (typeof FIELDS)[number];\n// Mirrors sqlite's bm25(content, 10.0, 5.0, 1.0, ...) column-weight intent; DuckDB's match_bm25 has no\n// field_weights argument (verified 1.5.5), so each field is scored separately and combined here.\nconst FIELD_WEIGHT: Record<Field, number> = { title: 10, summary: 5, text: 1 };\n\n// FTS5 operator syntax (sqlite.org/fts5.html sec. 3) that words/substrings below would otherwise silently\n// treat as literal terms (PRINCIPLES: no-silent-modes). Checked with quoted spans blanked; those go through contains() and are supported.\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: 'initial-token operator', re: /(?:^|\\s)\\^\\S+/ },\n { label: 'column filter', re: /(?:^|\\s)[\\p{L}_]\\w*\\s*:/u },\n];\n\nfunction unsupportedOperator(terms: string): { label: string; token: string } | null {\n const withoutPhrases = terms.replace(/\"[^\"]*\"/g, ' ');\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\ninterface FtsIndexState {\n stale: boolean;\n}\n\n// A run whose script marks no word boundaries makes match_bm25's whitespace tokenizer index the whole run as one token\n// (same gap FTS5 has without the `_seg` sidecar), so such runs -- and any author-quoted phrase, any script -- go through contains() instead.\nfunction splitTerms(terms: string): { words: string[]; substrings: string[] } {\n const words: string[] = [];\n const substrings: string[] = [];\n const withoutPhrases = terms.replace(/\"([^\"]*)\"/g, (_m, inner: string) => {\n const phrase = inner.trim();\n if (phrase.length === 0) return ' ';\n substrings.push(phrase);\n if (!hasUnspacedRun(phrase)) for (const w of phrase.split(/\\s+/)) if (w.length > 0) words.push(w);\n return ' ';\n });\n for (const tok of withoutPhrases.split(/\\s+/)) {\n if (tok.length === 0) continue;\n if (hasUnspacedRun(tok)) substrings.push(tok);\n else words.push(tok);\n }\n return { words, substrings };\n}\n\n// No incremental update (verified 1.5.5: PRAGMA create_fts_index is rebuild-only), so this pays the full rebuild -- but only\n// once per store instance's first lexical query, and only when content actually changed since (FtsIndexState / store.ts's markStale()).\nasync function ensureFtsFresh(conn: Connection, state: FtsIndexState): Promise<void> {\n if (!state.stale) return;\n await conn.exec('INSTALL fts; LOAD fts;');\n await withTransaction(conn, async () => {\n // stopwords='none': sqlite's porter/unicode61 tokenizer never removes stopwords either (verified 1.5.5); the fts extension's\n // default 571-word list would otherwise drop common words (e.g. \"and\") from the index but not match_bm25's conjunctive gate, breaking multi-word matches.\n await conn.exec(`PRAGMA create_fts_index('content', 'path', 'title', 'summary', 'text', stopwords='none', overwrite=1)`);\n });\n state.stale = false;\n}\n\n// Field-weighted fragments (title 10 / summary 5 / text 1), with params emitted in the exact left-to-right order the assembled SQL needs for positional `?` binding.\n// score/gate params stay in separate arrays until joined at the end, since the SQL puts every score part before every gate part; sharing one array would shift binds once a query has both words and substrings (PRINCIPLES: no-silent-modes).\nfunction buildScoreAndGate(words: string[], substrings: string[]): { scoreSql: string; gateSql: string; params: unknown[] } {\n const scoreParts: string[] = [];\n const gateParts: string[] = [];\n const scoreParams: unknown[] = [];\n const gateParams: unknown[] = [];\n\n if (words.length > 0) {\n const wordQuery = words.join(' ');\n for (const field of FIELDS) {\n scoreParts.push(`${FIELD_WEIGHT[field]}.0 * COALESCE(fts_main_content.match_bm25(content.\"path\", ?, fields := '${field}', conjunctive := false), 0)`);\n scoreParams.push(wordQuery);\n }\n gateParts.push(`fts_main_content.match_bm25(content.\"path\", ?, conjunctive := true) IS NOT NULL`);\n gateParams.push(wordQuery);\n }\n\n for (const raw of substrings) {\n const needle = raw.toLowerCase();\n for (const field of FIELDS) {\n scoreParts.push(`${FIELD_WEIGHT[field]}.0 * CAST(contains(lower(content.${field}), ?) AS INTEGER)`);\n scoreParams.push(needle);\n }\n gateParts.push(`contains(lower(content.title) || ' ' || lower(content.summary) || ' ' || lower(content.text), ?)`);\n gateParams.push(needle);\n }\n\n return { scoreSql: scoreParts.join(' + '), gateSql: gateParts.join(' AND '), params: [...scoreParams, ...gateParams] };\n}\n\n// Ranked word-match query, scoped by the caller-built SQL fragments (same fragments sqlite's queryLexical takes).\n// `hit` is always NULL: no snippet() equivalent exists, so every row goes through the caller's JS excerpt fallback (commands/search.ts).\nexport async function queryLexical(conn: Connection, terms: string, opts: LexicalQueryOptions, state: FtsIndexState): Promise<LexicalHit[]> {\n const unsupported = unsupportedOperator(terms);\n if (unsupported !== null) {\n throw new SenseError('STORE_CAPABILITY_MISSING', `store \"duckdb\" does not implement FTS5's ${unsupported.label} (\"${unsupported.token}\") in this build; rephrase \"${terms.trim()}\" without it, or set \"store\" to \"sqlite\" in this tree's config to search it as written`);\n }\n const { whereJoin, whereCond, scopeCond, limit } = opts;\n const { words, substrings } = splitTerms(terms);\n if (words.length === 0 && substrings.length === 0) return [];\n\n if (words.length > 0) await ensureFtsFresh(conn, state);\n\n const { scoreSql, gateSql, params } = buildScoreAndGate(words, substrings);\n const sql = `SELECT path, NULL AS hit FROM (\n SELECT content.\"path\" AS path, (${scoreSql}) AS score\n FROM content\n ${whereJoin}\n WHERE ${gateSql}\n ${whereCond} ${scopeCond}\n ) sq ORDER BY score DESC, path LIMIT ?`;\n const stmt = await conn.prepare(sql);\n return (await stmt.all(...params, limit)) as unknown as LexicalHit[];\n}\n\n// One instance per store: `stale` starts true (a fresh connection can't know whether an on-disk fts schema matches `content`),\n// and store.ts's reconcile wrapper flips it back to true whenever content changes, so the next query rebuilds first.\nexport function createLexicalIndex(conn: Connection): { query: (terms: string, opts: LexicalQueryOptions) => Promise<LexicalHit[]>; markStale: () => void } {\n const state: FtsIndexState = { stale: true };\n return {\n query: (terms, opts) => queryLexical(conn, terms, opts, state),\n markStale: () => {\n state.stale = true;\n },\n };\n}\n"],"names":["SenseError","hasUnspacedRun","withTransaction","FIELDS","FIELD_WEIGHT","title","summary","text","FTS5_OPERATORS","label","re","unsupportedOperator","terms","withoutPhrases","replace","m","match","token","trim","splitTerms","words","substrings","_m","inner","phrase","length","push","w","split","tok","ensureFtsFresh","conn","state","stale","exec","buildScoreAndGate","scoreParts","gateParts","scoreParams","gateParams","wordQuery","join","field","raw","needle","toLowerCase","scoreSql","gateSql","params","queryLexical","opts","unsupported","whereJoin","whereCond","scopeCond","limit","sql","stmt","prepare","all","createLexicalIndex","query","markStale"],"mappings":"AAAA,yFAAyF;AACzF,8DAA8D;AAC9D,SAASA,UAAU,QAAQ,kBAAkB;AAC7C,SAASC,cAAc,QAAQ,wBAAwB;AACvD,SAASC,eAAe,QAAQ,oBAAoB;AAGpD,MAAMC,SAAS;IAAC;IAAS;IAAW;CAAO;AAE3C,uGAAuG;AACvG,iGAAiG;AACjG,MAAMC,eAAsC;IAAEC,OAAO;IAAIC,SAAS;IAAGC,MAAM;AAAE;AAE7E,0GAA0G;AAC1G,0IAA0I;AAC1I,MAAMC,iBAAuD;IAC3D;QAAEC,OAAO;QAAgBC,IAAI;IAA4B;IACzD;QAAED,OAAO;QAAoBC,IAAI;IAAiC;IAClE;QAAED,OAAO;QAAiBC,IAAI;IAAiB;IAC/C;QAAED,OAAO;QAA0BC,IAAI;IAAgB;IACvD;QAAED,OAAO;QAAiBC,IAAI;IAA2B;CAC1D;AAED,SAASC,oBAAoBC,KAAa;IACxC,MAAMC,iBAAiBD,MAAME,OAAO,CAAC,YAAY;IACjD,KAAK,MAAM,EAAEL,KAAK,EAAEC,EAAE,EAAE,IAAIF,eAAgB;QAC1C,MAAMO,IAAIF,eAAeG,KAAK,CAACN;QAC/B,IAAIK,GAAG,OAAO;YAAEN;YAAOQ,OAAOF,CAAC,CAAC,EAAE,CAACG,IAAI;QAAG;IAC5C;IACA,OAAO;AACT;AAMA,uHAAuH;AACvH,6IAA6I;AAC7I,SAASC,WAAWP,KAAa;IAC/B,MAAMQ,QAAkB,EAAE;IAC1B,MAAMC,aAAuB,EAAE;IAC/B,MAAMR,iBAAiBD,MAAME,OAAO,CAAC,cAAc,CAACQ,IAAIC;QACtD,MAAMC,SAASD,MAAML,IAAI;QACzB,IAAIM,OAAOC,MAAM,KAAK,GAAG,OAAO;QAChCJ,WAAWK,IAAI,CAACF;QAChB,IAAI,CAACvB,eAAeuB,SAAS;YAAA,KAAK,MAAMG,KAAKH,OAAOI,KAAK,CAAC,OAAQ,IAAID,EAAEF,MAAM,GAAG,GAAGL,MAAMM,IAAI,CAACC;QAAE;QACjG,OAAO;IACT;IACA,KAAK,MAAME,OAAOhB,eAAee,KAAK,CAAC,OAAQ;QAC7C,IAAIC,IAAIJ,MAAM,KAAK,GAAG;QACtB,IAAIxB,eAAe4B,MAAMR,WAAWK,IAAI,CAACG;aACpCT,MAAMM,IAAI,CAACG;IAClB;IACA,OAAO;QAAET;QAAOC;IAAW;AAC7B;AAEA,6HAA6H;AAC7H,wIAAwI;AACxI,eAAeS,eAAeC,IAAgB,EAAEC,KAAoB;IAClE,IAAI,CAACA,MAAMC,KAAK,EAAE;IAClB,MAAMF,KAAKG,IAAI,CAAC;IAChB,MAAMhC,gBAAgB6B,MAAM;QAC1B,6HAA6H;QAC7H,0JAA0J;QAC1J,MAAMA,KAAKG,IAAI,CAAC,CAAC,qGAAqG,CAAC;IACzH;IACAF,MAAMC,KAAK,GAAG;AAChB;AAEA,qKAAqK;AACrK,+OAA+O;AAC/O,SAASE,kBAAkBf,KAAe,EAAEC,UAAoB;IAC9D,MAAMe,aAAuB,EAAE;IAC/B,MAAMC,YAAsB,EAAE;IAC9B,MAAMC,cAAyB,EAAE;IACjC,MAAMC,aAAwB,EAAE;IAEhC,IAAInB,MAAMK,MAAM,GAAG,GAAG;QACpB,MAAMe,YAAYpB,MAAMqB,IAAI,CAAC;QAC7B,KAAK,MAAMC,SAASvC,OAAQ;YAC1BiC,WAAWV,IAAI,CAAC,GAAGtB,YAAY,CAACsC,MAAM,CAAC,wEAAwE,EAAEA,MAAM,4BAA4B,CAAC;YACpJJ,YAAYZ,IAAI,CAACc;QACnB;QACAH,UAAUX,IAAI,CAAC,CAAC,+EAA+E,CAAC;QAChGa,WAAWb,IAAI,CAACc;IAClB;IAEA,KAAK,MAAMG,OAAOtB,WAAY;QAC5B,MAAMuB,SAASD,IAAIE,WAAW;QAC9B,KAAK,MAAMH,SAASvC,OAAQ;YAC1BiC,WAAWV,IAAI,CAAC,GAAGtB,YAAY,CAACsC,MAAM,CAAC,iCAAiC,EAAEA,MAAM,iBAAiB,CAAC;YAClGJ,YAAYZ,IAAI,CAACkB;QACnB;QACAP,UAAUX,IAAI,CAAC,CAAC,gGAAgG,CAAC;QACjHa,WAAWb,IAAI,CAACkB;IAClB;IAEA,OAAO;QAAEE,UAAUV,WAAWK,IAAI,CAAC;QAAQM,SAASV,UAAUI,IAAI,CAAC;QAAUO,QAAQ;eAAIV;eAAgBC;SAAW;IAAC;AACvH;AAEA,kHAAkH;AAClH,yIAAyI;AACzI,OAAO,eAAeU,aAAalB,IAAgB,EAAEnB,KAAa,EAAEsC,IAAyB,EAAElB,KAAoB;IACjH,MAAMmB,cAAcxC,oBAAoBC;IACxC,IAAIuC,gBAAgB,MAAM;QACxB,MAAM,IAAInD,WAAW,4BAA4B,CAAC,yCAAyC,EAAEmD,YAAY1C,KAAK,CAAC,GAAG,EAAE0C,YAAYlC,KAAK,CAAC,4BAA4B,EAAEL,MAAMM,IAAI,GAAG,sFAAsF,CAAC;IAC1Q;IACA,MAAM,EAAEkC,SAAS,EAAEC,SAAS,EAAEC,SAAS,EAAEC,KAAK,EAAE,GAAGL;IACnD,MAAM,EAAE9B,KAAK,EAAEC,UAAU,EAAE,GAAGF,WAAWP;IACzC,IAAIQ,MAAMK,MAAM,KAAK,KAAKJ,WAAWI,MAAM,KAAK,GAAG,OAAO,EAAE;IAE5D,IAAIL,MAAMK,MAAM,GAAG,GAAG,MAAMK,eAAeC,MAAMC;IAEjD,MAAM,EAAEc,QAAQ,EAAEC,OAAO,EAAEC,MAAM,EAAE,GAAGb,kBAAkBf,OAAOC;IAC/D,MAAMmC,MAAM,CAAC;oCACqB,EAAEV,SAAS;;IAE3C,EAAEM,UAAU;UACN,EAAEL,QAAQ;IAChB,EAAEM,UAAU,CAAC,EAAEC,UAAU;wCACW,CAAC;IACvC,MAAMG,OAAO,MAAM1B,KAAK2B,OAAO,CAACF;IAChC,OAAQ,MAAMC,KAAKE,GAAG,IAAIX,QAAQO;AACpC;AAEA,+HAA+H;AAC/H,qHAAqH;AACrH,OAAO,SAASK,mBAAmB7B,IAAgB;IACjD,MAAMC,QAAuB;QAAEC,OAAO;IAAK;IAC3C,OAAO;QACL4B,OAAO,CAACjD,OAAOsC,OAASD,aAAalB,MAAMnB,OAAOsC,MAAMlB;QACxD8B,WAAW;YACT9B,MAAMC,KAAK,GAAG;QAChB;IACF;AACF"}
@@ -6,13 +6,8 @@ const DUCKDB = {
6
6
  sizeHint: '~110MB'
7
7
  };
8
8
  let duckdbApiPromise;
9
- // The one accessor every consumer of @duckdb/node-api (open.ts, sql-functions.ts, vectors.ts)
10
- // must go through, rather than each doing its own `import('@duckdb/node-api')`: once a fresh
11
- // install has happened in this process, Node keeps returning that pre-install "not found" miss
12
- // to any NEW bare-specifier lookup of the same path, so a second independent import elsewhere
13
- // in the process fails even though the package is now genuinely on disk (the same stale-cache
14
- // problem loadOrInstall works around for its own retry). Caching the resolved module here means
15
- // only the first caller ever resolves the bare specifier at all.
9
+ // The one accessor every consumer of @duckdb/node-api must go through: after a fresh install in this process, Node keeps returning
10
+ // the pre-install "not found" miss to any new bare-specifier lookup of the same path, so caching the resolved module here means only the first caller ever resolves it.
16
11
  export function duckdbApi() {
17
12
  if (!duckdbApiPromise) duckdbApiPromise = loadOrInstall(DUCKDB, packageNodeModules());
18
13
  return duckdbApiPromise;
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/duckdb/native.ts"],"sourcesContent":["import { loadOrInstall, type NativeDescriptor, packageNodeModules } from '../native.ts';\n\nexport const DUCKDB_PACKAGE = '@duckdb/node-api';\n\nconst DUCKDB: NativeDescriptor = { store: 'duckdb', pkg: DUCKDB_PACKAGE, sizeHint: '~110MB' };\n\nlet duckdbApiPromise: Promise<typeof import('@duckdb/node-api')> | undefined;\n\n// The one accessor every consumer of @duckdb/node-api (open.ts, sql-functions.ts, vectors.ts)\n// must go through, rather than each doing its own `import('@duckdb/node-api')`: once a fresh\n// install has happened in this process, Node keeps returning that pre-install \"not found\" miss\n// to any NEW bare-specifier lookup of the same path, so a second independent import elsewhere\n// in the process fails even though the package is now genuinely on disk (the same stale-cache\n// problem loadOrInstall works around for its own retry). Caching the resolved module here means\n// only the first caller ever resolves the bare specifier at all.\nexport function duckdbApi(): Promise<typeof import('@duckdb/node-api')> {\n if (!duckdbApiPromise) duckdbApiPromise = loadOrInstall<typeof import('@duckdb/node-api')>(DUCKDB, packageNodeModules());\n return duckdbApiPromise;\n}\n"],"names":["loadOrInstall","packageNodeModules","DUCKDB_PACKAGE","DUCKDB","store","pkg","sizeHint","duckdbApiPromise","duckdbApi"],"mappings":"AAAA,SAASA,aAAa,EAAyBC,kBAAkB,QAAQ,eAAe;AAExF,OAAO,MAAMC,iBAAiB,mBAAmB;AAEjD,MAAMC,SAA2B;IAAEC,OAAO;IAAUC,KAAKH;IAAgBI,UAAU;AAAS;AAE5F,IAAIC;AAEJ,8FAA8F;AAC9F,6FAA6F;AAC7F,+FAA+F;AAC/F,8FAA8F;AAC9F,8FAA8F;AAC9F,gGAAgG;AAChG,iEAAiE;AACjE,OAAO,SAASC;IACd,IAAI,CAACD,kBAAkBA,mBAAmBP,cAAiDG,QAAQF;IACnG,OAAOM;AACT"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/duckdb/native.ts"],"sourcesContent":["import { loadOrInstall, type NativeDescriptor, packageNodeModules } from '../native.ts';\n\nexport const DUCKDB_PACKAGE = '@duckdb/node-api';\n\nconst DUCKDB: NativeDescriptor = { store: 'duckdb', pkg: DUCKDB_PACKAGE, sizeHint: '~110MB' };\n\nlet duckdbApiPromise: Promise<typeof import('@duckdb/node-api')> | undefined;\n\n// The one accessor every consumer of @duckdb/node-api must go through: after a fresh install in this process, Node keeps returning\n// the pre-install \"not found\" miss to any new bare-specifier lookup of the same path, so caching the resolved module here means only the first caller ever resolves it.\nexport function duckdbApi(): Promise<typeof import('@duckdb/node-api')> {\n if (!duckdbApiPromise) duckdbApiPromise = loadOrInstall<typeof import('@duckdb/node-api')>(DUCKDB, packageNodeModules());\n return duckdbApiPromise;\n}\n"],"names":["loadOrInstall","packageNodeModules","DUCKDB_PACKAGE","DUCKDB","store","pkg","sizeHint","duckdbApiPromise","duckdbApi"],"mappings":"AAAA,SAASA,aAAa,EAAyBC,kBAAkB,QAAQ,eAAe;AAExF,OAAO,MAAMC,iBAAiB,mBAAmB;AAEjD,MAAMC,SAA2B;IAAEC,OAAO;IAAUC,KAAKH;IAAgBI,UAAU;AAAS;AAE5F,IAAIC;AAEJ,mIAAmI;AACnI,wKAAwK;AACxK,OAAO,SAASC;IACd,IAAI,CAACD,kBAAkBA,mBAAmBP,cAAiDG,QAAQF;IACnG,OAAOM;AACT"}
@@ -1,36 +1,31 @@
1
1
  import { mkdirSync } from 'node:fs';
2
2
  import { join } from 'node:path';
3
3
  import { featureSignature, STATE_DIR } from '../../config/index.js';
4
+ import { rekeyChunkText } from '../../embed/handoff.js';
4
5
  import { STORE_DIMS } from '../../embed/types.js';
5
6
  import { SenseError } from '../../errors.js';
6
7
  import { activeFeatures, FEATURES } from '../../features/index.js';
8
+ import { clearCache } from '../cache.js';
7
9
  import { getMeta, setMeta } from '../shared.js';
8
- import { clearCache } from '../sqlite/open.js';
9
10
  import { createConnection } from './connection.js';
10
11
  import { DUCKDB_PACKAGE, duckdbApi } from './native.js';
11
12
  import { reconcile } from './reconcile.js';
12
13
  import { registerFunctions } from './sql-functions.js';
13
14
  import { createStore } from './store.js';
14
15
  export const DB_FILENAME = 'cache.duckdb';
15
- // Independent of sqlite's SCHEMA_VERSION -- the two stores' cache shapes evolve separately
16
- // (VARIANT frontmatter columns instead of untyped ones), and the store name already joins the
17
- // feature signature (see featureSignature), so a config's `store` key switching cleanly
18
- // rebuilds rather than reusing the other engine's stale cache shape.
16
+ // Independent of sqlite's SCHEMA_VERSION -- the two stores' cache shapes evolve separately (VARIANT frontmatter columns vs untyped).
17
+ // The store name already joins the feature signature, so switching a config's `store` key rebuilds rather than reusing the other engine's cache.
19
18
  export const SCHEMA_VERSION = '2';
20
- // `content` is a plain table (not FTS-virtual): D1 builds DuckDB's fts index over it lazily,
21
- // only when a lexical query actually runs (see duckdb/lexical.ts), and reads it directly for
22
- // contains() substring/phrase/unspaced-script verification either way. No tokenizer resolution
23
- // -- this store always uses the fts extension's default (porter) stemmer, unlike sqlite's
24
- // configurable `content.tokenize` (not read here; D1 does not extend that knob to DuckDB).
19
+ // `content` is a plain table (not FTS-virtual): the fts index is built over it lazily, only when a lexical query runs (lexical.ts), and read directly for contains() verification either way.
20
+ // No tokenizer resolution: this store always uses the fts extension's default (porter) stemmer; sqlite's configurable `content.tokenize` is not read here.
25
21
  async function ensureSchema(conn, cfg) {
26
22
  await conn.exec(`CREATE TABLE IF NOT EXISTS frontmatter ("path" TEXT PRIMARY KEY, "_mtime" DOUBLE, "_ctime" DOUBLE, "_size" INTEGER, "_parse_error" TEXT)`);
27
23
  await conn.exec(`CREATE TABLE IF NOT EXISTS content ("path" TEXT PRIMARY KEY, title TEXT, summary TEXT, text TEXT)`);
28
24
  await conn.exec(`CREATE TABLE IF NOT EXISTS preset_files ("path" TEXT, preset TEXT, PRIMARY KEY ("path", preset))`);
29
25
  await conn.exec('CREATE INDEX IF NOT EXISTS preset_files_preset ON preset_files(preset)');
30
26
  for (const feature of activeFeatures(cfg)){
31
- // Native FLOAT[STORE_DIMS] instead of the embed feature's engine-neutral BLOB+scale DDL
32
- // (see duckdb/vectors.ts). `scale` is kept, unused, so the feature's shared reconcile-time
33
- // INSERT/DELETE (src/features/embed.ts) still names a column that exists on both stores.
27
+ // Native FLOAT[STORE_DIMS] instead of the embed feature's engine-neutral BLOB+scale DDL (vectors.ts). `scale` is kept, unused,
28
+ // so the feature's shared reconcile-time INSERT/DELETE (features/embed.ts) names a column that exists on both stores.
34
29
  if (feature.name === 'embed') {
35
30
  await conn.exec(`CREATE TABLE IF NOT EXISTS embeddings ("path" TEXT, chunk INTEGER, start_line INTEGER, end_line INTEGER, scale REAL, vector FLOAT[${STORE_DIMS}], PRIMARY KEY ("path", chunk))`);
36
31
  continue;
@@ -46,11 +41,8 @@ async function connect(cfg) {
46
41
  recursive: true
47
42
  });
48
43
  const dbPath = join(stateDir, DB_FILENAME);
49
- // Dynamic, not a top-level import: sqlite trees must never even attempt to resolve this
50
- // optional peer dependency, so nothing in this store's module graph imports it as a value
51
- // until a duckdb tree is actually opened (types-only imports elsewhere are erased and cost
52
- // nothing either way). Installed on first use if missing, and shared with sql-functions.ts
53
- // and vectors.ts (see native.ts's duckdbApi).
44
+ // Dynamic, not a top-level import: sqlite trees must never attempt to resolve this optional peer dependency, so nothing imports
45
+ // it as a value until a duckdb tree opens (types-only imports are erased). Installed on first use if missing, shared with sql-functions.ts and vectors.ts via native.ts's duckdbApi.
54
46
  let DuckDBInstance;
55
47
  try {
56
48
  ({ DuckDBInstance } = await duckdbApi());
@@ -61,10 +53,8 @@ async function connect(cfg) {
61
53
  let duckdb;
62
54
  let instance;
63
55
  try {
64
- // On-disk files default to an older storage format for cross-version compatibility, which
65
- // rejects VARIANT columns ("VARIANT columns are not supported in storage versions prior to
66
- // v1.5.0"); this store's dynamic frontmatter columns need VARIANT (see reconcile.ts), so
67
- // the format floor is pinned explicitly. :memory: databases are unaffected either way.
56
+ // On-disk files default to an older storage format for cross-version compatibility, which rejects VARIANT columns ("VARIANT
57
+ // columns are not supported in storage versions prior to v1.5.0"); this store's dynamic frontmatter columns need VARIANT (reconcile.ts), so the floor is pinned explicitly.
68
58
  instance = await DuckDBInstance.create(dbPath, {
69
59
  storage_compatibility_version: 'v1.5.0'
70
60
  });
@@ -117,8 +107,11 @@ async function connect(cfg) {
117
107
  }
118
108
  export async function openDuckdb(cfg) {
119
109
  const { instance, duckdb, conn, cfg: resolvedCfg, dbPath, parsed, warnings } = await connect(cfg);
110
+ const store = createStore(instance, duckdb, conn, resolvedCfg, resolvedCfg.baseDir);
111
+ // reconcile ran before this object existed, so its chunk text is keyed by the connection.
112
+ rekeyChunkText(conn, store);
120
113
  return {
121
- store: createStore(instance, duckdb, conn, resolvedCfg, resolvedCfg.baseDir),
114
+ store: store,
122
115
  cfg: resolvedCfg,
123
116
  dbPath,
124
117
  parsed,
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/duckdb/open.ts"],"sourcesContent":["import { mkdirSync } from 'node:fs';\nimport { join } from 'node:path';\nimport type { DuckDBConnection, DuckDBInstance } from '@duckdb/node-api';\nimport type { Config, ResolvedConfig } from '../../config/index.ts';\nimport { featureSignature, STATE_DIR } from '../../config/index.ts';\nimport { STORE_DIMS } from '../../embed/types.ts';\nimport { SenseError } from '../../errors.ts';\nimport { activeFeatures, FEATURES } from '../../features/index.ts';\nimport { getMeta, setMeta } from '../shared.ts';\nimport { clearCache } from '../sqlite/open.ts';\nimport type { Connection, Store } from '../types.ts';\nimport { createConnection } from './connection.ts';\nimport { DUCKDB_PACKAGE, duckdbApi } from './native.ts';\nimport { reconcile } from './reconcile.ts';\nimport { registerFunctions } from './sql-functions.ts';\nimport { createStore } from './store.ts';\n\nexport const DB_FILENAME = 'cache.duckdb';\n// Independent of sqlite's SCHEMA_VERSION -- the two stores' cache shapes evolve separately\n// (VARIANT frontmatter columns instead of untyped ones), and the store name already joins the\n// feature signature (see featureSignature), so a config's `store` key switching cleanly\n// rebuilds rather than reusing the other engine's stale cache shape.\nexport const SCHEMA_VERSION = '2';\n\nexport interface OpenResult {\n store: Store;\n cfg: ResolvedConfig;\n dbPath: string;\n parsed: number;\n warnings: string[];\n}\n\n// `content` is a plain table (not FTS-virtual): D1 builds DuckDB's fts index over it lazily,\n// only when a lexical query actually runs (see duckdb/lexical.ts), and reads it directly for\n// contains() substring/phrase/unspaced-script verification either way. No tokenizer resolution\n// -- this store always uses the fts extension's default (porter) stemmer, unlike sqlite's\n// configurable `content.tokenize` (not read here; D1 does not extend that knob to DuckDB).\nasync function ensureSchema(conn: Connection, cfg: Config): Promise<void> {\n await conn.exec(`CREATE TABLE IF NOT EXISTS frontmatter (\"path\" TEXT PRIMARY KEY, \"_mtime\" DOUBLE, \"_ctime\" DOUBLE, \"_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)`);\n await conn.exec(`CREATE TABLE IF NOT EXISTS preset_files (\"path\" TEXT, preset TEXT, PRIMARY KEY (\"path\", preset))`);\n await conn.exec('CREATE INDEX IF NOT EXISTS preset_files_preset ON preset_files(preset)');\n for (const feature of activeFeatures(cfg)) {\n // Native FLOAT[STORE_DIMS] instead of the embed feature's engine-neutral BLOB+scale DDL\n // (see duckdb/vectors.ts). `scale` is kept, unused, so the feature's shared reconcile-time\n // INSERT/DELETE (src/features/embed.ts) still names a column that exists on both stores.\n if (feature.name === 'embed') {\n await conn.exec(`CREATE TABLE IF NOT EXISTS embeddings (\"path\" TEXT, chunk INTEGER, start_line INTEGER, end_line INTEGER, scale REAL, vector FLOAT[${STORE_DIMS}], PRIMARY KEY (\"path\", chunk))`);\n continue;\n }\n await feature.schema(conn);\n }\n if ((await getMeta(conn, 'schema_version')) === null) await setMeta(conn, 'schema_version', SCHEMA_VERSION);\n if ((await getMeta(conn, 'features')) === null) await setMeta(conn, 'features', featureSignature(cfg, FEATURES));\n}\n\nasync function connect(cfg: ResolvedConfig): Promise<{ instance: DuckDBInstance; duckdb: DuckDBConnection; conn: Connection; cfg: ResolvedConfig; dbPath: string; parsed: number; warnings: string[] }> {\n const stateDir = join(cfg.baseDir, STATE_DIR);\n mkdirSync(stateDir, { recursive: true });\n const dbPath = join(stateDir, DB_FILENAME);\n\n // Dynamic, not a top-level import: sqlite trees must never even attempt to resolve this\n // optional peer dependency, so nothing in this store's module graph imports it as a value\n // until a duckdb tree is actually opened (types-only imports elsewhere are erased and cost\n // nothing either way). Installed on first use if missing, and shared with sql-functions.ts\n // and vectors.ts (see native.ts's duckdbApi).\n let DuckDBInstance: typeof import('@duckdb/node-api').DuckDBInstance;\n try {\n ({ DuckDBInstance } = await duckdbApi());\n } catch (err) {\n if (err instanceof SenseError) throw err;\n throw new SenseError('STORE_DEPENDENCY_MISSING', `store \"duckdb\" needs the ${DUCKDB_PACKAGE} package (${(err as Error).message})`);\n }\n\n let duckdb: DuckDBConnection;\n let instance: DuckDBInstance | undefined;\n try {\n // On-disk files default to an older storage format for cross-version compatibility, which\n // rejects VARIANT columns (\"VARIANT columns are not supported in storage versions prior to\n // v1.5.0\"); this store's dynamic frontmatter columns need VARIANT (see reconcile.ts), so\n // the format floor is pinned explicitly. :memory: databases are unaffected either way.\n instance = await DuckDBInstance.create(dbPath, { storage_compatibility_version: 'v1.5.0' });\n duckdb = await instance.connect();\n } catch (err) {\n // create() may have succeeded before connect() failed: close it, or its WAL stays open.\n instance?.closeSync();\n throw new SenseError('STORE_DEPENDENCY_MISSING', `store \"duckdb\" failed to open ${dbPath}: ${(err as Error).message}`);\n }\n // A throw below (schema, reconcile's COLUMN_LIMIT) would leak the open instance, whose WAL\n // then locks the .duckdb file undeletable on Windows. closed marks the rebuild branch.\n let closed = false;\n try {\n await registerFunctions(duckdb);\n const conn = createConnection(duckdb);\n\n await conn.exec('CREATE TABLE IF NOT EXISTS meta (key TEXT PRIMARY KEY, value TEXT)');\n\n const version = await getMeta(conn, 'schema_version');\n const features = await getMeta(conn, 'features');\n const wantFeatures = featureSignature(cfg, FEATURES);\n if ((version !== null && version !== SCHEMA_VERSION) || (features !== null && features !== wantFeatures)) {\n // Reconcile only reparses changed files, so a stale cache can't be patched incrementally --\n // rebuild instead, same as sqlite's open() (see its comment for why this is announced).\n const reason = version !== null && version !== SCHEMA_VERSION ? 'cache format changed (new sensemaking version)' : 'config change (features, tokenizer, or presets)';\n console.error(`sense: ${reason}; rebuilding the index`);\n // Close before clearCache deletes the files underneath the still-open instance.\n closed = true;\n duckdb.disconnectSync();\n instance.closeSync();\n clearCache(cfg);\n return connect(cfg);\n }\n\n await ensureSchema(conn, cfg);\n\n const { parsed, warnings } = await reconcile(conn, cfg, cfg.baseDir);\n return { instance, duckdb, conn, cfg, dbPath, parsed, warnings };\n } catch (err) {\n if (!closed) {\n duckdb.disconnectSync();\n instance?.closeSync();\n }\n throw err;\n }\n}\n\nexport async function openDuckdb(cfg: ResolvedConfig): Promise<OpenResult> {\n const { instance, duckdb, conn, cfg: resolvedCfg, dbPath, parsed, warnings } = await connect(cfg);\n return { store: createStore(instance, duckdb, conn, resolvedCfg, resolvedCfg.baseDir), cfg: resolvedCfg, dbPath, parsed, warnings };\n}\n"],"names":["mkdirSync","join","featureSignature","STATE_DIR","STORE_DIMS","SenseError","activeFeatures","FEATURES","getMeta","setMeta","clearCache","createConnection","DUCKDB_PACKAGE","duckdbApi","reconcile","registerFunctions","createStore","DB_FILENAME","SCHEMA_VERSION","ensureSchema","conn","cfg","exec","feature","name","schema","connect","stateDir","baseDir","recursive","dbPath","DuckDBInstance","err","message","duckdb","instance","create","storage_compatibility_version","closeSync","closed","version","features","wantFeatures","reason","console","error","disconnectSync","parsed","warnings","openDuckdb","resolvedCfg","store"],"mappings":"AAAA,SAASA,SAAS,QAAQ,UAAU;AACpC,SAASC,IAAI,QAAQ,YAAY;AAGjC,SAASC,gBAAgB,EAAEC,SAAS,QAAQ,wBAAwB;AACpE,SAASC,UAAU,QAAQ,uBAAuB;AAClD,SAASC,UAAU,QAAQ,kBAAkB;AAC7C,SAASC,cAAc,EAAEC,QAAQ,QAAQ,0BAA0B;AACnE,SAASC,OAAO,EAAEC,OAAO,QAAQ,eAAe;AAChD,SAASC,UAAU,QAAQ,oBAAoB;AAE/C,SAASC,gBAAgB,QAAQ,kBAAkB;AACnD,SAASC,cAAc,EAAEC,SAAS,QAAQ,cAAc;AACxD,SAASC,SAAS,QAAQ,iBAAiB;AAC3C,SAASC,iBAAiB,QAAQ,qBAAqB;AACvD,SAASC,WAAW,QAAQ,aAAa;AAEzC,OAAO,MAAMC,cAAc,eAAe;AAC1C,2FAA2F;AAC3F,8FAA8F;AAC9F,wFAAwF;AACxF,qEAAqE;AACrE,OAAO,MAAMC,iBAAiB,IAAI;AAUlC,6FAA6F;AAC7F,6FAA6F;AAC7F,+FAA+F;AAC/F,0FAA0F;AAC1F,2FAA2F;AAC3F,eAAeC,aAAaC,IAAgB,EAAEC,GAAW;IACvD,MAAMD,KAAKE,IAAI,CAAC,CAAC,wIAAwI,CAAC;IAC1J,MAAMF,KAAKE,IAAI,CAAC,CAAC,iGAAiG,CAAC;IACnH,MAAMF,KAAKE,IAAI,CAAC,CAAC,gGAAgG,CAAC;IAClH,MAAMF,KAAKE,IAAI,CAAC;IAChB,KAAK,MAAMC,WAAWjB,eAAee,KAAM;QACzC,wFAAwF;QACxF,2FAA2F;QAC3F,yFAAyF;QACzF,IAAIE,QAAQC,IAAI,KAAK,SAAS;YAC5B,MAAMJ,KAAKE,IAAI,CAAC,CAAC,kIAAkI,EAAElB,WAAW,+BAA+B,CAAC;YAChM;QACF;QACA,MAAMmB,QAAQE,MAAM,CAACL;IACvB;IACA,IAAI,AAAC,MAAMZ,QAAQY,MAAM,sBAAuB,MAAM,MAAMX,QAAQW,MAAM,kBAAkBF;IAC5F,IAAI,AAAC,MAAMV,QAAQY,MAAM,gBAAiB,MAAM,MAAMX,QAAQW,MAAM,YAAYlB,iBAAiBmB,KAAKd;AACxG;AAEA,eAAemB,QAAQL,GAAmB;IACxC,MAAMM,WAAW1B,KAAKoB,IAAIO,OAAO,EAAEzB;IACnCH,UAAU2B,UAAU;QAAEE,WAAW;IAAK;IACtC,MAAMC,SAAS7B,KAAK0B,UAAUV;IAE9B,wFAAwF;IACxF,0FAA0F;IAC1F,2FAA2F;IAC3F,2FAA2F;IAC3F,8CAA8C;IAC9C,IAAIc;IACJ,IAAI;QACD,CAAA,EAAEA,cAAc,EAAE,GAAG,MAAMlB,WAAU;IACxC,EAAE,OAAOmB,KAAK;QACZ,IAAIA,eAAe3B,YAAY,MAAM2B;QACrC,MAAM,IAAI3B,WAAW,4BAA4B,CAAC,yBAAyB,EAAEO,eAAe,UAAU,EAAE,AAACoB,IAAcC,OAAO,CAAC,CAAC,CAAC;IACnI;IAEA,IAAIC;IACJ,IAAIC;IACJ,IAAI;QACF,0FAA0F;QAC1F,2FAA2F;QAC3F,yFAAyF;QACzF,uFAAuF;QACvFA,WAAW,MAAMJ,eAAeK,MAAM,CAACN,QAAQ;YAAEO,+BAA+B;QAAS;QACzFH,SAAS,MAAMC,SAAST,OAAO;IACjC,EAAE,OAAOM,KAAK;QACZ,wFAAwF;QACxFG,qBAAAA,+BAAAA,SAAUG,SAAS;QACnB,MAAM,IAAIjC,WAAW,4BAA4B,CAAC,8BAA8B,EAAEyB,OAAO,EAAE,EAAE,AAACE,IAAcC,OAAO,EAAE;IACvH;IACA,2FAA2F;IAC3F,uFAAuF;IACvF,IAAIM,SAAS;IACb,IAAI;QACF,MAAMxB,kBAAkBmB;QACxB,MAAMd,OAAOT,iBAAiBuB;QAE9B,MAAMd,KAAKE,IAAI,CAAC;QAEhB,MAAMkB,UAAU,MAAMhC,QAAQY,MAAM;QACpC,MAAMqB,WAAW,MAAMjC,QAAQY,MAAM;QACrC,MAAMsB,eAAexC,iBAAiBmB,KAAKd;QAC3C,IAAI,AAACiC,YAAY,QAAQA,YAAYtB,kBAAoBuB,aAAa,QAAQA,aAAaC,cAAe;YACxG,4FAA4F;YAC5F,wFAAwF;YACxF,MAAMC,SAASH,YAAY,QAAQA,YAAYtB,iBAAiB,mDAAmD;YACnH0B,QAAQC,KAAK,CAAC,CAAC,OAAO,EAAEF,OAAO,sBAAsB,CAAC;YACtD,gFAAgF;YAChFJ,SAAS;YACTL,OAAOY,cAAc;YACrBX,SAASG,SAAS;YAClB5B,WAAWW;YACX,OAAOK,QAAQL;QACjB;QAEA,MAAMF,aAAaC,MAAMC;QAEzB,MAAM,EAAE0B,MAAM,EAAEC,QAAQ,EAAE,GAAG,MAAMlC,UAAUM,MAAMC,KAAKA,IAAIO,OAAO;QACnE,OAAO;YAAEO;YAAUD;YAAQd;YAAMC;YAAKS;YAAQiB;YAAQC;QAAS;IACjE,EAAE,OAAOhB,KAAK;QACZ,IAAI,CAACO,QAAQ;YACXL,OAAOY,cAAc;YACrBX,qBAAAA,+BAAAA,SAAUG,SAAS;QACrB;QACA,MAAMN;IACR;AACF;AAEA,OAAO,eAAeiB,WAAW5B,GAAmB;IAClD,MAAM,EAAEc,QAAQ,EAAED,MAAM,EAAEd,IAAI,EAAEC,KAAK6B,WAAW,EAAEpB,MAAM,EAAEiB,MAAM,EAAEC,QAAQ,EAAE,GAAG,MAAMtB,QAAQL;IAC7F,OAAO;QAAE8B,OAAOnC,YAAYmB,UAAUD,QAAQd,MAAM8B,aAAaA,YAAYtB,OAAO;QAAGP,KAAK6B;QAAapB;QAAQiB;QAAQC;IAAS;AACpI"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/duckdb/open.ts"],"sourcesContent":["import { mkdirSync } from 'node:fs';\nimport { join } from 'node:path';\nimport type { DuckDBConnection, DuckDBInstance } from '@duckdb/node-api';\nimport type { Config, ResolvedConfig } from '../../config/index.ts';\nimport { featureSignature, STATE_DIR } from '../../config/index.ts';\nimport { rekeyChunkText } from '../../embed/handoff.ts';\nimport { STORE_DIMS } from '../../embed/types.ts';\nimport { SenseError } from '../../errors.ts';\nimport { activeFeatures, FEATURES } from '../../features/index.ts';\nimport { clearCache } from '../cache.ts';\nimport { getMeta, setMeta } from '../shared.ts';\nimport type { Connection, Store } from '../types.ts';\nimport { createConnection } from './connection.ts';\nimport { DUCKDB_PACKAGE, duckdbApi } from './native.ts';\nimport { reconcile } from './reconcile.ts';\nimport { registerFunctions } from './sql-functions.ts';\nimport { createStore } from './store.ts';\n\nexport const DB_FILENAME = 'cache.duckdb';\n// Independent of sqlite's SCHEMA_VERSION -- the two stores' cache shapes evolve separately (VARIANT frontmatter columns vs untyped).\n// The store name already joins the feature signature, so switching a config's `store` key rebuilds rather than reusing the other engine's cache.\nexport const SCHEMA_VERSION = '2';\n\nexport interface OpenResult {\n store: Store;\n cfg: ResolvedConfig;\n dbPath: string;\n parsed: number;\n warnings: string[];\n}\n\n// `content` is a plain table (not FTS-virtual): the fts index is built over it lazily, only when a lexical query runs (lexical.ts), and read directly for contains() verification either way.\n// No tokenizer resolution: this store always uses the fts extension's default (porter) stemmer; sqlite's configurable `content.tokenize` is not read here.\nasync function ensureSchema(conn: Connection, cfg: Config): Promise<void> {\n await conn.exec(`CREATE TABLE IF NOT EXISTS frontmatter (\"path\" TEXT PRIMARY KEY, \"_mtime\" DOUBLE, \"_ctime\" DOUBLE, \"_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)`);\n await conn.exec(`CREATE TABLE IF NOT EXISTS preset_files (\"path\" TEXT, preset TEXT, PRIMARY KEY (\"path\", preset))`);\n await conn.exec('CREATE INDEX IF NOT EXISTS preset_files_preset ON preset_files(preset)');\n for (const feature of activeFeatures(cfg)) {\n // Native FLOAT[STORE_DIMS] instead of the embed feature's engine-neutral BLOB+scale DDL (vectors.ts). `scale` is kept, unused,\n // so the feature's shared reconcile-time INSERT/DELETE (features/embed.ts) names a column that exists on both stores.\n if (feature.name === 'embed') {\n await conn.exec(`CREATE TABLE IF NOT EXISTS embeddings (\"path\" TEXT, chunk INTEGER, start_line INTEGER, end_line INTEGER, scale REAL, vector FLOAT[${STORE_DIMS}], PRIMARY KEY (\"path\", chunk))`);\n continue;\n }\n await feature.schema(conn);\n }\n if ((await getMeta(conn, 'schema_version')) === null) await setMeta(conn, 'schema_version', SCHEMA_VERSION);\n if ((await getMeta(conn, 'features')) === null) await setMeta(conn, 'features', featureSignature(cfg, FEATURES));\n}\n\nasync function connect(cfg: ResolvedConfig): Promise<{ instance: DuckDBInstance; duckdb: DuckDBConnection; conn: Connection; cfg: ResolvedConfig; dbPath: string; parsed: number; warnings: string[] }> {\n const stateDir = join(cfg.baseDir, STATE_DIR);\n mkdirSync(stateDir, { recursive: true });\n const dbPath = join(stateDir, DB_FILENAME);\n\n // Dynamic, not a top-level import: sqlite trees must never attempt to resolve this optional peer dependency, so nothing imports\n // it as a value until a duckdb tree opens (types-only imports are erased). Installed on first use if missing, shared with sql-functions.ts and vectors.ts via native.ts's duckdbApi.\n let DuckDBInstance: typeof import('@duckdb/node-api').DuckDBInstance;\n try {\n ({ DuckDBInstance } = await duckdbApi());\n } catch (err) {\n if (err instanceof SenseError) throw err;\n throw new SenseError('STORE_DEPENDENCY_MISSING', `store \"duckdb\" needs the ${DUCKDB_PACKAGE} package (${(err as Error).message})`);\n }\n\n let duckdb: DuckDBConnection;\n let instance: DuckDBInstance | undefined;\n try {\n // On-disk files default to an older storage format for cross-version compatibility, which rejects VARIANT columns (\"VARIANT\n // columns are not supported in storage versions prior to v1.5.0\"); this store's dynamic frontmatter columns need VARIANT (reconcile.ts), so the floor is pinned explicitly.\n instance = await DuckDBInstance.create(dbPath, { storage_compatibility_version: 'v1.5.0' });\n duckdb = await instance.connect();\n } catch (err) {\n // create() may have succeeded before connect() failed: close it, or its WAL stays open.\n instance?.closeSync();\n throw new SenseError('STORE_DEPENDENCY_MISSING', `store \"duckdb\" failed to open ${dbPath}: ${(err as Error).message}`);\n }\n // A throw below (schema, reconcile's COLUMN_LIMIT) would leak the open instance, whose WAL\n // then locks the .duckdb file undeletable on Windows. closed marks the rebuild branch.\n let closed = false;\n try {\n await registerFunctions(duckdb);\n const conn = createConnection(duckdb);\n\n await conn.exec('CREATE TABLE IF NOT EXISTS meta (key TEXT PRIMARY KEY, value TEXT)');\n\n const version = await getMeta(conn, 'schema_version');\n const features = await getMeta(conn, 'features');\n const wantFeatures = featureSignature(cfg, FEATURES);\n if ((version !== null && version !== SCHEMA_VERSION) || (features !== null && features !== wantFeatures)) {\n // Reconcile only reparses changed files, so a stale cache can't be patched incrementally --\n // rebuild instead, same as sqlite's open() (see its comment for why this is announced).\n const reason = version !== null && version !== SCHEMA_VERSION ? 'cache format changed (new sensemaking version)' : 'config change (features, tokenizer, or presets)';\n console.error(`sense: ${reason}; rebuilding the index`);\n // Close before clearCache deletes the files underneath the still-open instance.\n closed = true;\n duckdb.disconnectSync();\n instance.closeSync();\n clearCache(cfg);\n return connect(cfg);\n }\n\n await ensureSchema(conn, cfg);\n\n const { parsed, warnings } = await reconcile(conn, cfg, cfg.baseDir);\n return { instance, duckdb, conn, cfg, dbPath, parsed, warnings };\n } catch (err) {\n if (!closed) {\n duckdb.disconnectSync();\n instance?.closeSync();\n }\n throw err;\n }\n}\n\nexport async function openDuckdb(cfg: ResolvedConfig): Promise<OpenResult> {\n const { instance, duckdb, conn, cfg: resolvedCfg, dbPath, parsed, warnings } = await connect(cfg);\n const store = createStore(instance, duckdb, conn, resolvedCfg, resolvedCfg.baseDir);\n // reconcile ran before this object existed, so its chunk text is keyed by the connection.\n rekeyChunkText(conn, store);\n return { store: store, cfg: resolvedCfg, dbPath, parsed, warnings };\n}\n"],"names":["mkdirSync","join","featureSignature","STATE_DIR","rekeyChunkText","STORE_DIMS","SenseError","activeFeatures","FEATURES","clearCache","getMeta","setMeta","createConnection","DUCKDB_PACKAGE","duckdbApi","reconcile","registerFunctions","createStore","DB_FILENAME","SCHEMA_VERSION","ensureSchema","conn","cfg","exec","feature","name","schema","connect","stateDir","baseDir","recursive","dbPath","DuckDBInstance","err","message","duckdb","instance","create","storage_compatibility_version","closeSync","closed","version","features","wantFeatures","reason","console","error","disconnectSync","parsed","warnings","openDuckdb","resolvedCfg","store"],"mappings":"AAAA,SAASA,SAAS,QAAQ,UAAU;AACpC,SAASC,IAAI,QAAQ,YAAY;AAGjC,SAASC,gBAAgB,EAAEC,SAAS,QAAQ,wBAAwB;AACpE,SAASC,cAAc,QAAQ,yBAAyB;AACxD,SAASC,UAAU,QAAQ,uBAAuB;AAClD,SAASC,UAAU,QAAQ,kBAAkB;AAC7C,SAASC,cAAc,EAAEC,QAAQ,QAAQ,0BAA0B;AACnE,SAASC,UAAU,QAAQ,cAAc;AACzC,SAASC,OAAO,EAAEC,OAAO,QAAQ,eAAe;AAEhD,SAASC,gBAAgB,QAAQ,kBAAkB;AACnD,SAASC,cAAc,EAAEC,SAAS,QAAQ,cAAc;AACxD,SAASC,SAAS,QAAQ,iBAAiB;AAC3C,SAASC,iBAAiB,QAAQ,qBAAqB;AACvD,SAASC,WAAW,QAAQ,aAAa;AAEzC,OAAO,MAAMC,cAAc,eAAe;AAC1C,qIAAqI;AACrI,iJAAiJ;AACjJ,OAAO,MAAMC,iBAAiB,IAAI;AAUlC,8LAA8L;AAC9L,2JAA2J;AAC3J,eAAeC,aAAaC,IAAgB,EAAEC,GAAW;IACvD,MAAMD,KAAKE,IAAI,CAAC,CAAC,wIAAwI,CAAC;IAC1J,MAAMF,KAAKE,IAAI,CAAC,CAAC,iGAAiG,CAAC;IACnH,MAAMF,KAAKE,IAAI,CAAC,CAAC,gGAAgG,CAAC;IAClH,MAAMF,KAAKE,IAAI,CAAC;IAChB,KAAK,MAAMC,WAAWjB,eAAee,KAAM;QACzC,+HAA+H;QAC/H,sHAAsH;QACtH,IAAIE,QAAQC,IAAI,KAAK,SAAS;YAC5B,MAAMJ,KAAKE,IAAI,CAAC,CAAC,kIAAkI,EAAElB,WAAW,+BAA+B,CAAC;YAChM;QACF;QACA,MAAMmB,QAAQE,MAAM,CAACL;IACvB;IACA,IAAI,AAAC,MAAMX,QAAQW,MAAM,sBAAuB,MAAM,MAAMV,QAAQU,MAAM,kBAAkBF;IAC5F,IAAI,AAAC,MAAMT,QAAQW,MAAM,gBAAiB,MAAM,MAAMV,QAAQU,MAAM,YAAYnB,iBAAiBoB,KAAKd;AACxG;AAEA,eAAemB,QAAQL,GAAmB;IACxC,MAAMM,WAAW3B,KAAKqB,IAAIO,OAAO,EAAE1B;IACnCH,UAAU4B,UAAU;QAAEE,WAAW;IAAK;IACtC,MAAMC,SAAS9B,KAAK2B,UAAUV;IAE9B,gIAAgI;IAChI,qLAAqL;IACrL,IAAIc;IACJ,IAAI;QACD,CAAA,EAAEA,cAAc,EAAE,GAAG,MAAMlB,WAAU;IACxC,EAAE,OAAOmB,KAAK;QACZ,IAAIA,eAAe3B,YAAY,MAAM2B;QACrC,MAAM,IAAI3B,WAAW,4BAA4B,CAAC,yBAAyB,EAAEO,eAAe,UAAU,EAAE,AAACoB,IAAcC,OAAO,CAAC,CAAC,CAAC;IACnI;IAEA,IAAIC;IACJ,IAAIC;IACJ,IAAI;QACF,4HAA4H;QAC5H,4KAA4K;QAC5KA,WAAW,MAAMJ,eAAeK,MAAM,CAACN,QAAQ;YAAEO,+BAA+B;QAAS;QACzFH,SAAS,MAAMC,SAAST,OAAO;IACjC,EAAE,OAAOM,KAAK;QACZ,wFAAwF;QACxFG,qBAAAA,+BAAAA,SAAUG,SAAS;QACnB,MAAM,IAAIjC,WAAW,4BAA4B,CAAC,8BAA8B,EAAEyB,OAAO,EAAE,EAAE,AAACE,IAAcC,OAAO,EAAE;IACvH;IACA,2FAA2F;IAC3F,uFAAuF;IACvF,IAAIM,SAAS;IACb,IAAI;QACF,MAAMxB,kBAAkBmB;QACxB,MAAMd,OAAOT,iBAAiBuB;QAE9B,MAAMd,KAAKE,IAAI,CAAC;QAEhB,MAAMkB,UAAU,MAAM/B,QAAQW,MAAM;QACpC,MAAMqB,WAAW,MAAMhC,QAAQW,MAAM;QACrC,MAAMsB,eAAezC,iBAAiBoB,KAAKd;QAC3C,IAAI,AAACiC,YAAY,QAAQA,YAAYtB,kBAAoBuB,aAAa,QAAQA,aAAaC,cAAe;YACxG,4FAA4F;YAC5F,wFAAwF;YACxF,MAAMC,SAASH,YAAY,QAAQA,YAAYtB,iBAAiB,mDAAmD;YACnH0B,QAAQC,KAAK,CAAC,CAAC,OAAO,EAAEF,OAAO,sBAAsB,CAAC;YACtD,gFAAgF;YAChFJ,SAAS;YACTL,OAAOY,cAAc;YACrBX,SAASG,SAAS;YAClB9B,WAAWa;YACX,OAAOK,QAAQL;QACjB;QAEA,MAAMF,aAAaC,MAAMC;QAEzB,MAAM,EAAE0B,MAAM,EAAEC,QAAQ,EAAE,GAAG,MAAMlC,UAAUM,MAAMC,KAAKA,IAAIO,OAAO;QACnE,OAAO;YAAEO;YAAUD;YAAQd;YAAMC;YAAKS;YAAQiB;YAAQC;QAAS;IACjE,EAAE,OAAOhB,KAAK;QACZ,IAAI,CAACO,QAAQ;YACXL,OAAOY,cAAc;YACrBX,qBAAAA,+BAAAA,SAAUG,SAAS;QACrB;QACA,MAAMN;IACR;AACF;AAEA,OAAO,eAAeiB,WAAW5B,GAAmB;IAClD,MAAM,EAAEc,QAAQ,EAAED,MAAM,EAAEd,IAAI,EAAEC,KAAK6B,WAAW,EAAEpB,MAAM,EAAEiB,MAAM,EAAEC,QAAQ,EAAE,GAAG,MAAMtB,QAAQL;IAC7F,MAAM8B,QAAQnC,YAAYmB,UAAUD,QAAQd,MAAM8B,aAAaA,YAAYtB,OAAO;IAClF,0FAA0F;IAC1FzB,eAAeiB,MAAM+B;IACrB,OAAO;QAAEA,OAAOA;QAAO9B,KAAK6B;QAAapB;QAAQiB;QAAQC;IAAS;AACpE"}
@@ -5,14 +5,8 @@ import { listFiles, RESERVED_COLUMNS } from '../../scan/index.js';
5
5
  import { reparseFiles } from '../../scan/reparse.js';
6
6
  import { getColumns, quoteIdent } from '../shared.js';
7
7
  import { withTransaction } from '../transaction.js';
8
- // Mirrors src/store/sqlite/reconcile.ts's frontmatter-upsert shape (dynamic ALTER TABLE per
9
- // discovered key, ON CONFLICT upsert keeping the row's identity stable). `content` is a plain
10
- // table, not FTS-virtual (D1: DuckDB's fts index is built lazily, on first lexical query, from
11
- // this table -- see duckdb/lexical.ts), so its rows are maintained here unconditionally. The one
12
- // forced divergence for frontmatter itself is the ALTER TABLE type: DuckDB requires a declared
13
- // column type where sqlite accepts none, so every dynamic frontmatter column is VARIANT (the
14
- // only DuckDB type that can hold the different JS types mapValue() produces across files for
15
- // the same key).
8
+ // Mirrors sqlite/reconcile.ts's frontmatter-upsert shape (dynamic ALTER TABLE per discovered key, ON CONFLICT upsert). `content` is
9
+ // a plain table, not FTS-virtual, so rows are maintained here unconditionally; DuckDB's ALTER TABLE needs a declared type, so every dynamic frontmatter column is VARIANT (holds mapValue()'s mixed JS types for one key across files).
16
10
  const CORE_FRONTMATTER_COLUMNS = new Set([
17
11
  'path',
18
12
  '_mtime',
@@ -31,9 +25,8 @@ function contentRow(doc) {
31
25
  doc.search.text
32
26
  ];
33
27
  }
34
- // No compile-time column cap in DuckDB (unlike SQLite's SQLITE_MAX_COLUMN); kept as a sanity
35
- // fence anyway so a runaway frontmatter generator fails with a clear message instead of an
36
- // unbounded ALTER TABLE loop.
28
+ // No compile-time column cap in DuckDB (unlike SQLite's SQLITE_MAX_COLUMN); kept as a sanity fence anyway
29
+ // so a runaway frontmatter generator fails with a clear message instead of an unbounded ALTER TABLE loop.
37
30
  const MAX_FRONTMATTER_COLUMNS = 10000;
38
31
  export async function reconcile(conn, cfg, baseDir) {
39
32
  const files = listFiles(cfg, baseDir);
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/duckdb/reconcile.ts"],"sourcesContent":["import type { Config } from '../../config/index.ts';\nimport { SenseError } from '../../errors.ts';\nimport { activeFeatures } from '../../features/index.ts';\nimport type { ExtractedDoc, ReconcileDelta } from '../../features/types.ts';\nimport { progress } from '../../output/progress.ts';\nimport type { ParsedDoc } from '../../scan/index.ts';\nimport { listFiles, RESERVED_COLUMNS } from '../../scan/index.ts';\nimport { reparseFiles } from '../../scan/reparse.ts';\nimport { getColumns, quoteIdent } from '../shared.ts';\nimport { withTransaction } from '../transaction.ts';\nimport type { Connection } from '../types.ts';\n\n// Mirrors src/store/sqlite/reconcile.ts's frontmatter-upsert shape (dynamic ALTER TABLE per\n// discovered key, ON CONFLICT upsert keeping the row's identity stable). `content` is a plain\n// table, not FTS-virtual (D1: DuckDB's fts index is built lazily, on first lexical query, from\n// this table -- see duckdb/lexical.ts), so its rows are maintained here unconditionally. The one\n// forced divergence for frontmatter itself is the ALTER TABLE type: DuckDB requires a declared\n// column type where sqlite accepts none, so every dynamic frontmatter column is VARIANT (the\n// only DuckDB type that can hold the different JS types mapValue() produces across files for\n// the same key).\nconst CORE_FRONTMATTER_COLUMNS = new Set(['path', '_mtime', '_ctime', '_size', '_parse_error']);\n\n// No rowid coupling needed (unlike sqlite's content, which links to frontmatter's rowid):\n// `path` is content's own primary key, so this is a plain per-doc row.\nconst INSERT_CONTENT_SQL = `INSERT INTO content (\"path\", title, summary, text) VALUES (?, ?, ?, ?)`;\n\nfunction contentRow(doc: ParsedDoc): unknown[] {\n return [doc.relPath, doc.search.title, doc.search.summary, doc.search.text];\n}\n\n// No compile-time column cap in DuckDB (unlike SQLite's SQLITE_MAX_COLUMN); kept as a sanity\n// fence anyway so a runaway frontmatter generator fails with a clear message instead of an\n// unbounded ALTER TABLE loop.\nconst MAX_FRONTMATTER_COLUMNS = 10_000;\n\nexport async function reconcile(conn: Connection, cfg: Config, baseDir: string): Promise<{ parsed: number; warnings: string[] }> {\n const files = listFiles(cfg, baseDir);\n const currentSet = new Set(files.map((f) => f.relPath));\n\n const existingStmt = await conn.prepare('SELECT \"path\", \"_mtime\", \"_size\" FROM frontmatter');\n const existingRows = (await existingStmt.all()) as Array<{ path: string; _mtime: number; _size: number }>;\n const existing = new Map(existingRows.map((r) => [r.path, r]));\n const vanished = existingRows.filter((r) => !currentSet.has(r.path)).map((r) => r.path);\n\n const toReparse = files.filter((f) => {\n const row = existing.get(f.relPath);\n return !row || row._mtime !== f.mtimeMs || row._size !== f.size;\n });\n\n if (vanished.length === 0 && toReparse.length === 0) return { parsed: 0, warnings: [] };\n\n const features = activeFeatures(cfg);\n const seenColumns = await getColumns(conn);\n\n const report = progress('reparsing files', toReparse.length);\n const { docs: parsedDocs, warnings, newColumns } = await reparseFiles(toReparse, features, cfg, seenColumns, report.tick);\n report.finish();\n for (const col of newColumns) seenColumns.add(col);\n\n const allColumns = [...seenColumns];\n if (allColumns.length > MAX_FRONTMATTER_COLUMNS) {\n throw new SenseError('COLUMN_LIMIT', `frontmatter would need ${allColumns.length} columns, crossing this store's sanity limit (${MAX_FRONTMATTER_COLUMNS}). Narrow the presets' include globs so fewer/other files are indexed, or fix whatever is generating unbounded frontmatter keys.`);\n }\n const writableColumns = allColumns.filter((c) => CORE_FRONTMATTER_COLUMNS.has(c) || !RESERVED_COLUMNS.has(c));\n const insertSql = `INSERT INTO frontmatter (${writableColumns.map(quoteIdent).join(', ')}) VALUES (${writableColumns.map(() => '?').join(', ')}) ON CONFLICT(\"path\") DO UPDATE SET ${writableColumns\n .filter((c) => c !== 'path')\n .map((c) => `${quoteIdent(c)} = excluded.${quoteIdent(c)}`)\n .join(', ')}`;\n\n const added = toReparse.filter((f) => !existing.has(f.relPath)).map((f) => f.relPath);\n const delta: ReconcileDelta = { files, reparsed: parsedDocs.map((d) => d.relPath), added, vanished };\n const addedSet = new Set(added);\n const reparsedExisting = parsedDocs.map((d) => d.relPath).filter((p) => !addedSet.has(p));\n\n await withTransaction(conn, async () => {\n // DuckDB's ALTER TABLE ADD COLUMN requires a type; VARIANT is the only one that can hold\n // the mixed bigint/number/string/null shapes mapValue() produces for one key across files.\n for (const col of newColumns) await conn.exec(`ALTER TABLE frontmatter ADD COLUMN ${quoteIdent(col)} VARIANT`);\n\n if (parsedDocs.length > 0) {\n const rows = parsedDocs.map((doc) =>\n writableColumns.map((col) => {\n if (col === 'path') return doc.relPath;\n if (col === '_mtime') return doc.mtimeMs;\n if (col === '_ctime') return doc.ctimeMs;\n if (col === '_size') return doc.size;\n if (col === '_parse_error') return doc.parseError;\n return doc.data[col] ?? null;\n })\n );\n await conn.runBatch(insertSql, rows);\n }\n\n const contentTouched = [...vanished, ...reparsedExisting];\n if (contentTouched.length > 0)\n await conn.runBatch(\n 'DELETE FROM content WHERE \"path\" = ?',\n contentTouched.map((p) => [p])\n );\n if (parsedDocs.length > 0) await conn.runBatch(INSERT_CONTENT_SQL, parsedDocs.map(contentRow));\n\n if (vanished.length > 0)\n await conn.runBatch(\n 'DELETE FROM frontmatter WHERE \"path\" = ?',\n vanished.map((p) => [p])\n );\n\n // A preset edit forces a full rebuild, so an unchanged doc's coverage is already correct;\n // new docs have nothing to clear, which keeps cold builds linear.\n const presetTouched = [...vanished, ...reparsedExisting];\n if (presetTouched.length > 0)\n await conn.runBatch(\n 'DELETE FROM preset_files WHERE \"path\" = ?',\n presetTouched.map((p) => [p])\n );\n const presetRows: unknown[][] = [];\n for (const doc of parsedDocs) for (const presetName of doc.presets) presetRows.push([doc.relPath, presetName]);\n if (presetRows.length > 0) await conn.runBatch('INSERT INTO preset_files (\"path\", preset) VALUES (?, ?)', presetRows);\n\n const removedPaths = [...vanished, ...reparsedExisting];\n if (removedPaths.length > 0) for (const feature of features) await feature.remove?.(conn, removedPaths, delta);\n for (const feature of features) {\n const docsForFeature: ExtractedDoc[] = parsedDocs.map((doc) => ({ path: doc.relPath, extracted: doc.extracted[feature.name] }));\n await feature.store?.(conn, docsForFeature, delta);\n }\n for (const feature of features) await feature.afterReconcile?.(conn, delta);\n });\n\n return { parsed: parsedDocs.length, warnings };\n}\n"],"names":["SenseError","activeFeatures","progress","listFiles","RESERVED_COLUMNS","reparseFiles","getColumns","quoteIdent","withTransaction","CORE_FRONTMATTER_COLUMNS","Set","INSERT_CONTENT_SQL","contentRow","doc","relPath","search","title","summary","text","MAX_FRONTMATTER_COLUMNS","reconcile","conn","cfg","baseDir","files","currentSet","map","f","existingStmt","prepare","existingRows","all","existing","Map","r","path","vanished","filter","has","toReparse","row","get","_mtime","mtimeMs","_size","size","length","parsed","warnings","features","seenColumns","report","docs","parsedDocs","newColumns","tick","finish","col","add","allColumns","writableColumns","c","insertSql","join","added","delta","reparsed","d","addedSet","reparsedExisting","p","feature","exec","rows","ctimeMs","parseError","data","runBatch","contentTouched","presetTouched","presetRows","presetName","presets","push","removedPaths","remove","docsForFeature","extracted","name","store","afterReconcile"],"mappings":"AACA,SAASA,UAAU,QAAQ,kBAAkB;AAC7C,SAASC,cAAc,QAAQ,0BAA0B;AAEzD,SAASC,QAAQ,QAAQ,2BAA2B;AAEpD,SAASC,SAAS,EAAEC,gBAAgB,QAAQ,sBAAsB;AAClE,SAASC,YAAY,QAAQ,wBAAwB;AACrD,SAASC,UAAU,EAAEC,UAAU,QAAQ,eAAe;AACtD,SAASC,eAAe,QAAQ,oBAAoB;AAGpD,4FAA4F;AAC5F,8FAA8F;AAC9F,+FAA+F;AAC/F,iGAAiG;AACjG,+FAA+F;AAC/F,6FAA6F;AAC7F,6FAA6F;AAC7F,iBAAiB;AACjB,MAAMC,2BAA2B,IAAIC,IAAI;IAAC;IAAQ;IAAU;IAAU;IAAS;CAAe;AAE9F,0FAA0F;AAC1F,uEAAuE;AACvE,MAAMC,qBAAqB,CAAC,sEAAsE,CAAC;AAEnG,SAASC,WAAWC,GAAc;IAChC,OAAO;QAACA,IAAIC,OAAO;QAAED,IAAIE,MAAM,CAACC,KAAK;QAAEH,IAAIE,MAAM,CAACE,OAAO;QAAEJ,IAAIE,MAAM,CAACG,IAAI;KAAC;AAC7E;AAEA,6FAA6F;AAC7F,2FAA2F;AAC3F,8BAA8B;AAC9B,MAAMC,0BAA0B;AAEhC,OAAO,eAAeC,UAAUC,IAAgB,EAAEC,GAAW,EAAEC,OAAe;IAC5E,MAAMC,QAAQrB,UAAUmB,KAAKC;IAC7B,MAAME,aAAa,IAAIf,IAAIc,MAAME,GAAG,CAAC,CAACC,IAAMA,EAAEb,OAAO;IAErD,MAAMc,eAAe,MAAMP,KAAKQ,OAAO,CAAC;IACxC,MAAMC,eAAgB,MAAMF,aAAaG,GAAG;IAC5C,MAAMC,WAAW,IAAIC,IAAIH,aAAaJ,GAAG,CAAC,CAACQ,IAAM;YAACA,EAAEC,IAAI;YAAED;SAAE;IAC5D,MAAME,WAAWN,aAAaO,MAAM,CAAC,CAACH,IAAM,CAACT,WAAWa,GAAG,CAACJ,EAAEC,IAAI,GAAGT,GAAG,CAAC,CAACQ,IAAMA,EAAEC,IAAI;IAEtF,MAAMI,YAAYf,MAAMa,MAAM,CAAC,CAACV;QAC9B,MAAMa,MAAMR,SAASS,GAAG,CAACd,EAAEb,OAAO;QAClC,OAAO,CAAC0B,OAAOA,IAAIE,MAAM,KAAKf,EAAEgB,OAAO,IAAIH,IAAII,KAAK,KAAKjB,EAAEkB,IAAI;IACjE;IAEA,IAAIT,SAASU,MAAM,KAAK,KAAKP,UAAUO,MAAM,KAAK,GAAG,OAAO;QAAEC,QAAQ;QAAGC,UAAU,EAAE;IAAC;IAEtF,MAAMC,WAAWhD,eAAeqB;IAChC,MAAM4B,cAAc,MAAM5C,WAAWe;IAErC,MAAM8B,SAASjD,SAAS,mBAAmBqC,UAAUO,MAAM;IAC3D,MAAM,EAAEM,MAAMC,UAAU,EAAEL,QAAQ,EAAEM,UAAU,EAAE,GAAG,MAAMjD,aAAakC,WAAWU,UAAU3B,KAAK4B,aAAaC,OAAOI,IAAI;IACxHJ,OAAOK,MAAM;IACb,KAAK,MAAMC,OAAOH,WAAYJ,YAAYQ,GAAG,CAACD;IAE9C,MAAME,aAAa;WAAIT;KAAY;IACnC,IAAIS,WAAWb,MAAM,GAAG3B,yBAAyB;QAC/C,MAAM,IAAInB,WAAW,gBAAgB,CAAC,uBAAuB,EAAE2D,WAAWb,MAAM,CAAC,8CAA8C,EAAE3B,wBAAwB,gIAAgI,CAAC;IAC5R;IACA,MAAMyC,kBAAkBD,WAAWtB,MAAM,CAAC,CAACwB,IAAMpD,yBAAyB6B,GAAG,CAACuB,MAAM,CAACzD,iBAAiBkC,GAAG,CAACuB;IAC1G,MAAMC,YAAY,CAAC,yBAAyB,EAAEF,gBAAgBlC,GAAG,CAACnB,YAAYwD,IAAI,CAAC,MAAM,UAAU,EAAEH,gBAAgBlC,GAAG,CAAC,IAAM,KAAKqC,IAAI,CAAC,MAAM,oCAAoC,EAAEH,gBAClLvB,MAAM,CAAC,CAACwB,IAAMA,MAAM,QACpBnC,GAAG,CAAC,CAACmC,IAAM,GAAGtD,WAAWsD,GAAG,YAAY,EAAEtD,WAAWsD,IAAI,EACzDE,IAAI,CAAC,OAAO;IAEf,MAAMC,QAAQzB,UAAUF,MAAM,CAAC,CAACV,IAAM,CAACK,SAASM,GAAG,CAACX,EAAEb,OAAO,GAAGY,GAAG,CAAC,CAACC,IAAMA,EAAEb,OAAO;IACpF,MAAMmD,QAAwB;QAAEzC;QAAO0C,UAAUb,WAAW3B,GAAG,CAAC,CAACyC,IAAMA,EAAErD,OAAO;QAAGkD;QAAO5B;IAAS;IACnG,MAAMgC,WAAW,IAAI1D,IAAIsD;IACzB,MAAMK,mBAAmBhB,WAAW3B,GAAG,CAAC,CAACyC,IAAMA,EAAErD,OAAO,EAAEuB,MAAM,CAAC,CAACiC,IAAM,CAACF,SAAS9B,GAAG,CAACgC;IAEtF,MAAM9D,gBAAgBa,MAAM;YA8CyCkD,iBAK7BA;QAlDtC,yFAAyF;QACzF,2FAA2F;QAC3F,KAAK,MAAMd,OAAOH,WAAY,MAAMjC,KAAKmD,IAAI,CAAC,CAAC,mCAAmC,EAAEjE,WAAWkD,KAAK,QAAQ,CAAC;QAE7G,IAAIJ,WAAWP,MAAM,GAAG,GAAG;YACzB,MAAM2B,OAAOpB,WAAW3B,GAAG,CAAC,CAACb,MAC3B+C,gBAAgBlC,GAAG,CAAC,CAAC+B;wBAMZ5C;oBALP,IAAI4C,QAAQ,QAAQ,OAAO5C,IAAIC,OAAO;oBACtC,IAAI2C,QAAQ,UAAU,OAAO5C,IAAI8B,OAAO;oBACxC,IAAIc,QAAQ,UAAU,OAAO5C,IAAI6D,OAAO;oBACxC,IAAIjB,QAAQ,SAAS,OAAO5C,IAAIgC,IAAI;oBACpC,IAAIY,QAAQ,gBAAgB,OAAO5C,IAAI8D,UAAU;oBACjD,QAAO9D,gBAAAA,IAAI+D,IAAI,CAACnB,IAAI,cAAb5C,2BAAAA,gBAAiB;gBAC1B;YAEF,MAAMQ,KAAKwD,QAAQ,CAACf,WAAWW;QACjC;QAEA,MAAMK,iBAAiB;eAAI1C;eAAaiC;SAAiB;QACzD,IAAIS,eAAehC,MAAM,GAAG,GAC1B,MAAMzB,KAAKwD,QAAQ,CACjB,wCACAC,eAAepD,GAAG,CAAC,CAAC4C,IAAM;gBAACA;aAAE;QAEjC,IAAIjB,WAAWP,MAAM,GAAG,GAAG,MAAMzB,KAAKwD,QAAQ,CAAClE,oBAAoB0C,WAAW3B,GAAG,CAACd;QAElF,IAAIwB,SAASU,MAAM,GAAG,GACpB,MAAMzB,KAAKwD,QAAQ,CACjB,4CACAzC,SAASV,GAAG,CAAC,CAAC4C,IAAM;gBAACA;aAAE;QAG3B,0FAA0F;QAC1F,kEAAkE;QAClE,MAAMS,gBAAgB;eAAI3C;eAAaiC;SAAiB;QACxD,IAAIU,cAAcjC,MAAM,GAAG,GACzB,MAAMzB,KAAKwD,QAAQ,CACjB,6CACAE,cAAcrD,GAAG,CAAC,CAAC4C,IAAM;gBAACA;aAAE;QAEhC,MAAMU,aAA0B,EAAE;QAClC,KAAK,MAAMnE,OAAOwC,WAAY,KAAK,MAAM4B,cAAcpE,IAAIqE,OAAO,CAAEF,WAAWG,IAAI,CAAC;YAACtE,IAAIC,OAAO;YAAEmE;SAAW;QAC7G,IAAID,WAAWlC,MAAM,GAAG,GAAG,MAAMzB,KAAKwD,QAAQ,CAAC,2DAA2DG;QAE1G,MAAMI,eAAe;eAAIhD;eAAaiC;SAAiB;QACvD,IAAIe,aAAatC,MAAM,GAAG,GAAG,KAAK,MAAMyB,WAAWtB,SAAU,QAAMsB,kBAAAA,QAAQc,MAAM,cAAdd,sCAAAA,qBAAAA,SAAiBlD,MAAM+D,cAAcnB;QACxG,KAAK,MAAMM,WAAWtB,SAAU;gBAExBsB;YADN,MAAMe,iBAAiCjC,WAAW3B,GAAG,CAAC,CAACb,MAAS,CAAA;oBAAEsB,MAAMtB,IAAIC,OAAO;oBAAEyE,WAAW1E,IAAI0E,SAAS,CAAChB,QAAQiB,IAAI,CAAC;gBAAC,CAAA;YAC5H,QAAMjB,iBAAAA,QAAQkB,KAAK,cAAblB,qCAAAA,oBAAAA,SAAgBlD,MAAMiE,gBAAgBrB;QAC9C;QACA,KAAK,MAAMM,WAAWtB,SAAU,QAAMsB,0BAAAA,QAAQmB,cAAc,cAAtBnB,8CAAAA,6BAAAA,SAAyBlD,MAAM4C;IACvE;IAEA,OAAO;QAAElB,QAAQM,WAAWP,MAAM;QAAEE;IAAS;AAC/C"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/duckdb/reconcile.ts"],"sourcesContent":["import type { Config } from '../../config/index.ts';\nimport { SenseError } from '../../errors.ts';\nimport { activeFeatures } from '../../features/index.ts';\nimport type { ExtractedDoc, ReconcileDelta } from '../../features/types.ts';\nimport { progress } from '../../output/progress.ts';\nimport type { ParsedDoc } from '../../scan/index.ts';\nimport { listFiles, RESERVED_COLUMNS } from '../../scan/index.ts';\nimport { reparseFiles } from '../../scan/reparse.ts';\nimport { getColumns, quoteIdent } from '../shared.ts';\nimport { withTransaction } from '../transaction.ts';\nimport type { Connection } from '../types.ts';\n\n// Mirrors sqlite/reconcile.ts's frontmatter-upsert shape (dynamic ALTER TABLE per discovered key, ON CONFLICT upsert). `content` is\n// a plain table, not FTS-virtual, so rows are maintained here unconditionally; DuckDB's ALTER TABLE needs a declared type, so every dynamic frontmatter column is VARIANT (holds mapValue()'s mixed JS types for one key across files).\nconst CORE_FRONTMATTER_COLUMNS = new Set(['path', '_mtime', '_ctime', '_size', '_parse_error']);\n\n// No rowid coupling needed (unlike sqlite's content, which links to frontmatter's rowid):\n// `path` is content's own primary key, so this is a plain per-doc row.\nconst INSERT_CONTENT_SQL = `INSERT INTO content (\"path\", title, summary, text) VALUES (?, ?, ?, ?)`;\n\nfunction contentRow(doc: ParsedDoc): unknown[] {\n return [doc.relPath, doc.search.title, doc.search.summary, doc.search.text];\n}\n\n// No compile-time column cap in DuckDB (unlike SQLite's SQLITE_MAX_COLUMN); kept as a sanity fence anyway\n// so a runaway frontmatter generator fails with a clear message instead of an unbounded ALTER TABLE loop.\nconst MAX_FRONTMATTER_COLUMNS = 10_000;\n\nexport async function reconcile(conn: Connection, cfg: Config, baseDir: string): Promise<{ parsed: number; warnings: string[] }> {\n const files = listFiles(cfg, baseDir);\n const currentSet = new Set(files.map((f) => f.relPath));\n\n const existingStmt = await conn.prepare('SELECT \"path\", \"_mtime\", \"_size\" FROM frontmatter');\n const existingRows = (await existingStmt.all()) as Array<{ path: string; _mtime: number; _size: number }>;\n const existing = new Map(existingRows.map((r) => [r.path, r]));\n const vanished = existingRows.filter((r) => !currentSet.has(r.path)).map((r) => r.path);\n\n const toReparse = files.filter((f) => {\n const row = existing.get(f.relPath);\n return !row || row._mtime !== f.mtimeMs || row._size !== f.size;\n });\n\n if (vanished.length === 0 && toReparse.length === 0) return { parsed: 0, warnings: [] };\n\n const features = activeFeatures(cfg);\n const seenColumns = await getColumns(conn);\n\n const report = progress('reparsing files', toReparse.length);\n const { docs: parsedDocs, warnings, newColumns } = await reparseFiles(toReparse, features, cfg, seenColumns, report.tick);\n report.finish();\n for (const col of newColumns) seenColumns.add(col);\n\n const allColumns = [...seenColumns];\n if (allColumns.length > MAX_FRONTMATTER_COLUMNS) {\n throw new SenseError('COLUMN_LIMIT', `frontmatter would need ${allColumns.length} columns, crossing this store's sanity limit (${MAX_FRONTMATTER_COLUMNS}). Narrow the presets' include globs so fewer/other files are indexed, or fix whatever is generating unbounded frontmatter keys.`);\n }\n const writableColumns = allColumns.filter((c) => CORE_FRONTMATTER_COLUMNS.has(c) || !RESERVED_COLUMNS.has(c));\n const insertSql = `INSERT INTO frontmatter (${writableColumns.map(quoteIdent).join(', ')}) VALUES (${writableColumns.map(() => '?').join(', ')}) ON CONFLICT(\"path\") DO UPDATE SET ${writableColumns\n .filter((c) => c !== 'path')\n .map((c) => `${quoteIdent(c)} = excluded.${quoteIdent(c)}`)\n .join(', ')}`;\n\n const added = toReparse.filter((f) => !existing.has(f.relPath)).map((f) => f.relPath);\n const delta: ReconcileDelta = { files, reparsed: parsedDocs.map((d) => d.relPath), added, vanished };\n const addedSet = new Set(added);\n const reparsedExisting = parsedDocs.map((d) => d.relPath).filter((p) => !addedSet.has(p));\n\n await withTransaction(conn, async () => {\n // DuckDB's ALTER TABLE ADD COLUMN requires a type; VARIANT is the only one that can hold\n // the mixed bigint/number/string/null shapes mapValue() produces for one key across files.\n for (const col of newColumns) await conn.exec(`ALTER TABLE frontmatter ADD COLUMN ${quoteIdent(col)} VARIANT`);\n\n if (parsedDocs.length > 0) {\n const rows = parsedDocs.map((doc) =>\n writableColumns.map((col) => {\n if (col === 'path') return doc.relPath;\n if (col === '_mtime') return doc.mtimeMs;\n if (col === '_ctime') return doc.ctimeMs;\n if (col === '_size') return doc.size;\n if (col === '_parse_error') return doc.parseError;\n return doc.data[col] ?? null;\n })\n );\n await conn.runBatch(insertSql, rows);\n }\n\n const contentTouched = [...vanished, ...reparsedExisting];\n if (contentTouched.length > 0)\n await conn.runBatch(\n 'DELETE FROM content WHERE \"path\" = ?',\n contentTouched.map((p) => [p])\n );\n if (parsedDocs.length > 0) await conn.runBatch(INSERT_CONTENT_SQL, parsedDocs.map(contentRow));\n\n if (vanished.length > 0)\n await conn.runBatch(\n 'DELETE FROM frontmatter WHERE \"path\" = ?',\n vanished.map((p) => [p])\n );\n\n // A preset edit forces a full rebuild, so an unchanged doc's coverage is already correct;\n // new docs have nothing to clear, which keeps cold builds linear.\n const presetTouched = [...vanished, ...reparsedExisting];\n if (presetTouched.length > 0)\n await conn.runBatch(\n 'DELETE FROM preset_files WHERE \"path\" = ?',\n presetTouched.map((p) => [p])\n );\n const presetRows: unknown[][] = [];\n for (const doc of parsedDocs) for (const presetName of doc.presets) presetRows.push([doc.relPath, presetName]);\n if (presetRows.length > 0) await conn.runBatch('INSERT INTO preset_files (\"path\", preset) VALUES (?, ?)', presetRows);\n\n const removedPaths = [...vanished, ...reparsedExisting];\n if (removedPaths.length > 0) for (const feature of features) await feature.remove?.(conn, removedPaths, delta);\n for (const feature of features) {\n const docsForFeature: ExtractedDoc[] = parsedDocs.map((doc) => ({ path: doc.relPath, extracted: doc.extracted[feature.name] }));\n await feature.store?.(conn, docsForFeature, delta);\n }\n for (const feature of features) await feature.afterReconcile?.(conn, delta);\n });\n\n return { parsed: parsedDocs.length, warnings };\n}\n"],"names":["SenseError","activeFeatures","progress","listFiles","RESERVED_COLUMNS","reparseFiles","getColumns","quoteIdent","withTransaction","CORE_FRONTMATTER_COLUMNS","Set","INSERT_CONTENT_SQL","contentRow","doc","relPath","search","title","summary","text","MAX_FRONTMATTER_COLUMNS","reconcile","conn","cfg","baseDir","files","currentSet","map","f","existingStmt","prepare","existingRows","all","existing","Map","r","path","vanished","filter","has","toReparse","row","get","_mtime","mtimeMs","_size","size","length","parsed","warnings","features","seenColumns","report","docs","parsedDocs","newColumns","tick","finish","col","add","allColumns","writableColumns","c","insertSql","join","added","delta","reparsed","d","addedSet","reparsedExisting","p","feature","exec","rows","ctimeMs","parseError","data","runBatch","contentTouched","presetTouched","presetRows","presetName","presets","push","removedPaths","remove","docsForFeature","extracted","name","store","afterReconcile"],"mappings":"AACA,SAASA,UAAU,QAAQ,kBAAkB;AAC7C,SAASC,cAAc,QAAQ,0BAA0B;AAEzD,SAASC,QAAQ,QAAQ,2BAA2B;AAEpD,SAASC,SAAS,EAAEC,gBAAgB,QAAQ,sBAAsB;AAClE,SAASC,YAAY,QAAQ,wBAAwB;AACrD,SAASC,UAAU,EAAEC,UAAU,QAAQ,eAAe;AACtD,SAASC,eAAe,QAAQ,oBAAoB;AAGpD,oIAAoI;AACpI,wOAAwO;AACxO,MAAMC,2BAA2B,IAAIC,IAAI;IAAC;IAAQ;IAAU;IAAU;IAAS;CAAe;AAE9F,0FAA0F;AAC1F,uEAAuE;AACvE,MAAMC,qBAAqB,CAAC,sEAAsE,CAAC;AAEnG,SAASC,WAAWC,GAAc;IAChC,OAAO;QAACA,IAAIC,OAAO;QAAED,IAAIE,MAAM,CAACC,KAAK;QAAEH,IAAIE,MAAM,CAACE,OAAO;QAAEJ,IAAIE,MAAM,CAACG,IAAI;KAAC;AAC7E;AAEA,0GAA0G;AAC1G,0GAA0G;AAC1G,MAAMC,0BAA0B;AAEhC,OAAO,eAAeC,UAAUC,IAAgB,EAAEC,GAAW,EAAEC,OAAe;IAC5E,MAAMC,QAAQrB,UAAUmB,KAAKC;IAC7B,MAAME,aAAa,IAAIf,IAAIc,MAAME,GAAG,CAAC,CAACC,IAAMA,EAAEb,OAAO;IAErD,MAAMc,eAAe,MAAMP,KAAKQ,OAAO,CAAC;IACxC,MAAMC,eAAgB,MAAMF,aAAaG,GAAG;IAC5C,MAAMC,WAAW,IAAIC,IAAIH,aAAaJ,GAAG,CAAC,CAACQ,IAAM;YAACA,EAAEC,IAAI;YAAED;SAAE;IAC5D,MAAME,WAAWN,aAAaO,MAAM,CAAC,CAACH,IAAM,CAACT,WAAWa,GAAG,CAACJ,EAAEC,IAAI,GAAGT,GAAG,CAAC,CAACQ,IAAMA,EAAEC,IAAI;IAEtF,MAAMI,YAAYf,MAAMa,MAAM,CAAC,CAACV;QAC9B,MAAMa,MAAMR,SAASS,GAAG,CAACd,EAAEb,OAAO;QAClC,OAAO,CAAC0B,OAAOA,IAAIE,MAAM,KAAKf,EAAEgB,OAAO,IAAIH,IAAII,KAAK,KAAKjB,EAAEkB,IAAI;IACjE;IAEA,IAAIT,SAASU,MAAM,KAAK,KAAKP,UAAUO,MAAM,KAAK,GAAG,OAAO;QAAEC,QAAQ;QAAGC,UAAU,EAAE;IAAC;IAEtF,MAAMC,WAAWhD,eAAeqB;IAChC,MAAM4B,cAAc,MAAM5C,WAAWe;IAErC,MAAM8B,SAASjD,SAAS,mBAAmBqC,UAAUO,MAAM;IAC3D,MAAM,EAAEM,MAAMC,UAAU,EAAEL,QAAQ,EAAEM,UAAU,EAAE,GAAG,MAAMjD,aAAakC,WAAWU,UAAU3B,KAAK4B,aAAaC,OAAOI,IAAI;IACxHJ,OAAOK,MAAM;IACb,KAAK,MAAMC,OAAOH,WAAYJ,YAAYQ,GAAG,CAACD;IAE9C,MAAME,aAAa;WAAIT;KAAY;IACnC,IAAIS,WAAWb,MAAM,GAAG3B,yBAAyB;QAC/C,MAAM,IAAInB,WAAW,gBAAgB,CAAC,uBAAuB,EAAE2D,WAAWb,MAAM,CAAC,8CAA8C,EAAE3B,wBAAwB,gIAAgI,CAAC;IAC5R;IACA,MAAMyC,kBAAkBD,WAAWtB,MAAM,CAAC,CAACwB,IAAMpD,yBAAyB6B,GAAG,CAACuB,MAAM,CAACzD,iBAAiBkC,GAAG,CAACuB;IAC1G,MAAMC,YAAY,CAAC,yBAAyB,EAAEF,gBAAgBlC,GAAG,CAACnB,YAAYwD,IAAI,CAAC,MAAM,UAAU,EAAEH,gBAAgBlC,GAAG,CAAC,IAAM,KAAKqC,IAAI,CAAC,MAAM,oCAAoC,EAAEH,gBAClLvB,MAAM,CAAC,CAACwB,IAAMA,MAAM,QACpBnC,GAAG,CAAC,CAACmC,IAAM,GAAGtD,WAAWsD,GAAG,YAAY,EAAEtD,WAAWsD,IAAI,EACzDE,IAAI,CAAC,OAAO;IAEf,MAAMC,QAAQzB,UAAUF,MAAM,CAAC,CAACV,IAAM,CAACK,SAASM,GAAG,CAACX,EAAEb,OAAO,GAAGY,GAAG,CAAC,CAACC,IAAMA,EAAEb,OAAO;IACpF,MAAMmD,QAAwB;QAAEzC;QAAO0C,UAAUb,WAAW3B,GAAG,CAAC,CAACyC,IAAMA,EAAErD,OAAO;QAAGkD;QAAO5B;IAAS;IACnG,MAAMgC,WAAW,IAAI1D,IAAIsD;IACzB,MAAMK,mBAAmBhB,WAAW3B,GAAG,CAAC,CAACyC,IAAMA,EAAErD,OAAO,EAAEuB,MAAM,CAAC,CAACiC,IAAM,CAACF,SAAS9B,GAAG,CAACgC;IAEtF,MAAM9D,gBAAgBa,MAAM;YA8CyCkD,iBAK7BA;QAlDtC,yFAAyF;QACzF,2FAA2F;QAC3F,KAAK,MAAMd,OAAOH,WAAY,MAAMjC,KAAKmD,IAAI,CAAC,CAAC,mCAAmC,EAAEjE,WAAWkD,KAAK,QAAQ,CAAC;QAE7G,IAAIJ,WAAWP,MAAM,GAAG,GAAG;YACzB,MAAM2B,OAAOpB,WAAW3B,GAAG,CAAC,CAACb,MAC3B+C,gBAAgBlC,GAAG,CAAC,CAAC+B;wBAMZ5C;oBALP,IAAI4C,QAAQ,QAAQ,OAAO5C,IAAIC,OAAO;oBACtC,IAAI2C,QAAQ,UAAU,OAAO5C,IAAI8B,OAAO;oBACxC,IAAIc,QAAQ,UAAU,OAAO5C,IAAI6D,OAAO;oBACxC,IAAIjB,QAAQ,SAAS,OAAO5C,IAAIgC,IAAI;oBACpC,IAAIY,QAAQ,gBAAgB,OAAO5C,IAAI8D,UAAU;oBACjD,QAAO9D,gBAAAA,IAAI+D,IAAI,CAACnB,IAAI,cAAb5C,2BAAAA,gBAAiB;gBAC1B;YAEF,MAAMQ,KAAKwD,QAAQ,CAACf,WAAWW;QACjC;QAEA,MAAMK,iBAAiB;eAAI1C;eAAaiC;SAAiB;QACzD,IAAIS,eAAehC,MAAM,GAAG,GAC1B,MAAMzB,KAAKwD,QAAQ,CACjB,wCACAC,eAAepD,GAAG,CAAC,CAAC4C,IAAM;gBAACA;aAAE;QAEjC,IAAIjB,WAAWP,MAAM,GAAG,GAAG,MAAMzB,KAAKwD,QAAQ,CAAClE,oBAAoB0C,WAAW3B,GAAG,CAACd;QAElF,IAAIwB,SAASU,MAAM,GAAG,GACpB,MAAMzB,KAAKwD,QAAQ,CACjB,4CACAzC,SAASV,GAAG,CAAC,CAAC4C,IAAM;gBAACA;aAAE;QAG3B,0FAA0F;QAC1F,kEAAkE;QAClE,MAAMS,gBAAgB;eAAI3C;eAAaiC;SAAiB;QACxD,IAAIU,cAAcjC,MAAM,GAAG,GACzB,MAAMzB,KAAKwD,QAAQ,CACjB,6CACAE,cAAcrD,GAAG,CAAC,CAAC4C,IAAM;gBAACA;aAAE;QAEhC,MAAMU,aAA0B,EAAE;QAClC,KAAK,MAAMnE,OAAOwC,WAAY,KAAK,MAAM4B,cAAcpE,IAAIqE,OAAO,CAAEF,WAAWG,IAAI,CAAC;YAACtE,IAAIC,OAAO;YAAEmE;SAAW;QAC7G,IAAID,WAAWlC,MAAM,GAAG,GAAG,MAAMzB,KAAKwD,QAAQ,CAAC,2DAA2DG;QAE1G,MAAMI,eAAe;eAAIhD;eAAaiC;SAAiB;QACvD,IAAIe,aAAatC,MAAM,GAAG,GAAG,KAAK,MAAMyB,WAAWtB,SAAU,QAAMsB,kBAAAA,QAAQc,MAAM,cAAdd,sCAAAA,qBAAAA,SAAiBlD,MAAM+D,cAAcnB;QACxG,KAAK,MAAMM,WAAWtB,SAAU;gBAExBsB;YADN,MAAMe,iBAAiCjC,WAAW3B,GAAG,CAAC,CAACb,MAAS,CAAA;oBAAEsB,MAAMtB,IAAIC,OAAO;oBAAEyE,WAAW1E,IAAI0E,SAAS,CAAChB,QAAQiB,IAAI,CAAC;gBAAC,CAAA;YAC5H,QAAMjB,iBAAAA,QAAQkB,KAAK,cAAblB,qCAAAA,oBAAAA,SAAgBlD,MAAMiE,gBAAgBrB;QAC9C;QACA,KAAK,MAAMM,WAAWtB,SAAU,QAAMsB,0BAAAA,QAAQmB,cAAc,cAAtBnB,8CAAAA,6BAAAA,SAAyBlD,MAAM4C;IACvE;IAEA,OAAO;QAAElB,QAAQM,WAAWP,MAAM;QAAEE;IAAS;AAC/C"}
@@ -1,11 +1,8 @@
1
1
  import { segmentMatch } from '../../text/segment.js';
2
2
  import { basenameImpl, hasImpl } from '../sql-functions.js';
3
3
  import { duckdbApi } from './native.js';
4
- // Dynamic, not a top-level import: sqlite trees must never even attempt to resolve this optional
5
- // peer dependency (see open.ts's connect()). Goes through native.ts's shared duckdbApi(), not a
6
- // bare `import('@duckdb/node-api')` of its own -- by the time this runs, open.ts has already
7
- // resolved the package, possibly via an on-demand install, and a second independent bare-specifier
8
- // lookup in the same process would hit Node's stale pre-install resolution miss (see native.ts).
4
+ // Dynamic, not a top-level import: sqlite trees must never attempt to resolve this optional peer dependency (open.ts's connect()).
5
+ // Goes through native.ts's shared duckdbApi(), not its own bare import, since a second independent lookup would hit Node's stale pre-install resolution miss.
9
6
  export async function registerFunctions(conn) {
10
7
  const { DuckDBIntegerType, DuckDBScalarFunction, DuckDBVarCharType } = await duckdbApi();
11
8
  conn.registerScalarFunction(DuckDBScalarFunction.create({
@@ -36,10 +33,8 @@ export async function registerFunctions(conn) {
36
33
  outputVector.flush();
37
34
  }
38
35
  }));
39
- // segment() rewrites unspaced-script runs into the grapheme phrase FTS5's `_seg` sidecars
40
- // need (see src/text/segment.ts); this store's own lexical index uses contains() instead (D1)
41
- // and never calls this, but hand-written raw SQL naming it still gets the real implementation
42
- // rather than a silent passthrough (PRINCIPLES: no-silent-modes).
36
+ // segment() rewrites unspaced-script runs into the grapheme phrase FTS5's `_seg` sidecars need (text/segment.ts); this store's own
37
+ // lexical index uses contains() instead and never calls this, but hand-written raw SQL naming it gets the real implementation, not a silent passthrough (PRINCIPLES: no-silent-modes).
43
38
  conn.registerScalarFunction(DuckDBScalarFunction.create({
44
39
  name: 'segment',
45
40
  parameterTypes: [
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/duckdb/sql-functions.ts"],"sourcesContent":["import type { DuckDBConnection } from '@duckdb/node-api';\nimport { segmentMatch } from '../../text/segment.ts';\nimport { basenameImpl, hasImpl } from '../sql-functions.ts';\nimport { duckdbApi } from './native.ts';\n\n// Dynamic, not a top-level import: sqlite trees must never even attempt to resolve this optional\n// peer dependency (see open.ts's connect()). Goes through native.ts's shared duckdbApi(), not a\n// bare `import('@duckdb/node-api')` of its own -- by the time this runs, open.ts has already\n// resolved the package, possibly via an on-demand install, and a second independent bare-specifier\n// lookup in the same process would hit Node's stale pre-install resolution miss (see native.ts).\nexport async function registerFunctions(conn: DuckDBConnection): Promise<void> {\n const { DuckDBIntegerType, DuckDBScalarFunction, DuckDBVarCharType } = await duckdbApi();\n conn.registerScalarFunction(\n DuckDBScalarFunction.create({\n name: 'has',\n parameterTypes: [DuckDBVarCharType.instance, DuckDBVarCharType.instance],\n returnType: DuckDBIntegerType.instance,\n mainFunction(_info, inputChunk, outputVector) {\n const fieldVec = inputChunk.getColumnVector(0);\n const valueVec = inputChunk.getColumnVector(1);\n for (let i = 0; i < inputChunk.rowCount; i++) outputVector.setItem(i, hasImpl(fieldVec.getItem(i), valueVec.getItem(i)));\n outputVector.flush();\n },\n })\n );\n conn.registerScalarFunction(\n DuckDBScalarFunction.create({\n name: 'basename',\n parameterTypes: [DuckDBVarCharType.instance],\n varArgsType: DuckDBVarCharType.instance,\n returnType: DuckDBVarCharType.instance,\n mainFunction(_info, inputChunk, outputVector) {\n const pathVec = inputChunk.getColumnVector(0);\n const suffixVec = inputChunk.columnCount > 1 ? inputChunk.getColumnVector(1) : null;\n for (let i = 0; i < inputChunk.rowCount; i++) outputVector.setItem(i, basenameImpl(pathVec.getItem(i), suffixVec ? suffixVec.getItem(i) : undefined));\n outputVector.flush();\n },\n })\n );\n // segment() rewrites unspaced-script runs into the grapheme phrase FTS5's `_seg` sidecars\n // need (see src/text/segment.ts); this store's own lexical index uses contains() instead (D1)\n // and never calls this, but hand-written raw SQL naming it still gets the real implementation\n // rather than a silent passthrough (PRINCIPLES: no-silent-modes).\n conn.registerScalarFunction(\n DuckDBScalarFunction.create({\n name: 'segment',\n parameterTypes: [DuckDBVarCharType.instance],\n returnType: DuckDBVarCharType.instance,\n mainFunction(_info, inputChunk, outputVector) {\n const textVec = inputChunk.getColumnVector(0);\n for (let i = 0; i < inputChunk.rowCount; i++) outputVector.setItem(i, segmentMatch(String(textVec.getItem(i) ?? '')));\n outputVector.flush();\n },\n })\n );\n}\n"],"names":["segmentMatch","basenameImpl","hasImpl","duckdbApi","registerFunctions","conn","DuckDBIntegerType","DuckDBScalarFunction","DuckDBVarCharType","registerScalarFunction","create","name","parameterTypes","instance","returnType","mainFunction","_info","inputChunk","outputVector","fieldVec","getColumnVector","valueVec","i","rowCount","setItem","getItem","flush","varArgsType","pathVec","suffixVec","columnCount","undefined","textVec","String"],"mappings":"AACA,SAASA,YAAY,QAAQ,wBAAwB;AACrD,SAASC,YAAY,EAAEC,OAAO,QAAQ,sBAAsB;AAC5D,SAASC,SAAS,QAAQ,cAAc;AAExC,iGAAiG;AACjG,gGAAgG;AAChG,6FAA6F;AAC7F,mGAAmG;AACnG,iGAAiG;AACjG,OAAO,eAAeC,kBAAkBC,IAAsB;IAC5D,MAAM,EAAEC,iBAAiB,EAAEC,oBAAoB,EAAEC,iBAAiB,EAAE,GAAG,MAAML;IAC7EE,KAAKI,sBAAsB,CACzBF,qBAAqBG,MAAM,CAAC;QAC1BC,MAAM;QACNC,gBAAgB;YAACJ,kBAAkBK,QAAQ;YAAEL,kBAAkBK,QAAQ;SAAC;QACxEC,YAAYR,kBAAkBO,QAAQ;QACtCE,cAAaC,KAAK,EAAEC,UAAU,EAAEC,YAAY;YAC1C,MAAMC,WAAWF,WAAWG,eAAe,CAAC;YAC5C,MAAMC,WAAWJ,WAAWG,eAAe,CAAC;YAC5C,IAAK,IAAIE,IAAI,GAAGA,IAAIL,WAAWM,QAAQ,EAAED,IAAKJ,aAAaM,OAAO,CAACF,GAAGpB,QAAQiB,SAASM,OAAO,CAACH,IAAID,SAASI,OAAO,CAACH;YACpHJ,aAAaQ,KAAK;QACpB;IACF;IAEFrB,KAAKI,sBAAsB,CACzBF,qBAAqBG,MAAM,CAAC;QAC1BC,MAAM;QACNC,gBAAgB;YAACJ,kBAAkBK,QAAQ;SAAC;QAC5Cc,aAAanB,kBAAkBK,QAAQ;QACvCC,YAAYN,kBAAkBK,QAAQ;QACtCE,cAAaC,KAAK,EAAEC,UAAU,EAAEC,YAAY;YAC1C,MAAMU,UAAUX,WAAWG,eAAe,CAAC;YAC3C,MAAMS,YAAYZ,WAAWa,WAAW,GAAG,IAAIb,WAAWG,eAAe,CAAC,KAAK;YAC/E,IAAK,IAAIE,IAAI,GAAGA,IAAIL,WAAWM,QAAQ,EAAED,IAAKJ,aAAaM,OAAO,CAACF,GAAGrB,aAAa2B,QAAQH,OAAO,CAACH,IAAIO,YAAYA,UAAUJ,OAAO,CAACH,KAAKS;YAC1Ib,aAAaQ,KAAK;QACpB;IACF;IAEF,0FAA0F;IAC1F,8FAA8F;IAC9F,8FAA8F;IAC9F,kEAAkE;IAClErB,KAAKI,sBAAsB,CACzBF,qBAAqBG,MAAM,CAAC;QAC1BC,MAAM;QACNC,gBAAgB;YAACJ,kBAAkBK,QAAQ;SAAC;QAC5CC,YAAYN,kBAAkBK,QAAQ;QACtCE,cAAaC,KAAK,EAAEC,UAAU,EAAEC,YAAY;gBAEgDc;YAD1F,MAAMA,UAAUf,WAAWG,eAAe,CAAC;YAC3C,IAAK,IAAIE,IAAI,GAAGA,IAAIL,WAAWM,QAAQ,EAAED,IAAKJ,aAAaM,OAAO,CAACF,GAAGtB,aAAaiC,QAAOD,mBAAAA,QAAQP,OAAO,CAACH,gBAAhBU,8BAAAA,mBAAsB;YAChHd,aAAaQ,KAAK;QACpB;IACF;AAEJ"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/duckdb/sql-functions.ts"],"sourcesContent":["import type { DuckDBConnection } from '@duckdb/node-api';\nimport { segmentMatch } from '../../text/segment.ts';\nimport { basenameImpl, hasImpl } from '../sql-functions.ts';\nimport { duckdbApi } from './native.ts';\n\n// Dynamic, not a top-level import: sqlite trees must never attempt to resolve this optional peer dependency (open.ts's connect()).\n// Goes through native.ts's shared duckdbApi(), not its own bare import, since a second independent lookup would hit Node's stale pre-install resolution miss.\nexport async function registerFunctions(conn: DuckDBConnection): Promise<void> {\n const { DuckDBIntegerType, DuckDBScalarFunction, DuckDBVarCharType } = await duckdbApi();\n conn.registerScalarFunction(\n DuckDBScalarFunction.create({\n name: 'has',\n parameterTypes: [DuckDBVarCharType.instance, DuckDBVarCharType.instance],\n returnType: DuckDBIntegerType.instance,\n mainFunction(_info, inputChunk, outputVector) {\n const fieldVec = inputChunk.getColumnVector(0);\n const valueVec = inputChunk.getColumnVector(1);\n for (let i = 0; i < inputChunk.rowCount; i++) outputVector.setItem(i, hasImpl(fieldVec.getItem(i), valueVec.getItem(i)));\n outputVector.flush();\n },\n })\n );\n conn.registerScalarFunction(\n DuckDBScalarFunction.create({\n name: 'basename',\n parameterTypes: [DuckDBVarCharType.instance],\n varArgsType: DuckDBVarCharType.instance,\n returnType: DuckDBVarCharType.instance,\n mainFunction(_info, inputChunk, outputVector) {\n const pathVec = inputChunk.getColumnVector(0);\n const suffixVec = inputChunk.columnCount > 1 ? inputChunk.getColumnVector(1) : null;\n for (let i = 0; i < inputChunk.rowCount; i++) outputVector.setItem(i, basenameImpl(pathVec.getItem(i), suffixVec ? suffixVec.getItem(i) : undefined));\n outputVector.flush();\n },\n })\n );\n // segment() rewrites unspaced-script runs into the grapheme phrase FTS5's `_seg` sidecars need (text/segment.ts); this store's own\n // lexical index uses contains() instead and never calls this, but hand-written raw SQL naming it gets the real implementation, not a silent passthrough (PRINCIPLES: no-silent-modes).\n conn.registerScalarFunction(\n DuckDBScalarFunction.create({\n name: 'segment',\n parameterTypes: [DuckDBVarCharType.instance],\n returnType: DuckDBVarCharType.instance,\n mainFunction(_info, inputChunk, outputVector) {\n const textVec = inputChunk.getColumnVector(0);\n for (let i = 0; i < inputChunk.rowCount; i++) outputVector.setItem(i, segmentMatch(String(textVec.getItem(i) ?? '')));\n outputVector.flush();\n },\n })\n );\n}\n"],"names":["segmentMatch","basenameImpl","hasImpl","duckdbApi","registerFunctions","conn","DuckDBIntegerType","DuckDBScalarFunction","DuckDBVarCharType","registerScalarFunction","create","name","parameterTypes","instance","returnType","mainFunction","_info","inputChunk","outputVector","fieldVec","getColumnVector","valueVec","i","rowCount","setItem","getItem","flush","varArgsType","pathVec","suffixVec","columnCount","undefined","textVec","String"],"mappings":"AACA,SAASA,YAAY,QAAQ,wBAAwB;AACrD,SAASC,YAAY,EAAEC,OAAO,QAAQ,sBAAsB;AAC5D,SAASC,SAAS,QAAQ,cAAc;AAExC,mIAAmI;AACnI,8JAA8J;AAC9J,OAAO,eAAeC,kBAAkBC,IAAsB;IAC5D,MAAM,EAAEC,iBAAiB,EAAEC,oBAAoB,EAAEC,iBAAiB,EAAE,GAAG,MAAML;IAC7EE,KAAKI,sBAAsB,CACzBF,qBAAqBG,MAAM,CAAC;QAC1BC,MAAM;QACNC,gBAAgB;YAACJ,kBAAkBK,QAAQ;YAAEL,kBAAkBK,QAAQ;SAAC;QACxEC,YAAYR,kBAAkBO,QAAQ;QACtCE,cAAaC,KAAK,EAAEC,UAAU,EAAEC,YAAY;YAC1C,MAAMC,WAAWF,WAAWG,eAAe,CAAC;YAC5C,MAAMC,WAAWJ,WAAWG,eAAe,CAAC;YAC5C,IAAK,IAAIE,IAAI,GAAGA,IAAIL,WAAWM,QAAQ,EAAED,IAAKJ,aAAaM,OAAO,CAACF,GAAGpB,QAAQiB,SAASM,OAAO,CAACH,IAAID,SAASI,OAAO,CAACH;YACpHJ,aAAaQ,KAAK;QACpB;IACF;IAEFrB,KAAKI,sBAAsB,CACzBF,qBAAqBG,MAAM,CAAC;QAC1BC,MAAM;QACNC,gBAAgB;YAACJ,kBAAkBK,QAAQ;SAAC;QAC5Cc,aAAanB,kBAAkBK,QAAQ;QACvCC,YAAYN,kBAAkBK,QAAQ;QACtCE,cAAaC,KAAK,EAAEC,UAAU,EAAEC,YAAY;YAC1C,MAAMU,UAAUX,WAAWG,eAAe,CAAC;YAC3C,MAAMS,YAAYZ,WAAWa,WAAW,GAAG,IAAIb,WAAWG,eAAe,CAAC,KAAK;YAC/E,IAAK,IAAIE,IAAI,GAAGA,IAAIL,WAAWM,QAAQ,EAAED,IAAKJ,aAAaM,OAAO,CAACF,GAAGrB,aAAa2B,QAAQH,OAAO,CAACH,IAAIO,YAAYA,UAAUJ,OAAO,CAACH,KAAKS;YAC1Ib,aAAaQ,KAAK;QACpB;IACF;IAEF,mIAAmI;IACnI,uLAAuL;IACvLrB,KAAKI,sBAAsB,CACzBF,qBAAqBG,MAAM,CAAC;QAC1BC,MAAM;QACNC,gBAAgB;YAACJ,kBAAkBK,QAAQ;SAAC;QAC5CC,YAAYN,kBAAkBK,QAAQ;QACtCE,cAAaC,KAAK,EAAEC,UAAU,EAAEC,YAAY;gBAEgDc;YAD1F,MAAMA,UAAUf,WAAWG,eAAe,CAAC;YAC3C,IAAK,IAAIE,IAAI,GAAGA,IAAIL,WAAWM,QAAQ,EAAED,IAAKJ,aAAaM,OAAO,CAACF,GAAGtB,aAAaiC,QAAOD,mBAAAA,QAAQP,OAAO,CAACH,gBAAhBU,8BAAAA,mBAAsB;YAChHd,aAAaQ,KAAK;QACpB;IACF;AAEJ"}
@@ -7,22 +7,15 @@ import { fieldStats } from './fieldStats.js';
7
7
  import { createLexicalIndex } from './lexical.js';
8
8
  import { reconcile } from './reconcile.js';
9
9
  import { scanCandidates, scanSimilar, writeVectorBatch } from './vectors.js';
10
- // Portable surface, links/sections/tags/rank, raw sql passthrough, lexical (D1: fts BM25 +
11
- // contains(), see lexical.ts), and vectors (D2: native FLOAT[N] + array_cosine_similarity, see
12
- // vectors.ts) are implemented. 'snippets' is declined: no snippet()-equivalent exists, so every
13
- // lexical hit relies on the caller's JS excerpt fallback (commands/search.ts) rather than a
14
- // bounded engine-native one. 'phrases' means quoted-phrase matching only (types.ts):
15
- // prefix/boolean/NEAR/column-filter FTS5 syntax is not claimed here -- lexical.ts rejects it
16
- // rather than answer it differently.
10
+ // Portable surface, links/sections/tags/rank, raw sql passthrough, lexical (fts BM25 + contains(), lexical.ts), and vectors (native
11
+ // FLOAT[N] + array_cosine_similarity, vectors.ts) are implemented. 'snippets' is declined (JS excerpt fallback handles it); 'phrases' means quoted-phrase only -- FTS5 operator syntax is rejected, not answered differently.
17
12
  export const CAPABILITIES = new Set([
18
13
  'lexical',
19
14
  'phrases',
20
15
  'vectors'
21
16
  ]);
22
- // Shares one Connection instance (conn) with open()'s own reconcile call so transaction depth
23
- // (see transaction.ts) is tracked against the same object everywhere. The instance is the
24
- // native database handle that owns the WAL: close() must close it, not just disconnect the
25
- // connection, or DuckDB never checkpoints and the next open reads a mismatched WAL.
17
+ // Shares one Connection instance (conn) with open()'s own reconcile call so transaction depth (transaction.ts) is tracked against the same object everywhere.
18
+ // The instance is the native handle that owns the WAL: close() must close it, not just disconnect, or DuckDB never checkpoints and the next open reads a mismatched WAL.
26
19
  export function createStore(instance, duckdb, conn, cfg, baseDir) {
27
20
  const lex = createLexicalIndex(conn);
28
21
  return {
@@ -58,9 +51,8 @@ export function createStore(instance, duckdb, conn, cfg, baseDir) {
58
51
  lexical: {
59
52
  query: lex.query
60
53
  },
61
- // The column's fixed DDL width (STORE_DIMS) is what every scan binds against, not the
62
- // interface's per-call storeDims -- see vectors.ts's padded() for why a shorter vector is
63
- // still correct against a wider column.
54
+ // The column's fixed DDL width (STORE_DIMS) is what every scan binds against, not the interface's per-call storeDims --
55
+ // see vectors.ts's padded() for why a shorter vector remains correct against a wider column.
64
56
  vectors: {
65
57
  pending: ()=>pendingRows(conn),
66
58
  writeVectors: (rows)=>writeVectorBatch(duckdb, conn, STORE_DIMS, rows),
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/duckdb/store.ts"],"sourcesContent":["import type { DuckDBConnection, DuckDBInstance } from '@duckdb/node-api';\nimport type { Config } from '../../config/index.ts';\nimport { STORE_DIMS } from '../../embed/types.ts';\nimport { getColumns } from '../shared.ts';\nimport { withTransaction } from '../transaction.ts';\nimport type { Capability, Connection, Statement, Store } from '../types.ts';\nimport { hasVectorRow, pendingRows } from '../vectors.ts';\nimport { storeRowToJs } from './connection.ts';\nimport { fieldStats } from './fieldStats.ts';\nimport { createLexicalIndex } from './lexical.ts';\nimport { reconcile } from './reconcile.ts';\nimport { scanCandidates, scanSimilar, writeVectorBatch } from './vectors.ts';\n\n// Portable surface, links/sections/tags/rank, raw sql passthrough, lexical (D1: fts BM25 +\n// contains(), see lexical.ts), and vectors (D2: native FLOAT[N] + array_cosine_similarity, see\n// vectors.ts) are implemented. 'snippets' is declined: no snippet()-equivalent exists, so every\n// lexical hit relies on the caller's JS excerpt fallback (commands/search.ts) rather than a\n// bounded engine-native one. 'phrases' means quoted-phrase matching only (types.ts):\n// prefix/boolean/NEAR/column-filter FTS5 syntax is not claimed here -- lexical.ts rejects it\n// rather than answer it differently.\nexport const CAPABILITIES: ReadonlySet<Capability> = new Set(['lexical', 'phrases', 'vectors']);\n\n// Shares one Connection instance (conn) with open()'s own reconcile call so transaction depth\n// (see transaction.ts) is tracked against the same object everywhere. The instance is the\n// native database handle that owns the WAL: close() must close it, not just disconnect the\n// connection, or DuckDB never checkpoints and the next open reads a mismatched WAL.\nexport function createStore(instance: DuckDBInstance, duckdb: DuckDBConnection, conn: Connection, cfg: Config, baseDir: string): Store {\n const lex = createLexicalIndex(conn);\n return {\n name: 'duckdb',\n capabilities: CAPABILITIES,\n async exec(sql: string): Promise<void> {\n await conn.exec(sql);\n },\n async prepare(sql: string): Promise<Statement> {\n return conn.prepare(sql);\n },\n async runBatch(sql: string, paramRows: unknown[][]): Promise<void> {\n await conn.runBatch(sql, paramRows);\n },\n async transaction<T>(fn: () => Promise<T>): Promise<T> {\n return withTransaction(conn, fn);\n },\n async reconcile() {\n const result = await reconcile(conn, cfg, baseDir);\n // content may have changed; the fts index is rebuilt lazily, on the next lexical query\n // that needs it, not here (see lexical.ts's FtsIndexState).\n lex.markStale();\n return result;\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: lex.query,\n },\n // The column's fixed DDL width (STORE_DIMS) is what every scan binds against, not the\n // interface's per-call storeDims -- see vectors.ts's padded() for why a shorter vector is\n // still correct against a wider column.\n vectors: {\n pending: () => pendingRows(conn),\n writeVectors: (rows) => writeVectorBatch(duckdb, conn, STORE_DIMS, rows),\n candidates: (qv, _storeDims, fetch, allowed) => scanCandidates(duckdb, qv, STORE_DIMS, fetch, allowed),\n similar: (path, opts) => scanSimilar(duckdb, conn, STORE_DIMS, path, opts),\n hasVector: (path) => hasVectorRow(conn, path),\n },\n // No comparable setting to report yet.\n async engineStatus() {\n return {};\n },\n raw: {\n async prepare(sql: string) {\n const prepared = await duckdb.prepare(sql);\n return {\n columns: () => {\n const out: Array<{ name: string }> = [];\n for (let i = 0; i < prepared.columnCount; i++) out.push({ name: prepared.columnName(i) });\n return out;\n },\n // Real incremental streaming: DuckDB reads its own result in row-group chunks, so\n // each readUntil only fetches what isn't already in hand.\n iterate: async function* (...params: unknown[]) {\n if (params.length > 0) prepared.bind(params as Parameters<typeof prepared.bind>[0]);\n const reader = await prepared.streamAndRead();\n let yielded = 0;\n do {\n await reader.readUntil(reader.currentRowCount + 2048);\n const rows = reader.getRowObjectsJS() as Array<Record<string, unknown>>;\n for (; yielded < rows.length; yielded++) yield storeRowToJs(rows[yielded]);\n } while (!reader.done);\n },\n };\n },\n },\n async close() {\n // Order matters: the connection must be gone before the instance closes the WAL.\n duckdb.disconnectSync();\n instance.closeSync();\n },\n };\n}\n"],"names":["STORE_DIMS","getColumns","withTransaction","hasVectorRow","pendingRows","storeRowToJs","fieldStats","createLexicalIndex","reconcile","scanCandidates","scanSimilar","writeVectorBatch","CAPABILITIES","Set","createStore","instance","duckdb","conn","cfg","baseDir","lex","name","capabilities","exec","sql","prepare","runBatch","paramRows","transaction","fn","result","markStale","docs","columns","scopeWhere","lexical","query","vectors","pending","writeVectors","rows","candidates","qv","_storeDims","fetch","allowed","similar","path","opts","hasVector","engineStatus","raw","prepared","out","i","columnCount","push","columnName","iterate","params","length","bind","reader","streamAndRead","yielded","readUntil","currentRowCount","getRowObjectsJS","done","close","disconnectSync","closeSync"],"mappings":"AAEA,SAASA,UAAU,QAAQ,uBAAuB;AAClD,SAASC,UAAU,QAAQ,eAAe;AAC1C,SAASC,eAAe,QAAQ,oBAAoB;AAEpD,SAASC,YAAY,EAAEC,WAAW,QAAQ,gBAAgB;AAC1D,SAASC,YAAY,QAAQ,kBAAkB;AAC/C,SAASC,UAAU,QAAQ,kBAAkB;AAC7C,SAASC,kBAAkB,QAAQ,eAAe;AAClD,SAASC,SAAS,QAAQ,iBAAiB;AAC3C,SAASC,cAAc,EAAEC,WAAW,EAAEC,gBAAgB,QAAQ,eAAe;AAE7E,2FAA2F;AAC3F,+FAA+F;AAC/F,gGAAgG;AAChG,4FAA4F;AAC5F,qFAAqF;AACrF,6FAA6F;AAC7F,qCAAqC;AACrC,OAAO,MAAMC,eAAwC,IAAIC,IAAI;IAAC;IAAW;IAAW;CAAU,EAAE;AAEhG,8FAA8F;AAC9F,0FAA0F;AAC1F,2FAA2F;AAC3F,oFAAoF;AACpF,OAAO,SAASC,YAAYC,QAAwB,EAAEC,MAAwB,EAAEC,IAAgB,EAAEC,GAAW,EAAEC,OAAe;IAC5H,MAAMC,MAAMb,mBAAmBU;IAC/B,OAAO;QACLI,MAAM;QACNC,cAAcV;QACd,MAAMW,MAAKC,GAAW;YACpB,MAAMP,KAAKM,IAAI,CAACC;QAClB;QACA,MAAMC,SAAQD,GAAW;YACvB,OAAOP,KAAKQ,OAAO,CAACD;QACtB;QACA,MAAME,UAASF,GAAW,EAAEG,SAAsB;YAChD,MAAMV,KAAKS,QAAQ,CAACF,KAAKG;QAC3B;QACA,MAAMC,aAAeC,EAAoB;YACvC,OAAO3B,gBAAgBe,MAAMY;QAC/B;QACA,MAAMrB;YACJ,MAAMsB,SAAS,MAAMtB,UAAUS,MAAMC,KAAKC;YAC1C,uFAAuF;YACvF,4DAA4D;YAC5DC,IAAIW,SAAS;YACb,OAAOD;QACT;QACAE,MAAM;YACJ,MAAMC;gBACJ,OAAO;uBAAK,MAAMhC,WAAWgB;iBAAO;YACtC;YACAX,YAAY,CAAC2B,SAASC,aAAe5B,WAAWW,MAAMgB,SAASC;QACjE;QACAC,SAAS;YACPC,OAAOhB,IAAIgB,KAAK;QAClB;QACA,sFAAsF;QACtF,0FAA0F;QAC1F,wCAAwC;QACxCC,SAAS;YACPC,SAAS,IAAMlC,YAAYa;YAC3BsB,cAAc,CAACC,OAAS7B,iBAAiBK,QAAQC,MAAMjB,YAAYwC;YACnEC,YAAY,CAACC,IAAIC,YAAYC,OAAOC,UAAYpC,eAAeO,QAAQ0B,IAAI1C,YAAY4C,OAAOC;YAC9FC,SAAS,CAACC,MAAMC,OAAStC,YAAYM,QAAQC,MAAMjB,YAAY+C,MAAMC;YACrEC,WAAW,CAACF,OAAS5C,aAAac,MAAM8B;QAC1C;QACA,uCAAuC;QACvC,MAAMG;YACJ,OAAO,CAAC;QACV;QACAC,KAAK;YACH,MAAM1B,SAAQD,GAAW;gBACvB,MAAM4B,WAAW,MAAMpC,OAAOS,OAAO,CAACD;gBACtC,OAAO;oBACLS,SAAS;wBACP,MAAMoB,MAA+B,EAAE;wBACvC,IAAK,IAAIC,IAAI,GAAGA,IAAIF,SAASG,WAAW,EAAED,IAAKD,IAAIG,IAAI,CAAC;4BAAEnC,MAAM+B,SAASK,UAAU,CAACH;wBAAG;wBACvF,OAAOD;oBACT;oBACA,kFAAkF;oBAClF,0DAA0D;oBAC1DK,SAAS,gBAAiB,GAAGC,MAAiB;wBAC5C,IAAIA,OAAOC,MAAM,GAAG,GAAGR,SAASS,IAAI,CAACF;wBACrC,MAAMG,SAAS,MAAMV,SAASW,aAAa;wBAC3C,IAAIC,UAAU;wBACd,GAAG;4BACD,MAAMF,OAAOG,SAAS,CAACH,OAAOI,eAAe,GAAG;4BAChD,MAAM1B,OAAOsB,OAAOK,eAAe;4BACnC,MAAOH,UAAUxB,KAAKoB,MAAM,EAAEI,UAAW,MAAM3D,aAAamC,IAAI,CAACwB,QAAQ;wBAC3E,QAAS,CAACF,OAAOM,IAAI,CAAE;oBACzB;gBACF;YACF;QACF;QACA,MAAMC;YACJ,iFAAiF;YACjFrD,OAAOsD,cAAc;YACrBvD,SAASwD,SAAS;QACpB;IACF;AACF"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/store/duckdb/store.ts"],"sourcesContent":["import type { DuckDBConnection, DuckDBInstance } from '@duckdb/node-api';\nimport type { Config } from '../../config/index.ts';\nimport { STORE_DIMS } from '../../embed/types.ts';\nimport { getColumns } from '../shared.ts';\nimport { withTransaction } from '../transaction.ts';\nimport type { Capability, Connection, Statement, Store } from '../types.ts';\nimport { hasVectorRow, pendingRows } from '../vectors.ts';\nimport { storeRowToJs } from './connection.ts';\nimport { fieldStats } from './fieldStats.ts';\nimport { createLexicalIndex } from './lexical.ts';\nimport { reconcile } from './reconcile.ts';\nimport { scanCandidates, scanSimilar, writeVectorBatch } from './vectors.ts';\n\n// Portable surface, links/sections/tags/rank, raw sql passthrough, lexical (fts BM25 + contains(), lexical.ts), and vectors (native\n// FLOAT[N] + array_cosine_similarity, vectors.ts) are implemented. 'snippets' is declined (JS excerpt fallback handles it); 'phrases' means quoted-phrase only -- FTS5 operator syntax is rejected, not answered differently.\nexport const CAPABILITIES: ReadonlySet<Capability> = new Set(['lexical', 'phrases', 'vectors']);\n\n// Shares one Connection instance (conn) with open()'s own reconcile call so transaction depth (transaction.ts) is tracked against the same object everywhere.\n// The instance is the native handle that owns the WAL: close() must close it, not just disconnect, or DuckDB never checkpoints and the next open reads a mismatched WAL.\nexport function createStore(instance: DuckDBInstance, duckdb: DuckDBConnection, conn: Connection, cfg: Config, baseDir: string): Store {\n const lex = createLexicalIndex(conn);\n return {\n name: 'duckdb',\n capabilities: CAPABILITIES,\n async exec(sql: string): Promise<void> {\n await conn.exec(sql);\n },\n async prepare(sql: string): Promise<Statement> {\n return conn.prepare(sql);\n },\n async runBatch(sql: string, paramRows: unknown[][]): Promise<void> {\n await conn.runBatch(sql, paramRows);\n },\n async transaction<T>(fn: () => Promise<T>): Promise<T> {\n return withTransaction(conn, fn);\n },\n async reconcile() {\n const result = await reconcile(conn, cfg, baseDir);\n // content may have changed; the fts index is rebuilt lazily, on the next lexical query\n // that needs it, not here (see lexical.ts's FtsIndexState).\n lex.markStale();\n return result;\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: lex.query,\n },\n // The column's fixed DDL width (STORE_DIMS) is what every scan binds against, not the interface's per-call storeDims --\n // see vectors.ts's padded() for why a shorter vector remains correct against a wider column.\n vectors: {\n pending: () => pendingRows(conn),\n writeVectors: (rows) => writeVectorBatch(duckdb, conn, STORE_DIMS, rows),\n candidates: (qv, _storeDims, fetch, allowed) => scanCandidates(duckdb, qv, STORE_DIMS, fetch, allowed),\n similar: (path, opts) => scanSimilar(duckdb, conn, STORE_DIMS, path, opts),\n hasVector: (path) => hasVectorRow(conn, path),\n },\n // No comparable setting to report yet.\n async engineStatus() {\n return {};\n },\n raw: {\n async prepare(sql: string) {\n const prepared = await duckdb.prepare(sql);\n return {\n columns: () => {\n const out: Array<{ name: string }> = [];\n for (let i = 0; i < prepared.columnCount; i++) out.push({ name: prepared.columnName(i) });\n return out;\n },\n // Real incremental streaming: DuckDB reads its own result in row-group chunks, so\n // each readUntil only fetches what isn't already in hand.\n iterate: async function* (...params: unknown[]) {\n if (params.length > 0) prepared.bind(params as Parameters<typeof prepared.bind>[0]);\n const reader = await prepared.streamAndRead();\n let yielded = 0;\n do {\n await reader.readUntil(reader.currentRowCount + 2048);\n const rows = reader.getRowObjectsJS() as Array<Record<string, unknown>>;\n for (; yielded < rows.length; yielded++) yield storeRowToJs(rows[yielded]);\n } while (!reader.done);\n },\n };\n },\n },\n async close() {\n // Order matters: the connection must be gone before the instance closes the WAL.\n duckdb.disconnectSync();\n instance.closeSync();\n },\n };\n}\n"],"names":["STORE_DIMS","getColumns","withTransaction","hasVectorRow","pendingRows","storeRowToJs","fieldStats","createLexicalIndex","reconcile","scanCandidates","scanSimilar","writeVectorBatch","CAPABILITIES","Set","createStore","instance","duckdb","conn","cfg","baseDir","lex","name","capabilities","exec","sql","prepare","runBatch","paramRows","transaction","fn","result","markStale","docs","columns","scopeWhere","lexical","query","vectors","pending","writeVectors","rows","candidates","qv","_storeDims","fetch","allowed","similar","path","opts","hasVector","engineStatus","raw","prepared","out","i","columnCount","push","columnName","iterate","params","length","bind","reader","streamAndRead","yielded","readUntil","currentRowCount","getRowObjectsJS","done","close","disconnectSync","closeSync"],"mappings":"AAEA,SAASA,UAAU,QAAQ,uBAAuB;AAClD,SAASC,UAAU,QAAQ,eAAe;AAC1C,SAASC,eAAe,QAAQ,oBAAoB;AAEpD,SAASC,YAAY,EAAEC,WAAW,QAAQ,gBAAgB;AAC1D,SAASC,YAAY,QAAQ,kBAAkB;AAC/C,SAASC,UAAU,QAAQ,kBAAkB;AAC7C,SAASC,kBAAkB,QAAQ,eAAe;AAClD,SAASC,SAAS,QAAQ,iBAAiB;AAC3C,SAASC,cAAc,EAAEC,WAAW,EAAEC,gBAAgB,QAAQ,eAAe;AAE7E,oIAAoI;AACpI,8NAA8N;AAC9N,OAAO,MAAMC,eAAwC,IAAIC,IAAI;IAAC;IAAW;IAAW;CAAU,EAAE;AAEhG,8JAA8J;AAC9J,yKAAyK;AACzK,OAAO,SAASC,YAAYC,QAAwB,EAAEC,MAAwB,EAAEC,IAAgB,EAAEC,GAAW,EAAEC,OAAe;IAC5H,MAAMC,MAAMb,mBAAmBU;IAC/B,OAAO;QACLI,MAAM;QACNC,cAAcV;QACd,MAAMW,MAAKC,GAAW;YACpB,MAAMP,KAAKM,IAAI,CAACC;QAClB;QACA,MAAMC,SAAQD,GAAW;YACvB,OAAOP,KAAKQ,OAAO,CAACD;QACtB;QACA,MAAME,UAASF,GAAW,EAAEG,SAAsB;YAChD,MAAMV,KAAKS,QAAQ,CAACF,KAAKG;QAC3B;QACA,MAAMC,aAAeC,EAAoB;YACvC,OAAO3B,gBAAgBe,MAAMY;QAC/B;QACA,MAAMrB;YACJ,MAAMsB,SAAS,MAAMtB,UAAUS,MAAMC,KAAKC;YAC1C,uFAAuF;YACvF,4DAA4D;YAC5DC,IAAIW,SAAS;YACb,OAAOD;QACT;QACAE,MAAM;YACJ,MAAMC;gBACJ,OAAO;uBAAK,MAAMhC,WAAWgB;iBAAO;YACtC;YACAX,YAAY,CAAC2B,SAASC,aAAe5B,WAAWW,MAAMgB,SAASC;QACjE;QACAC,SAAS;YACPC,OAAOhB,IAAIgB,KAAK;QAClB;QACA,wHAAwH;QACxH,6FAA6F;QAC7FC,SAAS;YACPC,SAAS,IAAMlC,YAAYa;YAC3BsB,cAAc,CAACC,OAAS7B,iBAAiBK,QAAQC,MAAMjB,YAAYwC;YACnEC,YAAY,CAACC,IAAIC,YAAYC,OAAOC,UAAYpC,eAAeO,QAAQ0B,IAAI1C,YAAY4C,OAAOC;YAC9FC,SAAS,CAACC,MAAMC,OAAStC,YAAYM,QAAQC,MAAMjB,YAAY+C,MAAMC;YACrEC,WAAW,CAACF,OAAS5C,aAAac,MAAM8B;QAC1C;QACA,uCAAuC;QACvC,MAAMG;YACJ,OAAO,CAAC;QACV;QACAC,KAAK;YACH,MAAM1B,SAAQD,GAAW;gBACvB,MAAM4B,WAAW,MAAMpC,OAAOS,OAAO,CAACD;gBACtC,OAAO;oBACLS,SAAS;wBACP,MAAMoB,MAA+B,EAAE;wBACvC,IAAK,IAAIC,IAAI,GAAGA,IAAIF,SAASG,WAAW,EAAED,IAAKD,IAAIG,IAAI,CAAC;4BAAEnC,MAAM+B,SAASK,UAAU,CAACH;wBAAG;wBACvF,OAAOD;oBACT;oBACA,kFAAkF;oBAClF,0DAA0D;oBAC1DK,SAAS,gBAAiB,GAAGC,MAAiB;wBAC5C,IAAIA,OAAOC,MAAM,GAAG,GAAGR,SAASS,IAAI,CAACF;wBACrC,MAAMG,SAAS,MAAMV,SAASW,aAAa;wBAC3C,IAAIC,UAAU;wBACd,GAAG;4BACD,MAAMF,OAAOG,SAAS,CAACH,OAAOI,eAAe,GAAG;4BAChD,MAAM1B,OAAOsB,OAAOK,eAAe;4BACnC,MAAOH,UAAUxB,KAAKoB,MAAM,EAAEI,UAAW,MAAM3D,aAAamC,IAAI,CAACwB,QAAQ;wBAC3E,QAAS,CAACF,OAAOM,IAAI,CAAE;oBACzB;gBACF;YACF;QACF;QACA,MAAMC;YACJ,iFAAiF;YACjFrD,OAAOsD,cAAc;YACrBvD,SAASwD,SAAS;QACpB;IACF;AACF"}
@@ -2,24 +2,15 @@ import { withTransaction } from '../transaction.js';
2
2
  import { asCosine, sampleEvenly } from '../vectors.js';
3
3
  import { rewriteUpdate } from './batch.js';
4
4
  import { duckdbApi } from './native.js';
5
- // This store keeps vectors as native FLOAT[dims] arrays (dims fixed at DDL time by
6
- // duckdb/open.ts's ensureSchema, from embed/types.ts's STORE_DIMS) and scans them with
7
- // array_cosine_similarity, pushing top-k into SQL instead of pulling every row into JS --
8
- // sqlite's int8+scale BLOB scan (src/store/sqlite/vectors.ts) is the other store's own
9
- // representation and stays as it is. Every function here takes `dims` as a parameter rather
10
- // than importing STORE_DIMS directly, so it is testable at any width (store.ts wires the real
11
- // constant at the call site).
12
- //
13
- // A value import of '@duckdb/node-api' must never sit at module top level, or a sqlite-only
14
- // tree would resolve this optional dependency just by importing store/index.ts -- native.ts's
15
- // shared duckdbApi() (not a bare import of its own) is what keeps that lazy and reuses whatever
16
- // open.ts already resolved, including via an on-demand install (see native.ts).
5
+ // This store keeps vectors as native FLOAT[dims] arrays (dims fixed at DDL time by open.ts's ensureSchema, from embed/types.ts's STORE_DIMS) and
6
+ // scans them with array_cosine_similarity, pushing top-k into SQL instead of pulling every row into JS; sqlite's int8+scale BLOB scan is its own representation. Every function here takes `dims` as a parameter rather than importing STORE_DIMS directly, so it stays testable at any width.
7
+ // A value import of '@duckdb/node-api' must never sit at module top level, or a sqlite-only tree would resolve this optional dependency
8
+ // just by importing store/index.ts; native.ts's shared duckdbApi() keeps that lazy and reuses whatever open.ts already resolved.
17
9
  // A bind position whose type is left to auto-inference (safe for plain strings/numbers; only
18
10
  // the vector ARRAY positions below need an explicit type -- see writeVectorBatch's comment).
19
11
  const untyped = undefined;
20
- // The DDL-fixed array width can exceed a vector's actual length (a hypothetical model sliced
21
- // under the column's width); zero-padding leaves cosine scores unchanged since the added
22
- // dimensions contribute nothing to either vector's dot product or norm.
12
+ // The DDL-fixed array width can exceed a vector's actual length (a hypothetical model sliced under the column's width);
13
+ // zero-padding leaves cosine scores unchanged since the added dimensions contribute nothing to either vector's dot product or norm.
23
14
  function padded(values, dims) {
24
15
  const out = new Array(dims).fill(0);
25
16
  for(let d = 0; d < Math.min(values.length, dims); d++)out[d] = values[d];
@@ -30,19 +21,14 @@ function inClause(column, count) {
30
21
  length: count
31
22
  }, ()=>'?').join(', ')})`;
32
23
  }
33
- // Rows arrive int8-quantized with a per-vector scale (the wire format embed/query.ts produces
34
- // for every store, see toStore); dequantizing into a plain float array is this store's own
35
- // representation choice, so both stores are handed the same vectors and diverge only here.
24
+ // Rows arrive int8-quantized with a per-vector scale (the wire format embed/query.ts produces for every store, see toStore);
25
+ // dequantizing into a plain float array is this store's own representation choice, so both stores are handed the same vectors and diverge only here.
36
26
  function dequantize(row, dims) {
37
27
  const q = new Int8Array(row.vector.buffer, row.vector.byteOffset, row.vector.byteLength);
38
28
  return padded(Array.from(q, (v)=>v * row.scale), dims);
39
29
  }
40
- // One runBatch-equivalent call per provider batch: a single multi-row UPDATE (batch.ts's
41
- // rewriteUpdate, the same rewrite Connection.runBatch uses), bound directly against the native
42
- // connection because the vector column needs an explicit ARRAY(DOUBLE, dims) bind type --
43
- // automatic type inference on the raw values misreads an all-integer first component (a real
44
- // vector's exact 0.0 is common) as an INTEGER array and silently truncates the rest of the row
45
- // (verified against a live connection: an untyped bind of [0, 0.35, 0.1, -0.2] stores [0,0,0,0]).
30
+ // One runBatch-equivalent call per provider batch: a single multi-row UPDATE (batch.ts's rewriteUpdate), bound directly against the
31
+ // native connection because the vector column needs an explicit ARRAY(DOUBLE, dims) bind type -- auto inference misreads an all-integer first component as INTEGER and silently truncates the rest (verified live: an untyped bind of [0, 0.35, 0.1, -0.2] stores [0,0,0,0]).
46
32
  export async function writeVectorBatch(duckdb, conn, dims, rows) {
47
33
  if (rows.length === 0) return;
48
34
  const { arrayValue, ARRAY, DOUBLE } = await duckdbApi();
@@ -64,9 +50,8 @@ export async function writeVectorBatch(duckdb, conn, dims, rows) {
64
50
  }
65
51
  });
66
52
  }
67
- // Best chunk per file by cosine, its line range riding along, entirely in SQL: arg_max picks
68
- // the winning chunk's line range from the same row max(score) came from, per path. An empty
69
- // `allowed` set matches sqlite's scan (every row filtered out) without touching the table.
53
+ // Best chunk per file by cosine, its line range riding along, entirely in SQL: arg_max picks the winning chunk's line range from the same row max(score) came from, per path.
54
+ // An empty `allowed` set matches sqlite's scan (every row filtered out) without touching the table.
70
55
  export async function scanCandidates(duckdb, qv, dims, fetch, allowed) {
71
56
  if (allowed && allowed.size === 0) return [];
72
57
  const { arrayValue, ARRAY, DOUBLE } = await duckdbApi();
@@ -106,10 +91,8 @@ export async function scanCandidates(duckdb, qv, dims, fetch, allowed) {
106
91
  stmt.destroySync();
107
92
  }
108
93
  }
109
- // Note-to-note similarity is the max cosine over (target chunk, other chunk) pairs, pushed
110
- // into one native scan: the sampled target vectors become a small VALUES list, cross-joined
111
- // against every stored vector and reduced with max()/GROUP BY, so the O(target x stored) cost
112
- // the plan measures (12.7s at 201 chunks/note in JS) runs vectorized instead of scalar.
94
+ // Note-to-note similarity is the max cosine over (target chunk, other chunk) pairs, pushed into one native scan: the sampled target vectors become
95
+ // a small VALUES list, cross-joined against every stored vector and reduced with max()/GROUP BY, so the O(target x stored) cost (12.7s at 201 chunks/note in JS) runs vectorized instead of scalar.
113
96
  export async function scanSimilar(duckdb, conn, dims, path, opts) {
114
97
  if (opts.allowed && opts.allowed.size === 0) return [];
115
98
  const targetStmt = await conn.prepare('SELECT vector FROM embeddings WHERE "path" = ? AND vector IS NOT NULL ORDER BY chunk');