sensemaking 0.15.4 → 0.16.0

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 (213) hide show
  1. package/README.md +33 -5
  2. package/dist/cjs/cli/download.js +12 -11
  3. package/dist/cjs/cli/download.js.map +1 -1
  4. package/dist/cjs/cli/index.d.cts +1 -1
  5. package/dist/cjs/cli/index.d.ts +1 -1
  6. package/dist/cjs/cli/index.js +1 -1
  7. package/dist/cjs/cli/index.js.map +1 -1
  8. package/dist/cjs/cli/init.js +31 -5
  9. package/dist/cjs/cli/init.js.map +1 -1
  10. package/dist/cjs/cli/status.js +343 -99
  11. package/dist/cjs/cli/status.js.map +1 -1
  12. package/dist/cjs/commands/related.js +17 -13
  13. package/dist/cjs/commands/related.js.map +1 -1
  14. package/dist/cjs/commands/search.js +25 -96
  15. package/dist/cjs/commands/search.js.map +1 -1
  16. package/dist/cjs/commands/signals.d.cts +18 -0
  17. package/dist/cjs/commands/signals.d.ts +18 -0
  18. package/dist/cjs/commands/signals.js +299 -0
  19. package/dist/cjs/commands/signals.js.map +1 -0
  20. package/dist/cjs/commands/status.d.cts +2 -2
  21. package/dist/cjs/commands/status.d.ts +2 -2
  22. package/dist/cjs/commands/status.js +1 -1
  23. package/dist/cjs/commands/status.js.map +1 -1
  24. package/dist/cjs/config/access.d.cts +5 -2
  25. package/dist/cjs/config/access.d.ts +5 -2
  26. package/dist/cjs/config/access.js +30 -12
  27. package/dist/cjs/config/access.js.map +1 -1
  28. package/dist/cjs/config/index.d.cts +4 -1
  29. package/dist/cjs/config/index.d.ts +4 -1
  30. package/dist/cjs/config/index.js +12 -2
  31. package/dist/cjs/config/index.js.map +1 -1
  32. package/dist/cjs/config/load.d.cts +6 -1
  33. package/dist/cjs/config/load.d.ts +6 -1
  34. package/dist/cjs/config/load.js +83 -8
  35. package/dist/cjs/config/load.js.map +1 -1
  36. package/dist/cjs/config/resolve.js +1 -1
  37. package/dist/cjs/config/resolve.js.map +1 -1
  38. package/dist/cjs/config/signals.d.cts +4 -0
  39. package/dist/cjs/config/signals.d.ts +4 -0
  40. package/dist/cjs/config/signals.js +30 -0
  41. package/dist/cjs/config/signals.js.map +1 -0
  42. package/dist/cjs/config/types.d.cts +6 -4
  43. package/dist/cjs/config/types.d.ts +6 -4
  44. package/dist/cjs/config/types.js +1 -1
  45. package/dist/cjs/config/types.js.map +1 -1
  46. package/dist/cjs/config/validate.js +120 -11
  47. package/dist/cjs/config/validate.js.map +1 -1
  48. package/dist/cjs/db/open.d.cts +1 -1
  49. package/dist/cjs/db/open.d.ts +1 -1
  50. package/dist/cjs/db/open.js +6 -1
  51. package/dist/cjs/db/open.js.map +1 -1
  52. package/dist/cjs/db/reconcile.d.cts +1 -0
  53. package/dist/cjs/db/reconcile.d.ts +1 -0
  54. package/dist/cjs/db/reconcile.js +15 -0
  55. package/dist/cjs/db/reconcile.js.map +1 -1
  56. package/dist/cjs/embed/cohere.d.cts +2 -0
  57. package/dist/cjs/embed/cohere.d.ts +2 -0
  58. package/dist/cjs/embed/cohere.js +238 -0
  59. package/dist/cjs/embed/cohere.js.map +1 -0
  60. package/dist/cjs/embed/http.d.cts +2 -0
  61. package/dist/cjs/embed/http.d.ts +2 -0
  62. package/dist/cjs/embed/http.js +258 -0
  63. package/dist/cjs/embed/http.js.map +1 -0
  64. package/dist/cjs/embed/identity.d.cts +13 -0
  65. package/dist/cjs/embed/identity.d.ts +13 -0
  66. package/dist/cjs/embed/identity.js +194 -0
  67. package/dist/cjs/embed/identity.js.map +1 -0
  68. package/dist/cjs/embed/langfit.d.cts +6 -0
  69. package/dist/cjs/embed/langfit.d.ts +6 -0
  70. package/dist/cjs/embed/langfit.js +326 -0
  71. package/dist/cjs/embed/langfit.js.map +1 -0
  72. package/dist/cjs/embed/languages.d.cts +3 -0
  73. package/dist/cjs/embed/languages.d.ts +3 -0
  74. package/dist/cjs/embed/languages.js +98 -0
  75. package/dist/cjs/embed/languages.js.map +1 -0
  76. package/dist/cjs/embed/openai.d.cts +2 -0
  77. package/dist/cjs/embed/openai.d.ts +2 -0
  78. package/dist/cjs/embed/openai.js +268 -0
  79. package/dist/cjs/embed/openai.js.map +1 -0
  80. package/dist/cjs/embed/query.d.cts +21 -0
  81. package/dist/cjs/embed/query.d.ts +21 -0
  82. package/dist/cjs/embed/query.js +610 -0
  83. package/dist/cjs/embed/query.js.map +1 -0
  84. package/dist/cjs/embed/registry.d.cts +3 -0
  85. package/dist/cjs/embed/registry.d.ts +3 -0
  86. package/dist/cjs/embed/registry.js +114 -0
  87. package/dist/cjs/embed/registry.js.map +1 -0
  88. package/dist/cjs/embed/static.d.cts +2 -0
  89. package/dist/cjs/embed/static.d.ts +2 -0
  90. package/dist/cjs/embed/static.js +367 -0
  91. package/dist/cjs/embed/static.js.map +1 -0
  92. package/dist/cjs/embed/store.d.cts +8 -0
  93. package/dist/cjs/embed/store.d.ts +8 -0
  94. package/dist/cjs/embed/store.js +498 -0
  95. package/dist/cjs/embed/store.js.map +1 -0
  96. package/dist/cjs/embed/types.d.cts +8 -0
  97. package/dist/cjs/embed/types.d.ts +8 -0
  98. package/dist/cjs/embed/types.js +7 -0
  99. package/dist/cjs/embed/types.js.map +1 -0
  100. package/dist/cjs/errors.d.cts +1 -1
  101. package/dist/cjs/errors.d.ts +1 -1
  102. package/dist/cjs/errors.js.map +1 -1
  103. package/dist/cjs/features/embed.d.cts +4 -27
  104. package/dist/cjs/features/embed.d.ts +4 -27
  105. package/dist/cjs/features/embed.js +3 -969
  106. package/dist/cjs/features/embed.js.map +1 -1
  107. package/dist/cjs/index.d.cts +1 -1
  108. package/dist/cjs/index.d.ts +1 -1
  109. package/dist/cjs/index.js.map +1 -1
  110. package/dist/cjs/output.d.cts +2 -2
  111. package/dist/cjs/output.d.ts +2 -2
  112. package/dist/cjs/output.js +9 -2
  113. package/dist/cjs/output.js.map +1 -1
  114. package/dist/cjs/scan.js +8 -5
  115. package/dist/cjs/scan.js.map +1 -1
  116. package/dist/cjs/segment.js +67 -14
  117. package/dist/cjs/segment.js.map +1 -1
  118. package/dist/esm/cli/download.js +9 -8
  119. package/dist/esm/cli/download.js.map +1 -1
  120. package/dist/esm/cli/index.d.ts +1 -1
  121. package/dist/esm/cli/index.js +1 -1
  122. package/dist/esm/cli/index.js.map +1 -1
  123. package/dist/esm/cli/init.js +32 -6
  124. package/dist/esm/cli/init.js.map +1 -1
  125. package/dist/esm/cli/status.js +36 -8
  126. package/dist/esm/cli/status.js.map +1 -1
  127. package/dist/esm/commands/related.js +14 -10
  128. package/dist/esm/commands/related.js.map +1 -1
  129. package/dist/esm/commands/search.js +27 -78
  130. package/dist/esm/commands/search.js.map +1 -1
  131. package/dist/esm/commands/signals.d.ts +18 -0
  132. package/dist/esm/commands/signals.js +78 -0
  133. package/dist/esm/commands/signals.js.map +1 -0
  134. package/dist/esm/commands/status.d.ts +2 -2
  135. package/dist/esm/commands/status.js +2 -2
  136. package/dist/esm/commands/status.js.map +1 -1
  137. package/dist/esm/config/access.d.ts +5 -2
  138. package/dist/esm/config/access.js +31 -12
  139. package/dist/esm/config/access.js.map +1 -1
  140. package/dist/esm/config/index.d.ts +4 -1
  141. package/dist/esm/config/index.js +2 -1
  142. package/dist/esm/config/index.js.map +1 -1
  143. package/dist/esm/config/load.d.ts +6 -1
  144. package/dist/esm/config/load.js +67 -9
  145. package/dist/esm/config/load.js.map +1 -1
  146. package/dist/esm/config/resolve.js +2 -2
  147. package/dist/esm/config/resolve.js.map +1 -1
  148. package/dist/esm/config/signals.d.ts +4 -0
  149. package/dist/esm/config/signals.js +14 -0
  150. package/dist/esm/config/signals.js.map +1 -0
  151. package/dist/esm/config/types.d.ts +6 -4
  152. package/dist/esm/config/types.js +1 -1
  153. package/dist/esm/config/types.js.map +1 -1
  154. package/dist/esm/config/validate.js +58 -11
  155. package/dist/esm/config/validate.js.map +1 -1
  156. package/dist/esm/db/open.d.ts +1 -1
  157. package/dist/esm/db/open.js +7 -2
  158. package/dist/esm/db/open.js.map +1 -1
  159. package/dist/esm/db/reconcile.d.ts +1 -0
  160. package/dist/esm/db/reconcile.js +10 -0
  161. package/dist/esm/db/reconcile.js.map +1 -1
  162. package/dist/esm/embed/cohere.d.ts +2 -0
  163. package/dist/esm/embed/cohere.js +46 -0
  164. package/dist/esm/embed/cohere.js.map +1 -0
  165. package/dist/esm/embed/http.d.ts +2 -0
  166. package/dist/esm/embed/http.js +33 -0
  167. package/dist/esm/embed/http.js.map +1 -0
  168. package/dist/esm/embed/identity.d.ts +13 -0
  169. package/dist/esm/embed/identity.js +92 -0
  170. package/dist/esm/embed/identity.js.map +1 -0
  171. package/dist/esm/embed/langfit.d.ts +6 -0
  172. package/dist/esm/embed/langfit.js +44 -0
  173. package/dist/esm/embed/langfit.js.map +1 -0
  174. package/dist/esm/embed/languages.d.ts +3 -0
  175. package/dist/esm/embed/languages.js +80 -0
  176. package/dist/esm/embed/languages.js.map +1 -0
  177. package/dist/esm/embed/openai.d.ts +2 -0
  178. package/dist/esm/embed/openai.js +49 -0
  179. package/dist/esm/embed/openai.js.map +1 -0
  180. package/dist/esm/embed/query.d.ts +21 -0
  181. package/dist/esm/embed/query.js +181 -0
  182. package/dist/esm/embed/query.js.map +1 -0
  183. package/dist/esm/embed/registry.d.ts +3 -0
  184. package/dist/esm/embed/registry.js +45 -0
  185. package/dist/esm/embed/registry.js.map +1 -0
  186. package/dist/esm/embed/static.d.ts +2 -0
  187. package/dist/esm/embed/static.js +72 -0
  188. package/dist/esm/embed/static.js.map +1 -0
  189. package/dist/esm/embed/store.d.ts +8 -0
  190. package/dist/esm/embed/store.js +123 -0
  191. package/dist/esm/embed/store.js.map +1 -0
  192. package/dist/esm/embed/types.d.ts +8 -0
  193. package/dist/esm/embed/types.js +3 -0
  194. package/dist/esm/embed/types.js.map +1 -0
  195. package/dist/esm/errors.d.ts +1 -1
  196. package/dist/esm/errors.js.map +1 -1
  197. package/dist/esm/features/embed.d.ts +4 -27
  198. package/dist/esm/features/embed.js +0 -346
  199. package/dist/esm/features/embed.js.map +1 -1
  200. package/dist/esm/index.d.ts +1 -1
  201. package/dist/esm/index.js.map +1 -1
  202. package/dist/esm/output.d.ts +2 -2
  203. package/dist/esm/output.js +6 -2
  204. package/dist/esm/output.js.map +1 -1
  205. package/dist/esm/scan.js +7 -4
  206. package/dist/esm/scan.js.map +1 -1
  207. package/dist/esm/segment.js +49 -19
  208. package/dist/esm/segment.js.map +1 -1
  209. package/package.json +9 -1
  210. package/schema.json +14 -7
  211. package/skills/sense/SKILL.md +21 -5
  212. package/skills/sense-setup/EXAMPLES.md +6 -6
  213. package/skills/sense-setup/SKILL.md +6 -4
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/config/validate.ts"],"sourcesContent":["import { SenseError } from '../errors.ts';\nimport type { Config } from './types.ts';\n\n// Shape check for hand-edited files, so a typo names itself instead of surfacing as a\n// TypeError. Unknown top-level keys warn (forward compat); unknown keys inside a block error.\nconst KNOWN_KEYS = new Set(['$schema', 'version', 'presets', 'features', 'embed', 'content', 'queries']);\nconst KNOWN_PRESET_KEYS = new Set(['include', 'exclude', 'k', 'semantic', 'where']);\nconst KNOWN_FEATURE_KEYS = new Set(['links', 'sections', 'tags', 'rank']);\nconst KNOWN_EMBED_KEYS = new Set(['model', 'type', 'url', 'key']);\nconst KNOWN_CONTENT_KEYS = new Set(['tokenize']);\nconst SAVED_SEARCH_KEYS = new Set(['search', 'preset', 'include', 'exclude', 'where', 'k']);\n\nexport function unknownConfigKeys(cfg: Record<string, unknown>): string[] {\n return Object.keys(cfg).filter((k) => !KNOWN_KEYS.has(k));\n}\n\nfunction isNonEmptyStringArray(value: unknown): value is string[] {\n return Array.isArray(value) && value.length > 0 && value.every((g) => typeof g === 'string' && g.length > 0);\n}\n\n// Pre-v3 shape check, just enough for migrateConfig to run safely (it reads scan.include\n// directly). The full v3 shape is checked by validateConfig once migration has produced it.\nexport function validateLegacyScan(parsed: unknown, configPath: string): void {\n const cfg = typeof parsed === 'object' && parsed !== null && !Array.isArray(parsed) ? (parsed as Record<string, unknown>) : null;\n const scan = cfg?.scan as { include?: unknown } | undefined;\n if (!cfg || !scan || !isNonEmptyStringArray(scan.include)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: scan.include must be a non-empty array of glob strings`);\n }\n}\n\n// Only the shape is checked here. Whether the linked SQLite accepts the tokenizer is settled\n// by probing it in db.ts, so this never has to carry a table of which version added what.\nfunction validateContentBlock(value: unknown, configPath: string): void {\n if (typeof value !== 'object' || value === null || Array.isArray(value)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: content must be an object of { tokenize?: string }`);\n }\n const block = value as Record<string, unknown>;\n const unknown = Object.keys(block).filter((k) => !KNOWN_CONTENT_KEYS.has(k));\n if (unknown.length > 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: content has unknown key(s) ${unknown.join(', ')}; content takes tokenize`);\n }\n if (block.tokenize !== undefined && (typeof block.tokenize !== 'string' || block.tokenize.trim() === '')) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: content.tokenize must be a non-empty string, e.g. \"trigram\" or \"unicode61 tokenchars '-_'\"`);\n }\n}\n\nfunction validateFeaturesBlock(value: unknown, configPath: string): void {\n if (typeof value !== 'object' || value === null || Array.isArray(value)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: features must be an object of name -> boolean`);\n }\n const block = value as Record<string, unknown>;\n const unknown = Object.keys(block).filter((k) => !KNOWN_FEATURE_KEYS.has(k));\n if (unknown.length > 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: features has unknown key(s) ${unknown.join(', ')}; embed is not a features key -- see the top-level \"embed\" block`);\n }\n for (const [name, v] of Object.entries(block)) {\n if (typeof v !== 'boolean') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: features.${name} must be a boolean`);\n }\n }\n}\n\nfunction validateEmbedBlock(value: unknown, configPath: string): void {\n if (typeof value !== 'object' || value === null || Array.isArray(value)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed must be an object of { model?, type?: \"static\"|\"api\", url?, key? }`);\n }\n const embed = value as Record<string, unknown>;\n const unknown = Object.keys(embed).filter((k) => !KNOWN_EMBED_KEYS.has(k));\n if (unknown.length > 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed has unknown key(s) ${unknown.join(', ')}`);\n }\n if (embed.model !== undefined && typeof embed.model !== 'string') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed.model must be a string`);\n }\n // The block exists to name a model; provider settings without one would read as \"vectors\n // configured\" while embedding stays off, which is the ambiguity v4 removed.\n if (embed.model === undefined || embed.model === '') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed.model is required when the \"embed\" block is present (it is what turns vectors on); remove the block to index without vectors`);\n }\n if (embed.type !== undefined && embed.type !== 'static' && embed.type !== 'api') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed.type must be \"static\" or \"api\"`);\n }\n if (embed.url !== undefined && typeof embed.url !== 'string') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed.url must be a string`);\n }\n if (embed.key !== undefined && typeof embed.key !== 'string') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed.key must be a string`);\n }\n}\n\nfunction validatePreset(name: string, value: unknown, configPath: string): void {\n if (typeof value !== 'object' || value === null || Array.isArray(value)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name} must be an object`);\n }\n const preset = value as Record<string, unknown>;\n const unknown = Object.keys(preset).filter((k) => !KNOWN_PRESET_KEYS.has(k));\n if (unknown.length > 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name} has unknown key(s) ${unknown.join(', ')}; a preset takes include, exclude, k, semantic, where`);\n }\n if (!isNonEmptyStringArray(preset.include)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.include must be a non-empty array of glob strings`);\n }\n if (preset.exclude !== undefined && !isNonEmptyStringArray(preset.exclude)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.exclude must be a non-empty array of glob strings`);\n }\n if (preset.semantic !== undefined && typeof preset.semantic !== 'boolean') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.semantic must be a boolean`);\n }\n if (preset.k !== undefined && (typeof preset.k !== 'number' || !Number.isInteger(preset.k) || preset.k <= 0)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.k must be a positive integer`);\n }\n if (preset.where !== undefined && typeof preset.where !== 'string') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.where must be a SQL condition string`);\n }\n}\n\n// A queries.<name> entry: { sql } or a saved search { search, preset?, include?, exclude?, where?, k? }.\nfunction validateSavedQuery(name: string, value: unknown, configPath: string): void {\n // A bare string used to mean SQL. It now fails rather than being inferred: an entry says\n // which of the two verbs it runs, the same choice the CLI makes explicit.\n if (typeof value === 'string') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name} must say which verb it runs: { \"sql\": ${JSON.stringify(value.length > 40 ? `${value.slice(0, 37)}...` : value)} } to run it as SQL, or { \"search\": \"...\" } for a ranked search`);\n }\n if (typeof value !== 'object' || value === null || Array.isArray(value)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name} must be { sql } or { search, preset?, include?, exclude?, where?, k? }`);\n }\n const entry = value as Record<string, unknown>;\n if ('sql' in entry) {\n const unknown = Object.keys(entry).filter((k) => k !== 'sql');\n if (unknown.length > 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name} has unknown key(s) ${unknown.join(', ')}; a { sql } query takes only sql`);\n }\n if (typeof entry.sql !== 'string' || entry.sql.trim() === '') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.sql must be a non-empty string`);\n }\n return;\n }\n if ('search' in entry) {\n const unknown = Object.keys(entry).filter((k) => !SAVED_SEARCH_KEYS.has(k));\n if (unknown.length > 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name} has unknown key(s) ${unknown.join(', ')}; a saved search takes search, preset, include, exclude, where, k`);\n }\n // A saved entry saves a question; a scope without a question is just flags.\n if (typeof entry.search !== 'string' || entry.search.trim() === '') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.search must be non-empty text`);\n }\n if (entry.preset !== undefined && (typeof entry.preset !== 'string' || entry.preset.length === 0)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.preset must be a preset name`);\n }\n if (entry.include !== undefined && !isNonEmptyStringArray(entry.include)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.include must be a non-empty array of glob strings`);\n }\n if (entry.exclude !== undefined && !isNonEmptyStringArray(entry.exclude)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.exclude must be a non-empty array of glob strings`);\n }\n if (entry.where !== undefined && typeof entry.where !== 'string') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.where must be a SQL condition string`);\n }\n if (entry.k !== undefined && (typeof entry.k !== 'number' || !Number.isInteger(entry.k) || entry.k <= 0)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.k must be a positive integer`);\n }\n return;\n }\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name} must be { sql } or { search, preset?, include?, exclude?, where?, k? }`);\n}\n\nexport function validateConfig(parsed: unknown, configPath: string): Config {\n if (typeof parsed !== 'object' || parsed === null || Array.isArray(parsed)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: config must be a JSON object`);\n }\n const cfg = parsed as Record<string, unknown>;\n\n // `checks` is rejected by name, not warned: silence would hide that assertions are gone.\n if (cfg.checks !== undefined) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: checks was removed in v3 -- sense check no longer asserts on saved queries; a returned row set is the reader's judgment`);\n }\n\n const presets = cfg.presets as Record<string, unknown> | undefined;\n if (!presets || typeof presets !== 'object' || Array.isArray(presets) || Object.keys(presets).length === 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets must be a non-empty object of preset name -> { include, exclude?, k?, semantic?, where? }`);\n }\n for (const [name, value] of Object.entries(presets)) {\n validatePreset(name, value, configPath);\n }\n if (presets.default === undefined) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets must include a \"default\" preset`);\n }\n\n if (cfg.queries === undefined) cfg.queries = {};\n if (typeof cfg.queries !== 'object' || cfg.queries === null || Array.isArray(cfg.queries)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries must be an object of name -> { sql } or { search }`);\n }\n const queries = cfg.queries as Record<string, unknown>;\n for (const [name, value] of Object.entries(queries)) {\n validateSavedQuery(name, value, configPath);\n }\n\n if (cfg.features !== undefined) {\n validateFeaturesBlock(cfg.features, configPath);\n }\n if (cfg.embed !== undefined) {\n validateEmbedBlock(cfg.embed, configPath);\n }\n if (cfg.content !== undefined) {\n validateContentBlock(cfg.content, configPath);\n }\n\n return cfg as unknown as Config;\n}\n"],"names":["SenseError","KNOWN_KEYS","Set","KNOWN_PRESET_KEYS","KNOWN_FEATURE_KEYS","KNOWN_EMBED_KEYS","KNOWN_CONTENT_KEYS","SAVED_SEARCH_KEYS","unknownConfigKeys","cfg","Object","keys","filter","k","has","isNonEmptyStringArray","value","Array","isArray","length","every","g","validateLegacyScan","parsed","configPath","scan","include","validateContentBlock","block","unknown","join","tokenize","undefined","trim","validateFeaturesBlock","name","v","entries","validateEmbedBlock","embed","model","type","url","key","validatePreset","preset","exclude","semantic","Number","isInteger","where","validateSavedQuery","JSON","stringify","slice","entry","sql","search","validateConfig","checks","presets","default","queries","features","content"],"mappings":"AAAA,SAASA,UAAU,QAAQ,eAAe;AAG1C,sFAAsF;AACtF,8FAA8F;AAC9F,MAAMC,aAAa,IAAIC,IAAI;IAAC;IAAW;IAAW;IAAW;IAAY;IAAS;IAAW;CAAU;AACvG,MAAMC,oBAAoB,IAAID,IAAI;IAAC;IAAW;IAAW;IAAK;IAAY;CAAQ;AAClF,MAAME,qBAAqB,IAAIF,IAAI;IAAC;IAAS;IAAY;IAAQ;CAAO;AACxE,MAAMG,mBAAmB,IAAIH,IAAI;IAAC;IAAS;IAAQ;IAAO;CAAM;AAChE,MAAMI,qBAAqB,IAAIJ,IAAI;IAAC;CAAW;AAC/C,MAAMK,oBAAoB,IAAIL,IAAI;IAAC;IAAU;IAAU;IAAW;IAAW;IAAS;CAAI;AAE1F,OAAO,SAASM,kBAAkBC,GAA4B;IAC5D,OAAOC,OAAOC,IAAI,CAACF,KAAKG,MAAM,CAAC,CAACC,IAAM,CAACZ,WAAWa,GAAG,CAACD;AACxD;AAEA,SAASE,sBAAsBC,KAAc;IAC3C,OAAOC,MAAMC,OAAO,CAACF,UAAUA,MAAMG,MAAM,GAAG,KAAKH,MAAMI,KAAK,CAAC,CAACC,IAAM,OAAOA,MAAM,YAAYA,EAAEF,MAAM,GAAG;AAC5G;AAEA,yFAAyF;AACzF,4FAA4F;AAC5F,OAAO,SAASG,mBAAmBC,MAAe,EAAEC,UAAkB;IACpE,MAAMf,MAAM,OAAOc,WAAW,YAAYA,WAAW,QAAQ,CAACN,MAAMC,OAAO,CAACK,UAAWA,SAAqC;IAC5H,MAAME,OAAOhB,gBAAAA,0BAAAA,IAAKgB,IAAI;IACtB,IAAI,CAAChB,OAAO,CAACgB,QAAQ,CAACV,sBAAsBU,KAAKC,OAAO,GAAG;QACzD,MAAM,IAAI1B,WAAW,kBAAkB,GAAGwB,WAAW,wDAAwD,CAAC;IAChH;AACF;AAEA,6FAA6F;AAC7F,0FAA0F;AAC1F,SAASG,qBAAqBX,KAAc,EAAEQ,UAAkB;IAC9D,IAAI,OAAOR,UAAU,YAAYA,UAAU,QAAQC,MAAMC,OAAO,CAACF,QAAQ;QACvE,MAAM,IAAIhB,WAAW,kBAAkB,GAAGwB,WAAW,oDAAoD,CAAC;IAC5G;IACA,MAAMI,QAAQZ;IACd,MAAMa,UAAUnB,OAAOC,IAAI,CAACiB,OAAOhB,MAAM,CAAC,CAACC,IAAM,CAACP,mBAAmBQ,GAAG,CAACD;IACzE,IAAIgB,QAAQV,MAAM,GAAG,GAAG;QACtB,MAAM,IAAInB,WAAW,kBAAkB,GAAGwB,WAAW,6BAA6B,EAAEK,QAAQC,IAAI,CAAC,MAAM,wBAAwB,CAAC;IAClI;IACA,IAAIF,MAAMG,QAAQ,KAAKC,aAAc,CAAA,OAAOJ,MAAMG,QAAQ,KAAK,YAAYH,MAAMG,QAAQ,CAACE,IAAI,OAAO,EAAC,GAAI;QACxG,MAAM,IAAIjC,WAAW,kBAAkB,GAAGwB,WAAW,4FAA4F,CAAC;IACpJ;AACF;AAEA,SAASU,sBAAsBlB,KAAc,EAAEQ,UAAkB;IAC/D,IAAI,OAAOR,UAAU,YAAYA,UAAU,QAAQC,MAAMC,OAAO,CAACF,QAAQ;QACvE,MAAM,IAAIhB,WAAW,kBAAkB,GAAGwB,WAAW,+CAA+C,CAAC;IACvG;IACA,MAAMI,QAAQZ;IACd,MAAMa,UAAUnB,OAAOC,IAAI,CAACiB,OAAOhB,MAAM,CAAC,CAACC,IAAM,CAACT,mBAAmBU,GAAG,CAACD;IACzE,IAAIgB,QAAQV,MAAM,GAAG,GAAG;QACtB,MAAM,IAAInB,WAAW,kBAAkB,GAAGwB,WAAW,8BAA8B,EAAEK,QAAQC,IAAI,CAAC,MAAM,gEAAgE,CAAC;IAC3K;IACA,KAAK,MAAM,CAACK,MAAMC,EAAE,IAAI1B,OAAO2B,OAAO,CAACT,OAAQ;QAC7C,IAAI,OAAOQ,MAAM,WAAW;YAC1B,MAAM,IAAIpC,WAAW,kBAAkB,GAAGwB,WAAW,WAAW,EAAEW,KAAK,kBAAkB,CAAC;QAC5F;IACF;AACF;AAEA,SAASG,mBAAmBtB,KAAc,EAAEQ,UAAkB;IAC5D,IAAI,OAAOR,UAAU,YAAYA,UAAU,QAAQC,MAAMC,OAAO,CAACF,QAAQ;QACvE,MAAM,IAAIhB,WAAW,kBAAkB,GAAGwB,WAAW,0EAA0E,CAAC;IAClI;IACA,MAAMe,QAAQvB;IACd,MAAMa,UAAUnB,OAAOC,IAAI,CAAC4B,OAAO3B,MAAM,CAAC,CAACC,IAAM,CAACR,iBAAiBS,GAAG,CAACD;IACvE,IAAIgB,QAAQV,MAAM,GAAG,GAAG;QACtB,MAAM,IAAInB,WAAW,kBAAkB,GAAGwB,WAAW,2BAA2B,EAAEK,QAAQC,IAAI,CAAC,OAAO;IACxG;IACA,IAAIS,MAAMC,KAAK,KAAKR,aAAa,OAAOO,MAAMC,KAAK,KAAK,UAAU;QAChE,MAAM,IAAIxC,WAAW,kBAAkB,GAAGwB,WAAW,8BAA8B,CAAC;IACtF;IACA,yFAAyF;IACzF,4EAA4E;IAC5E,IAAIe,MAAMC,KAAK,KAAKR,aAAaO,MAAMC,KAAK,KAAK,IAAI;QACnD,MAAM,IAAIxC,WAAW,kBAAkB,GAAGwB,WAAW,oIAAoI,CAAC;IAC5L;IACA,IAAIe,MAAME,IAAI,KAAKT,aAAaO,MAAME,IAAI,KAAK,YAAYF,MAAME,IAAI,KAAK,OAAO;QAC/E,MAAM,IAAIzC,WAAW,kBAAkB,GAAGwB,WAAW,sCAAsC,CAAC;IAC9F;IACA,IAAIe,MAAMG,GAAG,KAAKV,aAAa,OAAOO,MAAMG,GAAG,KAAK,UAAU;QAC5D,MAAM,IAAI1C,WAAW,kBAAkB,GAAGwB,WAAW,4BAA4B,CAAC;IACpF;IACA,IAAIe,MAAMI,GAAG,KAAKX,aAAa,OAAOO,MAAMI,GAAG,KAAK,UAAU;QAC5D,MAAM,IAAI3C,WAAW,kBAAkB,GAAGwB,WAAW,4BAA4B,CAAC;IACpF;AACF;AAEA,SAASoB,eAAeT,IAAY,EAAEnB,KAAc,EAAEQ,UAAkB;IACtE,IAAI,OAAOR,UAAU,YAAYA,UAAU,QAAQC,MAAMC,OAAO,CAACF,QAAQ;QACvE,MAAM,IAAIhB,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,kBAAkB,CAAC;IAC3F;IACA,MAAMU,SAAS7B;IACf,MAAMa,UAAUnB,OAAOC,IAAI,CAACkC,QAAQjC,MAAM,CAAC,CAACC,IAAM,CAACV,kBAAkBW,GAAG,CAACD;IACzE,IAAIgB,QAAQV,MAAM,GAAG,GAAG;QACtB,MAAM,IAAInB,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,oBAAoB,EAAEN,QAAQC,IAAI,CAAC,MAAM,qDAAqD,CAAC;IACvK;IACA,IAAI,CAACf,sBAAsB8B,OAAOnB,OAAO,GAAG;QAC1C,MAAM,IAAI1B,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,kDAAkD,CAAC;IAC3H;IACA,IAAIU,OAAOC,OAAO,KAAKd,aAAa,CAACjB,sBAAsB8B,OAAOC,OAAO,GAAG;QAC1E,MAAM,IAAI9C,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,kDAAkD,CAAC;IAC3H;IACA,IAAIU,OAAOE,QAAQ,KAAKf,aAAa,OAAOa,OAAOE,QAAQ,KAAK,WAAW;QACzE,MAAM,IAAI/C,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,2BAA2B,CAAC;IACpG;IACA,IAAIU,OAAOhC,CAAC,KAAKmB,aAAc,CAAA,OAAOa,OAAOhC,CAAC,KAAK,YAAY,CAACmC,OAAOC,SAAS,CAACJ,OAAOhC,CAAC,KAAKgC,OAAOhC,CAAC,IAAI,CAAA,GAAI;QAC5G,MAAM,IAAIb,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,6BAA6B,CAAC;IACtG;IACA,IAAIU,OAAOK,KAAK,KAAKlB,aAAa,OAAOa,OAAOK,KAAK,KAAK,UAAU;QAClE,MAAM,IAAIlD,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,qCAAqC,CAAC;IAC9G;AACF;AAEA,yGAAyG;AACzG,SAASgB,mBAAmBhB,IAAY,EAAEnB,KAAc,EAAEQ,UAAkB;IAC1E,yFAAyF;IACzF,0EAA0E;IAC1E,IAAI,OAAOR,UAAU,UAAU;QAC7B,MAAM,IAAIhB,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,uCAAuC,EAAEiB,KAAKC,SAAS,CAACrC,MAAMG,MAAM,GAAG,KAAK,GAAGH,MAAMsC,KAAK,CAAC,GAAG,IAAI,GAAG,CAAC,GAAGtC,OAAO,+DAA+D,CAAC;IACxP;IACA,IAAI,OAAOA,UAAU,YAAYA,UAAU,QAAQC,MAAMC,OAAO,CAACF,QAAQ;QACvE,MAAM,IAAIhB,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,uEAAuE,CAAC;IAChJ;IACA,MAAMoB,QAAQvC;IACd,IAAI,SAASuC,OAAO;QAClB,MAAM1B,UAAUnB,OAAOC,IAAI,CAAC4C,OAAO3C,MAAM,CAAC,CAACC,IAAMA,MAAM;QACvD,IAAIgB,QAAQV,MAAM,GAAG,GAAG;YACtB,MAAM,IAAInB,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,oBAAoB,EAAEN,QAAQC,IAAI,CAAC,MAAM,gCAAgC,CAAC;QAClJ;QACA,IAAI,OAAOyB,MAAMC,GAAG,KAAK,YAAYD,MAAMC,GAAG,CAACvB,IAAI,OAAO,IAAI;YAC5D,MAAM,IAAIjC,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,+BAA+B,CAAC;QACxG;QACA;IACF;IACA,IAAI,YAAYoB,OAAO;QACrB,MAAM1B,UAAUnB,OAAOC,IAAI,CAAC4C,OAAO3C,MAAM,CAAC,CAACC,IAAM,CAACN,kBAAkBO,GAAG,CAACD;QACxE,IAAIgB,QAAQV,MAAM,GAAG,GAAG;YACtB,MAAM,IAAInB,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,oBAAoB,EAAEN,QAAQC,IAAI,CAAC,MAAM,iEAAiE,CAAC;QACnL;QACA,4EAA4E;QAC5E,IAAI,OAAOyB,MAAME,MAAM,KAAK,YAAYF,MAAME,MAAM,CAACxB,IAAI,OAAO,IAAI;YAClE,MAAM,IAAIjC,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,8BAA8B,CAAC;QACvG;QACA,IAAIoB,MAAMV,MAAM,KAAKb,aAAc,CAAA,OAAOuB,MAAMV,MAAM,KAAK,YAAYU,MAAMV,MAAM,CAAC1B,MAAM,KAAK,CAAA,GAAI;YACjG,MAAM,IAAInB,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,6BAA6B,CAAC;QACtG;QACA,IAAIoB,MAAM7B,OAAO,KAAKM,aAAa,CAACjB,sBAAsBwC,MAAM7B,OAAO,GAAG;YACxE,MAAM,IAAI1B,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,kDAAkD,CAAC;QAC3H;QACA,IAAIoB,MAAMT,OAAO,KAAKd,aAAa,CAACjB,sBAAsBwC,MAAMT,OAAO,GAAG;YACxE,MAAM,IAAI9C,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,kDAAkD,CAAC;QAC3H;QACA,IAAIoB,MAAML,KAAK,KAAKlB,aAAa,OAAOuB,MAAML,KAAK,KAAK,UAAU;YAChE,MAAM,IAAIlD,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,qCAAqC,CAAC;QAC9G;QACA,IAAIoB,MAAM1C,CAAC,KAAKmB,aAAc,CAAA,OAAOuB,MAAM1C,CAAC,KAAK,YAAY,CAACmC,OAAOC,SAAS,CAACM,MAAM1C,CAAC,KAAK0C,MAAM1C,CAAC,IAAI,CAAA,GAAI;YACxG,MAAM,IAAIb,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,6BAA6B,CAAC;QACtG;QACA;IACF;IACA,MAAM,IAAInC,WAAW,kBAAkB,GAAGwB,WAAW,UAAU,EAAEW,KAAK,uEAAuE,CAAC;AAChJ;AAEA,OAAO,SAASuB,eAAenC,MAAe,EAAEC,UAAkB;IAChE,IAAI,OAAOD,WAAW,YAAYA,WAAW,QAAQN,MAAMC,OAAO,CAACK,SAAS;QAC1E,MAAM,IAAIvB,WAAW,kBAAkB,GAAGwB,WAAW,8BAA8B,CAAC;IACtF;IACA,MAAMf,MAAMc;IAEZ,yFAAyF;IACzF,IAAId,IAAIkD,MAAM,KAAK3B,WAAW;QAC5B,MAAM,IAAIhC,WAAW,kBAAkB,GAAGwB,WAAW,yHAAyH,CAAC;IACjL;IAEA,MAAMoC,UAAUnD,IAAImD,OAAO;IAC3B,IAAI,CAACA,WAAW,OAAOA,YAAY,YAAY3C,MAAMC,OAAO,CAAC0C,YAAYlD,OAAOC,IAAI,CAACiD,SAASzC,MAAM,KAAK,GAAG;QAC1G,MAAM,IAAInB,WAAW,kBAAkB,GAAGwB,WAAW,mGAAmG,CAAC;IAC3J;IACA,KAAK,MAAM,CAACW,MAAMnB,MAAM,IAAIN,OAAO2B,OAAO,CAACuB,SAAU;QACnDhB,eAAeT,MAAMnB,OAAOQ;IAC9B;IACA,IAAIoC,QAAQC,OAAO,KAAK7B,WAAW;QACjC,MAAM,IAAIhC,WAAW,kBAAkB,GAAGwB,WAAW,yCAAyC,CAAC;IACjG;IAEA,IAAIf,IAAIqD,OAAO,KAAK9B,WAAWvB,IAAIqD,OAAO,GAAG,CAAC;IAC9C,IAAI,OAAOrD,IAAIqD,OAAO,KAAK,YAAYrD,IAAIqD,OAAO,KAAK,QAAQ7C,MAAMC,OAAO,CAACT,IAAIqD,OAAO,GAAG;QACzF,MAAM,IAAI9D,WAAW,kBAAkB,GAAGwB,WAAW,4DAA4D,CAAC;IACpH;IACA,MAAMsC,UAAUrD,IAAIqD,OAAO;IAC3B,KAAK,MAAM,CAAC3B,MAAMnB,MAAM,IAAIN,OAAO2B,OAAO,CAACyB,SAAU;QACnDX,mBAAmBhB,MAAMnB,OAAOQ;IAClC;IAEA,IAAIf,IAAIsD,QAAQ,KAAK/B,WAAW;QAC9BE,sBAAsBzB,IAAIsD,QAAQ,EAAEvC;IACtC;IACA,IAAIf,IAAI8B,KAAK,KAAKP,WAAW;QAC3BM,mBAAmB7B,IAAI8B,KAAK,EAAEf;IAChC;IACA,IAAIf,IAAIuD,OAAO,KAAKhC,WAAW;QAC7BL,qBAAqBlB,IAAIuD,OAAO,EAAExC;IACpC;IAEA,OAAOf;AACT"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/config/validate.ts"],"sourcesContent":["import { SenseError } from '../errors.ts';\nimport { SIGNAL_NAMES, SIGNAL_PREREQUISITES, type SignalName } from './signals.ts';\nimport type { Config } from './types.ts';\n\n// Shape check for hand-edited files, so a typo names itself instead of surfacing as a\n// TypeError. Unknown top-level keys warn (forward compat); unknown keys inside a block error.\nconst KNOWN_KEYS = new Set(['$schema', 'version', 'presets', 'features', 'embed', 'content', 'queries']);\nconst KNOWN_PRESET_KEYS = new Set(['include', 'exclude', 'k', 'signals', 'where']);\nconst KNOWN_FEATURE_KEYS = new Set(['links', 'sections', 'tags', 'rank']);\nconst KNOWN_EMBED_KEYS = new Set(['model', 'provider', 'url', 'key', 'languages']);\nconst KNOWN_CONTENT_KEYS = new Set(['tokenize']);\nconst SAVED_SEARCH_KEYS = new Set(['search', 'preset', 'include', 'exclude', 'where', 'k']);\n\nexport function unknownConfigKeys(cfg: Record<string, unknown>): string[] {\n return Object.keys(cfg).filter((k) => !KNOWN_KEYS.has(k));\n}\n\nfunction isNonEmptyStringArray(value: unknown): value is string[] {\n return Array.isArray(value) && value.length > 0 && value.every((g) => typeof g === 'string' && g.length > 0);\n}\n\n// Pre-v3 shape check, just enough for migrateConfig to run safely (it reads scan.include\n// directly). The full v3 shape is checked by validateConfig once migration has produced it.\nexport function validateLegacyScan(parsed: unknown, configPath: string): void {\n const cfg = typeof parsed === 'object' && parsed !== null && !Array.isArray(parsed) ? (parsed as Record<string, unknown>) : null;\n const scan = cfg?.scan as { include?: unknown } | undefined;\n if (!cfg || !scan || !isNonEmptyStringArray(scan.include)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: scan.include must be a non-empty array of glob strings`);\n }\n}\n\n// Only the shape is checked here. Whether the linked SQLite accepts the tokenizer is settled\n// by probing it in db.ts, so this never has to carry a table of which version added what.\nfunction validateContentBlock(value: unknown, configPath: string): void {\n if (typeof value !== 'object' || value === null || Array.isArray(value)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: content must be an object of { tokenize?: string }`);\n }\n const block = value as Record<string, unknown>;\n const unknown = Object.keys(block).filter((k) => !KNOWN_CONTENT_KEYS.has(k));\n if (unknown.length > 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: content has unknown key(s) ${unknown.join(', ')}; content takes tokenize`);\n }\n if (block.tokenize !== undefined && (typeof block.tokenize !== 'string' || block.tokenize.trim() === '')) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: content.tokenize must be a non-empty string, e.g. \"trigram\" or \"unicode61 tokenchars '-_'\"`);\n }\n}\n\nfunction validateFeaturesBlock(value: unknown, configPath: string): void {\n if (typeof value !== 'object' || value === null || Array.isArray(value)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: features must be an object of name -> boolean`);\n }\n const block = value as Record<string, unknown>;\n const unknown = Object.keys(block).filter((k) => !KNOWN_FEATURE_KEYS.has(k));\n if (unknown.length > 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: features has unknown key(s) ${unknown.join(', ')}; embed is not a features key -- see the top-level \"embed\" block`);\n }\n for (const [name, v] of Object.entries(block)) {\n if (typeof v !== 'boolean') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: features.${name} must be a boolean`);\n }\n }\n}\n\nfunction validateEmbedBlock(value: unknown, configPath: string): void {\n if (typeof value !== 'object' || value === null || Array.isArray(value)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed must be an object of { model?, provider?: \"static\"|\"openai\"|\"cohere\", url?, key? }`);\n }\n const embed = value as Record<string, unknown>;\n const unknown = Object.keys(embed).filter((k) => !KNOWN_EMBED_KEYS.has(k));\n if (unknown.length > 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed has unknown key(s) ${unknown.join(', ')}`);\n }\n if (embed.model !== undefined && typeof embed.model !== 'string') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed.model must be a string`);\n }\n // The block exists to name a model; provider settings without one would read as \"vectors\n // configured\" while embedding stays off, which is the ambiguity v4 removed.\n if (embed.model === undefined || embed.model === '') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed.model is required when the \"embed\" block is present (it is what turns vectors on); remove the block to index without vectors`);\n }\n if (embed.provider !== undefined && embed.provider !== 'static' && embed.provider !== 'openai' && embed.provider !== 'cohere') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed.provider must be \"static\", \"openai\", or \"cohere\"`);\n }\n if (embed.url !== undefined && typeof embed.url !== 'string') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed.url must be a string`);\n }\n if (embed.key !== undefined && typeof embed.key !== 'string') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed.key must be a string`);\n }\n // openai has no default host, so a missing url would otherwise only surface at the first\n // vector search; a config error belongs at config time.\n if (embed.provider === 'openai' && (embed.url === undefined || embed.url === '')) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed.provider \"openai\" requires embed.url (e.g. \"http://localhost:11434/v1\" for Ollama)`);\n }\n if (embed.languages !== undefined && (!Array.isArray(embed.languages) || embed.languages.length === 0 || embed.languages.some((l) => typeof l !== 'string' || l === ''))) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: embed.languages must be a non-empty array of language tags (e.g. [\"en\", \"zh\"])`);\n }\n}\n\nfunction isSignalName(value: unknown): value is SignalName {\n return typeof value === 'string' && (SIGNAL_NAMES as readonly string[]).includes(value);\n}\n\nfunction isSignalWeight(value: unknown): value is number {\n return typeof value === 'number' && Number.isFinite(value) && value > 0;\n}\n\n// Shape (an object of signal name -> weight) plus the one cross-signal prerequisite that\n// doesn't need the top-level embed block: links is seeded by word-match rows, so declaring it\n// without words is always empty.\nfunction validatePresetSignals(name: string, value: unknown, configPath: string): void {\n if (typeof value !== 'object' || value === null || Array.isArray(value) || Object.keys(value).length === 0) {\n throw new SenseError(\n 'CONFIG_INVALID',\n `${configPath}: presets.${name}.signals must be a non-empty object of signal name -> weight (a finite number > 0), naming ${SIGNAL_NAMES.join(', ')} -- a preset that searches nothing is a mistake; name at least \"words\", or omit \"signals\" to use every signal whose prerequisites hold at weight 1`\n );\n }\n const entries = Object.entries(value as Record<string, unknown>);\n const unknown = entries.filter(([k]) => !isSignalName(k)).map(([k]) => k);\n if (unknown.length > 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.signals names unknown signal(s) ${unknown.join(', ')}; valid signals are ${SIGNAL_NAMES.join(', ')}`);\n }\n for (const [signal, weight] of entries) {\n if (!isSignalWeight(weight)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.signals.${signal} must be a finite number > 0`);\n }\n }\n const names = new Set(entries.map(([k]) => k));\n for (const [signal, requires] of Object.entries(SIGNAL_PREREQUISITES) as Array<[SignalName, SignalName]>) {\n if (names.has(signal) && !names.has(requires)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.signals has \"${signal}\" without \"${requires}\"; add \"${requires}\" to presets.${name}.signals, or drop \"${signal}\"`);\n }\n }\n}\n\nfunction validatePreset(name: string, value: unknown, configPath: string): void {\n if (typeof value !== 'object' || value === null || Array.isArray(value)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name} must be an object`);\n }\n const preset = value as Record<string, unknown>;\n const unknown = Object.keys(preset).filter((k) => !KNOWN_PRESET_KEYS.has(k));\n if (unknown.length > 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name} has unknown key(s) ${unknown.join(', ')}; a preset takes include, exclude, k, signals, where`);\n }\n if (!isNonEmptyStringArray(preset.include)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.include must be a non-empty array of glob strings`);\n }\n if (preset.exclude !== undefined && !isNonEmptyStringArray(preset.exclude)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.exclude must be a non-empty array of glob strings`);\n }\n if (preset.signals !== undefined) validatePresetSignals(name, preset.signals, configPath);\n if (preset.k !== undefined && (typeof preset.k !== 'number' || !Number.isInteger(preset.k) || preset.k <= 0)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.k must be a positive integer`);\n }\n if (preset.where !== undefined && typeof preset.where !== 'string') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.where must be a SQL condition string`);\n }\n}\n\n// A queries.<name> entry: { sql } or a saved search { search, preset?, include?, exclude?, where?, k? }.\nfunction validateSavedQuery(name: string, value: unknown, configPath: string): void {\n // A bare string used to mean SQL. It now fails rather than being inferred: an entry says\n // which of the two verbs it runs, the same choice the CLI makes explicit.\n if (typeof value === 'string') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name} must say which verb it runs: { \"sql\": ${JSON.stringify(value.length > 40 ? `${value.slice(0, 37)}...` : value)} } to run it as SQL, or { \"search\": \"...\" } for a ranked search`);\n }\n if (typeof value !== 'object' || value === null || Array.isArray(value)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name} must be { sql } or { search, preset?, include?, exclude?, where?, k? }`);\n }\n const entry = value as Record<string, unknown>;\n if ('sql' in entry) {\n const unknown = Object.keys(entry).filter((k) => k !== 'sql');\n if (unknown.length > 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name} has unknown key(s) ${unknown.join(', ')}; a { sql } query takes only sql`);\n }\n if (typeof entry.sql !== 'string' || entry.sql.trim() === '') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.sql must be a non-empty string`);\n }\n return;\n }\n if ('search' in entry) {\n const unknown = Object.keys(entry).filter((k) => !SAVED_SEARCH_KEYS.has(k));\n if (unknown.length > 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name} has unknown key(s) ${unknown.join(', ')}; a saved search takes search, preset, include, exclude, where, k`);\n }\n // A saved entry saves a question; a scope without a question is just flags.\n if (typeof entry.search !== 'string' || entry.search.trim() === '') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.search must be non-empty text`);\n }\n if (entry.preset !== undefined && (typeof entry.preset !== 'string' || entry.preset.length === 0)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.preset must be a preset name`);\n }\n if (entry.include !== undefined && !isNonEmptyStringArray(entry.include)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.include must be a non-empty array of glob strings`);\n }\n if (entry.exclude !== undefined && !isNonEmptyStringArray(entry.exclude)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.exclude must be a non-empty array of glob strings`);\n }\n if (entry.where !== undefined && typeof entry.where !== 'string') {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.where must be a SQL condition string`);\n }\n if (entry.k !== undefined && (typeof entry.k !== 'number' || !Number.isInteger(entry.k) || entry.k <= 0)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name}.k must be a positive integer`);\n }\n return;\n }\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries.${name} must be { sql } or { search, preset?, include?, exclude?, where?, k? }`);\n}\n\nexport function validateConfig(parsed: unknown, configPath: string): Config {\n if (typeof parsed !== 'object' || parsed === null || Array.isArray(parsed)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: config must be a JSON object`);\n }\n const cfg = parsed as Record<string, unknown>;\n\n // `checks` is rejected by name, not warned: silence would hide that assertions are gone.\n if (cfg.checks !== undefined) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: checks was removed in v3 -- sense check no longer asserts on saved queries; a returned row set is the reader's judgment`);\n }\n\n const presets = cfg.presets as Record<string, unknown> | undefined;\n if (!presets || typeof presets !== 'object' || Array.isArray(presets) || Object.keys(presets).length === 0) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets must be a non-empty object of preset name -> { include, exclude?, k?, signals?, where? }`);\n }\n for (const [name, value] of Object.entries(presets)) {\n validatePreset(name, value, configPath);\n }\n if (presets.default === undefined) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets must include a \"default\" preset`);\n }\n\n if (cfg.queries === undefined) cfg.queries = {};\n if (typeof cfg.queries !== 'object' || cfg.queries === null || Array.isArray(cfg.queries)) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: queries must be an object of name -> { sql } or { search }`);\n }\n const queries = cfg.queries as Record<string, unknown>;\n for (const [name, value] of Object.entries(queries)) {\n validateSavedQuery(name, value, configPath);\n }\n\n if (cfg.features !== undefined) {\n validateFeaturesBlock(cfg.features, configPath);\n }\n if (cfg.embed !== undefined) {\n validateEmbedBlock(cfg.embed, configPath);\n }\n if (cfg.content !== undefined) {\n validateContentBlock(cfg.content, configPath);\n }\n\n // vectors' prerequisite is a named model, checked here once embed is known valid (or absent).\n if (cfg.embed === undefined) {\n for (const [name, value] of Object.entries(presets)) {\n const signals = (value as Record<string, unknown>).signals as Record<string, unknown> | undefined;\n if (signals && 'vectors' in signals) {\n throw new SenseError('CONFIG_INVALID', `${configPath}: presets.${name}.signals includes \"vectors\", but no \"embed\" block names a model; add one, or drop \"vectors\" from presets.${name}.signals`);\n }\n }\n }\n\n return cfg as unknown as Config;\n}\n"],"names":["SenseError","SIGNAL_NAMES","SIGNAL_PREREQUISITES","KNOWN_KEYS","Set","KNOWN_PRESET_KEYS","KNOWN_FEATURE_KEYS","KNOWN_EMBED_KEYS","KNOWN_CONTENT_KEYS","SAVED_SEARCH_KEYS","unknownConfigKeys","cfg","Object","keys","filter","k","has","isNonEmptyStringArray","value","Array","isArray","length","every","g","validateLegacyScan","parsed","configPath","scan","include","validateContentBlock","block","unknown","join","tokenize","undefined","trim","validateFeaturesBlock","name","v","entries","validateEmbedBlock","embed","model","provider","url","key","languages","some","l","isSignalName","includes","isSignalWeight","Number","isFinite","validatePresetSignals","map","signal","weight","names","requires","validatePreset","preset","exclude","signals","isInteger","where","validateSavedQuery","JSON","stringify","slice","entry","sql","search","validateConfig","checks","presets","default","queries","features","content"],"mappings":"AAAA,SAASA,UAAU,QAAQ,eAAe;AAC1C,SAASC,YAAY,EAAEC,oBAAoB,QAAyB,eAAe;AAGnF,sFAAsF;AACtF,8FAA8F;AAC9F,MAAMC,aAAa,IAAIC,IAAI;IAAC;IAAW;IAAW;IAAW;IAAY;IAAS;IAAW;CAAU;AACvG,MAAMC,oBAAoB,IAAID,IAAI;IAAC;IAAW;IAAW;IAAK;IAAW;CAAQ;AACjF,MAAME,qBAAqB,IAAIF,IAAI;IAAC;IAAS;IAAY;IAAQ;CAAO;AACxE,MAAMG,mBAAmB,IAAIH,IAAI;IAAC;IAAS;IAAY;IAAO;IAAO;CAAY;AACjF,MAAMI,qBAAqB,IAAIJ,IAAI;IAAC;CAAW;AAC/C,MAAMK,oBAAoB,IAAIL,IAAI;IAAC;IAAU;IAAU;IAAW;IAAW;IAAS;CAAI;AAE1F,OAAO,SAASM,kBAAkBC,GAA4B;IAC5D,OAAOC,OAAOC,IAAI,CAACF,KAAKG,MAAM,CAAC,CAACC,IAAM,CAACZ,WAAWa,GAAG,CAACD;AACxD;AAEA,SAASE,sBAAsBC,KAAc;IAC3C,OAAOC,MAAMC,OAAO,CAACF,UAAUA,MAAMG,MAAM,GAAG,KAAKH,MAAMI,KAAK,CAAC,CAACC,IAAM,OAAOA,MAAM,YAAYA,EAAEF,MAAM,GAAG;AAC5G;AAEA,yFAAyF;AACzF,4FAA4F;AAC5F,OAAO,SAASG,mBAAmBC,MAAe,EAAEC,UAAkB;IACpE,MAAMf,MAAM,OAAOc,WAAW,YAAYA,WAAW,QAAQ,CAACN,MAAMC,OAAO,CAACK,UAAWA,SAAqC;IAC5H,MAAME,OAAOhB,gBAAAA,0BAAAA,IAAKgB,IAAI;IACtB,IAAI,CAAChB,OAAO,CAACgB,QAAQ,CAACV,sBAAsBU,KAAKC,OAAO,GAAG;QACzD,MAAM,IAAI5B,WAAW,kBAAkB,GAAG0B,WAAW,wDAAwD,CAAC;IAChH;AACF;AAEA,6FAA6F;AAC7F,0FAA0F;AAC1F,SAASG,qBAAqBX,KAAc,EAAEQ,UAAkB;IAC9D,IAAI,OAAOR,UAAU,YAAYA,UAAU,QAAQC,MAAMC,OAAO,CAACF,QAAQ;QACvE,MAAM,IAAIlB,WAAW,kBAAkB,GAAG0B,WAAW,oDAAoD,CAAC;IAC5G;IACA,MAAMI,QAAQZ;IACd,MAAMa,UAAUnB,OAAOC,IAAI,CAACiB,OAAOhB,MAAM,CAAC,CAACC,IAAM,CAACP,mBAAmBQ,GAAG,CAACD;IACzE,IAAIgB,QAAQV,MAAM,GAAG,GAAG;QACtB,MAAM,IAAIrB,WAAW,kBAAkB,GAAG0B,WAAW,6BAA6B,EAAEK,QAAQC,IAAI,CAAC,MAAM,wBAAwB,CAAC;IAClI;IACA,IAAIF,MAAMG,QAAQ,KAAKC,aAAc,CAAA,OAAOJ,MAAMG,QAAQ,KAAK,YAAYH,MAAMG,QAAQ,CAACE,IAAI,OAAO,EAAC,GAAI;QACxG,MAAM,IAAInC,WAAW,kBAAkB,GAAG0B,WAAW,4FAA4F,CAAC;IACpJ;AACF;AAEA,SAASU,sBAAsBlB,KAAc,EAAEQ,UAAkB;IAC/D,IAAI,OAAOR,UAAU,YAAYA,UAAU,QAAQC,MAAMC,OAAO,CAACF,QAAQ;QACvE,MAAM,IAAIlB,WAAW,kBAAkB,GAAG0B,WAAW,+CAA+C,CAAC;IACvG;IACA,MAAMI,QAAQZ;IACd,MAAMa,UAAUnB,OAAOC,IAAI,CAACiB,OAAOhB,MAAM,CAAC,CAACC,IAAM,CAACT,mBAAmBU,GAAG,CAACD;IACzE,IAAIgB,QAAQV,MAAM,GAAG,GAAG;QACtB,MAAM,IAAIrB,WAAW,kBAAkB,GAAG0B,WAAW,8BAA8B,EAAEK,QAAQC,IAAI,CAAC,MAAM,gEAAgE,CAAC;IAC3K;IACA,KAAK,MAAM,CAACK,MAAMC,EAAE,IAAI1B,OAAO2B,OAAO,CAACT,OAAQ;QAC7C,IAAI,OAAOQ,MAAM,WAAW;YAC1B,MAAM,IAAItC,WAAW,kBAAkB,GAAG0B,WAAW,WAAW,EAAEW,KAAK,kBAAkB,CAAC;QAC5F;IACF;AACF;AAEA,SAASG,mBAAmBtB,KAAc,EAAEQ,UAAkB;IAC5D,IAAI,OAAOR,UAAU,YAAYA,UAAU,QAAQC,MAAMC,OAAO,CAACF,QAAQ;QACvE,MAAM,IAAIlB,WAAW,kBAAkB,GAAG0B,WAAW,0FAA0F,CAAC;IAClJ;IACA,MAAMe,QAAQvB;IACd,MAAMa,UAAUnB,OAAOC,IAAI,CAAC4B,OAAO3B,MAAM,CAAC,CAACC,IAAM,CAACR,iBAAiBS,GAAG,CAACD;IACvE,IAAIgB,QAAQV,MAAM,GAAG,GAAG;QACtB,MAAM,IAAIrB,WAAW,kBAAkB,GAAG0B,WAAW,2BAA2B,EAAEK,QAAQC,IAAI,CAAC,OAAO;IACxG;IACA,IAAIS,MAAMC,KAAK,KAAKR,aAAa,OAAOO,MAAMC,KAAK,KAAK,UAAU;QAChE,MAAM,IAAI1C,WAAW,kBAAkB,GAAG0B,WAAW,8BAA8B,CAAC;IACtF;IACA,yFAAyF;IACzF,4EAA4E;IAC5E,IAAIe,MAAMC,KAAK,KAAKR,aAAaO,MAAMC,KAAK,KAAK,IAAI;QACnD,MAAM,IAAI1C,WAAW,kBAAkB,GAAG0B,WAAW,oIAAoI,CAAC;IAC5L;IACA,IAAIe,MAAME,QAAQ,KAAKT,aAAaO,MAAME,QAAQ,KAAK,YAAYF,MAAME,QAAQ,KAAK,YAAYF,MAAME,QAAQ,KAAK,UAAU;QAC7H,MAAM,IAAI3C,WAAW,kBAAkB,GAAG0B,WAAW,wDAAwD,CAAC;IAChH;IACA,IAAIe,MAAMG,GAAG,KAAKV,aAAa,OAAOO,MAAMG,GAAG,KAAK,UAAU;QAC5D,MAAM,IAAI5C,WAAW,kBAAkB,GAAG0B,WAAW,4BAA4B,CAAC;IACpF;IACA,IAAIe,MAAMI,GAAG,KAAKX,aAAa,OAAOO,MAAMI,GAAG,KAAK,UAAU;QAC5D,MAAM,IAAI7C,WAAW,kBAAkB,GAAG0B,WAAW,4BAA4B,CAAC;IACpF;IACA,yFAAyF;IACzF,wDAAwD;IACxD,IAAIe,MAAME,QAAQ,KAAK,YAAaF,CAAAA,MAAMG,GAAG,KAAKV,aAAaO,MAAMG,GAAG,KAAK,EAAC,GAAI;QAChF,MAAM,IAAI5C,WAAW,kBAAkB,GAAG0B,WAAW,0FAA0F,CAAC;IAClJ;IACA,IAAIe,MAAMK,SAAS,KAAKZ,aAAc,CAAA,CAACf,MAAMC,OAAO,CAACqB,MAAMK,SAAS,KAAKL,MAAMK,SAAS,CAACzB,MAAM,KAAK,KAAKoB,MAAMK,SAAS,CAACC,IAAI,CAAC,CAACC,IAAM,OAAOA,MAAM,YAAYA,MAAM,GAAE,GAAI;QACxK,MAAM,IAAIhD,WAAW,kBAAkB,GAAG0B,WAAW,gFAAgF,CAAC;IACxI;AACF;AAEA,SAASuB,aAAa/B,KAAc;IAClC,OAAO,OAAOA,UAAU,YAAY,AAACjB,aAAmCiD,QAAQ,CAAChC;AACnF;AAEA,SAASiC,eAAejC,KAAc;IACpC,OAAO,OAAOA,UAAU,YAAYkC,OAAOC,QAAQ,CAACnC,UAAUA,QAAQ;AACxE;AAEA,yFAAyF;AACzF,8FAA8F;AAC9F,iCAAiC;AACjC,SAASoC,sBAAsBjB,IAAY,EAAEnB,KAAc,EAAEQ,UAAkB;IAC7E,IAAI,OAAOR,UAAU,YAAYA,UAAU,QAAQC,MAAMC,OAAO,CAACF,UAAUN,OAAOC,IAAI,CAACK,OAAOG,MAAM,KAAK,GAAG;QAC1G,MAAM,IAAIrB,WACR,kBACA,GAAG0B,WAAW,UAAU,EAAEW,KAAK,2FAA2F,EAAEpC,aAAa+B,IAAI,CAAC,MAAM,kJAAkJ,CAAC;IAE3S;IACA,MAAMO,UAAU3B,OAAO2B,OAAO,CAACrB;IAC/B,MAAMa,UAAUQ,QAAQzB,MAAM,CAAC,CAAC,CAACC,EAAE,GAAK,CAACkC,aAAalC,IAAIwC,GAAG,CAAC,CAAC,CAACxC,EAAE,GAAKA;IACvE,IAAIgB,QAAQV,MAAM,GAAG,GAAG;QACtB,MAAM,IAAIrB,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,iCAAiC,EAAEN,QAAQC,IAAI,CAAC,MAAM,oBAAoB,EAAE/B,aAAa+B,IAAI,CAAC,OAAO;IAC7K;IACA,KAAK,MAAM,CAACwB,QAAQC,OAAO,IAAIlB,QAAS;QACtC,IAAI,CAACY,eAAeM,SAAS;YAC3B,MAAM,IAAIzD,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,SAAS,EAAEmB,OAAO,4BAA4B,CAAC;QACvH;IACF;IACA,MAAME,QAAQ,IAAItD,IAAImC,QAAQgB,GAAG,CAAC,CAAC,CAACxC,EAAE,GAAKA;IAC3C,KAAK,MAAM,CAACyC,QAAQG,SAAS,IAAI/C,OAAO2B,OAAO,CAACrC,sBAA0D;QACxG,IAAIwD,MAAM1C,GAAG,CAACwC,WAAW,CAACE,MAAM1C,GAAG,CAAC2C,WAAW;YAC7C,MAAM,IAAI3D,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,cAAc,EAAEmB,OAAO,WAAW,EAAEG,SAAS,QAAQ,EAAEA,SAAS,aAAa,EAAEtB,KAAK,mBAAmB,EAAEmB,OAAO,CAAC,CAAC;QAC1L;IACF;AACF;AAEA,SAASI,eAAevB,IAAY,EAAEnB,KAAc,EAAEQ,UAAkB;IACtE,IAAI,OAAOR,UAAU,YAAYA,UAAU,QAAQC,MAAMC,OAAO,CAACF,QAAQ;QACvE,MAAM,IAAIlB,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,kBAAkB,CAAC;IAC3F;IACA,MAAMwB,SAAS3C;IACf,MAAMa,UAAUnB,OAAOC,IAAI,CAACgD,QAAQ/C,MAAM,CAAC,CAACC,IAAM,CAACV,kBAAkBW,GAAG,CAACD;IACzE,IAAIgB,QAAQV,MAAM,GAAG,GAAG;QACtB,MAAM,IAAIrB,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,oBAAoB,EAAEN,QAAQC,IAAI,CAAC,MAAM,oDAAoD,CAAC;IACtK;IACA,IAAI,CAACf,sBAAsB4C,OAAOjC,OAAO,GAAG;QAC1C,MAAM,IAAI5B,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,kDAAkD,CAAC;IAC3H;IACA,IAAIwB,OAAOC,OAAO,KAAK5B,aAAa,CAACjB,sBAAsB4C,OAAOC,OAAO,GAAG;QAC1E,MAAM,IAAI9D,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,kDAAkD,CAAC;IAC3H;IACA,IAAIwB,OAAOE,OAAO,KAAK7B,WAAWoB,sBAAsBjB,MAAMwB,OAAOE,OAAO,EAAErC;IAC9E,IAAImC,OAAO9C,CAAC,KAAKmB,aAAc,CAAA,OAAO2B,OAAO9C,CAAC,KAAK,YAAY,CAACqC,OAAOY,SAAS,CAACH,OAAO9C,CAAC,KAAK8C,OAAO9C,CAAC,IAAI,CAAA,GAAI;QAC5G,MAAM,IAAIf,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,6BAA6B,CAAC;IACtG;IACA,IAAIwB,OAAOI,KAAK,KAAK/B,aAAa,OAAO2B,OAAOI,KAAK,KAAK,UAAU;QAClE,MAAM,IAAIjE,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,qCAAqC,CAAC;IAC9G;AACF;AAEA,yGAAyG;AACzG,SAAS6B,mBAAmB7B,IAAY,EAAEnB,KAAc,EAAEQ,UAAkB;IAC1E,yFAAyF;IACzF,0EAA0E;IAC1E,IAAI,OAAOR,UAAU,UAAU;QAC7B,MAAM,IAAIlB,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,uCAAuC,EAAE8B,KAAKC,SAAS,CAAClD,MAAMG,MAAM,GAAG,KAAK,GAAGH,MAAMmD,KAAK,CAAC,GAAG,IAAI,GAAG,CAAC,GAAGnD,OAAO,+DAA+D,CAAC;IACxP;IACA,IAAI,OAAOA,UAAU,YAAYA,UAAU,QAAQC,MAAMC,OAAO,CAACF,QAAQ;QACvE,MAAM,IAAIlB,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,uEAAuE,CAAC;IAChJ;IACA,MAAMiC,QAAQpD;IACd,IAAI,SAASoD,OAAO;QAClB,MAAMvC,UAAUnB,OAAOC,IAAI,CAACyD,OAAOxD,MAAM,CAAC,CAACC,IAAMA,MAAM;QACvD,IAAIgB,QAAQV,MAAM,GAAG,GAAG;YACtB,MAAM,IAAIrB,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,oBAAoB,EAAEN,QAAQC,IAAI,CAAC,MAAM,gCAAgC,CAAC;QAClJ;QACA,IAAI,OAAOsC,MAAMC,GAAG,KAAK,YAAYD,MAAMC,GAAG,CAACpC,IAAI,OAAO,IAAI;YAC5D,MAAM,IAAInC,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,+BAA+B,CAAC;QACxG;QACA;IACF;IACA,IAAI,YAAYiC,OAAO;QACrB,MAAMvC,UAAUnB,OAAOC,IAAI,CAACyD,OAAOxD,MAAM,CAAC,CAACC,IAAM,CAACN,kBAAkBO,GAAG,CAACD;QACxE,IAAIgB,QAAQV,MAAM,GAAG,GAAG;YACtB,MAAM,IAAIrB,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,oBAAoB,EAAEN,QAAQC,IAAI,CAAC,MAAM,iEAAiE,CAAC;QACnL;QACA,4EAA4E;QAC5E,IAAI,OAAOsC,MAAME,MAAM,KAAK,YAAYF,MAAME,MAAM,CAACrC,IAAI,OAAO,IAAI;YAClE,MAAM,IAAInC,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,8BAA8B,CAAC;QACvG;QACA,IAAIiC,MAAMT,MAAM,KAAK3B,aAAc,CAAA,OAAOoC,MAAMT,MAAM,KAAK,YAAYS,MAAMT,MAAM,CAACxC,MAAM,KAAK,CAAA,GAAI;YACjG,MAAM,IAAIrB,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,6BAA6B,CAAC;QACtG;QACA,IAAIiC,MAAM1C,OAAO,KAAKM,aAAa,CAACjB,sBAAsBqD,MAAM1C,OAAO,GAAG;YACxE,MAAM,IAAI5B,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,kDAAkD,CAAC;QAC3H;QACA,IAAIiC,MAAMR,OAAO,KAAK5B,aAAa,CAACjB,sBAAsBqD,MAAMR,OAAO,GAAG;YACxE,MAAM,IAAI9D,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,kDAAkD,CAAC;QAC3H;QACA,IAAIiC,MAAML,KAAK,KAAK/B,aAAa,OAAOoC,MAAML,KAAK,KAAK,UAAU;YAChE,MAAM,IAAIjE,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,qCAAqC,CAAC;QAC9G;QACA,IAAIiC,MAAMvD,CAAC,KAAKmB,aAAc,CAAA,OAAOoC,MAAMvD,CAAC,KAAK,YAAY,CAACqC,OAAOY,SAAS,CAACM,MAAMvD,CAAC,KAAKuD,MAAMvD,CAAC,IAAI,CAAA,GAAI;YACxG,MAAM,IAAIf,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,6BAA6B,CAAC;QACtG;QACA;IACF;IACA,MAAM,IAAIrC,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,uEAAuE,CAAC;AAChJ;AAEA,OAAO,SAASoC,eAAehD,MAAe,EAAEC,UAAkB;IAChE,IAAI,OAAOD,WAAW,YAAYA,WAAW,QAAQN,MAAMC,OAAO,CAACK,SAAS;QAC1E,MAAM,IAAIzB,WAAW,kBAAkB,GAAG0B,WAAW,8BAA8B,CAAC;IACtF;IACA,MAAMf,MAAMc;IAEZ,yFAAyF;IACzF,IAAId,IAAI+D,MAAM,KAAKxC,WAAW;QAC5B,MAAM,IAAIlC,WAAW,kBAAkB,GAAG0B,WAAW,yHAAyH,CAAC;IACjL;IAEA,MAAMiD,UAAUhE,IAAIgE,OAAO;IAC3B,IAAI,CAACA,WAAW,OAAOA,YAAY,YAAYxD,MAAMC,OAAO,CAACuD,YAAY/D,OAAOC,IAAI,CAAC8D,SAAStD,MAAM,KAAK,GAAG;QAC1G,MAAM,IAAIrB,WAAW,kBAAkB,GAAG0B,WAAW,kGAAkG,CAAC;IAC1J;IACA,KAAK,MAAM,CAACW,MAAMnB,MAAM,IAAIN,OAAO2B,OAAO,CAACoC,SAAU;QACnDf,eAAevB,MAAMnB,OAAOQ;IAC9B;IACA,IAAIiD,QAAQC,OAAO,KAAK1C,WAAW;QACjC,MAAM,IAAIlC,WAAW,kBAAkB,GAAG0B,WAAW,yCAAyC,CAAC;IACjG;IAEA,IAAIf,IAAIkE,OAAO,KAAK3C,WAAWvB,IAAIkE,OAAO,GAAG,CAAC;IAC9C,IAAI,OAAOlE,IAAIkE,OAAO,KAAK,YAAYlE,IAAIkE,OAAO,KAAK,QAAQ1D,MAAMC,OAAO,CAACT,IAAIkE,OAAO,GAAG;QACzF,MAAM,IAAI7E,WAAW,kBAAkB,GAAG0B,WAAW,4DAA4D,CAAC;IACpH;IACA,MAAMmD,UAAUlE,IAAIkE,OAAO;IAC3B,KAAK,MAAM,CAACxC,MAAMnB,MAAM,IAAIN,OAAO2B,OAAO,CAACsC,SAAU;QACnDX,mBAAmB7B,MAAMnB,OAAOQ;IAClC;IAEA,IAAIf,IAAImE,QAAQ,KAAK5C,WAAW;QAC9BE,sBAAsBzB,IAAImE,QAAQ,EAAEpD;IACtC;IACA,IAAIf,IAAI8B,KAAK,KAAKP,WAAW;QAC3BM,mBAAmB7B,IAAI8B,KAAK,EAAEf;IAChC;IACA,IAAIf,IAAIoE,OAAO,KAAK7C,WAAW;QAC7BL,qBAAqBlB,IAAIoE,OAAO,EAAErD;IACpC;IAEA,8FAA8F;IAC9F,IAAIf,IAAI8B,KAAK,KAAKP,WAAW;QAC3B,KAAK,MAAM,CAACG,MAAMnB,MAAM,IAAIN,OAAO2B,OAAO,CAACoC,SAAU;YACnD,MAAMZ,UAAU,AAAC7C,MAAkC6C,OAAO;YAC1D,IAAIA,WAAW,aAAaA,SAAS;gBACnC,MAAM,IAAI/D,WAAW,kBAAkB,GAAG0B,WAAW,UAAU,EAAEW,KAAK,yGAAyG,EAAEA,KAAK,QAAQ,CAAC;YACjM;QACF;IACF;IAEA,OAAO1B;AACT"}
@@ -1,7 +1,7 @@
1
1
  import { DatabaseSync } from 'node:sqlite';
2
2
  import type { ResolvedConfig } from '../config/index.js';
3
3
  export declare const DB_FILENAME = "cache.db";
4
- export declare const SCHEMA_VERSION = "16";
4
+ export declare const SCHEMA_VERSION = "17";
5
5
  export interface OpenResult {
6
6
  db: DatabaseSync;
7
7
  cfg: ResolvedConfig;
@@ -7,12 +7,12 @@ import { contentTokenize, featureSignature, STATE_DIR } from '../config/index.js
7
7
  import { SenseError } from '../errors.js';
8
8
  import { activeFeatures } from '../features/index.js';
9
9
  import { registerFunctions } from '../sql-functions.js';
10
- import { changedSignatureKeys, rebuildContentTable, reconcile, signatureDiff } from './reconcile.js';
10
+ import { changedSignatureKeys, embedIdentityAdopted, rebuildContentTable, reconcile, signatureDiff } from './reconcile.js';
11
11
  import { getMeta, setMeta } from './shared.js';
12
12
  export const DB_FILENAME = 'cache.db';
13
13
  // Cache shape version, independent of the config's own `version`. Bumping it rebuilds
14
14
  // existing trees on first query.
15
- export const SCHEMA_VERSION = '16';
15
+ export const SCHEMA_VERSION = '17';
16
16
  // Stemming is English-only, but the segmentation underneath it is what decides coverage:
17
17
  // unicode61 splits on spaces, so a language written without them (Chinese, Japanese, Thai)
18
18
  // indexes a whole run as one token and word search finds nothing. `content.tokenize` is how
@@ -107,6 +107,11 @@ export function open(cfg) {
107
107
  console.error('sense: config change (content tokenizer) rebuilds the text index; vectors, links, and sections are kept');
108
108
  db.exec('DROP TABLE content');
109
109
  tokenizeOnlyRebuild = true;
110
+ } else if (changedKeys.size === 1 && changedKeys.has('embed') && embedIdentityAdopted(features !== null && features !== void 0 ? features : '', wantFeatures)) {
111
+ // First sight of a resolved weight identity: the model itself hasn't changed, so
112
+ // adopt it into meta with no rebuild and no re-embed, mirroring the tokenize precedent.
113
+ console.error("sense: recorded the embedding model's resolved identity; vectors are unaffected");
114
+ setMeta(db, 'features', wantFeatures);
110
115
  } else {
111
116
  const changed = signatureDiff(features !== null && features !== void 0 ? features : '', wantFeatures);
112
117
  console.error(`sense: config change (${changed}) rebuilds the index`);
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/db/open.ts"],"sourcesContent":["// The Node floor (>=22.20) is explained here and nowhere else: 22.20 is the first release with\n// both FTS5 and row-returning INSERT ... RETURNING. Raise it only for a load-bearing capability.\nimport { mkdirSync, rmSync } from 'node:fs';\nimport { join } from 'node:path';\nimport { DatabaseSync } from 'node:sqlite';\nimport type { Config, ResolvedConfig } from '../config/index.ts';\nimport { contentTokenize, featureSignature, STATE_DIR } from '../config/index.ts';\nimport { SenseError } from '../errors.ts';\nimport { activeFeatures } from '../features/index.ts';\nimport { registerFunctions } from '../sql-functions.ts';\nimport { changedSignatureKeys, rebuildContentTable, reconcile, signatureDiff } from './reconcile.ts';\nimport { getMeta, setMeta } from './shared.ts';\n\nexport const DB_FILENAME = 'cache.db';\n// Cache shape version, independent of the config's own `version`. Bumping it rebuilds\n// existing trees on first query.\nexport const SCHEMA_VERSION = '16';\n\nexport interface OpenResult {\n db: DatabaseSync;\n cfg: ResolvedConfig;\n dbPath: string;\n parsed: number;\n warnings: string[];\n}\n\n// Stemming is English-only, but the segmentation underneath it is what decides coverage:\n// unicode61 splits on spaces, so a language written without them (Chinese, Japanese, Thai)\n// indexes a whole run as one token and word search finds nothing. `content.tokenize` is how\n// such a tree picks trigram instead.\nconst DEFAULT_TOKENIZE = 'porter unicode61';\n\n// FTS5 takes its tokenizer as a string literal inside DDL, where nothing can bind, so a\n// configured value has to be concatenated. Probing a throwaway table is what makes that safe\n// and is also the whole validation: anything the linked SQLite accepts passes, anything else\n// fails here with SQLite's own message rather than against the real table. It means no table\n// of which version added which tokenizer has to be maintained.\nfunction resolveTokenize(db: DatabaseSync, cfg: Config): string {\n const configured = contentTokenize(cfg);\n if (configured === undefined) return DEFAULT_TOKENIZE;\n const literal = configured.replace(/'/g, \"''\");\n try {\n db.exec('DROP TABLE IF EXISTS temp.sense_tokenize_probe');\n db.exec(`CREATE VIRTUAL TABLE temp.sense_tokenize_probe USING fts5(x, tokenize = '${literal}')`);\n db.exec('DROP TABLE IF EXISTS temp.sense_tokenize_probe');\n } catch (err) {\n throw new SenseError('CONFIG_INVALID', `content.tokenize \"${configured}\" is not a tokenizer this SQLite accepts (${(err as Error).message}); the built-in choices are unicode61, ascii, porter, and trigram, each with their own options`);\n }\n return literal;\n}\n\n// The tokenizer the content table was actually built with, from its own DDL -- the one\n// record that cannot desynchronize from the table. NULL when the table does not exist yet.\nfunction storedTokenize(db: DatabaseSync): string | null {\n const row = db.prepare(`SELECT sql FROM sqlite_master WHERE name = 'content'`).get() as { sql: string } | undefined;\n if (!row) return null;\n const m = row.sql.match(/tokenize = '((?:[^']|'')*)'/);\n return m ? m[1] : null;\n}\n\n// Content is a separate table (not a column on frontmatter) so `SELECT * FROM frontmatter`\n// can't dump file text into context. Features add their own tables after the core ones.\nfunction ensureSchema(db: DatabaseSync, cfg: Config, tokenize: string): void {\n db.exec(`CREATE TABLE IF NOT EXISTS frontmatter (\"path\" TEXT PRIMARY KEY, \"_mtime\" REAL, \"_ctime\" REAL, \"_size\" INTEGER, \"_parse_error\" TEXT)`);\n // IF NOT EXISTS is safe against a tokenizer change: open() compares the table's own DDL\n // against the resolved tokenizer before this runs, so a stale table is already gone by now.\n // The three `_seg` sidecars are appended after path, never inserted: bm25(content, ...) and\n // snippet(content, 2, ...) are documented against the first three columns and keep working\n // (FTS5 defaults the weights it was not given). Each carries its field's exploded unspaced\n // runs for text that needs it, and an empty string for text that does not.\n db.exec(`CREATE VIRTUAL TABLE IF NOT EXISTS content USING fts5(title, summary, text, path UNINDEXED, title_seg, summary_seg, text_seg, tokenize = '${tokenize}')`);\n // Coverage, not ownership: a path can appear under several presets. path leads the PK so the\n // per-doc delete is an index hit -- keyed the other way, cold builds went quadratic.\n db.exec(`CREATE TABLE IF NOT EXISTS preset_files (\"path\" TEXT, preset TEXT, PRIMARY KEY (\"path\", preset))`);\n db.exec('CREATE INDEX IF NOT EXISTS preset_files_preset ON preset_files(preset)');\n for (const feature of activeFeatures(cfg)) feature.schema(db);\n if (getMeta(db, 'schema_version') === null) setMeta(db, 'schema_version', SCHEMA_VERSION);\n if (getMeta(db, 'features') === null) setMeta(db, 'features', featureSignature(cfg));\n}\n\nexport function docCount(db: DatabaseSync): number {\n const row = db.prepare('SELECT COUNT(*) AS n FROM frontmatter').get() as { n: number };\n return row.n;\n}\n\nexport function open(cfg: ResolvedConfig): OpenResult {\n const stateDir = join(cfg.baseDir, STATE_DIR);\n mkdirSync(stateDir, { recursive: true });\n const dbPath = join(stateDir, DB_FILENAME);\n\n const db = new DatabaseSync(dbPath);\n db.exec('PRAGMA journal_mode = WAL');\n // Covers a concurrent watcher's bulk reconcile (~5s for 500 files at 26k notes). A query\n // that outwaits it still fails loudly.\n db.exec('PRAGMA busy_timeout = 30000');\n registerFunctions(db, contentTokenize(cfg) === undefined);\n\n db.exec('CREATE TABLE IF NOT EXISTS meta (key TEXT PRIMARY KEY, value TEXT)');\n\n // Before the rebuild branch below, never after: that branch deletes the cache, so a typo'd\n // tokenizer validated later would cost a full re-index (and re-embed) to reach its own error.\n const tokenize = resolveTokenize(db, cfg);\n\n // Schema-version or feature-set mismatch: reconcile only reparses changed files, so an\n // old cache can't be patched incrementally -- rebuild instead (cheap: nothing expensive lives here).\n const version = getMeta(db, 'schema_version');\n const features = getMeta(db, 'features');\n const wantFeatures = featureSignature(cfg);\n let tokenizeOnlyRebuild = false;\n if ((version !== null && version !== SCHEMA_VERSION) || (features !== null && features !== wantFeatures)) {\n // Indexing derives from presets, so a config edit rebuilding the cache must say so and\n // name what changed -- silent rebuilds make derived indexing look like a hang or a bug.\n if (version !== null && version !== SCHEMA_VERSION) {\n console.error('sense: cache format changed (new sensemaking version); rebuilding the index');\n db.close();\n clearCache(cfg);\n return open(cfg);\n }\n const changedKeys = changedSignatureKeys(features ?? '', wantFeatures);\n // Only the tokenizer moved: frontmatter, links, sections, and embeddings are file-derived\n // and tokenizer-independent, so they don't need re-deriving. Everything else -- a preset\n // edit, an embed model change -- still takes the full clear/reopen below.\n if (changedKeys.size === 1 && changedKeys.has('tokenize')) {\n console.error('sense: config change (content tokenizer) rebuilds the text index; vectors, links, and sections are kept');\n db.exec('DROP TABLE content');\n tokenizeOnlyRebuild = true;\n } else {\n const changed = signatureDiff(features ?? '', wantFeatures);\n console.error(`sense: config change (${changed}) rebuilds the index`);\n db.close();\n clearCache(cfg);\n return open(cfg);\n }\n }\n\n // Meta can lie after a crash between table creation and the signature write; the table's\n // own DDL cannot. A mismatch here rebuilds no matter what meta says. A tokenize-only rebuild\n // just dropped content above, so storedTokenize sees no table and this guard is skipped\n // naturally rather than needing its own case.\n const stored = storedTokenize(db);\n if (stored !== null && stored !== tokenize) {\n console.error('sense: cache was built with a different content tokenizer; rebuilding the index');\n db.close();\n clearCache(cfg);\n return open(cfg);\n }\n\n ensureSchema(db, cfg, tokenize);\n\n let rebuildWarnings: string[] = [];\n if (tokenizeOnlyRebuild) {\n rebuildWarnings = rebuildContentTable(db, cfg, cfg.baseDir);\n setMeta(db, 'features', wantFeatures);\n }\n\n // 3x the largest reconcile this cache has recorded, floored at 30s and capped at 10min.\n // Installed before reconcile() -- that call is the one that races a watcher's transaction.\n const recordedMaxMs = Number(getMeta(db, 'reconcile_max_ms') ?? '0');\n db.exec(`PRAGMA busy_timeout = ${Math.min(Math.max(30000, 3 * recordedMaxMs), 600_000)}`);\n\n const { parsed, warnings } = reconcile(db, cfg, cfg.baseDir);\n\n return { db, cfg, dbPath, parsed, warnings: [...rebuildWarnings, ...warnings] };\n}\n\n// Deletes the cache directory, and only that. The rebuild is not this function's job: the next\n// open() reconciles, which is what running any command already does, so a verb that bundled the\n// two described the half that was not its own. Manual reset for a doubted cache; the schema and\n// config-signature mismatches below reset themselves.\nexport function clearCache(cfg: ResolvedConfig): void {\n rmSync(join(cfg.baseDir, STATE_DIR), { recursive: true, force: true });\n}\n"],"names":["mkdirSync","rmSync","join","DatabaseSync","contentTokenize","featureSignature","STATE_DIR","SenseError","activeFeatures","registerFunctions","changedSignatureKeys","rebuildContentTable","reconcile","signatureDiff","getMeta","setMeta","DB_FILENAME","SCHEMA_VERSION","DEFAULT_TOKENIZE","resolveTokenize","db","cfg","configured","undefined","literal","replace","exec","err","message","storedTokenize","row","prepare","get","m","sql","match","ensureSchema","tokenize","feature","schema","docCount","n","open","stateDir","baseDir","recursive","dbPath","version","features","wantFeatures","tokenizeOnlyRebuild","console","error","close","clearCache","changedKeys","size","has","changed","stored","rebuildWarnings","recordedMaxMs","Number","Math","min","max","parsed","warnings","force"],"mappings":"AAAA,+FAA+F;AAC/F,iGAAiG;AACjG,SAASA,SAAS,EAAEC,MAAM,QAAQ,UAAU;AAC5C,SAASC,IAAI,QAAQ,YAAY;AACjC,SAASC,YAAY,QAAQ,cAAc;AAE3C,SAASC,eAAe,EAAEC,gBAAgB,EAAEC,SAAS,QAAQ,qBAAqB;AAClF,SAASC,UAAU,QAAQ,eAAe;AAC1C,SAASC,cAAc,QAAQ,uBAAuB;AACtD,SAASC,iBAAiB,QAAQ,sBAAsB;AACxD,SAASC,oBAAoB,EAAEC,mBAAmB,EAAEC,SAAS,EAAEC,aAAa,QAAQ,iBAAiB;AACrG,SAASC,OAAO,EAAEC,OAAO,QAAQ,cAAc;AAE/C,OAAO,MAAMC,cAAc,WAAW;AACtC,sFAAsF;AACtF,iCAAiC;AACjC,OAAO,MAAMC,iBAAiB,KAAK;AAUnC,yFAAyF;AACzF,2FAA2F;AAC3F,4FAA4F;AAC5F,qCAAqC;AACrC,MAAMC,mBAAmB;AAEzB,wFAAwF;AACxF,6FAA6F;AAC7F,6FAA6F;AAC7F,6FAA6F;AAC7F,+DAA+D;AAC/D,SAASC,gBAAgBC,EAAgB,EAAEC,GAAW;IACpD,MAAMC,aAAalB,gBAAgBiB;IACnC,IAAIC,eAAeC,WAAW,OAAOL;IACrC,MAAMM,UAAUF,WAAWG,OAAO,CAAC,MAAM;IACzC,IAAI;QACFL,GAAGM,IAAI,CAAC;QACRN,GAAGM,IAAI,CAAC,CAAC,yEAAyE,EAAEF,QAAQ,EAAE,CAAC;QAC/FJ,GAAGM,IAAI,CAAC;IACV,EAAE,OAAOC,KAAK;QACZ,MAAM,IAAIpB,WAAW,kBAAkB,CAAC,kBAAkB,EAAEe,WAAW,0CAA0C,EAAE,AAACK,IAAcC,OAAO,CAAC,8FAA8F,CAAC;IAC3O;IACA,OAAOJ;AACT;AAEA,uFAAuF;AACvF,2FAA2F;AAC3F,SAASK,eAAeT,EAAgB;IACtC,MAAMU,MAAMV,GAAGW,OAAO,CAAC,CAAC,oDAAoD,CAAC,EAAEC,GAAG;IAClF,IAAI,CAACF,KAAK,OAAO;IACjB,MAAMG,IAAIH,IAAII,GAAG,CAACC,KAAK,CAAC;IACxB,OAAOF,IAAIA,CAAC,CAAC,EAAE,GAAG;AACpB;AAEA,2FAA2F;AAC3F,wFAAwF;AACxF,SAASG,aAAahB,EAAgB,EAAEC,GAAW,EAAEgB,QAAgB;IACnEjB,GAAGM,IAAI,CAAC,CAAC,oIAAoI,CAAC;IAC9I,wFAAwF;IACxF,4FAA4F;IAC5F,4FAA4F;IAC5F,2FAA2F;IAC3F,2FAA2F;IAC3F,2EAA2E;IAC3EN,GAAGM,IAAI,CAAC,CAAC,0IAA0I,EAAEW,SAAS,EAAE,CAAC;IACjK,6FAA6F;IAC7F,qFAAqF;IACrFjB,GAAGM,IAAI,CAAC,CAAC,gGAAgG,CAAC;IAC1GN,GAAGM,IAAI,CAAC;IACR,KAAK,MAAMY,WAAW9B,eAAea,KAAMiB,QAAQC,MAAM,CAACnB;IAC1D,IAAIN,QAAQM,IAAI,sBAAsB,MAAML,QAAQK,IAAI,kBAAkBH;IAC1E,IAAIH,QAAQM,IAAI,gBAAgB,MAAML,QAAQK,IAAI,YAAYf,iBAAiBgB;AACjF;AAEA,OAAO,SAASmB,SAASpB,EAAgB;IACvC,MAAMU,MAAMV,GAAGW,OAAO,CAAC,yCAAyCC,GAAG;IACnE,OAAOF,IAAIW,CAAC;AACd;AAEA,OAAO,SAASC,KAAKrB,GAAmB;QAwETP;IAvE7B,MAAM6B,WAAWzC,KAAKmB,IAAIuB,OAAO,EAAEtC;IACnCN,UAAU2C,UAAU;QAAEE,WAAW;IAAK;IACtC,MAAMC,SAAS5C,KAAKyC,UAAU3B;IAE9B,MAAMI,KAAK,IAAIjB,aAAa2C;IAC5B1B,GAAGM,IAAI,CAAC;IACR,yFAAyF;IACzF,uCAAuC;IACvCN,GAAGM,IAAI,CAAC;IACRjB,kBAAkBW,IAAIhB,gBAAgBiB,SAASE;IAE/CH,GAAGM,IAAI,CAAC;IAER,2FAA2F;IAC3F,8FAA8F;IAC9F,MAAMW,WAAWlB,gBAAgBC,IAAIC;IAErC,uFAAuF;IACvF,qGAAqG;IACrG,MAAM0B,UAAUjC,QAAQM,IAAI;IAC5B,MAAM4B,WAAWlC,QAAQM,IAAI;IAC7B,MAAM6B,eAAe5C,iBAAiBgB;IACtC,IAAI6B,sBAAsB;IAC1B,IAAI,AAACH,YAAY,QAAQA,YAAY9B,kBAAoB+B,aAAa,QAAQA,aAAaC,cAAe;QACxG,uFAAuF;QACvF,wFAAwF;QACxF,IAAIF,YAAY,QAAQA,YAAY9B,gBAAgB;YAClDkC,QAAQC,KAAK,CAAC;YACdhC,GAAGiC,KAAK;YACRC,WAAWjC;YACX,OAAOqB,KAAKrB;QACd;QACA,MAAMkC,cAAc7C,qBAAqBsC,qBAAAA,sBAAAA,WAAY,IAAIC;QACzD,0FAA0F;QAC1F,yFAAyF;QACzF,0EAA0E;QAC1E,IAAIM,YAAYC,IAAI,KAAK,KAAKD,YAAYE,GAAG,CAAC,aAAa;YACzDN,QAAQC,KAAK,CAAC;YACdhC,GAAGM,IAAI,CAAC;YACRwB,sBAAsB;QACxB,OAAO;YACL,MAAMQ,UAAU7C,cAAcmC,qBAAAA,sBAAAA,WAAY,IAAIC;YAC9CE,QAAQC,KAAK,CAAC,CAAC,sBAAsB,EAAEM,QAAQ,oBAAoB,CAAC;YACpEtC,GAAGiC,KAAK;YACRC,WAAWjC;YACX,OAAOqB,KAAKrB;QACd;IACF;IAEA,yFAAyF;IACzF,6FAA6F;IAC7F,wFAAwF;IACxF,8CAA8C;IAC9C,MAAMsC,SAAS9B,eAAeT;IAC9B,IAAIuC,WAAW,QAAQA,WAAWtB,UAAU;QAC1Cc,QAAQC,KAAK,CAAC;QACdhC,GAAGiC,KAAK;QACRC,WAAWjC;QACX,OAAOqB,KAAKrB;IACd;IAEAe,aAAahB,IAAIC,KAAKgB;IAEtB,IAAIuB,kBAA4B,EAAE;IAClC,IAAIV,qBAAqB;QACvBU,kBAAkBjD,oBAAoBS,IAAIC,KAAKA,IAAIuB,OAAO;QAC1D7B,QAAQK,IAAI,YAAY6B;IAC1B;IAEA,wFAAwF;IACxF,2FAA2F;IAC3F,MAAMY,gBAAgBC,QAAOhD,WAAAA,QAAQM,IAAI,iCAAZN,sBAAAA,WAAmC;IAChEM,GAAGM,IAAI,CAAC,CAAC,sBAAsB,EAAEqC,KAAKC,GAAG,CAACD,KAAKE,GAAG,CAAC,OAAO,IAAIJ,gBAAgB,SAAU;IAExF,MAAM,EAAEK,MAAM,EAAEC,QAAQ,EAAE,GAAGvD,UAAUQ,IAAIC,KAAKA,IAAIuB,OAAO;IAE3D,OAAO;QAAExB;QAAIC;QAAKyB;QAAQoB;QAAQC,UAAU;eAAIP;eAAoBO;SAAS;IAAC;AAChF;AAEA,+FAA+F;AAC/F,gGAAgG;AAChG,gGAAgG;AAChG,sDAAsD;AACtD,OAAO,SAASb,WAAWjC,GAAmB;IAC5CpB,OAAOC,KAAKmB,IAAIuB,OAAO,EAAEtC,YAAY;QAAEuC,WAAW;QAAMuB,OAAO;IAAK;AACtE"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/db/open.ts"],"sourcesContent":["// The Node floor (>=22.20) is explained here and nowhere else: 22.20 is the first release with\n// both FTS5 and row-returning INSERT ... RETURNING. Raise it only for a load-bearing capability.\nimport { mkdirSync, rmSync } from 'node:fs';\nimport { join } from 'node:path';\nimport { DatabaseSync } from 'node:sqlite';\nimport type { Config, ResolvedConfig } from '../config/index.ts';\nimport { contentTokenize, featureSignature, STATE_DIR } from '../config/index.ts';\nimport { SenseError } from '../errors.ts';\nimport { activeFeatures } from '../features/index.ts';\nimport { registerFunctions } from '../sql-functions.ts';\nimport { changedSignatureKeys, embedIdentityAdopted, rebuildContentTable, reconcile, signatureDiff } from './reconcile.ts';\nimport { getMeta, setMeta } from './shared.ts';\n\nexport const DB_FILENAME = 'cache.db';\n// Cache shape version, independent of the config's own `version`. Bumping it rebuilds\n// existing trees on first query.\nexport const SCHEMA_VERSION = '17';\n\nexport interface OpenResult {\n db: DatabaseSync;\n cfg: ResolvedConfig;\n dbPath: string;\n parsed: number;\n warnings: string[];\n}\n\n// Stemming is English-only, but the segmentation underneath it is what decides coverage:\n// unicode61 splits on spaces, so a language written without them (Chinese, Japanese, Thai)\n// indexes a whole run as one token and word search finds nothing. `content.tokenize` is how\n// such a tree picks trigram instead.\nconst DEFAULT_TOKENIZE = 'porter unicode61';\n\n// FTS5 takes its tokenizer as a string literal inside DDL, where nothing can bind, so a\n// configured value has to be concatenated. Probing a throwaway table is what makes that safe\n// and is also the whole validation: anything the linked SQLite accepts passes, anything else\n// fails here with SQLite's own message rather than against the real table. It means no table\n// of which version added which tokenizer has to be maintained.\nfunction resolveTokenize(db: DatabaseSync, cfg: Config): string {\n const configured = contentTokenize(cfg);\n if (configured === undefined) return DEFAULT_TOKENIZE;\n const literal = configured.replace(/'/g, \"''\");\n try {\n db.exec('DROP TABLE IF EXISTS temp.sense_tokenize_probe');\n db.exec(`CREATE VIRTUAL TABLE temp.sense_tokenize_probe USING fts5(x, tokenize = '${literal}')`);\n db.exec('DROP TABLE IF EXISTS temp.sense_tokenize_probe');\n } catch (err) {\n throw new SenseError('CONFIG_INVALID', `content.tokenize \"${configured}\" is not a tokenizer this SQLite accepts (${(err as Error).message}); the built-in choices are unicode61, ascii, porter, and trigram, each with their own options`);\n }\n return literal;\n}\n\n// The tokenizer the content table was actually built with, from its own DDL -- the one\n// record that cannot desynchronize from the table. NULL when the table does not exist yet.\nfunction storedTokenize(db: DatabaseSync): string | null {\n const row = db.prepare(`SELECT sql FROM sqlite_master WHERE name = 'content'`).get() as { sql: string } | undefined;\n if (!row) return null;\n const m = row.sql.match(/tokenize = '((?:[^']|'')*)'/);\n return m ? m[1] : null;\n}\n\n// Content is a separate table (not a column on frontmatter) so `SELECT * FROM frontmatter`\n// can't dump file text into context. Features add their own tables after the core ones.\nfunction ensureSchema(db: DatabaseSync, cfg: Config, tokenize: string): void {\n db.exec(`CREATE TABLE IF NOT EXISTS frontmatter (\"path\" TEXT PRIMARY KEY, \"_mtime\" REAL, \"_ctime\" REAL, \"_size\" INTEGER, \"_parse_error\" TEXT)`);\n // IF NOT EXISTS is safe against a tokenizer change: open() compares the table's own DDL\n // against the resolved tokenizer before this runs, so a stale table is already gone by now.\n // The three `_seg` sidecars are appended after path, never inserted: bm25(content, ...) and\n // snippet(content, 2, ...) are documented against the first three columns and keep working\n // (FTS5 defaults the weights it was not given). Each carries its field's exploded unspaced\n // runs for text that needs it, and an empty string for text that does not.\n db.exec(`CREATE VIRTUAL TABLE IF NOT EXISTS content USING fts5(title, summary, text, path UNINDEXED, title_seg, summary_seg, text_seg, tokenize = '${tokenize}')`);\n // Coverage, not ownership: a path can appear under several presets. path leads the PK so the\n // per-doc delete is an index hit -- keyed the other way, cold builds went quadratic.\n db.exec(`CREATE TABLE IF NOT EXISTS preset_files (\"path\" TEXT, preset TEXT, PRIMARY KEY (\"path\", preset))`);\n db.exec('CREATE INDEX IF NOT EXISTS preset_files_preset ON preset_files(preset)');\n for (const feature of activeFeatures(cfg)) feature.schema(db);\n if (getMeta(db, 'schema_version') === null) setMeta(db, 'schema_version', SCHEMA_VERSION);\n if (getMeta(db, 'features') === null) setMeta(db, 'features', featureSignature(cfg));\n}\n\nexport function docCount(db: DatabaseSync): number {\n const row = db.prepare('SELECT COUNT(*) AS n FROM frontmatter').get() as { n: number };\n return row.n;\n}\n\nexport function open(cfg: ResolvedConfig): OpenResult {\n const stateDir = join(cfg.baseDir, STATE_DIR);\n mkdirSync(stateDir, { recursive: true });\n const dbPath = join(stateDir, DB_FILENAME);\n\n const db = new DatabaseSync(dbPath);\n db.exec('PRAGMA journal_mode = WAL');\n // Covers a concurrent watcher's bulk reconcile (~5s for 500 files at 26k notes). A query\n // that outwaits it still fails loudly.\n db.exec('PRAGMA busy_timeout = 30000');\n registerFunctions(db, contentTokenize(cfg) === undefined);\n\n db.exec('CREATE TABLE IF NOT EXISTS meta (key TEXT PRIMARY KEY, value TEXT)');\n\n // Before the rebuild branch below, never after: that branch deletes the cache, so a typo'd\n // tokenizer validated later would cost a full re-index (and re-embed) to reach its own error.\n const tokenize = resolveTokenize(db, cfg);\n\n // Schema-version or feature-set mismatch: reconcile only reparses changed files, so an\n // old cache can't be patched incrementally -- rebuild instead (cheap: nothing expensive lives here).\n const version = getMeta(db, 'schema_version');\n const features = getMeta(db, 'features');\n const wantFeatures = featureSignature(cfg);\n let tokenizeOnlyRebuild = false;\n if ((version !== null && version !== SCHEMA_VERSION) || (features !== null && features !== wantFeatures)) {\n // Indexing derives from presets, so a config edit rebuilding the cache must say so and\n // name what changed -- silent rebuilds make derived indexing look like a hang or a bug.\n if (version !== null && version !== SCHEMA_VERSION) {\n console.error('sense: cache format changed (new sensemaking version); rebuilding the index');\n db.close();\n clearCache(cfg);\n return open(cfg);\n }\n const changedKeys = changedSignatureKeys(features ?? '', wantFeatures);\n // Only the tokenizer moved: frontmatter, links, sections, and embeddings are file-derived\n // and tokenizer-independent, so they don't need re-deriving. Everything else -- a preset\n // edit, an embed model change -- still takes the full clear/reopen below.\n if (changedKeys.size === 1 && changedKeys.has('tokenize')) {\n console.error('sense: config change (content tokenizer) rebuilds the text index; vectors, links, and sections are kept');\n db.exec('DROP TABLE content');\n tokenizeOnlyRebuild = true;\n } else if (changedKeys.size === 1 && changedKeys.has('embed') && embedIdentityAdopted(features ?? '', wantFeatures)) {\n // First sight of a resolved weight identity: the model itself hasn't changed, so\n // adopt it into meta with no rebuild and no re-embed, mirroring the tokenize precedent.\n console.error(\"sense: recorded the embedding model's resolved identity; vectors are unaffected\");\n setMeta(db, 'features', wantFeatures);\n } else {\n const changed = signatureDiff(features ?? '', wantFeatures);\n console.error(`sense: config change (${changed}) rebuilds the index`);\n db.close();\n clearCache(cfg);\n return open(cfg);\n }\n }\n\n // Meta can lie after a crash between table creation and the signature write; the table's\n // own DDL cannot. A mismatch here rebuilds no matter what meta says. A tokenize-only rebuild\n // just dropped content above, so storedTokenize sees no table and this guard is skipped\n // naturally rather than needing its own case.\n const stored = storedTokenize(db);\n if (stored !== null && stored !== tokenize) {\n console.error('sense: cache was built with a different content tokenizer; rebuilding the index');\n db.close();\n clearCache(cfg);\n return open(cfg);\n }\n\n ensureSchema(db, cfg, tokenize);\n\n let rebuildWarnings: string[] = [];\n if (tokenizeOnlyRebuild) {\n rebuildWarnings = rebuildContentTable(db, cfg, cfg.baseDir);\n setMeta(db, 'features', wantFeatures);\n }\n\n // 3x the largest reconcile this cache has recorded, floored at 30s and capped at 10min.\n // Installed before reconcile() -- that call is the one that races a watcher's transaction.\n const recordedMaxMs = Number(getMeta(db, 'reconcile_max_ms') ?? '0');\n db.exec(`PRAGMA busy_timeout = ${Math.min(Math.max(30000, 3 * recordedMaxMs), 600_000)}`);\n\n const { parsed, warnings } = reconcile(db, cfg, cfg.baseDir);\n\n return { db, cfg, dbPath, parsed, warnings: [...rebuildWarnings, ...warnings] };\n}\n\n// Deletes the cache directory, and only that. The rebuild is not this function's job: the next\n// open() reconciles, which is what running any command already does, so a verb that bundled the\n// two described the half that was not its own. Manual reset for a doubted cache; the schema and\n// config-signature mismatches below reset themselves.\nexport function clearCache(cfg: ResolvedConfig): void {\n rmSync(join(cfg.baseDir, STATE_DIR), { recursive: true, force: true });\n}\n"],"names":["mkdirSync","rmSync","join","DatabaseSync","contentTokenize","featureSignature","STATE_DIR","SenseError","activeFeatures","registerFunctions","changedSignatureKeys","embedIdentityAdopted","rebuildContentTable","reconcile","signatureDiff","getMeta","setMeta","DB_FILENAME","SCHEMA_VERSION","DEFAULT_TOKENIZE","resolveTokenize","db","cfg","configured","undefined","literal","replace","exec","err","message","storedTokenize","row","prepare","get","m","sql","match","ensureSchema","tokenize","feature","schema","docCount","n","open","stateDir","baseDir","recursive","dbPath","version","features","wantFeatures","tokenizeOnlyRebuild","console","error","close","clearCache","changedKeys","size","has","changed","stored","rebuildWarnings","recordedMaxMs","Number","Math","min","max","parsed","warnings","force"],"mappings":"AAAA,+FAA+F;AAC/F,iGAAiG;AACjG,SAASA,SAAS,EAAEC,MAAM,QAAQ,UAAU;AAC5C,SAASC,IAAI,QAAQ,YAAY;AACjC,SAASC,YAAY,QAAQ,cAAc;AAE3C,SAASC,eAAe,EAAEC,gBAAgB,EAAEC,SAAS,QAAQ,qBAAqB;AAClF,SAASC,UAAU,QAAQ,eAAe;AAC1C,SAASC,cAAc,QAAQ,uBAAuB;AACtD,SAASC,iBAAiB,QAAQ,sBAAsB;AACxD,SAASC,oBAAoB,EAAEC,oBAAoB,EAAEC,mBAAmB,EAAEC,SAAS,EAAEC,aAAa,QAAQ,iBAAiB;AAC3H,SAASC,OAAO,EAAEC,OAAO,QAAQ,cAAc;AAE/C,OAAO,MAAMC,cAAc,WAAW;AACtC,sFAAsF;AACtF,iCAAiC;AACjC,OAAO,MAAMC,iBAAiB,KAAK;AAUnC,yFAAyF;AACzF,2FAA2F;AAC3F,4FAA4F;AAC5F,qCAAqC;AACrC,MAAMC,mBAAmB;AAEzB,wFAAwF;AACxF,6FAA6F;AAC7F,6FAA6F;AAC7F,6FAA6F;AAC7F,+DAA+D;AAC/D,SAASC,gBAAgBC,EAAgB,EAAEC,GAAW;IACpD,MAAMC,aAAanB,gBAAgBkB;IACnC,IAAIC,eAAeC,WAAW,OAAOL;IACrC,MAAMM,UAAUF,WAAWG,OAAO,CAAC,MAAM;IACzC,IAAI;QACFL,GAAGM,IAAI,CAAC;QACRN,GAAGM,IAAI,CAAC,CAAC,yEAAyE,EAAEF,QAAQ,EAAE,CAAC;QAC/FJ,GAAGM,IAAI,CAAC;IACV,EAAE,OAAOC,KAAK;QACZ,MAAM,IAAIrB,WAAW,kBAAkB,CAAC,kBAAkB,EAAEgB,WAAW,0CAA0C,EAAE,AAACK,IAAcC,OAAO,CAAC,8FAA8F,CAAC;IAC3O;IACA,OAAOJ;AACT;AAEA,uFAAuF;AACvF,2FAA2F;AAC3F,SAASK,eAAeT,EAAgB;IACtC,MAAMU,MAAMV,GAAGW,OAAO,CAAC,CAAC,oDAAoD,CAAC,EAAEC,GAAG;IAClF,IAAI,CAACF,KAAK,OAAO;IACjB,MAAMG,IAAIH,IAAII,GAAG,CAACC,KAAK,CAAC;IACxB,OAAOF,IAAIA,CAAC,CAAC,EAAE,GAAG;AACpB;AAEA,2FAA2F;AAC3F,wFAAwF;AACxF,SAASG,aAAahB,EAAgB,EAAEC,GAAW,EAAEgB,QAAgB;IACnEjB,GAAGM,IAAI,CAAC,CAAC,oIAAoI,CAAC;IAC9I,wFAAwF;IACxF,4FAA4F;IAC5F,4FAA4F;IAC5F,2FAA2F;IAC3F,2FAA2F;IAC3F,2EAA2E;IAC3EN,GAAGM,IAAI,CAAC,CAAC,0IAA0I,EAAEW,SAAS,EAAE,CAAC;IACjK,6FAA6F;IAC7F,qFAAqF;IACrFjB,GAAGM,IAAI,CAAC,CAAC,gGAAgG,CAAC;IAC1GN,GAAGM,IAAI,CAAC;IACR,KAAK,MAAMY,WAAW/B,eAAec,KAAMiB,QAAQC,MAAM,CAACnB;IAC1D,IAAIN,QAAQM,IAAI,sBAAsB,MAAML,QAAQK,IAAI,kBAAkBH;IAC1E,IAAIH,QAAQM,IAAI,gBAAgB,MAAML,QAAQK,IAAI,YAAYhB,iBAAiBiB;AACjF;AAEA,OAAO,SAASmB,SAASpB,EAAgB;IACvC,MAAMU,MAAMV,GAAGW,OAAO,CAAC,yCAAyCC,GAAG;IACnE,OAAOF,IAAIW,CAAC;AACd;AAEA,OAAO,SAASC,KAAKrB,GAAmB;QA6ETP;IA5E7B,MAAM6B,WAAW1C,KAAKoB,IAAIuB,OAAO,EAAEvC;IACnCN,UAAU4C,UAAU;QAAEE,WAAW;IAAK;IACtC,MAAMC,SAAS7C,KAAK0C,UAAU3B;IAE9B,MAAMI,KAAK,IAAIlB,aAAa4C;IAC5B1B,GAAGM,IAAI,CAAC;IACR,yFAAyF;IACzF,uCAAuC;IACvCN,GAAGM,IAAI,CAAC;IACRlB,kBAAkBY,IAAIjB,gBAAgBkB,SAASE;IAE/CH,GAAGM,IAAI,CAAC;IAER,2FAA2F;IAC3F,8FAA8F;IAC9F,MAAMW,WAAWlB,gBAAgBC,IAAIC;IAErC,uFAAuF;IACvF,qGAAqG;IACrG,MAAM0B,UAAUjC,QAAQM,IAAI;IAC5B,MAAM4B,WAAWlC,QAAQM,IAAI;IAC7B,MAAM6B,eAAe7C,iBAAiBiB;IACtC,IAAI6B,sBAAsB;IAC1B,IAAI,AAACH,YAAY,QAAQA,YAAY9B,kBAAoB+B,aAAa,QAAQA,aAAaC,cAAe;QACxG,uFAAuF;QACvF,wFAAwF;QACxF,IAAIF,YAAY,QAAQA,YAAY9B,gBAAgB;YAClDkC,QAAQC,KAAK,CAAC;YACdhC,GAAGiC,KAAK;YACRC,WAAWjC;YACX,OAAOqB,KAAKrB;QACd;QACA,MAAMkC,cAAc9C,qBAAqBuC,qBAAAA,sBAAAA,WAAY,IAAIC;QACzD,0FAA0F;QAC1F,yFAAyF;QACzF,0EAA0E;QAC1E,IAAIM,YAAYC,IAAI,KAAK,KAAKD,YAAYE,GAAG,CAAC,aAAa;YACzDN,QAAQC,KAAK,CAAC;YACdhC,GAAGM,IAAI,CAAC;YACRwB,sBAAsB;QACxB,OAAO,IAAIK,YAAYC,IAAI,KAAK,KAAKD,YAAYE,GAAG,CAAC,YAAY/C,qBAAqBsC,qBAAAA,sBAAAA,WAAY,IAAIC,eAAe;YACnH,iFAAiF;YACjF,wFAAwF;YACxFE,QAAQC,KAAK,CAAC;YACdrC,QAAQK,IAAI,YAAY6B;QAC1B,OAAO;YACL,MAAMS,UAAU7C,cAAcmC,qBAAAA,sBAAAA,WAAY,IAAIC;YAC9CE,QAAQC,KAAK,CAAC,CAAC,sBAAsB,EAAEM,QAAQ,oBAAoB,CAAC;YACpEtC,GAAGiC,KAAK;YACRC,WAAWjC;YACX,OAAOqB,KAAKrB;QACd;IACF;IAEA,yFAAyF;IACzF,6FAA6F;IAC7F,wFAAwF;IACxF,8CAA8C;IAC9C,MAAMsC,SAAS9B,eAAeT;IAC9B,IAAIuC,WAAW,QAAQA,WAAWtB,UAAU;QAC1Cc,QAAQC,KAAK,CAAC;QACdhC,GAAGiC,KAAK;QACRC,WAAWjC;QACX,OAAOqB,KAAKrB;IACd;IAEAe,aAAahB,IAAIC,KAAKgB;IAEtB,IAAIuB,kBAA4B,EAAE;IAClC,IAAIV,qBAAqB;QACvBU,kBAAkBjD,oBAAoBS,IAAIC,KAAKA,IAAIuB,OAAO;QAC1D7B,QAAQK,IAAI,YAAY6B;IAC1B;IAEA,wFAAwF;IACxF,2FAA2F;IAC3F,MAAMY,gBAAgBC,QAAOhD,WAAAA,QAAQM,IAAI,iCAAZN,sBAAAA,WAAmC;IAChEM,GAAGM,IAAI,CAAC,CAAC,sBAAsB,EAAEqC,KAAKC,GAAG,CAACD,KAAKE,GAAG,CAAC,OAAO,IAAIJ,gBAAgB,SAAU;IAExF,MAAM,EAAEK,MAAM,EAAEC,QAAQ,EAAE,GAAGvD,UAAUQ,IAAIC,KAAKA,IAAIuB,OAAO;IAE3D,OAAO;QAAExB;QAAIC;QAAKyB;QAAQoB;QAAQC,UAAU;eAAIP;eAAoBO;SAAS;IAAC;AAChF;AAEA,+FAA+F;AAC/F,gGAAgG;AAChG,gGAAgG;AAChG,sDAAsD;AACtD,OAAO,SAASb,WAAWjC,GAAmB;IAC5CrB,OAAOC,KAAKoB,IAAIuB,OAAO,EAAEvC,YAAY;QAAEwC,WAAW;QAAMuB,OAAO;IAAK;AACtE"}
@@ -5,5 +5,6 @@ export declare function reconcile(db: DatabaseSync, cfg: Config, baseDir: string
5
5
  warnings: string[];
6
6
  };
7
7
  export declare function changedSignatureKeys(before: string, after: string): Set<string>;
8
+ export declare function embedIdentityAdopted(before: string, after: string): boolean;
8
9
  export declare function signatureDiff(before: string, after: string): string;
9
10
  export declare function rebuildContentTable(db: DatabaseSync, cfg: Config, baseDir: string): string[];
@@ -178,6 +178,16 @@ export function changedSignatureKeys(before, after) {
178
178
  for (const key of a.keys())if (!b.has(key)) changed.add(key);
179
179
  return changed;
180
180
  }
181
+ // Whether the embed segment only gained its resolved weight identity: same provider and
182
+ // model, no identity recorded before, one now -- adopted into meta without a rebuild.
183
+ export function embedIdentityAdopted(before, after) {
184
+ const embedPart = (sig)=>sig.split('|').find((p)=>p.startsWith('embed:'));
185
+ const b = embedPart(before);
186
+ const a = embedPart(after);
187
+ if (b === undefined || a === undefined) return false;
188
+ const at = a.indexOf('@');
189
+ return b.indexOf('@') === -1 && at !== -1 && a.slice(0, at) === b;
190
+ }
181
191
  // Names what moved, for the rebuild notice.
182
192
  export function signatureDiff(before, after) {
183
193
  const changed = changedSignatureKeys(before, after);
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/db/reconcile.ts"],"sourcesContent":["import type { DatabaseSync, StatementSync } from 'node:sqlite';\nimport type { Config } from '../config/index.ts';\nimport { contentTokenize } from '../config/index.ts';\nimport { SenseError } from '../errors.ts';\nimport { activeFeatures } from '../features/index.ts';\nimport type { ReconcileDelta } from '../features/types.ts';\nimport { progress } from '../progress.ts';\nimport type { ParsedDoc } from '../scan.ts';\nimport { listFiles, parseFile, RESERVED_COLUMNS } from '../scan.ts';\nimport { segmentField } from '../segment.ts';\nimport { getColumns, getMeta, quoteIdent, setMeta } from './shared.ts';\n\n// Feature-owned columns (`_rank`) must stay out of the upsert: a reparse would null the last\n// computed value on every touch, not just the reconciles that recompute it.\nconst CORE_FRONTMATTER_COLUMNS = new Set(['path', '_mtime', '_ctime', '_size', '_parse_error']);\n\n// SQLite's compile-time SQLITE_MAX_COLUMN, default 2000 (https://www.sqlite.org/limits.html).\nconst MAX_FRONTMATTER_COLUMNS = 2000;\n\n// Shared by reconcile() and the tokenize-only rebuild in open(), so both prepare the same\n// literal instead of two copies drifting apart.\nconst INSERT_CONTENT_SQL = `INSERT INTO content (rowid, title, summary, text, \"path\", title_seg, summary_seg, text_seg) VALUES ((SELECT rowid FROM frontmatter WHERE \"path\" = ?), ?, ?, ?, ?, ?, ?, ?)`;\n\n// Shared by reconcile() and the tokenize-only rebuild in open(), so content population is\n// defined once. Assumes the frontmatter row for doc.relPath already exists (rowid lookup).\nfunction insertContentRow(insertBody: StatementSync, doc: ParsedDoc, segmenting: boolean): void {\n insertBody.run(doc.relPath, doc.search.title, doc.search.summary, doc.search.text, doc.relPath, segmenting ? segmentField(doc.search.title) : '', segmenting ? segmentField(doc.search.summary) : '', segmenting ? segmentField(doc.search.text) : '');\n}\n\nexport function reconcile(db: DatabaseSync, cfg: Config, baseDir: string): { parsed: number; warnings: string[] } {\n const files = listFiles(cfg, baseDir);\n const currentSet = new Set(files.map((f) => f.relPath));\n\n const existingRows = db.prepare(`SELECT \"path\", \"_mtime\", \"_size\" FROM frontmatter`).all() as Array<{\n path: string;\n _mtime: number;\n _size: number;\n }>;\n const existing = new Map(existingRows.map((r) => [r.path, r]));\n const vanished = existingRows.filter((r) => !currentSet.has(r.path)).map((r) => r.path);\n\n const toReparse = files.filter((f) => {\n const row = existing.get(f.relPath);\n return !row || row._mtime !== f.mtimeMs || row._size !== f.size;\n });\n\n if (vanished.length === 0 && toReparse.length === 0) return { parsed: 0, warnings: [] };\n\n const features = activeFeatures(cfg);\n const seenColumns = getColumns(db);\n const newColumns: string[] = [];\n const parsedDocs: ParsedDoc[] = [];\n const warnings: string[] = [];\n\n // Bulk reparses (a sync, a cold build) are the long silences a query can hit; short\n // reconciles stay silent (progress() has a threshold).\n const report = progress('reparsing files', toReparse.length);\n let parsedCount = 0;\n for (const file of toReparse) {\n // A doc only gets extract/store from features that apply to it (currently: embed, via\n // FileStat.embed -- true iff the config names an embedding model).\n const fileFeatures = features.filter((feature) => !feature.enabledForFile || feature.enabledForFile(cfg, file));\n const { doc, warnings: fileWarnings } = parseFile(file, fileFeatures);\n report.tick(++parsedCount);\n warnings.push(...fileWarnings);\n for (const key of Object.keys(doc.data)) {\n if (!seenColumns.has(key)) {\n seenColumns.add(key);\n newColumns.push(key);\n }\n }\n parsedDocs.push(doc);\n }\n report.finish();\n\n const allColumns = [...seenColumns];\n // Fence before ALTERing: SQLite's own failure past this point is a raw\n // \"too many columns on sqlite_altertab_frontmatter\" with no indication of the boundary or the levers.\n if (allColumns.length > MAX_FRONTMATTER_COLUMNS) {\n throw new SenseError(\n 'COLUMN_LIMIT',\n `frontmatter would need ${allColumns.length} columns, crossing SQLite's compile-time SQLITE_MAX_COLUMN limit (default ${MAX_FRONTMATTER_COLUMNS}; see https://www.sqlite.org/limits.html). Narrow the presets' include globs so fewer/other files are indexed, or fix whatever is generating unbounded frontmatter keys.`\n );\n }\n // Columns the frontmatter upsert actually writes: core + parsed frontmatter keys, never a\n // feature-owned reserved column (see CORE_FRONTMATTER_COLUMNS above).\n const writableColumns = allColumns.filter((c) => CORE_FRONTMATTER_COLUMNS.has(c) || !RESERVED_COLUMNS.has(c));\n // ON CONFLICT UPDATE (not OR REPLACE) keeps the row's rowid stable across reparses --\n // content rows are coupled to that rowid below.\n const insertSql = `INSERT INTO frontmatter (${writableColumns.map(quoteIdent).join(', ')}) VALUES (${writableColumns.map(() => '?').join(', ')}) ON CONFLICT(\"path\") DO UPDATE SET ${writableColumns\n .filter((c) => c !== 'path')\n .map((c) => `${quoteIdent(c)} = excluded.${quoteIdent(c)}`)\n .join(', ')}`;\n\n const added = toReparse.filter((f) => !existing.has(f.relPath)).map((f) => f.relPath);\n const delta: ReconcileDelta = { files, reparsed: parsedDocs.map((d) => d.relPath), added, vanished };\n\n const txStart = Date.now();\n db.exec('BEGIN');\n try {\n for (const col of newColumns) db.exec(`ALTER TABLE frontmatter ADD COLUMN ${quoteIdent(col)}`);\n // FTS5 has no upsert, so delete-before-insert into `content`; coupled to the\n // frontmatter rowid (indexed via its PRIMARY KEY) instead of the UNINDEXED `path`\n // column, which a per-row DELETE would otherwise scan the whole table to find.\n const delBody = db.prepare(`DELETE FROM content WHERE rowid = (SELECT rowid FROM frontmatter WHERE \"path\" = ?)`);\n const delPresetFiles = db.prepare(`DELETE FROM preset_files WHERE \"path\" = ?`);\n const insertPresetFile = db.prepare(`INSERT INTO preset_files (\"path\", preset) VALUES (?, ?)`);\n if (vanished.length > 0) {\n const del = db.prepare(`DELETE FROM frontmatter WHERE \"path\" = ?`);\n for (const path of vanished) {\n // content delete must run first: it looks up the frontmatter rowid by path, which\n // the frontmatter delete below would otherwise have already removed.\n delBody.run(path);\n del.run(path);\n delPresetFiles.run(path);\n for (const feature of features) feature.remove?.(db, path, delta);\n }\n }\n if (parsedDocs.length > 0) {\n const insert = db.prepare(insertSql);\n const insertBody = db.prepare(INSERT_CONTENT_SQL);\n // A non-default tokenizer means the tree has chosen its own scheme; a phrase query over\n // grapheme runs would be nonsense against trigram, so the sidecars stay empty.\n const segmenting = contentTokenize(cfg) === undefined;\n for (const doc of parsedDocs) {\n const values = writableColumns.map((col) => {\n if (col === 'path') return doc.relPath;\n if (col === '_mtime') return doc.mtimeMs;\n if (col === '_ctime') return doc.ctimeMs;\n if (col === '_size') return doc.size;\n // Written per parse, unlike _rank, which a feature pass owns and the upsert skips.\n if (col === '_parse_error') return doc.parseError;\n return doc.data[col] ?? null;\n });\n // Frontmatter upsert first: content's rowid lookup below depends on this row existing.\n insert.run(...values);\n // Only for docs that have rows: an unconditional delete made cold crawls quadratic.\n if (existing.has(doc.relPath)) {\n delBody.run(doc.relPath);\n for (const feature of features) feature.remove?.(db, doc.relPath, delta);\n }\n insertContentRow(insertBody, doc, segmenting);\n // A preset edit forces a full rebuild, so an unchanged doc's coverage is already\n // correct; new docs have nothing to clear, which keeps cold builds linear.\n if (existing.has(doc.relPath)) delPresetFiles.run(doc.relPath);\n for (const presetName of doc.presets) insertPresetFile.run(doc.relPath, presetName);\n for (const feature of features) feature.store?.(db, doc.relPath, doc.extracted[feature.name], delta);\n }\n }\n for (const feature of features) feature.afterReconcile?.(db, delta);\n db.exec('COMMIT');\n } catch (err) {\n db.exec('ROLLBACK');\n throw err;\n }\n\n // Reconcile's own write-transaction duration, for open()'s derived busy_timeout (F):\n // keep the observed max so a big watcher reconcile's lock hold is what the next open bounds its wait against.\n const durationMs = Date.now() - txStart;\n const prevRaw = getMeta(db, 'reconcile_max_ms');\n // -1, not 0, so a genuinely 0ms first reconcile (sub-millisecond, common on a tiny tree)\n // still gets recorded instead of losing to the \"nothing recorded yet\" default.\n const prevMax = prevRaw === null ? -1 : Number(prevRaw);\n if (durationMs > prevMax) setMeta(db, 'reconcile_max_ms', String(durationMs));\n\n return { parsed: parsedDocs.length, warnings };\n}\n\n// Segment keys that moved between two feature signatures (see config.featureSignature's\n// format: global features, embed provider, tokenize, then one segment per preset).\nexport function changedSignatureKeys(before: string, after: string): Set<string> {\n const keyOf = (part: string) => (part.startsWith('preset:') ? part.split(':').slice(0, 2).join(':') : part.split(':')[0]);\n const parse = (sig: string) => new Map(sig.split('|').map((part) => [keyOf(part), part]));\n const a = parse(before);\n const b = parse(after);\n const changed = new Set<string>();\n for (const [key, val] of b) if (a.get(key) !== val) changed.add(key);\n for (const key of a.keys()) if (!b.has(key)) changed.add(key);\n return changed;\n}\n\n// Names what moved, for the rebuild notice.\nexport function signatureDiff(before: string, after: string): string {\n const changed = changedSignatureKeys(before, after);\n const label = (key: string) => (key === 'embed' ? 'embed settings' : key === 'tokenize' ? 'content tokenizer' : key.startsWith('preset:') ? `preset \"${key.slice(7)}\"` : 'features');\n return changed.size === 0 ? 'features' : [...changed].map(label).join(', ');\n}\n\n// The tokenize-only rebuild in open(): content is dropped and repopulated from files already\n// listed in frontmatter, which itself is untouched. Frontmatter, links, sections, and\n// embeddings are file-derived and tokenizer-independent, so they survive. No feature extractors\n// run here -- doc.search (title/summary/text) is all content population needs. Returns\n// parseFile's per-file warnings (e.g. a bad date) so open() can surface them -- mtimes are\n// untouched, so reconcile() never reparses these files and would otherwise never emit them again.\nexport function rebuildContentTable(db: DatabaseSync, cfg: Config, baseDir: string): string[] {\n const known = new Set((db.prepare('SELECT \"path\" FROM frontmatter').all() as Array<{ path: string }>).map((r) => r.path));\n const files = listFiles(cfg, baseDir).filter((f) => known.has(f.relPath));\n const segmenting = contentTokenize(cfg) === undefined;\n const insertBody = db.prepare(INSERT_CONTENT_SQL);\n const warnings: string[] = [];\n db.exec('BEGIN');\n try {\n for (const file of files) {\n const { doc, warnings: fileWarnings } = parseFile(file);\n warnings.push(...fileWarnings);\n insertContentRow(insertBody, doc, segmenting);\n }\n db.exec('COMMIT');\n } catch (err) {\n db.exec('ROLLBACK');\n throw err;\n }\n return warnings;\n}\n"],"names":["contentTokenize","SenseError","activeFeatures","progress","listFiles","parseFile","RESERVED_COLUMNS","segmentField","getColumns","getMeta","quoteIdent","setMeta","CORE_FRONTMATTER_COLUMNS","Set","MAX_FRONTMATTER_COLUMNS","INSERT_CONTENT_SQL","insertContentRow","insertBody","doc","segmenting","run","relPath","search","title","summary","text","reconcile","db","cfg","baseDir","files","currentSet","map","f","existingRows","prepare","all","existing","Map","r","path","vanished","filter","has","toReparse","row","get","_mtime","mtimeMs","_size","size","length","parsed","warnings","features","seenColumns","newColumns","parsedDocs","report","parsedCount","file","fileFeatures","feature","enabledForFile","fileWarnings","tick","push","key","Object","keys","data","add","finish","allColumns","writableColumns","c","insertSql","join","added","delta","reparsed","d","txStart","Date","now","exec","col","delBody","delPresetFiles","insertPresetFile","del","remove","insert","undefined","values","ctimeMs","parseError","presetName","presets","store","extracted","name","afterReconcile","err","durationMs","prevRaw","prevMax","Number","String","changedSignatureKeys","before","after","keyOf","part","startsWith","split","slice","parse","sig","a","b","changed","val","signatureDiff","label","rebuildContentTable","known"],"mappings":"AAEA,SAASA,eAAe,QAAQ,qBAAqB;AACrD,SAASC,UAAU,QAAQ,eAAe;AAC1C,SAASC,cAAc,QAAQ,uBAAuB;AAEtD,SAASC,QAAQ,QAAQ,iBAAiB;AAE1C,SAASC,SAAS,EAAEC,SAAS,EAAEC,gBAAgB,QAAQ,aAAa;AACpE,SAASC,YAAY,QAAQ,gBAAgB;AAC7C,SAASC,UAAU,EAAEC,OAAO,EAAEC,UAAU,EAAEC,OAAO,QAAQ,cAAc;AAEvE,6FAA6F;AAC7F,4EAA4E;AAC5E,MAAMC,2BAA2B,IAAIC,IAAI;IAAC;IAAQ;IAAU;IAAU;IAAS;CAAe;AAE9F,8FAA8F;AAC9F,MAAMC,0BAA0B;AAEhC,0FAA0F;AAC1F,gDAAgD;AAChD,MAAMC,qBAAqB,CAAC,0KAA0K,CAAC;AAEvM,0FAA0F;AAC1F,2FAA2F;AAC3F,SAASC,iBAAiBC,UAAyB,EAAEC,GAAc,EAAEC,UAAmB;IACtFF,WAAWG,GAAG,CAACF,IAAIG,OAAO,EAAEH,IAAII,MAAM,CAACC,KAAK,EAAEL,IAAII,MAAM,CAACE,OAAO,EAAEN,IAAII,MAAM,CAACG,IAAI,EAAEP,IAAIG,OAAO,EAAEF,aAAaZ,aAAaW,IAAII,MAAM,CAACC,KAAK,IAAI,IAAIJ,aAAaZ,aAAaW,IAAII,MAAM,CAACE,OAAO,IAAI,IAAIL,aAAaZ,aAAaW,IAAII,MAAM,CAACG,IAAI,IAAI;AACrP;AAEA,OAAO,SAASC,UAAUC,EAAgB,EAAEC,GAAW,EAAEC,OAAe;IACtE,MAAMC,QAAQ1B,UAAUwB,KAAKC;IAC7B,MAAME,aAAa,IAAIlB,IAAIiB,MAAME,GAAG,CAAC,CAACC,IAAMA,EAAEZ,OAAO;IAErD,MAAMa,eAAeP,GAAGQ,OAAO,CAAC,CAAC,iDAAiD,CAAC,EAAEC,GAAG;IAKxF,MAAMC,WAAW,IAAIC,IAAIJ,aAAaF,GAAG,CAAC,CAACO,IAAM;YAACA,EAAEC,IAAI;YAAED;SAAE;IAC5D,MAAME,WAAWP,aAAaQ,MAAM,CAAC,CAACH,IAAM,CAACR,WAAWY,GAAG,CAACJ,EAAEC,IAAI,GAAGR,GAAG,CAAC,CAACO,IAAMA,EAAEC,IAAI;IAEtF,MAAMI,YAAYd,MAAMY,MAAM,CAAC,CAACT;QAC9B,MAAMY,MAAMR,SAASS,GAAG,CAACb,EAAEZ,OAAO;QAClC,OAAO,CAACwB,OAAOA,IAAIE,MAAM,KAAKd,EAAEe,OAAO,IAAIH,IAAII,KAAK,KAAKhB,EAAEiB,IAAI;IACjE;IAEA,IAAIT,SAASU,MAAM,KAAK,KAAKP,UAAUO,MAAM,KAAK,GAAG,OAAO;QAAEC,QAAQ;QAAGC,UAAU,EAAE;IAAC;IAEtF,MAAMC,WAAWpD,eAAe0B;IAChC,MAAM2B,cAAc/C,WAAWmB;IAC/B,MAAM6B,aAAuB,EAAE;IAC/B,MAAMC,aAA0B,EAAE;IAClC,MAAMJ,WAAqB,EAAE;IAE7B,oFAAoF;IACpF,uDAAuD;IACvD,MAAMK,SAASvD,SAAS,mBAAmByC,UAAUO,MAAM;IAC3D,IAAIQ,cAAc;IAClB,KAAK,MAAMC,QAAQhB,UAAW;QAC5B,sFAAsF;QACtF,mEAAmE;QACnE,MAAMiB,eAAeP,SAASZ,MAAM,CAAC,CAACoB,UAAY,CAACA,QAAQC,cAAc,IAAID,QAAQC,cAAc,CAACnC,KAAKgC;QACzG,MAAM,EAAE1C,GAAG,EAAEmC,UAAUW,YAAY,EAAE,GAAG3D,UAAUuD,MAAMC;QACxDH,OAAOO,IAAI,CAAC,EAAEN;QACdN,SAASa,IAAI,IAAIF;QACjB,KAAK,MAAMG,OAAOC,OAAOC,IAAI,CAACnD,IAAIoD,IAAI,EAAG;YACvC,IAAI,CAACf,YAAYZ,GAAG,CAACwB,MAAM;gBACzBZ,YAAYgB,GAAG,CAACJ;gBAChBX,WAAWU,IAAI,CAACC;YAClB;QACF;QACAV,WAAWS,IAAI,CAAChD;IAClB;IACAwC,OAAOc,MAAM;IAEb,MAAMC,aAAa;WAAIlB;KAAY;IACnC,uEAAuE;IACvE,sGAAsG;IACtG,IAAIkB,WAAWtB,MAAM,GAAGrC,yBAAyB;QAC/C,MAAM,IAAIb,WACR,gBACA,CAAC,uBAAuB,EAAEwE,WAAWtB,MAAM,CAAC,0EAA0E,EAAErC,wBAAwB,wKAAwK,CAAC;IAE7T;IACA,0FAA0F;IAC1F,sEAAsE;IACtE,MAAM4D,kBAAkBD,WAAW/B,MAAM,CAAC,CAACiC,IAAM/D,yBAAyB+B,GAAG,CAACgC,MAAM,CAACrE,iBAAiBqC,GAAG,CAACgC;IAC1G,sFAAsF;IACtF,gDAAgD;IAChD,MAAMC,YAAY,CAAC,yBAAyB,EAAEF,gBAAgB1C,GAAG,CAACtB,YAAYmE,IAAI,CAAC,MAAM,UAAU,EAAEH,gBAAgB1C,GAAG,CAAC,IAAM,KAAK6C,IAAI,CAAC,MAAM,oCAAoC,EAAEH,gBAClLhC,MAAM,CAAC,CAACiC,IAAMA,MAAM,QACpB3C,GAAG,CAAC,CAAC2C,IAAM,GAAGjE,WAAWiE,GAAG,YAAY,EAAEjE,WAAWiE,IAAI,EACzDE,IAAI,CAAC,OAAO;IAEf,MAAMC,QAAQlC,UAAUF,MAAM,CAAC,CAACT,IAAM,CAACI,SAASM,GAAG,CAACV,EAAEZ,OAAO,GAAGW,GAAG,CAAC,CAACC,IAAMA,EAAEZ,OAAO;IACpF,MAAM0D,QAAwB;QAAEjD;QAAOkD,UAAUvB,WAAWzB,GAAG,CAAC,CAACiD,IAAMA,EAAE5D,OAAO;QAAGyD;QAAOrC;IAAS;IAEnG,MAAMyC,UAAUC,KAAKC,GAAG;IACxBzD,GAAG0D,IAAI,CAAC;IACR,IAAI;YAkD8BvB;QAjDhC,KAAK,MAAMwB,OAAO9B,WAAY7B,GAAG0D,IAAI,CAAC,CAAC,mCAAmC,EAAE3E,WAAW4E,MAAM;QAC7F,6EAA6E;QAC7E,kFAAkF;QAClF,+EAA+E;QAC/E,MAAMC,UAAU5D,GAAGQ,OAAO,CAAC,CAAC,kFAAkF,CAAC;QAC/G,MAAMqD,iBAAiB7D,GAAGQ,OAAO,CAAC,CAAC,yCAAyC,CAAC;QAC7E,MAAMsD,mBAAmB9D,GAAGQ,OAAO,CAAC,CAAC,uDAAuD,CAAC;QAC7F,IAAIM,SAASU,MAAM,GAAG,GAAG;YACvB,MAAMuC,MAAM/D,GAAGQ,OAAO,CAAC,CAAC,wCAAwC,CAAC;YACjE,KAAK,MAAMK,QAAQC,SAAU;oBAMKqB;gBALhC,kFAAkF;gBAClF,qEAAqE;gBACrEyB,QAAQnE,GAAG,CAACoB;gBACZkD,IAAItE,GAAG,CAACoB;gBACRgD,eAAepE,GAAG,CAACoB;gBACnB,KAAK,MAAMsB,WAAWR,UAAUQ,kBAAAA,QAAQ6B,MAAM,cAAd7B,sCAAAA,qBAAAA,SAAiBnC,IAAIa,MAAMuC;YAC7D;QACF;QACA,IAAItB,WAAWN,MAAM,GAAG,GAAG;YACzB,MAAMyC,SAASjE,GAAGQ,OAAO,CAACyC;YAC1B,MAAM3D,aAAaU,GAAGQ,OAAO,CAACpB;YAC9B,wFAAwF;YACxF,+EAA+E;YAC/E,MAAMI,aAAanB,gBAAgB4B,SAASiE;YAC5C,KAAK,MAAM3E,OAAOuC,WAAY;oBAsBIK;gBArBhC,MAAMgC,SAASpB,gBAAgB1C,GAAG,CAAC,CAACsD;wBAO3BpE;oBANP,IAAIoE,QAAQ,QAAQ,OAAOpE,IAAIG,OAAO;oBACtC,IAAIiE,QAAQ,UAAU,OAAOpE,IAAI8B,OAAO;oBACxC,IAAIsC,QAAQ,UAAU,OAAOpE,IAAI6E,OAAO;oBACxC,IAAIT,QAAQ,SAAS,OAAOpE,IAAIgC,IAAI;oBACpC,mFAAmF;oBACnF,IAAIoC,QAAQ,gBAAgB,OAAOpE,IAAI8E,UAAU;oBACjD,QAAO9E,gBAAAA,IAAIoD,IAAI,CAACgB,IAAI,cAAbpE,2BAAAA,gBAAiB;gBAC1B;gBACA,uFAAuF;gBACvF0E,OAAOxE,GAAG,IAAI0E;gBACd,oFAAoF;gBACpF,IAAIzD,SAASM,GAAG,CAACzB,IAAIG,OAAO,GAAG;wBAEGyC;oBADhCyB,QAAQnE,GAAG,CAACF,IAAIG,OAAO;oBACvB,KAAK,MAAMyC,WAAWR,UAAUQ,mBAAAA,QAAQ6B,MAAM,cAAd7B,uCAAAA,sBAAAA,SAAiBnC,IAAIT,IAAIG,OAAO,EAAE0D;gBACpE;gBACA/D,iBAAiBC,YAAYC,KAAKC;gBAClC,iFAAiF;gBACjF,2EAA2E;gBAC3E,IAAIkB,SAASM,GAAG,CAACzB,IAAIG,OAAO,GAAGmE,eAAepE,GAAG,CAACF,IAAIG,OAAO;gBAC7D,KAAK,MAAM4E,cAAc/E,IAAIgF,OAAO,CAAET,iBAAiBrE,GAAG,CAACF,IAAIG,OAAO,EAAE4E;gBACxE,KAAK,MAAMnC,WAAWR,UAAUQ,iBAAAA,QAAQqC,KAAK,cAAbrC,qCAAAA,oBAAAA,SAAgBnC,IAAIT,IAAIG,OAAO,EAAEH,IAAIkF,SAAS,CAACtC,QAAQuC,IAAI,CAAC,EAAEtB;YAChG;QACF;QACA,KAAK,MAAMjB,WAAWR,UAAUQ,0BAAAA,QAAQwC,cAAc,cAAtBxC,8CAAAA,6BAAAA,SAAyBnC,IAAIoD;QAC7DpD,GAAG0D,IAAI,CAAC;IACV,EAAE,OAAOkB,KAAK;QACZ5E,GAAG0D,IAAI,CAAC;QACR,MAAMkB;IACR;IAEA,qFAAqF;IACrF,8GAA8G;IAC9G,MAAMC,aAAarB,KAAKC,GAAG,KAAKF;IAChC,MAAMuB,UAAUhG,QAAQkB,IAAI;IAC5B,yFAAyF;IACzF,+EAA+E;IAC/E,MAAM+E,UAAUD,YAAY,OAAO,CAAC,IAAIE,OAAOF;IAC/C,IAAID,aAAaE,SAAS/F,QAAQgB,IAAI,oBAAoBiF,OAAOJ;IAEjE,OAAO;QAAEpD,QAAQK,WAAWN,MAAM;QAAEE;IAAS;AAC/C;AAEA,wFAAwF;AACxF,mFAAmF;AACnF,OAAO,SAASwD,qBAAqBC,MAAc,EAAEC,KAAa;IAChE,MAAMC,QAAQ,CAACC,OAAkBA,KAAKC,UAAU,CAAC,aAAaD,KAAKE,KAAK,CAAC,KAAKC,KAAK,CAAC,GAAG,GAAGvC,IAAI,CAAC,OAAOoC,KAAKE,KAAK,CAAC,IAAI,CAAC,EAAE;IACxH,MAAME,QAAQ,CAACC,MAAgB,IAAIhF,IAAIgF,IAAIH,KAAK,CAAC,KAAKnF,GAAG,CAAC,CAACiF,OAAS;gBAACD,MAAMC;gBAAOA;aAAK;IACvF,MAAMM,IAAIF,MAAMP;IAChB,MAAMU,IAAIH,MAAMN;IAChB,MAAMU,UAAU,IAAI5G;IACpB,KAAK,MAAM,CAACsD,KAAKuD,IAAI,IAAIF,EAAG,IAAID,EAAEzE,GAAG,CAACqB,SAASuD,KAAKD,QAAQlD,GAAG,CAACJ;IAChE,KAAK,MAAMA,OAAOoD,EAAElD,IAAI,GAAI,IAAI,CAACmD,EAAE7E,GAAG,CAACwB,MAAMsD,QAAQlD,GAAG,CAACJ;IACzD,OAAOsD;AACT;AAEA,4CAA4C;AAC5C,OAAO,SAASE,cAAcb,MAAc,EAAEC,KAAa;IACzD,MAAMU,UAAUZ,qBAAqBC,QAAQC;IAC7C,MAAMa,QAAQ,CAACzD,MAAiBA,QAAQ,UAAU,mBAAmBA,QAAQ,aAAa,sBAAsBA,IAAI+C,UAAU,CAAC,aAAa,CAAC,QAAQ,EAAE/C,IAAIiD,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG;IACzK,OAAOK,QAAQvE,IAAI,KAAK,IAAI,aAAa;WAAIuE;KAAQ,CAACzF,GAAG,CAAC4F,OAAO/C,IAAI,CAAC;AACxE;AAEA,6FAA6F;AAC7F,sFAAsF;AACtF,gGAAgG;AAChG,uFAAuF;AACvF,2FAA2F;AAC3F,kGAAkG;AAClG,OAAO,SAASgD,oBAAoBlG,EAAgB,EAAEC,GAAW,EAAEC,OAAe;IAChF,MAAMiG,QAAQ,IAAIjH,IAAI,AAACc,GAAGQ,OAAO,CAAC,kCAAkCC,GAAG,GAA+BJ,GAAG,CAAC,CAACO,IAAMA,EAAEC,IAAI;IACvH,MAAMV,QAAQ1B,UAAUwB,KAAKC,SAASa,MAAM,CAAC,CAACT,IAAM6F,MAAMnF,GAAG,CAACV,EAAEZ,OAAO;IACvE,MAAMF,aAAanB,gBAAgB4B,SAASiE;IAC5C,MAAM5E,aAAaU,GAAGQ,OAAO,CAACpB;IAC9B,MAAMsC,WAAqB,EAAE;IAC7B1B,GAAG0D,IAAI,CAAC;IACR,IAAI;QACF,KAAK,MAAMzB,QAAQ9B,MAAO;YACxB,MAAM,EAAEZ,GAAG,EAAEmC,UAAUW,YAAY,EAAE,GAAG3D,UAAUuD;YAClDP,SAASa,IAAI,IAAIF;YACjBhD,iBAAiBC,YAAYC,KAAKC;QACpC;QACAQ,GAAG0D,IAAI,CAAC;IACV,EAAE,OAAOkB,KAAK;QACZ5E,GAAG0D,IAAI,CAAC;QACR,MAAMkB;IACR;IACA,OAAOlD;AACT"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/db/reconcile.ts"],"sourcesContent":["import type { DatabaseSync, StatementSync } from 'node:sqlite';\nimport type { Config } from '../config/index.ts';\nimport { contentTokenize } from '../config/index.ts';\nimport { SenseError } from '../errors.ts';\nimport { activeFeatures } from '../features/index.ts';\nimport type { ReconcileDelta } from '../features/types.ts';\nimport { progress } from '../progress.ts';\nimport type { ParsedDoc } from '../scan.ts';\nimport { listFiles, parseFile, RESERVED_COLUMNS } from '../scan.ts';\nimport { segmentField } from '../segment.ts';\nimport { getColumns, getMeta, quoteIdent, setMeta } from './shared.ts';\n\n// Feature-owned columns (`_rank`) must stay out of the upsert: a reparse would null the last\n// computed value on every touch, not just the reconciles that recompute it.\nconst CORE_FRONTMATTER_COLUMNS = new Set(['path', '_mtime', '_ctime', '_size', '_parse_error']);\n\n// SQLite's compile-time SQLITE_MAX_COLUMN, default 2000 (https://www.sqlite.org/limits.html).\nconst MAX_FRONTMATTER_COLUMNS = 2000;\n\n// Shared by reconcile() and the tokenize-only rebuild in open(), so both prepare the same\n// literal instead of two copies drifting apart.\nconst INSERT_CONTENT_SQL = `INSERT INTO content (rowid, title, summary, text, \"path\", title_seg, summary_seg, text_seg) VALUES ((SELECT rowid FROM frontmatter WHERE \"path\" = ?), ?, ?, ?, ?, ?, ?, ?)`;\n\n// Shared by reconcile() and the tokenize-only rebuild in open(), so content population is\n// defined once. Assumes the frontmatter row for doc.relPath already exists (rowid lookup).\nfunction insertContentRow(insertBody: StatementSync, doc: ParsedDoc, segmenting: boolean): void {\n insertBody.run(doc.relPath, doc.search.title, doc.search.summary, doc.search.text, doc.relPath, segmenting ? segmentField(doc.search.title) : '', segmenting ? segmentField(doc.search.summary) : '', segmenting ? segmentField(doc.search.text) : '');\n}\n\nexport function reconcile(db: DatabaseSync, cfg: Config, baseDir: string): { parsed: number; warnings: string[] } {\n const files = listFiles(cfg, baseDir);\n const currentSet = new Set(files.map((f) => f.relPath));\n\n const existingRows = db.prepare(`SELECT \"path\", \"_mtime\", \"_size\" FROM frontmatter`).all() as Array<{\n path: string;\n _mtime: number;\n _size: number;\n }>;\n const existing = new Map(existingRows.map((r) => [r.path, r]));\n const vanished = existingRows.filter((r) => !currentSet.has(r.path)).map((r) => r.path);\n\n const toReparse = files.filter((f) => {\n const row = existing.get(f.relPath);\n return !row || row._mtime !== f.mtimeMs || row._size !== f.size;\n });\n\n if (vanished.length === 0 && toReparse.length === 0) return { parsed: 0, warnings: [] };\n\n const features = activeFeatures(cfg);\n const seenColumns = getColumns(db);\n const newColumns: string[] = [];\n const parsedDocs: ParsedDoc[] = [];\n const warnings: string[] = [];\n\n // Bulk reparses (a sync, a cold build) are the long silences a query can hit; short\n // reconciles stay silent (progress() has a threshold).\n const report = progress('reparsing files', toReparse.length);\n let parsedCount = 0;\n for (const file of toReparse) {\n // A doc only gets extract/store from features that apply to it (currently: embed, via\n // FileStat.embed -- true iff the config names an embedding model).\n const fileFeatures = features.filter((feature) => !feature.enabledForFile || feature.enabledForFile(cfg, file));\n const { doc, warnings: fileWarnings } = parseFile(file, fileFeatures);\n report.tick(++parsedCount);\n warnings.push(...fileWarnings);\n for (const key of Object.keys(doc.data)) {\n if (!seenColumns.has(key)) {\n seenColumns.add(key);\n newColumns.push(key);\n }\n }\n parsedDocs.push(doc);\n }\n report.finish();\n\n const allColumns = [...seenColumns];\n // Fence before ALTERing: SQLite's own failure past this point is a raw\n // \"too many columns on sqlite_altertab_frontmatter\" with no indication of the boundary or the levers.\n if (allColumns.length > MAX_FRONTMATTER_COLUMNS) {\n throw new SenseError(\n 'COLUMN_LIMIT',\n `frontmatter would need ${allColumns.length} columns, crossing SQLite's compile-time SQLITE_MAX_COLUMN limit (default ${MAX_FRONTMATTER_COLUMNS}; see https://www.sqlite.org/limits.html). Narrow the presets' include globs so fewer/other files are indexed, or fix whatever is generating unbounded frontmatter keys.`\n );\n }\n // Columns the frontmatter upsert actually writes: core + parsed frontmatter keys, never a\n // feature-owned reserved column (see CORE_FRONTMATTER_COLUMNS above).\n const writableColumns = allColumns.filter((c) => CORE_FRONTMATTER_COLUMNS.has(c) || !RESERVED_COLUMNS.has(c));\n // ON CONFLICT UPDATE (not OR REPLACE) keeps the row's rowid stable across reparses --\n // content rows are coupled to that rowid below.\n const insertSql = `INSERT INTO frontmatter (${writableColumns.map(quoteIdent).join(', ')}) VALUES (${writableColumns.map(() => '?').join(', ')}) ON CONFLICT(\"path\") DO UPDATE SET ${writableColumns\n .filter((c) => c !== 'path')\n .map((c) => `${quoteIdent(c)} = excluded.${quoteIdent(c)}`)\n .join(', ')}`;\n\n const added = toReparse.filter((f) => !existing.has(f.relPath)).map((f) => f.relPath);\n const delta: ReconcileDelta = { files, reparsed: parsedDocs.map((d) => d.relPath), added, vanished };\n\n const txStart = Date.now();\n db.exec('BEGIN');\n try {\n for (const col of newColumns) db.exec(`ALTER TABLE frontmatter ADD COLUMN ${quoteIdent(col)}`);\n // FTS5 has no upsert, so delete-before-insert into `content`; coupled to the\n // frontmatter rowid (indexed via its PRIMARY KEY) instead of the UNINDEXED `path`\n // column, which a per-row DELETE would otherwise scan the whole table to find.\n const delBody = db.prepare(`DELETE FROM content WHERE rowid = (SELECT rowid FROM frontmatter WHERE \"path\" = ?)`);\n const delPresetFiles = db.prepare(`DELETE FROM preset_files WHERE \"path\" = ?`);\n const insertPresetFile = db.prepare(`INSERT INTO preset_files (\"path\", preset) VALUES (?, ?)`);\n if (vanished.length > 0) {\n const del = db.prepare(`DELETE FROM frontmatter WHERE \"path\" = ?`);\n for (const path of vanished) {\n // content delete must run first: it looks up the frontmatter rowid by path, which\n // the frontmatter delete below would otherwise have already removed.\n delBody.run(path);\n del.run(path);\n delPresetFiles.run(path);\n for (const feature of features) feature.remove?.(db, path, delta);\n }\n }\n if (parsedDocs.length > 0) {\n const insert = db.prepare(insertSql);\n const insertBody = db.prepare(INSERT_CONTENT_SQL);\n // A non-default tokenizer means the tree has chosen its own scheme; a phrase query over\n // grapheme runs would be nonsense against trigram, so the sidecars stay empty.\n const segmenting = contentTokenize(cfg) === undefined;\n for (const doc of parsedDocs) {\n const values = writableColumns.map((col) => {\n if (col === 'path') return doc.relPath;\n if (col === '_mtime') return doc.mtimeMs;\n if (col === '_ctime') return doc.ctimeMs;\n if (col === '_size') return doc.size;\n // Written per parse, unlike _rank, which a feature pass owns and the upsert skips.\n if (col === '_parse_error') return doc.parseError;\n return doc.data[col] ?? null;\n });\n // Frontmatter upsert first: content's rowid lookup below depends on this row existing.\n insert.run(...values);\n // Only for docs that have rows: an unconditional delete made cold crawls quadratic.\n if (existing.has(doc.relPath)) {\n delBody.run(doc.relPath);\n for (const feature of features) feature.remove?.(db, doc.relPath, delta);\n }\n insertContentRow(insertBody, doc, segmenting);\n // A preset edit forces a full rebuild, so an unchanged doc's coverage is already\n // correct; new docs have nothing to clear, which keeps cold builds linear.\n if (existing.has(doc.relPath)) delPresetFiles.run(doc.relPath);\n for (const presetName of doc.presets) insertPresetFile.run(doc.relPath, presetName);\n for (const feature of features) feature.store?.(db, doc.relPath, doc.extracted[feature.name], delta);\n }\n }\n for (const feature of features) feature.afterReconcile?.(db, delta);\n db.exec('COMMIT');\n } catch (err) {\n db.exec('ROLLBACK');\n throw err;\n }\n\n // Reconcile's own write-transaction duration, for open()'s derived busy_timeout (F):\n // keep the observed max so a big watcher reconcile's lock hold is what the next open bounds its wait against.\n const durationMs = Date.now() - txStart;\n const prevRaw = getMeta(db, 'reconcile_max_ms');\n // -1, not 0, so a genuinely 0ms first reconcile (sub-millisecond, common on a tiny tree)\n // still gets recorded instead of losing to the \"nothing recorded yet\" default.\n const prevMax = prevRaw === null ? -1 : Number(prevRaw);\n if (durationMs > prevMax) setMeta(db, 'reconcile_max_ms', String(durationMs));\n\n return { parsed: parsedDocs.length, warnings };\n}\n\n// Segment keys that moved between two feature signatures (see config.featureSignature's\n// format: global features, embed provider, tokenize, then one segment per preset).\nexport function changedSignatureKeys(before: string, after: string): Set<string> {\n const keyOf = (part: string) => (part.startsWith('preset:') ? part.split(':').slice(0, 2).join(':') : part.split(':')[0]);\n const parse = (sig: string) => new Map(sig.split('|').map((part) => [keyOf(part), part]));\n const a = parse(before);\n const b = parse(after);\n const changed = new Set<string>();\n for (const [key, val] of b) if (a.get(key) !== val) changed.add(key);\n for (const key of a.keys()) if (!b.has(key)) changed.add(key);\n return changed;\n}\n\n// Whether the embed segment only gained its resolved weight identity: same provider and\n// model, no identity recorded before, one now -- adopted into meta without a rebuild.\nexport function embedIdentityAdopted(before: string, after: string): boolean {\n const embedPart = (sig: string) => sig.split('|').find((p) => p.startsWith('embed:'));\n const b = embedPart(before);\n const a = embedPart(after);\n if (b === undefined || a === undefined) return false;\n const at = a.indexOf('@');\n return b.indexOf('@') === -1 && at !== -1 && a.slice(0, at) === b;\n}\n\n// Names what moved, for the rebuild notice.\nexport function signatureDiff(before: string, after: string): string {\n const changed = changedSignatureKeys(before, after);\n const label = (key: string) => (key === 'embed' ? 'embed settings' : key === 'tokenize' ? 'content tokenizer' : key.startsWith('preset:') ? `preset \"${key.slice(7)}\"` : 'features');\n return changed.size === 0 ? 'features' : [...changed].map(label).join(', ');\n}\n\n// The tokenize-only rebuild in open(): content is dropped and repopulated from files already\n// listed in frontmatter, which itself is untouched. Frontmatter, links, sections, and\n// embeddings are file-derived and tokenizer-independent, so they survive. No feature extractors\n// run here -- doc.search (title/summary/text) is all content population needs. Returns\n// parseFile's per-file warnings (e.g. a bad date) so open() can surface them -- mtimes are\n// untouched, so reconcile() never reparses these files and would otherwise never emit them again.\nexport function rebuildContentTable(db: DatabaseSync, cfg: Config, baseDir: string): string[] {\n const known = new Set((db.prepare('SELECT \"path\" FROM frontmatter').all() as Array<{ path: string }>).map((r) => r.path));\n const files = listFiles(cfg, baseDir).filter((f) => known.has(f.relPath));\n const segmenting = contentTokenize(cfg) === undefined;\n const insertBody = db.prepare(INSERT_CONTENT_SQL);\n const warnings: string[] = [];\n db.exec('BEGIN');\n try {\n for (const file of files) {\n const { doc, warnings: fileWarnings } = parseFile(file);\n warnings.push(...fileWarnings);\n insertContentRow(insertBody, doc, segmenting);\n }\n db.exec('COMMIT');\n } catch (err) {\n db.exec('ROLLBACK');\n throw err;\n }\n return warnings;\n}\n"],"names":["contentTokenize","SenseError","activeFeatures","progress","listFiles","parseFile","RESERVED_COLUMNS","segmentField","getColumns","getMeta","quoteIdent","setMeta","CORE_FRONTMATTER_COLUMNS","Set","MAX_FRONTMATTER_COLUMNS","INSERT_CONTENT_SQL","insertContentRow","insertBody","doc","segmenting","run","relPath","search","title","summary","text","reconcile","db","cfg","baseDir","files","currentSet","map","f","existingRows","prepare","all","existing","Map","r","path","vanished","filter","has","toReparse","row","get","_mtime","mtimeMs","_size","size","length","parsed","warnings","features","seenColumns","newColumns","parsedDocs","report","parsedCount","file","fileFeatures","feature","enabledForFile","fileWarnings","tick","push","key","Object","keys","data","add","finish","allColumns","writableColumns","c","insertSql","join","added","delta","reparsed","d","txStart","Date","now","exec","col","delBody","delPresetFiles","insertPresetFile","del","remove","insert","undefined","values","ctimeMs","parseError","presetName","presets","store","extracted","name","afterReconcile","err","durationMs","prevRaw","prevMax","Number","String","changedSignatureKeys","before","after","keyOf","part","startsWith","split","slice","parse","sig","a","b","changed","val","embedIdentityAdopted","embedPart","find","p","at","indexOf","signatureDiff","label","rebuildContentTable","known"],"mappings":"AAEA,SAASA,eAAe,QAAQ,qBAAqB;AACrD,SAASC,UAAU,QAAQ,eAAe;AAC1C,SAASC,cAAc,QAAQ,uBAAuB;AAEtD,SAASC,QAAQ,QAAQ,iBAAiB;AAE1C,SAASC,SAAS,EAAEC,SAAS,EAAEC,gBAAgB,QAAQ,aAAa;AACpE,SAASC,YAAY,QAAQ,gBAAgB;AAC7C,SAASC,UAAU,EAAEC,OAAO,EAAEC,UAAU,EAAEC,OAAO,QAAQ,cAAc;AAEvE,6FAA6F;AAC7F,4EAA4E;AAC5E,MAAMC,2BAA2B,IAAIC,IAAI;IAAC;IAAQ;IAAU;IAAU;IAAS;CAAe;AAE9F,8FAA8F;AAC9F,MAAMC,0BAA0B;AAEhC,0FAA0F;AAC1F,gDAAgD;AAChD,MAAMC,qBAAqB,CAAC,0KAA0K,CAAC;AAEvM,0FAA0F;AAC1F,2FAA2F;AAC3F,SAASC,iBAAiBC,UAAyB,EAAEC,GAAc,EAAEC,UAAmB;IACtFF,WAAWG,GAAG,CAACF,IAAIG,OAAO,EAAEH,IAAII,MAAM,CAACC,KAAK,EAAEL,IAAII,MAAM,CAACE,OAAO,EAAEN,IAAII,MAAM,CAACG,IAAI,EAAEP,IAAIG,OAAO,EAAEF,aAAaZ,aAAaW,IAAII,MAAM,CAACC,KAAK,IAAI,IAAIJ,aAAaZ,aAAaW,IAAII,MAAM,CAACE,OAAO,IAAI,IAAIL,aAAaZ,aAAaW,IAAII,MAAM,CAACG,IAAI,IAAI;AACrP;AAEA,OAAO,SAASC,UAAUC,EAAgB,EAAEC,GAAW,EAAEC,OAAe;IACtE,MAAMC,QAAQ1B,UAAUwB,KAAKC;IAC7B,MAAME,aAAa,IAAIlB,IAAIiB,MAAME,GAAG,CAAC,CAACC,IAAMA,EAAEZ,OAAO;IAErD,MAAMa,eAAeP,GAAGQ,OAAO,CAAC,CAAC,iDAAiD,CAAC,EAAEC,GAAG;IAKxF,MAAMC,WAAW,IAAIC,IAAIJ,aAAaF,GAAG,CAAC,CAACO,IAAM;YAACA,EAAEC,IAAI;YAAED;SAAE;IAC5D,MAAME,WAAWP,aAAaQ,MAAM,CAAC,CAACH,IAAM,CAACR,WAAWY,GAAG,CAACJ,EAAEC,IAAI,GAAGR,GAAG,CAAC,CAACO,IAAMA,EAAEC,IAAI;IAEtF,MAAMI,YAAYd,MAAMY,MAAM,CAAC,CAACT;QAC9B,MAAMY,MAAMR,SAASS,GAAG,CAACb,EAAEZ,OAAO;QAClC,OAAO,CAACwB,OAAOA,IAAIE,MAAM,KAAKd,EAAEe,OAAO,IAAIH,IAAII,KAAK,KAAKhB,EAAEiB,IAAI;IACjE;IAEA,IAAIT,SAASU,MAAM,KAAK,KAAKP,UAAUO,MAAM,KAAK,GAAG,OAAO;QAAEC,QAAQ;QAAGC,UAAU,EAAE;IAAC;IAEtF,MAAMC,WAAWpD,eAAe0B;IAChC,MAAM2B,cAAc/C,WAAWmB;IAC/B,MAAM6B,aAAuB,EAAE;IAC/B,MAAMC,aAA0B,EAAE;IAClC,MAAMJ,WAAqB,EAAE;IAE7B,oFAAoF;IACpF,uDAAuD;IACvD,MAAMK,SAASvD,SAAS,mBAAmByC,UAAUO,MAAM;IAC3D,IAAIQ,cAAc;IAClB,KAAK,MAAMC,QAAQhB,UAAW;QAC5B,sFAAsF;QACtF,mEAAmE;QACnE,MAAMiB,eAAeP,SAASZ,MAAM,CAAC,CAACoB,UAAY,CAACA,QAAQC,cAAc,IAAID,QAAQC,cAAc,CAACnC,KAAKgC;QACzG,MAAM,EAAE1C,GAAG,EAAEmC,UAAUW,YAAY,EAAE,GAAG3D,UAAUuD,MAAMC;QACxDH,OAAOO,IAAI,CAAC,EAAEN;QACdN,SAASa,IAAI,IAAIF;QACjB,KAAK,MAAMG,OAAOC,OAAOC,IAAI,CAACnD,IAAIoD,IAAI,EAAG;YACvC,IAAI,CAACf,YAAYZ,GAAG,CAACwB,MAAM;gBACzBZ,YAAYgB,GAAG,CAACJ;gBAChBX,WAAWU,IAAI,CAACC;YAClB;QACF;QACAV,WAAWS,IAAI,CAAChD;IAClB;IACAwC,OAAOc,MAAM;IAEb,MAAMC,aAAa;WAAIlB;KAAY;IACnC,uEAAuE;IACvE,sGAAsG;IACtG,IAAIkB,WAAWtB,MAAM,GAAGrC,yBAAyB;QAC/C,MAAM,IAAIb,WACR,gBACA,CAAC,uBAAuB,EAAEwE,WAAWtB,MAAM,CAAC,0EAA0E,EAAErC,wBAAwB,wKAAwK,CAAC;IAE7T;IACA,0FAA0F;IAC1F,sEAAsE;IACtE,MAAM4D,kBAAkBD,WAAW/B,MAAM,CAAC,CAACiC,IAAM/D,yBAAyB+B,GAAG,CAACgC,MAAM,CAACrE,iBAAiBqC,GAAG,CAACgC;IAC1G,sFAAsF;IACtF,gDAAgD;IAChD,MAAMC,YAAY,CAAC,yBAAyB,EAAEF,gBAAgB1C,GAAG,CAACtB,YAAYmE,IAAI,CAAC,MAAM,UAAU,EAAEH,gBAAgB1C,GAAG,CAAC,IAAM,KAAK6C,IAAI,CAAC,MAAM,oCAAoC,EAAEH,gBAClLhC,MAAM,CAAC,CAACiC,IAAMA,MAAM,QACpB3C,GAAG,CAAC,CAAC2C,IAAM,GAAGjE,WAAWiE,GAAG,YAAY,EAAEjE,WAAWiE,IAAI,EACzDE,IAAI,CAAC,OAAO;IAEf,MAAMC,QAAQlC,UAAUF,MAAM,CAAC,CAACT,IAAM,CAACI,SAASM,GAAG,CAACV,EAAEZ,OAAO,GAAGW,GAAG,CAAC,CAACC,IAAMA,EAAEZ,OAAO;IACpF,MAAM0D,QAAwB;QAAEjD;QAAOkD,UAAUvB,WAAWzB,GAAG,CAAC,CAACiD,IAAMA,EAAE5D,OAAO;QAAGyD;QAAOrC;IAAS;IAEnG,MAAMyC,UAAUC,KAAKC,GAAG;IACxBzD,GAAG0D,IAAI,CAAC;IACR,IAAI;YAkD8BvB;QAjDhC,KAAK,MAAMwB,OAAO9B,WAAY7B,GAAG0D,IAAI,CAAC,CAAC,mCAAmC,EAAE3E,WAAW4E,MAAM;QAC7F,6EAA6E;QAC7E,kFAAkF;QAClF,+EAA+E;QAC/E,MAAMC,UAAU5D,GAAGQ,OAAO,CAAC,CAAC,kFAAkF,CAAC;QAC/G,MAAMqD,iBAAiB7D,GAAGQ,OAAO,CAAC,CAAC,yCAAyC,CAAC;QAC7E,MAAMsD,mBAAmB9D,GAAGQ,OAAO,CAAC,CAAC,uDAAuD,CAAC;QAC7F,IAAIM,SAASU,MAAM,GAAG,GAAG;YACvB,MAAMuC,MAAM/D,GAAGQ,OAAO,CAAC,CAAC,wCAAwC,CAAC;YACjE,KAAK,MAAMK,QAAQC,SAAU;oBAMKqB;gBALhC,kFAAkF;gBAClF,qEAAqE;gBACrEyB,QAAQnE,GAAG,CAACoB;gBACZkD,IAAItE,GAAG,CAACoB;gBACRgD,eAAepE,GAAG,CAACoB;gBACnB,KAAK,MAAMsB,WAAWR,UAAUQ,kBAAAA,QAAQ6B,MAAM,cAAd7B,sCAAAA,qBAAAA,SAAiBnC,IAAIa,MAAMuC;YAC7D;QACF;QACA,IAAItB,WAAWN,MAAM,GAAG,GAAG;YACzB,MAAMyC,SAASjE,GAAGQ,OAAO,CAACyC;YAC1B,MAAM3D,aAAaU,GAAGQ,OAAO,CAACpB;YAC9B,wFAAwF;YACxF,+EAA+E;YAC/E,MAAMI,aAAanB,gBAAgB4B,SAASiE;YAC5C,KAAK,MAAM3E,OAAOuC,WAAY;oBAsBIK;gBArBhC,MAAMgC,SAASpB,gBAAgB1C,GAAG,CAAC,CAACsD;wBAO3BpE;oBANP,IAAIoE,QAAQ,QAAQ,OAAOpE,IAAIG,OAAO;oBACtC,IAAIiE,QAAQ,UAAU,OAAOpE,IAAI8B,OAAO;oBACxC,IAAIsC,QAAQ,UAAU,OAAOpE,IAAI6E,OAAO;oBACxC,IAAIT,QAAQ,SAAS,OAAOpE,IAAIgC,IAAI;oBACpC,mFAAmF;oBACnF,IAAIoC,QAAQ,gBAAgB,OAAOpE,IAAI8E,UAAU;oBACjD,QAAO9E,gBAAAA,IAAIoD,IAAI,CAACgB,IAAI,cAAbpE,2BAAAA,gBAAiB;gBAC1B;gBACA,uFAAuF;gBACvF0E,OAAOxE,GAAG,IAAI0E;gBACd,oFAAoF;gBACpF,IAAIzD,SAASM,GAAG,CAACzB,IAAIG,OAAO,GAAG;wBAEGyC;oBADhCyB,QAAQnE,GAAG,CAACF,IAAIG,OAAO;oBACvB,KAAK,MAAMyC,WAAWR,UAAUQ,mBAAAA,QAAQ6B,MAAM,cAAd7B,uCAAAA,sBAAAA,SAAiBnC,IAAIT,IAAIG,OAAO,EAAE0D;gBACpE;gBACA/D,iBAAiBC,YAAYC,KAAKC;gBAClC,iFAAiF;gBACjF,2EAA2E;gBAC3E,IAAIkB,SAASM,GAAG,CAACzB,IAAIG,OAAO,GAAGmE,eAAepE,GAAG,CAACF,IAAIG,OAAO;gBAC7D,KAAK,MAAM4E,cAAc/E,IAAIgF,OAAO,CAAET,iBAAiBrE,GAAG,CAACF,IAAIG,OAAO,EAAE4E;gBACxE,KAAK,MAAMnC,WAAWR,UAAUQ,iBAAAA,QAAQqC,KAAK,cAAbrC,qCAAAA,oBAAAA,SAAgBnC,IAAIT,IAAIG,OAAO,EAAEH,IAAIkF,SAAS,CAACtC,QAAQuC,IAAI,CAAC,EAAEtB;YAChG;QACF;QACA,KAAK,MAAMjB,WAAWR,UAAUQ,0BAAAA,QAAQwC,cAAc,cAAtBxC,8CAAAA,6BAAAA,SAAyBnC,IAAIoD;QAC7DpD,GAAG0D,IAAI,CAAC;IACV,EAAE,OAAOkB,KAAK;QACZ5E,GAAG0D,IAAI,CAAC;QACR,MAAMkB;IACR;IAEA,qFAAqF;IACrF,8GAA8G;IAC9G,MAAMC,aAAarB,KAAKC,GAAG,KAAKF;IAChC,MAAMuB,UAAUhG,QAAQkB,IAAI;IAC5B,yFAAyF;IACzF,+EAA+E;IAC/E,MAAM+E,UAAUD,YAAY,OAAO,CAAC,IAAIE,OAAOF;IAC/C,IAAID,aAAaE,SAAS/F,QAAQgB,IAAI,oBAAoBiF,OAAOJ;IAEjE,OAAO;QAAEpD,QAAQK,WAAWN,MAAM;QAAEE;IAAS;AAC/C;AAEA,wFAAwF;AACxF,mFAAmF;AACnF,OAAO,SAASwD,qBAAqBC,MAAc,EAAEC,KAAa;IAChE,MAAMC,QAAQ,CAACC,OAAkBA,KAAKC,UAAU,CAAC,aAAaD,KAAKE,KAAK,CAAC,KAAKC,KAAK,CAAC,GAAG,GAAGvC,IAAI,CAAC,OAAOoC,KAAKE,KAAK,CAAC,IAAI,CAAC,EAAE;IACxH,MAAME,QAAQ,CAACC,MAAgB,IAAIhF,IAAIgF,IAAIH,KAAK,CAAC,KAAKnF,GAAG,CAAC,CAACiF,OAAS;gBAACD,MAAMC;gBAAOA;aAAK;IACvF,MAAMM,IAAIF,MAAMP;IAChB,MAAMU,IAAIH,MAAMN;IAChB,MAAMU,UAAU,IAAI5G;IACpB,KAAK,MAAM,CAACsD,KAAKuD,IAAI,IAAIF,EAAG,IAAID,EAAEzE,GAAG,CAACqB,SAASuD,KAAKD,QAAQlD,GAAG,CAACJ;IAChE,KAAK,MAAMA,OAAOoD,EAAElD,IAAI,GAAI,IAAI,CAACmD,EAAE7E,GAAG,CAACwB,MAAMsD,QAAQlD,GAAG,CAACJ;IACzD,OAAOsD;AACT;AAEA,wFAAwF;AACxF,sFAAsF;AACtF,OAAO,SAASE,qBAAqBb,MAAc,EAAEC,KAAa;IAChE,MAAMa,YAAY,CAACN,MAAgBA,IAAIH,KAAK,CAAC,KAAKU,IAAI,CAAC,CAACC,IAAMA,EAAEZ,UAAU,CAAC;IAC3E,MAAMM,IAAII,UAAUd;IACpB,MAAMS,IAAIK,UAAUb;IACpB,IAAIS,MAAM3B,aAAa0B,MAAM1B,WAAW,OAAO;IAC/C,MAAMkC,KAAKR,EAAES,OAAO,CAAC;IACrB,OAAOR,EAAEQ,OAAO,CAAC,SAAS,CAAC,KAAKD,OAAO,CAAC,KAAKR,EAAEH,KAAK,CAAC,GAAGW,QAAQP;AAClE;AAEA,4CAA4C;AAC5C,OAAO,SAASS,cAAcnB,MAAc,EAAEC,KAAa;IACzD,MAAMU,UAAUZ,qBAAqBC,QAAQC;IAC7C,MAAMmB,QAAQ,CAAC/D,MAAiBA,QAAQ,UAAU,mBAAmBA,QAAQ,aAAa,sBAAsBA,IAAI+C,UAAU,CAAC,aAAa,CAAC,QAAQ,EAAE/C,IAAIiD,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG;IACzK,OAAOK,QAAQvE,IAAI,KAAK,IAAI,aAAa;WAAIuE;KAAQ,CAACzF,GAAG,CAACkG,OAAOrD,IAAI,CAAC;AACxE;AAEA,6FAA6F;AAC7F,sFAAsF;AACtF,gGAAgG;AAChG,uFAAuF;AACvF,2FAA2F;AAC3F,kGAAkG;AAClG,OAAO,SAASsD,oBAAoBxG,EAAgB,EAAEC,GAAW,EAAEC,OAAe;IAChF,MAAMuG,QAAQ,IAAIvH,IAAI,AAACc,GAAGQ,OAAO,CAAC,kCAAkCC,GAAG,GAA+BJ,GAAG,CAAC,CAACO,IAAMA,EAAEC,IAAI;IACvH,MAAMV,QAAQ1B,UAAUwB,KAAKC,SAASa,MAAM,CAAC,CAACT,IAAMmG,MAAMzF,GAAG,CAACV,EAAEZ,OAAO;IACvE,MAAMF,aAAanB,gBAAgB4B,SAASiE;IAC5C,MAAM5E,aAAaU,GAAGQ,OAAO,CAACpB;IAC9B,MAAMsC,WAAqB,EAAE;IAC7B1B,GAAG0D,IAAI,CAAC;IACR,IAAI;QACF,KAAK,MAAMzB,QAAQ9B,MAAO;YACxB,MAAM,EAAEZ,GAAG,EAAEmC,UAAUW,YAAY,EAAE,GAAG3D,UAAUuD;YAClDP,SAASa,IAAI,IAAIF;YACjBhD,iBAAiBC,YAAYC,KAAKC;QACpC;QACAQ,GAAG0D,IAAI,CAAC;IACV,EAAE,OAAOkB,KAAK;QACZ5E,GAAG0D,IAAI,CAAC;QACR,MAAMkB;IACR;IACA,OAAOlD;AACT"}
@@ -0,0 +1,2 @@
1
+ import type { EmbedProvider } from './types.js';
2
+ export declare function cohereProvider(model: string, url: string | undefined, keyEnv: string | undefined): Promise<EmbedProvider>;
@@ -0,0 +1,46 @@
1
+ import { SenseError } from '../errors.js';
2
+ import { fetchWithRetry } from './http.js';
3
+ const BATCH_CAP = 96;
4
+ // Cohere's native /v2/embed: input_type distinguishes doc vs query embeddings, which the
5
+ // OpenAI-compatible shape cannot express.
6
+ export async function cohereProvider(model, url, keyEnv) {
7
+ const base = (url !== null && url !== void 0 ? url : 'https://api.cohere.com').replace(/\/+$/, '');
8
+ const headers = {
9
+ 'content-type': 'application/json'
10
+ };
11
+ const key = keyEnv ? process.env[keyEnv] : undefined;
12
+ if (key) headers.authorization = `Bearer ${key}`;
13
+ async function post(texts, inputType) {
14
+ const res = await fetchWithRetry(`${base}/v2/embed`, {
15
+ method: 'POST',
16
+ headers,
17
+ body: JSON.stringify({
18
+ model,
19
+ input_type: inputType,
20
+ texts,
21
+ embedding_types: [
22
+ 'float'
23
+ ]
24
+ })
25
+ });
26
+ if (!res.ok) throw new SenseError('EMBED_MODEL', `${base}/v2/embed -> HTTP ${res.status}`);
27
+ const body = await res.json();
28
+ // Cohere returns one vector per text in order, with no index field to re-sort by; the
29
+ // count is what a caller mapping positionally still has to be able to trust.
30
+ const vectors = body.embeddings.float;
31
+ if (vectors.length !== texts.length) throw new SenseError('EMBED_MODEL', `${base}/v2/embed returned ${vectors.length} embeddings for ${texts.length} inputs`);
32
+ return vectors.map((v)=>Float32Array.from(v));
33
+ }
34
+ const dims = (await post([
35
+ 'dimension probe'
36
+ ], 'search_query'))[0].length;
37
+ return {
38
+ id: `cohere:${base}:${model}`,
39
+ dims,
40
+ batchCap: BATCH_CAP,
41
+ embedDocuments: (texts)=>post(texts, 'search_document'),
42
+ embedQuery: async (text)=>(await post([
43
+ text
44
+ ], 'search_query'))[0]
45
+ };
46
+ }
@@ -0,0 +1 @@
1
+ {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/embed/cohere.ts"],"sourcesContent":["import { SenseError } from '../errors.ts';\nimport { fetchWithRetry } from './http.ts';\nimport type { EmbedProvider } from './types.ts';\n\nconst BATCH_CAP = 96;\n\n// Cohere's native /v2/embed: input_type distinguishes doc vs query embeddings, which the\n// OpenAI-compatible shape cannot express.\nexport async function cohereProvider(model: string, url: string | undefined, keyEnv: string | undefined): Promise<EmbedProvider> {\n const base = (url ?? 'https://api.cohere.com').replace(/\\/+$/, '');\n const headers: Record<string, string> = { 'content-type': 'application/json' };\n const key = keyEnv ? process.env[keyEnv] : undefined;\n if (key) headers.authorization = `Bearer ${key}`;\n\n async function post(texts: string[], inputType: 'search_document' | 'search_query'): Promise<Float32Array[]> {\n const res = await fetchWithRetry(`${base}/v2/embed`, { method: 'POST', headers, body: JSON.stringify({ model, input_type: inputType, texts, embedding_types: ['float'] }) });\n if (!res.ok) throw new SenseError('EMBED_MODEL', `${base}/v2/embed -> HTTP ${res.status}`);\n const body = (await res.json()) as { embeddings: { float: number[][] } };\n // Cohere returns one vector per text in order, with no index field to re-sort by; the\n // count is what a caller mapping positionally still has to be able to trust.\n const vectors = body.embeddings.float;\n if (vectors.length !== texts.length) throw new SenseError('EMBED_MODEL', `${base}/v2/embed returned ${vectors.length} embeddings for ${texts.length} inputs`);\n return vectors.map((v) => Float32Array.from(v));\n }\n\n const dims = (await post(['dimension probe'], 'search_query'))[0].length;\n return { id: `cohere:${base}:${model}`, dims, batchCap: BATCH_CAP, embedDocuments: (texts) => post(texts, 'search_document'), embedQuery: async (text) => (await post([text], 'search_query'))[0] };\n}\n"],"names":["SenseError","fetchWithRetry","BATCH_CAP","cohereProvider","model","url","keyEnv","base","replace","headers","key","process","env","undefined","authorization","post","texts","inputType","res","method","body","JSON","stringify","input_type","embedding_types","ok","status","json","vectors","embeddings","float","length","map","v","Float32Array","from","dims","id","batchCap","embedDocuments","embedQuery","text"],"mappings":"AAAA,SAASA,UAAU,QAAQ,eAAe;AAC1C,SAASC,cAAc,QAAQ,YAAY;AAG3C,MAAMC,YAAY;AAElB,yFAAyF;AACzF,0CAA0C;AAC1C,OAAO,eAAeC,eAAeC,KAAa,EAAEC,GAAuB,EAAEC,MAA0B;IACrG,MAAMC,OAAO,CAACF,gBAAAA,iBAAAA,MAAO,0BAA0BG,OAAO,CAAC,QAAQ;IAC/D,MAAMC,UAAkC;QAAE,gBAAgB;IAAmB;IAC7E,MAAMC,MAAMJ,SAASK,QAAQC,GAAG,CAACN,OAAO,GAAGO;IAC3C,IAAIH,KAAKD,QAAQK,aAAa,GAAG,CAAC,OAAO,EAAEJ,KAAK;IAEhD,eAAeK,KAAKC,KAAe,EAAEC,SAA6C;QAChF,MAAMC,MAAM,MAAMjB,eAAe,GAAGM,KAAK,SAAS,CAAC,EAAE;YAAEY,QAAQ;YAAQV;YAASW,MAAMC,KAAKC,SAAS,CAAC;gBAAElB;gBAAOmB,YAAYN;gBAAWD;gBAAOQ,iBAAiB;oBAAC;iBAAQ;YAAC;QAAG;QAC1K,IAAI,CAACN,IAAIO,EAAE,EAAE,MAAM,IAAIzB,WAAW,eAAe,GAAGO,KAAK,kBAAkB,EAAEW,IAAIQ,MAAM,EAAE;QACzF,MAAMN,OAAQ,MAAMF,IAAIS,IAAI;QAC5B,sFAAsF;QACtF,6EAA6E;QAC7E,MAAMC,UAAUR,KAAKS,UAAU,CAACC,KAAK;QACrC,IAAIF,QAAQG,MAAM,KAAKf,MAAMe,MAAM,EAAE,MAAM,IAAI/B,WAAW,eAAe,GAAGO,KAAK,mBAAmB,EAAEqB,QAAQG,MAAM,CAAC,gBAAgB,EAAEf,MAAMe,MAAM,CAAC,OAAO,CAAC;QAC5J,OAAOH,QAAQI,GAAG,CAAC,CAACC,IAAMC,aAAaC,IAAI,CAACF;IAC9C;IAEA,MAAMG,OAAO,AAAC,CAAA,MAAMrB,KAAK;QAAC;KAAkB,EAAE,eAAc,CAAE,CAAC,EAAE,CAACgB,MAAM;IACxE,OAAO;QAAEM,IAAI,CAAC,OAAO,EAAE9B,KAAK,CAAC,EAAEH,OAAO;QAAEgC;QAAME,UAAUpC;QAAWqC,gBAAgB,CAACvB,QAAUD,KAAKC,OAAO;QAAoBwB,YAAY,OAAOC,OAAS,AAAC,CAAA,MAAM1B,KAAK;gBAAC0B;aAAK,EAAE,eAAc,CAAE,CAAC,EAAE;IAAC;AACpM"}
@@ -0,0 +1,2 @@
1
+ export declare function fetchWithRetry(url: string, init: RequestInit, baseDelayMs?: number): Promise<Response>;
2
+ export declare function probeReachable(url: string, timeoutMs?: number): Promise<boolean>;
@@ -0,0 +1,33 @@
1
+ const MAX_ATTEMPTS = 3;
2
+ function sleep(ms) {
3
+ return new Promise((resolve)=>setTimeout(resolve, ms));
4
+ }
5
+ // Shared retry glue for the HTTP providers: 429/5xx/network failure retry with exponential
6
+ // backoff honoring Retry-After; any other status returns immediately for the caller to handle.
7
+ export async function fetchWithRetry(url, init, baseDelayMs = 200) {
8
+ for(let attempt = 0; attempt < MAX_ATTEMPTS; attempt++){
9
+ try {
10
+ const res = await fetch(url, init);
11
+ if (res.ok || res.status !== 429 && res.status < 500 || attempt === MAX_ATTEMPTS - 1) return res;
12
+ const retryAfter = Number(res.headers.get('retry-after'));
13
+ // Capped: a server naming a huge Retry-After must not stall an embed run for it.
14
+ await sleep(retryAfter > 0 ? Math.min(retryAfter * 1000, 10000) : baseDelayMs * 2 ** attempt);
15
+ } catch (err) {
16
+ if (attempt === MAX_ATTEMPTS - 1) throw err;
17
+ await sleep(baseDelayMs * 2 ** attempt);
18
+ }
19
+ }
20
+ throw new Error('unreachable');
21
+ }
22
+ // Any HTTP response means the endpoint is up; only a network failure or the timeout counts as
23
+ // unreachable, so an air-gapped machine never waits more than timeoutMs on `sense status`.
24
+ export async function probeReachable(url, timeoutMs = 1500) {
25
+ try {
26
+ await fetch(url, {
27
+ signal: AbortSignal.timeout(timeoutMs)
28
+ });
29
+ return true;
30
+ } catch {
31
+ return false;
32
+ }
33
+ }
@@ -0,0 +1 @@
1
+ {"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/embed/http.ts"],"sourcesContent":["const MAX_ATTEMPTS = 3;\n\nfunction sleep(ms: number): Promise<void> {\n return new Promise((resolve) => setTimeout(resolve, ms));\n}\n\n// Shared retry glue for the HTTP providers: 429/5xx/network failure retry with exponential\n// backoff honoring Retry-After; any other status returns immediately for the caller to handle.\nexport async function fetchWithRetry(url: string, init: RequestInit, baseDelayMs = 200): Promise<Response> {\n for (let attempt = 0; attempt < MAX_ATTEMPTS; attempt++) {\n try {\n const res = await fetch(url, init);\n if (res.ok || (res.status !== 429 && res.status < 500) || attempt === MAX_ATTEMPTS - 1) return res;\n const retryAfter = Number(res.headers.get('retry-after'));\n // Capped: a server naming a huge Retry-After must not stall an embed run for it.\n await sleep(retryAfter > 0 ? Math.min(retryAfter * 1000, 10_000) : baseDelayMs * 2 ** attempt);\n } catch (err) {\n if (attempt === MAX_ATTEMPTS - 1) throw err;\n await sleep(baseDelayMs * 2 ** attempt);\n }\n }\n throw new Error('unreachable');\n}\n\n// Any HTTP response means the endpoint is up; only a network failure or the timeout counts as\n// unreachable, so an air-gapped machine never waits more than timeoutMs on `sense status`.\nexport async function probeReachable(url: string, timeoutMs = 1500): Promise<boolean> {\n try {\n await fetch(url, { signal: AbortSignal.timeout(timeoutMs) });\n return true;\n } catch {\n return false;\n }\n}\n"],"names":["MAX_ATTEMPTS","sleep","ms","Promise","resolve","setTimeout","fetchWithRetry","url","init","baseDelayMs","attempt","res","fetch","ok","status","retryAfter","Number","headers","get","Math","min","err","Error","probeReachable","timeoutMs","signal","AbortSignal","timeout"],"mappings":"AAAA,MAAMA,eAAe;AAErB,SAASC,MAAMC,EAAU;IACvB,OAAO,IAAIC,QAAQ,CAACC,UAAYC,WAAWD,SAASF;AACtD;AAEA,2FAA2F;AAC3F,+FAA+F;AAC/F,OAAO,eAAeI,eAAeC,GAAW,EAAEC,IAAiB,EAAEC,cAAc,GAAG;IACpF,IAAK,IAAIC,UAAU,GAAGA,UAAUV,cAAcU,UAAW;QACvD,IAAI;YACF,MAAMC,MAAM,MAAMC,MAAML,KAAKC;YAC7B,IAAIG,IAAIE,EAAE,IAAKF,IAAIG,MAAM,KAAK,OAAOH,IAAIG,MAAM,GAAG,OAAQJ,YAAYV,eAAe,GAAG,OAAOW;YAC/F,MAAMI,aAAaC,OAAOL,IAAIM,OAAO,CAACC,GAAG,CAAC;YAC1C,iFAAiF;YACjF,MAAMjB,MAAMc,aAAa,IAAII,KAAKC,GAAG,CAACL,aAAa,MAAM,SAAUN,cAAc,KAAKC;QACxF,EAAE,OAAOW,KAAK;YACZ,IAAIX,YAAYV,eAAe,GAAG,MAAMqB;YACxC,MAAMpB,MAAMQ,cAAc,KAAKC;QACjC;IACF;IACA,MAAM,IAAIY,MAAM;AAClB;AAEA,8FAA8F;AAC9F,2FAA2F;AAC3F,OAAO,eAAeC,eAAehB,GAAW,EAAEiB,YAAY,IAAI;IAChE,IAAI;QACF,MAAMZ,MAAML,KAAK;YAAEkB,QAAQC,YAAYC,OAAO,CAACH;QAAW;QAC1D,OAAO;IACT,EAAE,OAAM;QACN,OAAO;IACT;AACF"}
@@ -0,0 +1,13 @@
1
+ export declare const MODEL_FILES: string[];
2
+ export declare const MODEL_FILENAMES: string;
3
+ export declare function isDownloadable(model: string): boolean;
4
+ export declare function refsMainPath(model: string): string;
5
+ export declare function readRef(model: string): string | undefined;
6
+ export declare function writeRef(model: string, sha: string): void;
7
+ export declare function snapshotDir(model: string, sha: string): string;
8
+ export declare function languagesPath(model: string): string;
9
+ export declare function readLanguages(model: string): string[] | undefined;
10
+ export declare function writeLanguages(model: string, languages: string[]): void;
11
+ export declare function modelDir(model: string): string;
12
+ export declare function hasModelFiles(model: string): boolean;
13
+ export declare function modelIdentity(model: string): string | undefined;
@@ -0,0 +1,92 @@
1
+ import { existsSync, mkdirSync, readFileSync, statSync, writeFileSync } from 'node:fs';
2
+ import { homedir } from 'node:os';
3
+ import { join } from 'node:path';
4
+ // HF cache layout, no network: ~/.sense/models/models--<org>--<name>/{refs/main,
5
+ // snapshots/<sha>/<files>}, files stored directly (hf_hub's blobs/symlinks layer skipped).
6
+ export const MODEL_FILES = [
7
+ 'model.safetensors',
8
+ 'tokenizer.json'
9
+ ];
10
+ // The pair, for messages that name what a local model directory must contain.
11
+ export const MODEL_FILENAMES = MODEL_FILES.join(' and ');
12
+ // A Hugging Face repo id: one slash, HF's charset. Anything else is a path, used as one rather
13
+ // than resolved against the HF cache.
14
+ const HF_MODEL_ID = /^[A-Za-z0-9][\w.-]*\/[A-Za-z0-9][\w.-]*$/;
15
+ export function isDownloadable(model) {
16
+ return HF_MODEL_ID.test(model);
17
+ }
18
+ function repoDir(model) {
19
+ const [org, name] = model.split('/');
20
+ return join(homedir(), '.sense', 'models', `models--${org}--${name}`);
21
+ }
22
+ export function refsMainPath(model) {
23
+ return join(repoDir(model), 'refs', 'main');
24
+ }
25
+ // What `main` last resolved to, read from disk -- no network. Undefined until some resolution
26
+ // (a download or an explicit prefetch) has written it once.
27
+ export function readRef(model) {
28
+ const p = refsMainPath(model);
29
+ if (!existsSync(p)) return undefined;
30
+ const sha = readFileSync(p, 'utf8').trim();
31
+ return sha.length > 0 ? sha : undefined;
32
+ }
33
+ // A recorded ref pins until its snapshot directory is removed; nothing here re-resolves it.
34
+ export function writeRef(model, sha) {
35
+ mkdirSync(join(repoDir(model), 'refs'), {
36
+ recursive: true
37
+ });
38
+ writeFileSync(refsMainPath(model), sha);
39
+ }
40
+ export function snapshotDir(model, sha) {
41
+ return join(repoDir(model), 'snapshots', sha);
42
+ }
43
+ // Languages describe the repo, not one weight snapshot, so this lives beside refs/, not
44
+ // inside a snapshot dir.
45
+ export function languagesPath(model) {
46
+ return join(repoDir(model), 'languages.json');
47
+ }
48
+ // undefined means "never resolved yet"; an empty array is a resolved fact (checked, none
49
+ // declared) and is cached the same way, so a languageless model is not re-fetched every run.
50
+ export function readLanguages(model) {
51
+ const p = languagesPath(model);
52
+ if (!existsSync(p)) return undefined;
53
+ try {
54
+ return JSON.parse(readFileSync(p, 'utf8'));
55
+ } catch {
56
+ return undefined;
57
+ }
58
+ }
59
+ export function writeLanguages(model, languages) {
60
+ mkdirSync(repoDir(model), {
61
+ recursive: true
62
+ });
63
+ writeFileSync(languagesPath(model), JSON.stringify(languages));
64
+ }
65
+ // A local directory the caller manages, or the resolved snapshot once `main` is pinned; an HF
66
+ // id with no recorded ref yet has no known snapshot, so this names the (fileless) repo dir.
67
+ export function modelDir(model) {
68
+ if (!isDownloadable(model)) return model;
69
+ const sha = readRef(model);
70
+ return sha ? snapshotDir(model, sha) : repoDir(model);
71
+ }
72
+ // Both files, so an interrupted download reads as absent and the next one resumes it.
73
+ export function hasModelFiles(model) {
74
+ const dir = modelDir(model);
75
+ return MODEL_FILES.every((file)=>existsSync(join(dir, file)));
76
+ }
77
+ // Weight identity for the signature, no network: a HF id's resolved sha, or the two
78
+ // files' size+mtime for a local path. Undefined when nothing is known yet.
79
+ export function modelIdentity(model) {
80
+ if (isDownloadable(model)) return readRef(model);
81
+ const dir = modelDir(model);
82
+ const parts = [];
83
+ for (const file of MODEL_FILES){
84
+ try {
85
+ const st = statSync(join(dir, file));
86
+ parts.push(`${st.size}:${Math.trunc(st.mtimeMs)}`);
87
+ } catch {
88
+ return undefined;
89
+ }
90
+ }
91
+ return parts.join(',');
92
+ }