@murumets-ee/imports 0.49.0 → 0.51.0

package/dist/worker.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"worker.mjs","names":[],"sources":["../src/worker.ts"],"sourcesContent":["/**\n * Queue handler for `imports:run`. The route handler that creates the\n * `import_run` row enqueues this job carrying `{ importRunId }`; the\n * worker module here resolves the run, the transform, and the ES\n * client, then runs `runImport` and writes results back.\n *\n * Per PLAN-ECOMMERCE.md PR 7 (PoC scope): one job type, no resumability.\n * A failed handler attempt re-runs from row 1 on retry — fine for the\n * tens-of-MB feed sizes the PoC exercises.\n *\n * Per CLAUDE.md \"leaf packages\" boundary rule: this handler reads the\n * import_run row via an injected `AdminClient`, NOT a direct `db.select`.\n * Construction happens at plugin init in `plugin.ts`.\n */\n\nimport { type Logger, runAsCli } from '@murumets-ee/core'\nimport {\n  defineJob,\n  type JobDefinition,\n  type QueueJobDeadEvent,\n  type QueueTerminalNotifier,\n} from '@murumets-ee/queue/client'\nimport type { EsClientLike } from '@murumets-ee/search-elasticsearch'\nimport { eq } from 'drizzle-orm'\nimport type { PgColumn } from 'drizzle-orm/pg-core'\nimport { z } from 'zod'\nimport type { ImportRunClient, ImportRunStatus } from './entities/import-run.js'\nimport { type ReadFeedOptions, readPersistedFeedConfig } from './read-feed.js'\nimport {\n  publishImportRunFailed,\n  publishImportRunStarted,\n  publishImportRunSucceeded,\n} from './realtime/publish.js'\nimport { type ImportRunProgress, runImport } from './runner.js'\nimport { countDataRows } from './streaming.js'\nimport type { DirectionalPair, TransformRegistry } from './transform.js'\n\n/**\n * Structural subset of `JobContext` the handler reads. Declared\n * locally so we don't depend on a `JobHandler` import — the queue\n * package doesn't re-export it from any subpath today (matches the\n * `SendEmailJobContext` pattern in `@murumets-ee/notifications`).\n *\n * Function-param contravariance lets a handler returned with this\n * narrower shape be assigned to the wider `JobHandler<TPayload>` that\n * `registerJob` expects.\n */\nexport interface ImportsRunJobContext {\n  id: string\n  payload: ImportsRunJobPayload\n  updateProgress(data: ImportRunProgress): void\n}\n\n/**\n * Payload schema. `importRunId` is the `import_run.id` UUID — the worker\n * looks up everything else (filePath, transformName, params, …) from\n * that row. Keeping the payload tiny means the queue's progress JSON\n * column never bloats with feed metadata duplicated across `toolkit_jobs`.\n */\nexport const importsRunJobPayloadSchema = z.object({\n  importRunId: z.string().uuid(),\n})\nexport type ImportsRunJobPayload = z.infer<typeof importsRunJobPayloadSchema>\n\n/**\n * Job definition. Consumers register their handler against this with\n * `registerJob(importsRunJob, createRunImportHandler({...}))`.\n *\n * `defaultRetries: 0` — re-running a multi-batch import against the\n * same `import_run.id` while the previous handler may still be writing\n * is a footgun (duplicate batches, double-counted progress). The\n * operator handles retries explicitly via PR 8a's \"retry\" button by\n * creating a new `import_run` row. When the resumable design lands\n * (post-PoC), retries become safe to enable.\n */\nexport const importsRunJob: JobDefinition<ImportsRunJobPayload> = defineJob({\n  name: 'imports:run',\n  description: 'Stream a feed file, transform rows, bulk-write to Elasticsearch.',\n  schema: importsRunJobPayloadSchema,\n  defaultRetries: 0,\n})\n\n/**\n * Resolves the ES client at handler-invocation time. A function (rather\n * than the bare `EsClientLike`) so the consumer can lazy-construct the\n * client — typical Next.js setups create the ES connection in a\n * route-handler initialiser, not at plugin-init time.\n */\nexport type EsClientResolver = () => EsClientLike | Promise<EsClientLike>\n\n/**\n * Ensure the destination index/alias exists before the first bulk-write.\n * Invoked once per job, BEFORE {@link runImport}. Idempotent: the canonical\n * implementation (`ensureAliasedIndex` in `@murumets-ee/search-elasticsearch`)\n * no-ops when the alias already resolves.\n *\n * **Why it lives here.** This is the home of the foot-gun it guards: a worker\n * that bulk-writes to alias `<x>` before the alias exists makes ES\n * auto-create a physical `<x>` index with default `text` mappings, breaking\n * aggregations forever. The bootstrap used to run at app boot, which turned\n * \"ES is down\" into a full-admin 500 (issue #610). Moving it onto the import\n * path keeps the guarantee while scoping ES availability to imports/search —\n * the rest of the admin boots and runs with ES down.\n */\nexport type EnsureIndexResolver = () => Promise<unknown>\n\n/**\n * Resolves the value of `import_run.filePath` (whatever the upload\n * route persisted there — typically a storage adapter key) to a\n * readable LOCAL filesystem path that {@link runImport} can hand to\n * `node:fs.createReadStream`.\n *\n * **Why this exists:** the streaming reader (`streamFeed`) reads from\n * local disk via `createReadStream`. The upload route may persist a\n * remote-storage object key (R2, S3, …) on `import_run.filePath`\n * because that's what `@murumets-ee/storage` returns. Without a\n * resolver, `createReadStream('uploads/2026/05/<uuid>/feed.txt')`\n * crashes with `ENOENT`. The resolver is the documented integration\n * point — typical wiring downloads the storage object to a tmpfile\n * and returns its path.\n *\n * **`cleanup`** runs after the run finishes (success OR failure). The\n * worker awaits it best-effort — a failed cleanup logs but does not\n * crash the run.\n *\n * **No-resolver fallback:** when `resolveFilePath` is unset, the\n * worker treats `import_run.filePath` as already a local FS path\n * (back-compat with the original PoC design where uploads landed on\n * local disk). This stays valid for fixture-driven tests + on-disk\n * deployments.\n */\nexport type FilePathResolver = (\n  storageKey: string,\n) => Promise<{ localPath: string; cleanup?: () => Promise<void> }>\n\n/**\n * Context handed to a {@link PairSink} alongside the harvested pairs. Carries the\n * run identity so the sink can build `source` attribution\n * (`feed:<supplierId>:<importBatchId>`, D20) and scope its writes.\n */\nexport interface PairSinkContext {\n  importRunId: string\n  /** The `import_run.params` object — the sink reads `supplierId` / `importBatchId`. */\n  params: Record<string, unknown>\n  /**\n   * The run's structured logger (child), forwarded so the sink's INTERNAL telemetry —\n   * a params-parse miss that skips ingestion, an over-merge flag — is visible. Without it\n   * a misconfigured run looks successful from the worker's vantage (a params-miss returns\n   * normally, so the worker logs \"applied\"). Optional; the sink no-ops its logs if absent.\n   */\n  logger?: Logger\n}\n\n/**\n * Optional post-run consumer for the directional supersession pairs the runner\n * harvested as a side-output (commerce-demo PR6, SD007). Invoked ONCE after a\n * successful run with the full in-memory pair buffer (read-once / merge-after) so\n * the consumer (the commerce replacement-merge engine) can build equivalence groups\n * and bulk-write them to Postgres.\n *\n * **Why injected, not imported.** The merge writes the CONSUMER's Postgres tables,\n * but the runner + this worker live in the `@murumets-ee/imports` leaf, which must\n * not depend on commerce. The app composes the sink (commerce exports it, the app\n * passes it to `imports({ pairSink })`) — the same injection shape as `esClient` /\n * `resolveFilePath` / `ensureIndex`. The boundary holds.\n *\n * **Best-effort.** A sink throw is logged and SWALLOWED — the catalogue import\n * (the ES write) has already succeeded and must not be undone by a secondary\n * replacement-ingestion failure. The merge is idempotent, so a re-import recovers.\n */\nexport type PairSink = (pairs: readonly DirectionalPair[], ctx: PairSinkContext) => Promise<void>\n\nexport interface RunImportHandlerConfig {\n  /** AdminClient over the `import_run` entity. */\n  importRuns: ImportRunClient\n  /** Transform registry to dispatch against. Defaults to the process-global singleton. */\n  transforms: TransformRegistry\n  /** Resolver for the ES client. */\n  esClient: EsClientResolver\n  /** ES alias / index to bulk-write into. Per D6 callers always pass an alias. */\n  esIndex: string\n  /**\n   * Optional bootstrap for the destination index/alias, run once per job\n   * before any bulk-write. See {@link EnsureIndexResolver}. Omit for\n   * deployments that bootstrap the index out-of-band (e.g. a reindex job).\n   */\n  ensureIndex?: EnsureIndexResolver\n  /**\n   * Optional resolver for `import_run.filePath`. See {@link FilePathResolver}.\n   * Required when uploads land in remote storage (R2/S3); optional for\n   * on-disk PoC setups.\n   */\n  resolveFilePath?: FilePathResolver\n  /**\n   * Optional consumer for the directional supersession pairs harvested as a\n   * side-output (commerce-demo PR6, SD007). See {@link PairSink}. Omit for\n   * deployments that don't ingest replacement codes — the pairs are simply\n   * discarded after the run.\n   */\n  pairSink?: PairSink\n  /** Optional structured logger. Defaults to silent. */\n  logger?: Logger\n}\n\n/**\n * Build the `JobHandler` for {@link importsRunJob}. The returned\n * handler is what gets passed to `registerJob`.\n *\n * The body is wrapped in `runAsCli` so AdminClient calls inside have a\n * synthetic `cli` admin context — `auditable()` records `updatedBy:\n * 'cli'` rather than NULL, and the firewall checker passes. This is\n * the documented worker entry-point pattern (see `runAsCli` JSDoc in\n * `@murumets-ee/core`).\n */\nexport function createRunImportHandler(\n  config: RunImportHandlerConfig,\n): (job: ImportsRunJobContext) => Promise<void> {\n  const {\n    importRuns,\n    transforms,\n    esClient,\n    esIndex,\n    resolveFilePath,\n    ensureIndex,\n    pairSink,\n    logger,\n  } = config\n\n  return async (job: ImportsRunJobContext): Promise<void> =>\n    runAsCli(async () => {\n      const { importRunId } = job.payload\n      const log = logger?.child({ jobId: job.id, importRunId, type: 'imports:run' })\n\n      const run = await importRuns.findById(importRunId)\n      if (!run) {\n        // Row was deleted between enqueue and dispatch — nothing to do.\n        // Don't throw: a thrown handler retries with no payoff because\n        // the row stays gone. Log + return so the queue marks the job\n        // completed.\n        log?.warn('import_run row not found — skipping')\n        return\n      }\n\n      // Re-claim detection. The queue's `recoverStuckJobs` sweep flips a\n      // crashed worker's `toolkit_jobs` row from `processing` back to\n      // `pending` — but the queue layer has no knowledge of `import_run`,\n      // so this row stays at `status='running'` from the prior attempt.\n      // When the queue re-dispatches the job, we land here with a row\n      // that the operator-facing Active panel has been showing as live\n      // for however long lock-recovery took (default 30 min). The\n      // existing `update` below overwrites `queueJobId` + `startedAt` so\n      // there is no STATE corruption — but emit a structured warning so\n      // ops can correlate this re-run with the worker crash that\n      // produced it. The sibling `sweepStaleImportRuns` job below catches\n      // the case where re-dispatch never happens (queue marks the job\n      // dead, or the queue worker is offline long-term).\n      if (run.status === 'running') {\n        log?.warn(\n          {\n            previousQueueJobId: run.queueJobId,\n            previousStartedAt: run.startedAt,\n          },\n          'imports: re-claiming a row left in running by a prior worker — likely a crash recovered by queue lock-recovery',\n        )\n      }\n\n      // `transformName`, `filePath`, `label` are declared `field.text({\n      // required: true })` so the inferred DTO already types them as\n      // `string` — no cast needed. `params` is `field.json()` which\n      // infers as `JsonValue | undefined`; narrow to a record so the\n      // transform receives an object.\n      const transform = transforms.get(run.transformName)\n      if (!transform) {\n        await markFailed(importRuns, importRunId, `Unknown transform \"${run.transformName}\"`)\n        throw new Error(`No transform registered for name \"${run.transformName}\"`)\n      }\n      const params: Record<string, unknown> =\n        run.params && typeof run.params === 'object' && !Array.isArray(run.params)\n          ? (run.params as Record<string, unknown>)\n          : {}\n\n      // Reader config (format/delimiter/header/sheet) travels in\n      // `params.feed` for mapping-driven runs (commerce-demo PR3). Carmaker\n      // runs omit it → empty config → tab-delimited default, so the existing\n      // path is byte-for-byte unchanged. Re-validate here as defense-in-depth\n      // even though the enqueuing route already validated the saved mapping.\n      // Shared with the #842 audit, which has to rehydrate IDENTICALLY or it\n      // re-reads a past run with options that run never used.\n      const feedParsed = readPersistedFeedConfig(params)\n      if (!feedParsed.success) {\n        const reason = feedParsed.error.issues[0]?.message ?? 'invalid feed reader config'\n        await markFailed(importRuns, importRunId, `Invalid params.feed: ${reason}`)\n        throw new Error(`Invalid params.feed for import_run ${importRunId}: ${reason}`)\n      }\n      const feedConfig = feedParsed.data\n\n      await importRuns.update(importRunId, {\n        status: 'running' satisfies ImportRunStatus,\n        queueJobId: job.id,\n        startedAt: new Date(),\n      })\n      // Realtime: row transitioned pending → running. Subscribers\n      // refetch to surface the change in the imports admin page.\n      // Best-effort — a publish failure is logged inside `publishSafe`\n      // and does NOT roll back the row update.\n      publishImportRunStarted(importRunId)\n\n      let resolvedClient: EsClientLike\n      try {\n        resolvedClient = await esClient()\n      } catch (err) {\n        const reason = err instanceof Error ? err.message : String(err)\n        await markFailed(importRuns, importRunId, `Could not resolve ES client: ${reason}`)\n        throw err\n      }\n\n      // Ensure the destination alias exists BEFORE any bulk-write. Moved here\n      // from app boot (issue #610): the foot-gun is the worker auto-creating a\n      // mis-mapped physical index, so the guard belongs on the import path —\n      // not gating the whole app on ES. Idempotent + scoped: an ES outage (or\n      // the `ensureAliasedIndex` physical-collision guard firing) fails THIS\n      // job loudly while the rest of the admin keeps working.\n      if (ensureIndex) {\n        try {\n          await ensureIndex()\n        } catch (err) {\n          const reason = err instanceof Error ? err.message : String(err)\n          await markFailed(importRuns, importRunId, `Could not ensure ES index/alias: ${reason}`)\n          throw err\n        }\n      }\n\n      // Materialise the file locally if the consumer wired a resolver.\n      // The `cleanup` callback (when present) runs in the `finally`\n      // below regardless of run outcome so a tmpfile created here\n      // doesn't outlive the handler.\n      let localPath: string\n      let fileCleanup: (() => Promise<void>) | undefined\n      if (resolveFilePath) {\n        try {\n          const resolved = await resolveFilePath(run.filePath)\n          localPath = resolved.localPath\n          fileCleanup = resolved.cleanup\n        } catch (err) {\n          const reason = err instanceof Error ? err.message : String(err)\n          await markFailed(importRuns, importRunId, `Could not resolve filePath: ${reason}`)\n          throw err\n        }\n      } else {\n        // Back-compat: no resolver wired → treat `filePath` as already\n        // a local FS path. Fine for fixture-driven tests + the original\n        // on-disk PoC design.\n        localPath = run.filePath\n      }\n\n      // Pre-pass line count: a 16 MB tab-delimited feed counts in\n      // ~50ms; a 500 MB feed in <2s. Cheap relative to the full\n      // import. Gives the operator a denominator (`processed / total`)\n      // so the imports-page panel can render an actual percent\n      // instead of just \"rows so far\".\n      // Pre-pass row count is a `\\n`-tally over a delimited file — meaningless\n      // on an xlsx binary, so skip it there (the UI falls back to bare counts).\n      // `hasHeader` from the feed config keeps the header-exclusion correct for\n      // headerless mappings.\n      let totalRows: number | undefined\n      if (feedConfig.format !== 'xlsx') {\n        try {\n          totalRows = await countDataRows(\n            localPath,\n            feedConfig.hasHeader !== undefined ? { hasHeader: feedConfig.hasHeader } : {},\n          )\n          log?.info({ importRunId, totalRows }, 'imports: pre-pass row count')\n        } catch (err) {\n          // Don't fail the run on a count miss — the actual import is\n          // unaffected; the operator just loses the percent denominator.\n          log?.warn({ err }, 'imports: pre-pass row count failed — continuing without total')\n        }\n      }\n\n      let result: Awaited<ReturnType<typeof runImport>>\n      try {\n        // Build feed options with the conditional-spread idiom (NOT\n        // `...feedConfig`) so absent optionals stay ABSENT — spreading a\n        // `T | undefined` value would set the key to `undefined`, which\n        // `exactOptionalPropertyTypes` rejects against `ReadFeedOptions`.\n        const feedOpts: ReadFeedOptions = {\n          filePath: localPath,\n          ...(feedConfig.format !== undefined && { format: feedConfig.format }),\n          ...(feedConfig.delimiter !== undefined && { delimiter: feedConfig.delimiter }),\n          ...(feedConfig.hasHeader !== undefined && { hasHeader: feedConfig.hasHeader }),\n          ...(feedConfig.headerRowIndex !== undefined && {\n            headerRowIndex: feedConfig.headerRowIndex,\n          }),\n          ...(feedConfig.sheet !== undefined && { sheet: feedConfig.sheet }),\n        }\n        result = await runImport({\n          importRunId,\n          runLabel: run.label,\n          params,\n          transform,\n          feed: feedOpts,\n          esClient: resolvedClient,\n          esIndex,\n          onProgress: (progress) =>\n            job.updateProgress(totalRows !== undefined ? { ...progress, totalRows } : progress),\n          ...(log !== undefined && { logger: log }),\n        })\n      } catch (err) {\n        const reason = err instanceof Error ? err.message : String(err)\n        await markFailed(importRuns, importRunId, reason)\n        if (fileCleanup) {\n          await fileCleanup().catch((cleanupErr: unknown) => {\n            log?.warn({ err: cleanupErr }, 'filePath cleanup after failed run threw — ignoring')\n          })\n        }\n        throw err\n      }\n\n      // Cleanup runs before the success-update so a cleanup throw\n      // surfaces as a logged warning rather than a crashed handler at\n      // the worst possible moment (just after the row was about to be\n      // marked succeeded).\n      if (fileCleanup) {\n        await fileCleanup().catch((cleanupErr: unknown) => {\n          log?.warn({ err: cleanupErr }, 'filePath cleanup after successful run threw — ignoring')\n        })\n      }\n\n      await importRuns.update(importRunId, {\n        status: 'succeeded' satisfies ImportRunStatus,\n        finishedAt: new Date(),\n        totals: {\n          rowsRead: result.rowsRead,\n          rowsSucceeded: result.rowsSucceeded,\n          rowsFailed: result.rowsFailed,\n          rowsSkipped: result.rowsSkipped,\n          rowsWarned: result.rowsWarned,\n          batchesCompleted: result.batchesCompleted,\n          // Phase-2 core-charge attach (commerce-demo PR7) — durable on the run row so\n          // an operator sees how many `*S` deposits attached vs orphaned, not just the\n          // orphan warnings in `warningSummary`.\n          patchesApplied: result.patchesApplied,\n          patchesOrphaned: result.patchesOrphaned,\n        },\n        // The typed `ErrorTrackerSnapshot` is fully JSON-serialisable but\n        // doesn't structurally satisfy entity's recursive `JsonValue`\n        // (objects without an index signature aren't `Record<string,\n        // JsonValue>`). Round-trip through `JSON.parse(JSON.stringify(...))`\n        // to land as the recursive JSON shape — the runtime cost is one\n        // serialise per import_run completion, the type-cast cost is zero.\n        errorSummary: JSON.parse(JSON.stringify(result.errors)),\n        // Non-fatal warnings (commerce-demo PR4) — kept separate from\n        // `errorSummary` so a `no_purchase_price` advisory never reads as a\n        // failed row. Same serialise-roundtrip rationale as `errorSummary`.\n        warningSummary: JSON.parse(JSON.stringify(result.warnings)),\n      })\n      // Realtime: row transitioned to terminal-success. Subscribers\n      // refetch to move the row from Active → Recent.\n      publishImportRunSucceeded(importRunId, {\n        rowsRead: result.rowsRead,\n        rowsSucceeded: result.rowsSucceeded,\n        rowsFailed: result.rowsFailed,\n      })\n\n      // Side-output: hand the harvested supersession pairs to the injected sink\n      // (commerce-demo PR6, SD007) AFTER the catalogue run is marked succeeded — the\n      // ES import is primary; replacement ingestion is secondary. Best-effort: a sink\n      // throw is logged + swallowed so a Postgres hiccup can't undo a successful\n      // catalogue import (the merge is idempotent → a re-import recovers).\n      if (pairSink && result.pairs.length > 0) {\n        try {\n          await pairSink(result.pairs, {\n            importRunId,\n            params,\n            ...(log !== undefined && { logger: log }),\n          })\n          // \"invoked\" not \"applied\": the sink may internally skip (e.g. missing run params);\n          // it logs its own applied/skipped outcome via the forwarded logger.\n          log?.info({ importRunId, pairs: result.pairs.length }, 'imports: pairSink invoked')\n        } catch (err) {\n          log?.warn(\n            { importRunId, pairs: result.pairs.length, err },\n            // NOT \"skipped\": a sink may apply part of the batch before failing (the commerce\n            // replacement sink commits per batch), so some pairs can already have landed.\n            // \"Incomplete\" is the only claim this boundary can honestly make about a sink\n            // whose internals it does not know.\n            'imports: pairSink threw — replacement ingestion incomplete, catalogue import unaffected',\n          )\n        }\n      }\n\n      log?.info(\n        {\n          rowsRead: result.rowsRead,\n          rowsSucceeded: result.rowsSucceeded,\n          rowsFailed: result.rowsFailed,\n          rowsSkipped: result.rowsSkipped,\n          batches: result.batchesCompleted,\n        },\n        'import_run completed',\n      )\n    })\n}\n\n/** Write a `failed` row with a fatal error reason. Best-effort — logs but does not throw on DB failure. */\nasync function markFailed(importRuns: ImportRunClient, id: string, reason: string): Promise<void> {\n  try {\n    await importRuns.update(id, {\n      status: 'failed' satisfies ImportRunStatus,\n      finishedAt: new Date(),\n      errorSummary: { fatal: reason },\n    })\n    // Realtime: row transitioned to terminal-failure. Subscribers\n    // refetch to surface the failure in the Recent panel. Publish\n    // is best-effort and only fires when the persist succeeds —\n    // skipping it on persist failure keeps subscribers consistent\n    // with the DB (no event for a row that didn't transition).\n    publishImportRunFailed(id, reason)\n  } catch {\n    // The handler already threw / will throw — the queue will surface\n    // the original error. A second failure here just logs noise.\n  }\n}\n\nexport interface ImportsRunDeadListenerConfig {\n  importRuns: ImportRunClient\n  /** Optional structured logger. */\n  logger?: Logger\n}\n\n/**\n * Build a {@link QueueTerminalNotifier} that mirrors `imports:run`\n * dead-job events into `import_run.status='failed'`.\n *\n * **Why this exists.** The `imports:run` handler's normal failure path\n * already calls `markFailed` before re-throwing, so a synchronous\n * handler error reaches the operator-visible `failed` status without\n * help from anyone. The case THIS listener catches is the one\n * `markFailed` never gets to: the worker process is killed mid-handler\n * (OOM, container restart, kill -9), the queue's lock-recovery sweep\n * eventually reclaims the row, and the queue ultimately marks the job\n * `dead` (defaultRetries: 0 + repeated process death = dead-letter\n * with `import_run.status` still `running`). Without this listener\n * the row sits in the operator's Active panel forever.\n *\n * **Generic primitive.** This is a thin subscriber over the queue's\n * multi-subscriber `QueueTerminalNotifier` registry — the same shape\n * graphile-worker (`job:failed`) and BullMQ (`failed`) expose. Other\n * domain plugins that mirror queue state (ticketing-reply, future\n * exports/reports/...) can register their own listeners exactly the\n * same way without coordinating via a per-domain sweep job.\n *\n * **Filtering.** The listener is invoked for EVERY dead event in the\n * process. We early-return on any `jobType` other than `imports:run`\n * so the per-event cost on unrelated jobs is a single property read.\n *\n * **Idempotency.** A dead event delivered after the row has already\n * reached its own terminal state (succeeded / failed / cancelled) is\n * ignored — the only transition we make is `running → failed`. This\n * keeps a late replay or out-of-order delivery from clobbering a\n * row whose worker reached `markFailed` faster than the queue's\n * dead-letter dispatch.\n *\n * The handler is wrapped in `runAsCli` so AdminClient calls inside\n * have a synthetic CLI admin context — same pattern as\n * {@link createRunImportHandler}.\n */\nexport function createImportsRunDeadListener(\n  config: ImportsRunDeadListenerConfig,\n): QueueTerminalNotifier {\n  const { importRuns, logger } = config\n\n  return {\n    onJobDead: async (event: QueueJobDeadEvent): Promise<void> => {\n      if (event.jobType !== 'imports:run') return\n\n      await runAsCli(async () => {\n        const log = logger?.child({\n          event: 'imports.run.dead',\n          jobId: event.jobId,\n          type: 'imports:run',\n        })\n\n        // Look up the mirror row by `queueJobId`. Cap to 1 — the\n        // upload route writes one `import_run` per enqueued job.\n        //\n        // `AdminClient.getTable()` returns `PgTableWithColumns<any>`\n        // because entity tables have dynamic column shapes (see the\n        // `getTable()` JSDoc in `packages/entity/src/admin/client.ts`\n        // and the documented-`any`-cases list in CLAUDE.md). Cast\n        // the table to a structural shape covering only the column\n        // we touch, so both `eq()` and type-coverage see a typed\n        // operand instead of an `any`-derived property access.\n        const table = importRuns.getTable() as unknown as { queueJobId: PgColumn }\n        const matches = await importRuns.findMany({\n          where: eq(table.queueJobId, event.jobId),\n          limit: 1,\n        })\n        const row = matches[0]\n        if (!row) {\n          // No mirror row for this dead job. Either it was deleted\n          // by the operator, or the dead event is for a job\n          // unrelated to imports despite the type filter\n          // (shouldn't happen, but is cheap to log).\n          log?.warn('queue.job.dead with no matching import_run — skipping')\n          return\n        }\n\n        // Only transition rows that haven't already reached their\n        // own terminal state. A late dead-event must not overwrite\n        // a `succeeded` row that the worker reached via the normal\n        // path before the queue's dead-letter dispatch fired.\n        if (row.status !== 'running' && row.status !== 'pending') {\n          log?.info(\n            { importRunId: row.id, status: row.status },\n            'queue.job.dead arrived for a row already in a terminal state — leaving as-is',\n          )\n          return\n        }\n\n        // Dead-from-`pending` is the rarer + more interesting case:\n        // the worker process died (or never claimed) BEFORE it could\n        // flip the row to `running`. Log it at warn so ops can spot\n        // a queue+worker-startup pathology that the standard `running\n        // → failed` transition wouldn't surface. Dead-from-`running`\n        // is the documented \"crash mid-import\" case — info is enough.\n        if (row.status === 'pending') {\n          log?.warn(\n            { importRunId: row.id, attempts: event.attempts },\n            'queue.job.dead arrived for an import_run still in pending — worker crashed BEFORE claim/lookup',\n          )\n        }\n\n        await markFailed(importRuns, row.id, event.error)\n        log?.info(\n          { importRunId: row.id, attempts: event.attempts, priorStatus: row.status },\n          'queue.job.dead → import_run marked failed',\n        )\n      })\n    },\n  }\n}\n"],"mappings":"wQA2DA,MAAa,EAA6B,EAAE,OAAO,CACjD,YAAa,EAAE,OAAO,CAAC,CAAC,KAAK,CAC/B,CAAC,EAcY,EAAqD,EAAU,CAC1E,KAAM,cACN,YAAa,mEACb,OAAQ,EACR,eAAgB,CAClB,CAAC,EAsID,SAAgB,EACd,EAC8C,CAC9C,GAAM,CACJ,aACA,aACA,WACA,UACA,kBACA,cACA,WACA,UACE,EAEJ,OAAO,KAAO,IACZ,EAAS,SAAY,CACnB,GAAM,CAAE,eAAgB,EAAI,QACtB,EAAM,GAAQ,MAAM,CAAE,MAAO,EAAI,GAAI,cAAa,KAAM,aAAc,CAAC,EAEvE,EAAM,MAAM,EAAW,SAAS,CAAW,EACjD,GAAI,CAAC,EAAK,CAKR,GAAK,KAAK,qCAAqC,EAC/C,MACF,CAeI,EAAI,SAAW,WACjB,GAAK,KACH,CACE,mBAAoB,EAAI,WACxB,kBAAmB,EAAI,SACzB,EACA,gHACF,EAQF,IAAM,EAAY,EAAW,IAAI,EAAI,aAAa,EAClD,GAAI,CAAC,EAEH,MADA,MAAM,EAAW,EAAY,EAAa,sBAAsB,EAAI,cAAc,EAAE,EAC1E,MAAM,qCAAqC,EAAI,cAAc,EAAE,EAE3E,IAAM,EACJ,EAAI,QAAU,OAAO,EAAI,QAAW,UAAY,CAAC,MAAM,QAAQ,EAAI,MAAM,EACpE,EAAI,OACL,CAAC,EASD,EAAa,EAAwB,CAAM,EACjD,GAAI,CAAC,EAAW,QAAS,CACvB,IAAM,EAAS,EAAW,MAAM,OAAO,EAAE,EAAE,SAAW,6BAEtD,MADA,MAAM,EAAW,EAAY,EAAa,wBAAwB,GAAQ,EAChE,MAAM,sCAAsC,EAAY,IAAI,GAAQ,CAChF,CACA,IAAM,EAAa,EAAW,KAE9B,MAAM,EAAW,OAAO,EAAa,CACnC,OAAQ,UACR,WAAY,EAAI,GAChB,UAAW,IAAI,IACjB,CAAC,EAKD,EAAwB,CAAW,EAEnC,IAAI,EACJ,GAAI,CACF,EAAiB,MAAM,EAAS,CAClC,OAAS,EAAK,CAGZ,MADA,MAAM,EAAW,EAAY,EAAa,gCAD3B,aAAe,MAAQ,EAAI,QAAU,OAAO,CAAG,GACoB,EAC5E,CACR,CAQA,GAAI,EACF,GAAI,CACF,MAAM,EAAY,CACpB,OAAS,EAAK,CAGZ,MADA,MAAM,EAAW,EAAY,EAAa,oCAD3B,aAAe,MAAQ,EAAI,QAAU,OAAO,CAAG,GACwB,EAChF,CACR,CAOF,IAAI,EACA,EACJ,GAAI,EACF,GAAI,CACF,IAAM,EAAW,MAAM,EAAgB,EAAI,QAAQ,EACnD,EAAY,EAAS,UACrB,EAAc,EAAS,OACzB,OAAS,EAAK,CAGZ,MADA,MAAM,EAAW,EAAY,EAAa,+BAD3B,aAAe,MAAQ,EAAI,QAAU,OAAO,CAAG,GACmB,EAC3E,CACR,KAKA,GAAY,EAAI,SAYlB,IAAI,EACJ,GAAI,EAAW,SAAW,OACxB,GAAI,CACF,EAAY,MAAM,EAChB,EACA,EAAW,YAAc,IAAA,GAAkD,CAAC,EAAvC,CAAE,UAAW,EAAW,SAAU,CACzE,EACA,GAAK,KAAK,CAAE,cAAa,WAAU,EAAG,6BAA6B,CACrE,OAAS,EAAK,CAGZ,GAAK,KAAK,CAAE,KAAI,EAAG,+DAA+D,CACpF,CAGF,IAAI,EACJ,GAAI,CAKF,IAAM,EAA4B,CAChC,SAAU,EACV,GAAI,EAAW,SAAW,IAAA,IAAa,CAAE,OAAQ,EAAW,MAAO,EACnE,GAAI,EAAW,YAAc,IAAA,IAAa,CAAE,UAAW,EAAW,SAAU,EAC5E,GAAI,EAAW,YAAc,IAAA,IAAa,CAAE,UAAW,EAAW,SAAU,EAC5E,GAAI,EAAW,iBAAmB,IAAA,IAAa,CAC7C,eAAgB,EAAW,cAC7B,EACA,GAAI,EAAW,QAAU,IAAA,IAAa,CAAE,MAAO,EAAW,KAAM,CAClE,EACA,EAAS,MAAM,EAAU,CACvB,cACA,SAAU,EAAI,MACd,SACA,YACA,KAAM,EACN,SAAU,EACV,UACA,WAAa,GACX,EAAI,eAAe,IAAc,IAAA,GAAyC,EAA7B,CAAE,GAAG,EAAU,WAAU,CAAY,EACpF,GAAI,IAAQ,IAAA,IAAa,CAAE,OAAQ,CAAI,CACzC,CAAC,CACH,OAAS,EAAK,CAQZ,MANA,MAAM,EAAW,EAAY,EADd,aAAe,MAAQ,EAAI,QAAU,OAAO,CAAG,CACd,EAC5C,GACF,MAAM,EAAY,CAAC,CAAC,MAAO,GAAwB,CACjD,GAAK,KAAK,CAAE,IAAK,CAAW,EAAG,oDAAoD,CACrF,CAAC,EAEG,CACR,CAqDA,GA/CI,GACF,MAAM,EAAY,CAAC,CAAC,MAAO,GAAwB,CACjD,GAAK,KAAK,CAAE,IAAK,CAAW,EAAG,wDAAwD,CACzF,CAAC,EAGH,MAAM,EAAW,OAAO,EAAa,CACnC,OAAQ,YACR,WAAY,IAAI,KAChB,OAAQ,CACN,SAAU,EAAO,SACjB,cAAe,EAAO,cACtB,WAAY,EAAO,WACnB,YAAa,EAAO,YACpB,WAAY,EAAO,WACnB,iBAAkB,EAAO,iBAIzB,eAAgB,EAAO,eACvB,gBAAiB,EAAO,eAC1B,EAOA,aAAc,KAAK,MAAM,KAAK,UAAU,EAAO,MAAM,CAAC,EAItD,eAAgB,KAAK,MAAM,KAAK,UAAU,EAAO,QAAQ,CAAC,CAC5D,CAAC,EAGD,EAA0B,EAAa,CACrC,SAAU,EAAO,SACjB,cAAe,EAAO,cACtB,WAAY,EAAO,UACrB,CAAC,EAOG,GAAY,EAAO,MAAM,OAAS,EACpC,GAAI,CACF,MAAM,EAAS,EAAO,MAAO,CAC3B,cACA,SACA,GAAI,IAAQ,IAAA,IAAa,CAAE,OAAQ,CAAI,CACzC,CAAC,EAGD,GAAK,KAAK,CAAE,cAAa,MAAO,EAAO,MAAM,MAAO,EAAG,2BAA2B,CACpF,OAAS,EAAK,CACZ,GAAK,KACH,CAAE,cAAa,MAAO,EAAO,MAAM,OAAQ,KAAI,EAK/C,yFACF,CACF,CAGF,GAAK,KACH,CACE,SAAU,EAAO,SACjB,cAAe,EAAO,cACtB,WAAY,EAAO,WACnB,YAAa,EAAO,YACpB,QAAS,EAAO,gBAClB,EACA,sBACF,CACF,CAAC,CACL,CAGA,eAAe,EAAW,EAA6B,EAAY,EAA+B,CAChG,GAAI,CACF,MAAM,EAAW,OAAO,EAAI,CAC1B,OAAQ,SACR,WAAY,IAAI,KAChB,aAAc,CAAE,MAAO,CAAO,CAChC,CAAC,EAMD,EAAuB,EAAI,CAAM,CACnC,MAAQ,CAGR,CACF,CA6CA,SAAgB,EACd,EACuB,CACvB,GAAM,CAAE,aAAY,UAAW,EAE/B,MAAO,CACL,UAAW,KAAO,IAA4C,CACxD,EAAM,UAAY,eAEtB,MAAM,EAAS,SAAY,CACzB,IAAM,EAAM,GAAQ,MAAM,CACxB,MAAO,mBACP,MAAO,EAAM,MACb,KAAM,aACR,CAAC,EAYK,EAAQ,EAAW,SAAS,EAK5B,GAAM,MAJU,EAAW,SAAS,CACxC,MAAO,EAAG,EAAM,WAAY,EAAM,KAAK,EACvC,MAAO,CACT,CAAC,EAAA,CACmB,GACpB,GAAI,CAAC,EAAK,CAKR,GAAK,KAAK,uDAAuD,EACjE,MACF,CAMA,GAAI,EAAI,SAAW,WAAa,EAAI,SAAW,UAAW,CACxD,GAAK,KACH,CAAE,YAAa,EAAI,GAAI,OAAQ,EAAI,MAAO,EAC1C,8EACF,EACA,MACF,CAQI,EAAI,SAAW,WACjB,GAAK,KACH,CAAE,YAAa,EAAI,GAAI,SAAU,EAAM,QAAS,EAChD,gGACF,EAGF,MAAM,EAAW,EAAY,EAAI,GAAI,EAAM,KAAK,EAChD,GAAK,KACH,CAAE,YAAa,EAAI,GAAI,SAAU,EAAM,SAAU,YAAa,EAAI,MAAO,EACzE,2CACF,CACF,CAAC,CACH,CACF,CACF"}
+{"version":3,"file":"worker.mjs","names":[],"sources":["../src/worker.ts"],"sourcesContent":["/**\n * Queue handler for `imports:run`. The route handler that creates the\n * `import_run` row enqueues this job carrying `{ importRunId }`; the\n * worker module here resolves the run, the transform, and the ES\n * client, then runs `runImport` and writes results back.\n *\n * Per PLAN-ECOMMERCE.md PR 7 (PoC scope): one job type, no resumability.\n * A failed handler attempt re-runs from row 1 on retry — fine for the\n * tens-of-MB feed sizes the PoC exercises.\n *\n * Per CLAUDE.md \"leaf packages\" boundary rule: this handler reads the\n * import_run row via an injected `AdminClient`, NOT a direct `db.select`.\n * Construction happens at plugin init in `plugin.ts`.\n */\n\nimport { type Logger, runAsCli } from '@murumets-ee/core'\nimport {\n  defineJob,\n  type JobDefinition,\n  type QueueJobDeadEvent,\n  type QueueTerminalNotifier,\n} from '@murumets-ee/queue/client'\nimport type { EsClientLike } from '@murumets-ee/search-elasticsearch'\nimport { and, desc, eq } from 'drizzle-orm'\nimport type { PgColumn } from 'drizzle-orm/pg-core'\nimport { z } from 'zod'\nimport type { ImportRunClient, ImportRunStatus } from './entities/import-run.js'\nimport { ErrorTracker, mergeErrorTrackerSnapshots } from './error-tracker.js'\nimport { detectPriceAnomaly, parsePriceStats } from './price-stats.js'\nimport { type ReadFeedOptions, readPersistedFeedConfig } from './read-feed.js'\nimport {\n  publishImportRunFailed,\n  publishImportRunStarted,\n  publishImportRunSucceeded,\n} from './realtime/publish.js'\nimport { type ImportRunProgress, runImport } from './runner.js'\nimport { countDataRows } from './streaming.js'\nimport type { DirectionalPair, TransformRegistry } from './transform.js'\n\n/**\n * Structural subset of `JobContext` the handler reads. Declared\n * locally so we don't depend on a `JobHandler` import — the queue\n * package doesn't re-export it from any subpath today (matches the\n * `SendEmailJobContext` pattern in `@murumets-ee/notifications`).\n *\n * Function-param contravariance lets a handler returned with this\n * narrower shape be assigned to the wider `JobHandler<TPayload>` that\n * `registerJob` expects.\n */\nexport interface ImportsRunJobContext {\n  id: string\n  payload: ImportsRunJobPayload\n  updateProgress(data: ImportRunProgress): void\n}\n\n/**\n * Payload schema. `importRunId` is the `import_run.id` UUID — the worker\n * looks up everything else (filePath, transformName, params, …) from\n * that row. Keeping the payload tiny means the queue's progress JSON\n * column never bloats with feed metadata duplicated across `toolkit_jobs`.\n */\nexport const importsRunJobPayloadSchema = z.object({\n  importRunId: z.string().uuid(),\n})\nexport type ImportsRunJobPayload = z.infer<typeof importsRunJobPayloadSchema>\n\n/**\n * Job definition. Consumers register their handler against this with\n * `registerJob(importsRunJob, createRunImportHandler({...}))`.\n *\n * `defaultRetries: 0` — re-running a multi-batch import against the\n * same `import_run.id` while the previous handler may still be writing\n * is a footgun (duplicate batches, double-counted progress). The\n * operator handles retries explicitly via PR 8a's \"retry\" button by\n * creating a new `import_run` row. When the resumable design lands\n * (post-PoC), retries become safe to enable.\n */\nexport const importsRunJob: JobDefinition<ImportsRunJobPayload> = defineJob({\n  name: 'imports:run',\n  description: 'Stream a feed file, transform rows, bulk-write to Elasticsearch.',\n  schema: importsRunJobPayloadSchema,\n  defaultRetries: 0,\n})\n\n/**\n * Resolves the ES client at handler-invocation time. A function (rather\n * than the bare `EsClientLike`) so the consumer can lazy-construct the\n * client — typical Next.js setups create the ES connection in a\n * route-handler initialiser, not at plugin-init time.\n */\nexport type EsClientResolver = () => EsClientLike | Promise<EsClientLike>\n\n/**\n * Ensure the destination index/alias exists before the first bulk-write.\n * Invoked once per job, BEFORE {@link runImport}. Idempotent: the canonical\n * implementation (`ensureAliasedIndex` in `@murumets-ee/search-elasticsearch`)\n * no-ops when the alias already resolves.\n *\n * **Why it lives here.** This is the home of the foot-gun it guards: a worker\n * that bulk-writes to alias `<x>` before the alias exists makes ES\n * auto-create a physical `<x>` index with default `text` mappings, breaking\n * aggregations forever. The bootstrap used to run at app boot, which turned\n * \"ES is down\" into a full-admin 500 (issue #610). Moving it onto the import\n * path keeps the guarantee while scoping ES availability to imports/search —\n * the rest of the admin boots and runs with ES down.\n */\nexport type EnsureIndexResolver = () => Promise<unknown>\n\n/**\n * Resolves the value of `import_run.filePath` (whatever the upload\n * route persisted there — typically a storage adapter key) to a\n * readable LOCAL filesystem path that {@link runImport} can hand to\n * `node:fs.createReadStream`.\n *\n * **Why this exists:** the streaming reader (`streamFeed`) reads from\n * local disk via `createReadStream`. The upload route may persist a\n * remote-storage object key (R2, S3, …) on `import_run.filePath`\n * because that's what `@murumets-ee/storage` returns. Without a\n * resolver, `createReadStream('uploads/2026/05/<uuid>/feed.txt')`\n * crashes with `ENOENT`. The resolver is the documented integration\n * point — typical wiring downloads the storage object to a tmpfile\n * and returns its path.\n *\n * **`cleanup`** runs after the run finishes (success OR failure). The\n * worker awaits it best-effort — a failed cleanup logs but does not\n * crash the run.\n *\n * **No-resolver fallback:** when `resolveFilePath` is unset, the\n * worker treats `import_run.filePath` as already a local FS path\n * (back-compat with the original PoC design where uploads landed on\n * local disk). This stays valid for fixture-driven tests + on-disk\n * deployments.\n */\nexport type FilePathResolver = (\n  storageKey: string,\n) => Promise<{ localPath: string; cleanup?: () => Promise<void> }>\n\n/**\n * Context handed to a {@link PairSink} alongside the harvested pairs. Carries the\n * run identity so the sink can build `source` attribution\n * (`feed:<supplierId>:<importBatchId>`, D20) and scope its writes.\n */\nexport interface PairSinkContext {\n  importRunId: string\n  /** The `import_run.params` object — the sink reads `supplierId` / `importBatchId`. */\n  params: Record<string, unknown>\n  /**\n   * The run's structured logger (child), forwarded so the sink's INTERNAL telemetry —\n   * a params-parse miss that skips ingestion, an over-merge flag — is visible. Without it\n   * a misconfigured run looks successful from the worker's vantage (a params-miss returns\n   * normally, so the worker logs \"applied\"). Optional; the sink no-ops its logs if absent.\n   */\n  logger?: Logger\n}\n\n/**\n * Optional post-run consumer for the directional supersession pairs the runner\n * harvested as a side-output (commerce-demo PR6, SD007). Invoked ONCE after a\n * successful run with the full in-memory pair buffer (read-once / merge-after) so\n * the consumer (the commerce replacement-merge engine) can build equivalence groups\n * and bulk-write them to Postgres.\n *\n * **Why injected, not imported.** The merge writes the CONSUMER's Postgres tables,\n * but the runner + this worker live in the `@murumets-ee/imports` leaf, which must\n * not depend on commerce. The app composes the sink (commerce exports it, the app\n * passes it to `imports({ pairSink })`) — the same injection shape as `esClient` /\n * `resolveFilePath` / `ensureIndex`. The boundary holds.\n *\n * **Best-effort.** A sink throw is logged and SWALLOWED — the catalogue import\n * (the ES write) has already succeeded and must not be undone by a secondary\n * replacement-ingestion failure. The merge is idempotent, so a re-import recovers.\n */\nexport type PairSink = (pairs: readonly DirectionalPair[], ctx: PairSinkContext) => Promise<void>\n\nexport interface RunImportHandlerConfig {\n  /** AdminClient over the `import_run` entity. */\n  importRuns: ImportRunClient\n  /** Transform registry to dispatch against. Defaults to the process-global singleton. */\n  transforms: TransformRegistry\n  /** Resolver for the ES client. */\n  esClient: EsClientResolver\n  /** ES alias / index to bulk-write into. Per D6 callers always pass an alias. */\n  esIndex: string\n  /**\n   * Optional bootstrap for the destination index/alias, run once per job\n   * before any bulk-write. See {@link EnsureIndexResolver}. Omit for\n   * deployments that bootstrap the index out-of-band (e.g. a reindex job).\n   */\n  ensureIndex?: EnsureIndexResolver\n  /**\n   * Optional resolver for `import_run.filePath`. See {@link FilePathResolver}.\n   * Required when uploads land in remote storage (R2/S3); optional for\n   * on-disk PoC setups.\n   */\n  resolveFilePath?: FilePathResolver\n  /**\n   * Optional consumer for the directional supersession pairs harvested as a\n   * side-output (commerce-demo PR6, SD007). See {@link PairSink}. Omit for\n   * deployments that don't ingest replacement codes — the pairs are simply\n   * discarded after the run.\n   */\n  pairSink?: PairSink\n  /** Optional structured logger. Defaults to silent. */\n  logger?: Logger\n}\n\n/**\n * Build the `JobHandler` for {@link importsRunJob}. The returned\n * handler is what gets passed to `registerJob`.\n *\n * The body is wrapped in `runAsCli` so AdminClient calls inside have a\n * synthetic `cli` admin context — `auditable()` records `updatedBy:\n * 'cli'` rather than NULL, and the firewall checker passes. This is\n * the documented worker entry-point pattern (see `runAsCli` JSDoc in\n * `@murumets-ee/core`).\n */\nexport function createRunImportHandler(\n  config: RunImportHandlerConfig,\n): (job: ImportsRunJobContext) => Promise<void> {\n  const {\n    importRuns,\n    transforms,\n    esClient,\n    esIndex,\n    resolveFilePath,\n    ensureIndex,\n    pairSink,\n    logger,\n  } = config\n\n  return async (job: ImportsRunJobContext): Promise<void> =>\n    runAsCli(async () => {\n      const { importRunId } = job.payload\n      const log = logger?.child({ jobId: job.id, importRunId, type: 'imports:run' })\n\n      const run = await importRuns.findById(importRunId)\n      if (!run) {\n        // Row was deleted between enqueue and dispatch — nothing to do.\n        // Don't throw: a thrown handler retries with no payoff because\n        // the row stays gone. Log + return so the queue marks the job\n        // completed.\n        log?.warn('import_run row not found — skipping')\n        return\n      }\n\n      // Re-claim detection. The queue's `recoverStuckJobs` sweep flips a\n      // crashed worker's `toolkit_jobs` row from `processing` back to\n      // `pending` — but the queue layer has no knowledge of `import_run`,\n      // so this row stays at `status='running'` from the prior attempt.\n      // When the queue re-dispatches the job, we land here with a row\n      // that the operator-facing Active panel has been showing as live\n      // for however long lock-recovery took (default 30 min). The\n      // existing `update` below overwrites `queueJobId` + `startedAt` so\n      // there is no STATE corruption — but emit a structured warning so\n      // ops can correlate this re-run with the worker crash that\n      // produced it. The sibling `sweepStaleImportRuns` job below catches\n      // the case where re-dispatch never happens (queue marks the job\n      // dead, or the queue worker is offline long-term).\n      if (run.status === 'running') {\n        log?.warn(\n          {\n            previousQueueJobId: run.queueJobId,\n            previousStartedAt: run.startedAt,\n          },\n          'imports: re-claiming a row left in running by a prior worker — likely a crash recovered by queue lock-recovery',\n        )\n      }\n\n      // `transformName`, `filePath`, `label` are declared `field.text({\n      // required: true })` so the inferred DTO already types them as\n      // `string` — no cast needed. `params` is `field.json()` which\n      // infers as `JsonValue | undefined`; narrow to a record so the\n      // transform receives an object.\n      const transform = transforms.get(run.transformName)\n      if (!transform) {\n        await markFailed(importRuns, importRunId, `Unknown transform \"${run.transformName}\"`)\n        throw new Error(`No transform registered for name \"${run.transformName}\"`)\n      }\n      const params: Record<string, unknown> =\n        run.params && typeof run.params === 'object' && !Array.isArray(run.params)\n          ? (run.params as Record<string, unknown>)\n          : {}\n\n      // Reader config (format/delimiter/header/sheet) travels in\n      // `params.feed` for mapping-driven runs (commerce-demo PR3). Carmaker\n      // runs omit it → empty config → tab-delimited default, so the existing\n      // path is byte-for-byte unchanged. Re-validate here as defense-in-depth\n      // even though the enqueuing route already validated the saved mapping.\n      // Shared with the #842 audit, which has to rehydrate IDENTICALLY or it\n      // re-reads a past run with options that run never used.\n      const feedParsed = readPersistedFeedConfig(params)\n      if (!feedParsed.success) {\n        const reason = feedParsed.error.issues[0]?.message ?? 'invalid feed reader config'\n        await markFailed(importRuns, importRunId, `Invalid params.feed: ${reason}`)\n        throw new Error(`Invalid params.feed for import_run ${importRunId}: ${reason}`)\n      }\n      const feedConfig = feedParsed.data\n\n      await importRuns.update(importRunId, {\n        status: 'running' satisfies ImportRunStatus,\n        queueJobId: job.id,\n        startedAt: new Date(),\n      })\n      // Realtime: row transitioned pending → running. Subscribers\n      // refetch to surface the change in the imports admin page.\n      // Best-effort — a publish failure is logged inside `publishSafe`\n      // and does NOT roll back the row update.\n      publishImportRunStarted(importRunId)\n\n      let resolvedClient: EsClientLike\n      try {\n        resolvedClient = await esClient()\n      } catch (err) {\n        const reason = err instanceof Error ? err.message : String(err)\n        await markFailed(importRuns, importRunId, `Could not resolve ES client: ${reason}`)\n        throw err\n      }\n\n      // Ensure the destination alias exists BEFORE any bulk-write. Moved here\n      // from app boot (issue #610): the foot-gun is the worker auto-creating a\n      // mis-mapped physical index, so the guard belongs on the import path —\n      // not gating the whole app on ES. Idempotent + scoped: an ES outage (or\n      // the `ensureAliasedIndex` physical-collision guard firing) fails THIS\n      // job loudly while the rest of the admin keeps working.\n      if (ensureIndex) {\n        try {\n          await ensureIndex()\n        } catch (err) {\n          const reason = err instanceof Error ? err.message : String(err)\n          await markFailed(importRuns, importRunId, `Could not ensure ES index/alias: ${reason}`)\n          throw err\n        }\n      }\n\n      // Materialise the file locally if the consumer wired a resolver.\n      // The `cleanup` callback (when present) runs in the `finally`\n      // below regardless of run outcome so a tmpfile created here\n      // doesn't outlive the handler.\n      let localPath: string\n      let fileCleanup: (() => Promise<void>) | undefined\n      if (resolveFilePath) {\n        try {\n          const resolved = await resolveFilePath(run.filePath)\n          localPath = resolved.localPath\n          fileCleanup = resolved.cleanup\n        } catch (err) {\n          const reason = err instanceof Error ? err.message : String(err)\n          await markFailed(importRuns, importRunId, `Could not resolve filePath: ${reason}`)\n          throw err\n        }\n      } else {\n        // Back-compat: no resolver wired → treat `filePath` as already\n        // a local FS path. Fine for fixture-driven tests + the original\n        // on-disk PoC design.\n        localPath = run.filePath\n      }\n\n      // Pre-pass line count: a 16 MB tab-delimited feed counts in\n      // ~50ms; a 500 MB feed in <2s. Cheap relative to the full\n      // import. Gives the operator a denominator (`processed / total`)\n      // so the imports-page panel can render an actual percent\n      // instead of just \"rows so far\".\n      // Pre-pass row count is a `\\n`-tally over a delimited file — meaningless\n      // on an xlsx binary, so skip it there (the UI falls back to bare counts).\n      // `hasHeader` from the feed config keeps the header-exclusion correct for\n      // headerless mappings.\n      let totalRows: number | undefined\n      if (feedConfig.format !== 'xlsx') {\n        try {\n          totalRows = await countDataRows(\n            localPath,\n            feedConfig.hasHeader !== undefined ? { hasHeader: feedConfig.hasHeader } : {},\n          )\n          log?.info({ importRunId, totalRows }, 'imports: pre-pass row count')\n        } catch (err) {\n          // Don't fail the run on a count miss — the actual import is\n          // unaffected; the operator just loses the percent denominator.\n          log?.warn({ err }, 'imports: pre-pass row count failed — continuing without total')\n        }\n      }\n\n      let result: Awaited<ReturnType<typeof runImport>>\n      try {\n        // Build feed options with the conditional-spread idiom (NOT\n        // `...feedConfig`) so absent optionals stay ABSENT — spreading a\n        // `T | undefined` value would set the key to `undefined`, which\n        // `exactOptionalPropertyTypes` rejects against `ReadFeedOptions`.\n        const feedOpts: ReadFeedOptions = {\n          filePath: localPath,\n          ...(feedConfig.format !== undefined && { format: feedConfig.format }),\n          ...(feedConfig.delimiter !== undefined && { delimiter: feedConfig.delimiter }),\n          ...(feedConfig.hasHeader !== undefined && { hasHeader: feedConfig.hasHeader }),\n          ...(feedConfig.headerRowIndex !== undefined && {\n            headerRowIndex: feedConfig.headerRowIndex,\n          }),\n          ...(feedConfig.sheet !== undefined && { sheet: feedConfig.sheet }),\n        }\n        result = await runImport({\n          importRunId,\n          runLabel: run.label,\n          params,\n          transform,\n          feed: feedOpts,\n          esClient: resolvedClient,\n          esIndex,\n          onProgress: (progress) =>\n            job.updateProgress(totalRows !== undefined ? { ...progress, totalRows } : progress),\n          ...(log !== undefined && { logger: log }),\n        })\n      } catch (err) {\n        const reason = err instanceof Error ? err.message : String(err)\n        await markFailed(importRuns, importRunId, reason)\n        if (fileCleanup) {\n          await fileCleanup().catch((cleanupErr: unknown) => {\n            log?.warn({ err: cleanupErr }, 'filePath cleanup after failed run threw — ignoring')\n          })\n        }\n        throw err\n      }\n\n      // Cleanup runs before the success-update so a cleanup throw\n      // surfaces as a logged warning rather than a crashed handler at\n      // the worst possible moment (just after the row was about to be\n      // marked succeeded).\n      if (fileCleanup) {\n        await fileCleanup().catch((cleanupErr: unknown) => {\n          log?.warn({ err: cleanupErr }, 'filePath cleanup after successful run threw — ignoring')\n        })\n      }\n\n      // Value-drift detection (F010 / #602). The 6b header-presence drift check\n      // (commerce's register-handler.ts, upload-time) only catches a STRUCTURAL\n      // change — a mapped column vanishing from the header row. It cannot catch a\n      // supplier silently flipping their decimal/thousands separator: no header\n      // changes, and `import_mapping.numberFormat` stays the same, so every price\n      // parses as a numerically valid but silently WRONG value. Compare THIS run's\n      // price distribution against the most recent prior SUCCEEDED run for the same\n      // (transformName, brandId, supplierId) and, on a magnitude-scale shift, append\n      // a NON-FATAL `price_anomaly` warning. Unlike 6b's 409, this never blocks the\n      // run — a genuine repricing is a legitimate reason the median would move, so\n      // this is advisory only.\n      let warningsToWrite = result.warnings\n      const brandIdForDrift = typeof params.brandId === 'string' ? params.brandId : undefined\n      const supplierIdForDrift =\n        typeof params.supplierId === 'string' ? params.supplierId : undefined\n      if (brandIdForDrift && supplierIdForDrift && result.priceStats.median !== null) {\n        try {\n          // `getTable()` is `PgTableWithColumns<any>` (dynamic entity columns —\n          // see the documented-`any` list in CLAUDE.md); cast to the columns we\n          // filter/sort on, same pattern as `createImportsRunDeadListener` below.\n          const table = importRuns.getTable() as unknown as {\n            transformName: PgColumn\n            status: PgColumn\n            createdAt: PgColumn\n          }\n          // Bounded lookup (limit 50), filtered to this transform + succeeded runs\n          // in SQL, then narrowed to the same (brandId, supplierId) in memory —\n          // `params` is an opaque per-transform JSONB blob (CLAUDE.md \"no raw\n          // SQL\": no `params->>'brandId'` jsonb operator here). 50 candidates is\n          // far more than any operator re-runs between two imports for the same\n          // supplier/brand pair before this one.\n          const candidates = await importRuns.findMany({\n            where: and(eq(table.transformName, run.transformName), eq(table.status, 'succeeded')),\n            orderBy: desc(table.createdAt),\n            limit: 50,\n          })\n          const prior = candidates.find((c) => {\n            if (c.id === importRunId) return false\n            const p = c.params\n            if (!p || typeof p !== 'object' || Array.isArray(p)) return false\n            const pr = p as Record<string, unknown>\n            return pr.brandId === brandIdForDrift && pr.supplierId === supplierIdForDrift\n          })\n          // CLAUDE.md \"No Unbounded Fan-Out\": \"log() whatever you drop — silent\n          // truncation reads as 'covered everything' when it didn't.\" `transformName`\n          // is shared across every (brand, supplier) pair using that mapping (e.g.\n          // all 17 GIGA brands share `commerce:mapping`), so a genuinely prior run\n          // for THIS pair can fall outside the 50-candidate window as import cadence\n          // and supplier count grow. Distinguish that from \"no prior run exists\" —\n          // both look identical downstream (no anomaly check runs) but only one is\n          // an operator-actionable coverage gap.\n          if (!prior && candidates.length >= 50) {\n            log?.warn(\n              {\n                brandId: brandIdForDrift,\n                supplierId: supplierIdForDrift,\n                transformName: run.transformName,\n              },\n              'imports: price-drift check found no prior run for this (brand, supplier) within the 50 most recent succeeded runs for this transform — a real prior run may exist beyond the bound, not necessarily absent',\n            )\n          }\n          const priorStats = prior ? parsePriceStats(prior.priceStats) : null\n          const anomaly = priorStats ? detectPriceAnomaly(priorStats, result.priceStats) : null\n          if (anomaly) {\n            const higher = anomaly.ratio >= 1\n            const anomalyTracker = new ErrorTracker()\n            anomalyTracker.addError(\n              -1,\n              'price_anomaly',\n              `Median purchase price is ${(higher ? anomaly.ratio : 1 / anomaly.ratio).toFixed(1)}x ${higher ? 'higher' : 'lower'} than the previous successful run for this supplier/brand (previous run ${prior?.id ?? 'unknown'}: median ${anomaly.prevMedian}; this run: median ${anomaly.currMedian}, minor units). Check for a decimal or thousands-separator format change in the source feed.`,\n              'priceStats.median',\n              {\n                previousImportRunId: prior?.id ?? null,\n                previousMedian: anomaly.prevMedian,\n                currentMedian: anomaly.currMedian,\n                ratio: anomaly.ratio,\n              },\n            )\n            warningsToWrite = mergeErrorTrackerSnapshots(result.warnings, anomalyTracker.snapshot())\n            log?.warn(\n              { previousImportRunId: prior?.id, ratio: anomaly.ratio },\n              'imports: price_anomaly — median purchase price shifted by a magnitude vs the prior run',\n            )\n          }\n        } catch (err) {\n          // Best-effort — a comparison failure must never fail an otherwise-\n          // successful catalogue import. Same \"log + continue\" discipline as the\n          // pairSink / cleanup best-effort paths elsewhere in this handler.\n          log?.warn({ err }, 'imports: price-drift comparison failed — continuing without it')\n        }\n      }\n\n      await importRuns.update(importRunId, {\n        status: 'succeeded' satisfies ImportRunStatus,\n        finishedAt: new Date(),\n        totals: {\n          rowsRead: result.rowsRead,\n          rowsSucceeded: result.rowsSucceeded,\n          rowsFailed: result.rowsFailed,\n          rowsSkipped: result.rowsSkipped,\n          rowsWarned: result.rowsWarned,\n          batchesCompleted: result.batchesCompleted,\n          // Phase-2 core-charge attach (commerce-demo PR7) — durable on the run row so\n          // an operator sees how many `*S` deposits attached vs orphaned, not just the\n          // orphan warnings in `warningSummary`.\n          patchesApplied: result.patchesApplied,\n          patchesOrphaned: result.patchesOrphaned,\n        },\n        // The typed `ErrorTrackerSnapshot` is fully JSON-serialisable but\n        // doesn't structurally satisfy entity's recursive `JsonValue`\n        // (objects without an index signature aren't `Record<string,\n        // JsonValue>`). Round-trip through `JSON.parse(JSON.stringify(...))`\n        // to land as the recursive JSON shape — the runtime cost is one\n        // serialise per import_run completion, the type-cast cost is zero.\n        errorSummary: JSON.parse(JSON.stringify(result.errors)),\n        // Non-fatal warnings (commerce-demo PR4) — kept separate from\n        // `errorSummary` so a `no_purchase_price` advisory never reads as a\n        // failed row. Same serialise-roundtrip rationale as `errorSummary`.\n        // `warningsToWrite` is `result.warnings` UNLESS the price-drift check\n        // above appended a `price_anomaly` entry (F010 / #602).\n        warningSummary: JSON.parse(JSON.stringify(warningsToWrite)),\n        // Price-distribution snapshot (F010 / #602) — read back by the NEXT\n        // run for this (transformName, brandId, supplierId) to detect drift.\n        priceStats: JSON.parse(JSON.stringify(result.priceStats)),\n      })\n      // Realtime: row transitioned to terminal-success. Subscribers\n      // refetch to move the row from Active → Recent.\n      publishImportRunSucceeded(importRunId, {\n        rowsRead: result.rowsRead,\n        rowsSucceeded: result.rowsSucceeded,\n        rowsFailed: result.rowsFailed,\n      })\n\n      // Side-output: hand the harvested supersession pairs to the injected sink\n      // (commerce-demo PR6, SD007) AFTER the catalogue run is marked succeeded — the\n      // ES import is primary; replacement ingestion is secondary. Best-effort: a sink\n      // throw is logged + swallowed so a Postgres hiccup can't undo a successful\n      // catalogue import (the merge is idempotent → a re-import recovers).\n      if (pairSink && result.pairs.length > 0) {\n        try {\n          await pairSink(result.pairs, {\n            importRunId,\n            params,\n            ...(log !== undefined && { logger: log }),\n          })\n          // \"invoked\" not \"applied\": the sink may internally skip (e.g. missing run params);\n          // it logs its own applied/skipped outcome via the forwarded logger.\n          log?.info({ importRunId, pairs: result.pairs.length }, 'imports: pairSink invoked')\n        } catch (err) {\n          log?.warn(\n            { importRunId, pairs: result.pairs.length, err },\n            // NOT \"skipped\": a sink may apply part of the batch before failing (the commerce\n            // replacement sink commits per batch), so some pairs can already have landed.\n            // \"Incomplete\" is the only claim this boundary can honestly make about a sink\n            // whose internals it does not know.\n            'imports: pairSink threw — replacement ingestion incomplete, catalogue import unaffected',\n          )\n        }\n      }\n\n      log?.info(\n        {\n          rowsRead: result.rowsRead,\n          rowsSucceeded: result.rowsSucceeded,\n          rowsFailed: result.rowsFailed,\n          rowsSkipped: result.rowsSkipped,\n          batches: result.batchesCompleted,\n        },\n        'import_run completed',\n      )\n    })\n}\n\n/** Write a `failed` row with a fatal error reason. Best-effort — logs but does not throw on DB failure. */\nasync function markFailed(importRuns: ImportRunClient, id: string, reason: string): Promise<void> {\n  try {\n    await importRuns.update(id, {\n      status: 'failed' satisfies ImportRunStatus,\n      finishedAt: new Date(),\n      errorSummary: { fatal: reason },\n    })\n    // Realtime: row transitioned to terminal-failure. Subscribers\n    // refetch to surface the failure in the Recent panel. Publish\n    // is best-effort and only fires when the persist succeeds —\n    // skipping it on persist failure keeps subscribers consistent\n    // with the DB (no event for a row that didn't transition).\n    publishImportRunFailed(id, reason)\n  } catch {\n    // The handler already threw / will throw — the queue will surface\n    // the original error. A second failure here just logs noise.\n  }\n}\n\nexport interface ImportsRunDeadListenerConfig {\n  importRuns: ImportRunClient\n  /** Optional structured logger. */\n  logger?: Logger\n}\n\n/**\n * Build a {@link QueueTerminalNotifier} that mirrors `imports:run`\n * dead-job events into `import_run.status='failed'`.\n *\n * **Why this exists.** The `imports:run` handler's normal failure path\n * already calls `markFailed` before re-throwing, so a synchronous\n * handler error reaches the operator-visible `failed` status without\n * help from anyone. The case THIS listener catches is the one\n * `markFailed` never gets to: the worker process is killed mid-handler\n * (OOM, container restart, kill -9), the queue's lock-recovery sweep\n * eventually reclaims the row, and the queue ultimately marks the job\n * `dead` (defaultRetries: 0 + repeated process death = dead-letter\n * with `import_run.status` still `running`). Without this listener\n * the row sits in the operator's Active panel forever.\n *\n * **Generic primitive.** This is a thin subscriber over the queue's\n * multi-subscriber `QueueTerminalNotifier` registry — the same shape\n * graphile-worker (`job:failed`) and BullMQ (`failed`) expose. Other\n * domain plugins that mirror queue state (ticketing-reply, future\n * exports/reports/...) can register their own listeners exactly the\n * same way without coordinating via a per-domain sweep job.\n *\n * **Filtering.** The listener is invoked for EVERY dead event in the\n * process. We early-return on any `jobType` other than `imports:run`\n * so the per-event cost on unrelated jobs is a single property read.\n *\n * **Idempotency.** A dead event delivered after the row has already\n * reached its own terminal state (succeeded / failed / cancelled) is\n * ignored — the only transition we make is `running → failed`. This\n * keeps a late replay or out-of-order delivery from clobbering a\n * row whose worker reached `markFailed` faster than the queue's\n * dead-letter dispatch.\n *\n * The handler is wrapped in `runAsCli` so AdminClient calls inside\n * have a synthetic CLI admin context — same pattern as\n * {@link createRunImportHandler}.\n */\nexport function createImportsRunDeadListener(\n  config: ImportsRunDeadListenerConfig,\n): QueueTerminalNotifier {\n  const { importRuns, logger } = config\n\n  return {\n    onJobDead: async (event: QueueJobDeadEvent): Promise<void> => {\n      if (event.jobType !== 'imports:run') return\n\n      await runAsCli(async () => {\n        const log = logger?.child({\n          event: 'imports.run.dead',\n          jobId: event.jobId,\n          type: 'imports:run',\n        })\n\n        // Look up the mirror row by `queueJobId`. Cap to 1 — the\n        // upload route writes one `import_run` per enqueued job.\n        //\n        // `AdminClient.getTable()` returns `PgTableWithColumns<any>`\n        // because entity tables have dynamic column shapes (see the\n        // `getTable()` JSDoc in `packages/entity/src/admin/client.ts`\n        // and the documented-`any`-cases list in CLAUDE.md). Cast\n        // the table to a structural shape covering only the column\n        // we touch, so both `eq()` and type-coverage see a typed\n        // operand instead of an `any`-derived property access.\n        const table = importRuns.getTable() as unknown as { queueJobId: PgColumn }\n        const matches = await importRuns.findMany({\n          where: eq(table.queueJobId, event.jobId),\n          limit: 1,\n        })\n        const row = matches[0]\n        if (!row) {\n          // No mirror row for this dead job. Either it was deleted\n          // by the operator, or the dead event is for a job\n          // unrelated to imports despite the type filter\n          // (shouldn't happen, but is cheap to log).\n          log?.warn('queue.job.dead with no matching import_run — skipping')\n          return\n        }\n\n        // Only transition rows that haven't already reached their\n        // own terminal state. A late dead-event must not overwrite\n        // a `succeeded` row that the worker reached via the normal\n        // path before the queue's dead-letter dispatch fired.\n        if (row.status !== 'running' && row.status !== 'pending') {\n          log?.info(\n            { importRunId: row.id, status: row.status },\n            'queue.job.dead arrived for a row already in a terminal state — leaving as-is',\n          )\n          return\n        }\n\n        // Dead-from-`pending` is the rarer + more interesting case:\n        // the worker process died (or never claimed) BEFORE it could\n        // flip the row to `running`. Log it at warn so ops can spot\n        // a queue+worker-startup pathology that the standard `running\n        // → failed` transition wouldn't surface. Dead-from-`running`\n        // is the documented \"crash mid-import\" case — info is enough.\n        if (row.status === 'pending') {\n          log?.warn(\n            { importRunId: row.id, attempts: event.attempts },\n            'queue.job.dead arrived for an import_run still in pending — worker crashed BEFORE claim/lookup',\n          )\n        }\n\n        await markFailed(importRuns, row.id, event.error)\n        log?.info(\n          { importRunId: row.id, attempts: event.attempts, priorStatus: row.status },\n          'queue.job.dead → import_run marked failed',\n        )\n      })\n    },\n  }\n}\n"],"mappings":"4TA6DA,MAAa,EAA6B,EAAE,OAAO,CACjD,YAAa,EAAE,OAAO,CAAC,CAAC,KAAK,CAC/B,CAAC,EAcY,EAAqD,EAAU,CAC1E,KAAM,cACN,YAAa,mEACb,OAAQ,EACR,eAAgB,CAClB,CAAC,EAsID,SAAgB,EACd,EAC8C,CAC9C,GAAM,CACJ,aACA,aACA,WACA,UACA,kBACA,cACA,WACA,UACE,EAEJ,OAAO,KAAO,IACZ,EAAS,SAAY,CACnB,GAAM,CAAE,eAAgB,EAAI,QACtB,EAAM,GAAQ,MAAM,CAAE,MAAO,EAAI,GAAI,cAAa,KAAM,aAAc,CAAC,EAEvE,EAAM,MAAM,EAAW,SAAS,CAAW,EACjD,GAAI,CAAC,EAAK,CAKR,GAAK,KAAK,qCAAqC,EAC/C,MACF,CAeI,EAAI,SAAW,WACjB,GAAK,KACH,CACE,mBAAoB,EAAI,WACxB,kBAAmB,EAAI,SACzB,EACA,gHACF,EAQF,IAAM,EAAY,EAAW,IAAI,EAAI,aAAa,EAClD,GAAI,CAAC,EAEH,MADA,MAAM,EAAW,EAAY,EAAa,sBAAsB,EAAI,cAAc,EAAE,EAC1E,MAAM,qCAAqC,EAAI,cAAc,EAAE,EAE3E,IAAM,EACJ,EAAI,QAAU,OAAO,EAAI,QAAW,UAAY,CAAC,MAAM,QAAQ,EAAI,MAAM,EACpE,EAAI,OACL,CAAC,EASD,EAAa,EAAwB,CAAM,EACjD,GAAI,CAAC,EAAW,QAAS,CACvB,IAAM,EAAS,EAAW,MAAM,OAAO,EAAE,EAAE,SAAW,6BAEtD,MADA,MAAM,EAAW,EAAY,EAAa,wBAAwB,GAAQ,EAChE,MAAM,sCAAsC,EAAY,IAAI,GAAQ,CAChF,CACA,IAAM,EAAa,EAAW,KAE9B,MAAM,EAAW,OAAO,EAAa,CACnC,OAAQ,UACR,WAAY,EAAI,GAChB,UAAW,IAAI,IACjB,CAAC,EAKD,EAAwB,CAAW,EAEnC,IAAI,EACJ,GAAI,CACF,EAAiB,MAAM,EAAS,CAClC,OAAS,EAAK,CAGZ,MADA,MAAM,EAAW,EAAY,EAAa,gCAD3B,aAAe,MAAQ,EAAI,QAAU,OAAO,CAAG,GACoB,EAC5E,CACR,CAQA,GAAI,EACF,GAAI,CACF,MAAM,EAAY,CACpB,OAAS,EAAK,CAGZ,MADA,MAAM,EAAW,EAAY,EAAa,oCAD3B,aAAe,MAAQ,EAAI,QAAU,OAAO,CAAG,GACwB,EAChF,CACR,CAOF,IAAI,EACA,EACJ,GAAI,EACF,GAAI,CACF,IAAM,EAAW,MAAM,EAAgB,EAAI,QAAQ,EACnD,EAAY,EAAS,UACrB,EAAc,EAAS,OACzB,OAAS,EAAK,CAGZ,MADA,MAAM,EAAW,EAAY,EAAa,+BAD3B,aAAe,MAAQ,EAAI,QAAU,OAAO,CAAG,GACmB,EAC3E,CACR,KAKA,GAAY,EAAI,SAYlB,IAAI,EACJ,GAAI,EAAW,SAAW,OACxB,GAAI,CACF,EAAY,MAAM,EAChB,EACA,EAAW,YAAc,IAAA,GAAkD,CAAC,EAAvC,CAAE,UAAW,EAAW,SAAU,CACzE,EACA,GAAK,KAAK,CAAE,cAAa,WAAU,EAAG,6BAA6B,CACrE,OAAS,EAAK,CAGZ,GAAK,KAAK,CAAE,KAAI,EAAG,+DAA+D,CACpF,CAGF,IAAI,EACJ,GAAI,CAKF,IAAM,EAA4B,CAChC,SAAU,EACV,GAAI,EAAW,SAAW,IAAA,IAAa,CAAE,OAAQ,EAAW,MAAO,EACnE,GAAI,EAAW,YAAc,IAAA,IAAa,CAAE,UAAW,EAAW,SAAU,EAC5E,GAAI,EAAW,YAAc,IAAA,IAAa,CAAE,UAAW,EAAW,SAAU,EAC5E,GAAI,EAAW,iBAAmB,IAAA,IAAa,CAC7C,eAAgB,EAAW,cAC7B,EACA,GAAI,EAAW,QAAU,IAAA,IAAa,CAAE,MAAO,EAAW,KAAM,CAClE,EACA,EAAS,MAAM,EAAU,CACvB,cACA,SAAU,EAAI,MACd,SACA,YACA,KAAM,EACN,SAAU,EACV,UACA,WAAa,GACX,EAAI,eAAe,IAAc,IAAA,GAAyC,EAA7B,CAAE,GAAG,EAAU,WAAU,CAAY,EACpF,GAAI,IAAQ,IAAA,IAAa,CAAE,OAAQ,CAAI,CACzC,CAAC,CACH,OAAS,EAAK,CAQZ,MANA,MAAM,EAAW,EAAY,EADd,aAAe,MAAQ,EAAI,QAAU,OAAO,CAAG,CACd,EAC5C,GACF,MAAM,EAAY,CAAC,CAAC,MAAO,GAAwB,CACjD,GAAK,KAAK,CAAE,IAAK,CAAW,EAAG,oDAAoD,CACrF,CAAC,EAEG,CACR,CAMI,GACF,MAAM,EAAY,CAAC,CAAC,MAAO,GAAwB,CACjD,GAAK,KAAK,CAAE,IAAK,CAAW,EAAG,wDAAwD,CACzF,CAAC,EAcH,IAAI,EAAkB,EAAO,SACvB,EAAkB,OAAO,EAAO,SAAY,SAAW,EAAO,QAAU,IAAA,GACxE,EACJ,OAAO,EAAO,YAAe,SAAW,EAAO,WAAa,IAAA,GAC9D,GAAI,GAAmB,GAAsB,EAAO,WAAW,SAAW,KACxE,GAAI,CAIF,IAAM,EAAQ,EAAW,SAAS,EAW5B,EAAa,MAAM,EAAW,SAAS,CAC3C,MAAO,EAAI,EAAG,EAAM,cAAe,EAAI,aAAa,EAAG,EAAG,EAAM,OAAQ,WAAW,CAAC,EACpF,QAAS,EAAK,EAAM,SAAS,EAC7B,MAAO,EACT,CAAC,EACK,EAAQ,EAAW,KAAM,GAAM,CACnC,GAAI,EAAE,KAAO,EAAa,MAAO,GACjC,IAAM,EAAI,EAAE,OACZ,GAAI,CAAC,GAAK,OAAO,GAAM,UAAY,MAAM,QAAQ,CAAC,EAAG,MAAO,GAC5D,IAAM,EAAK,EACX,OAAO,EAAG,UAAY,GAAmB,EAAG,aAAe,CAC7D,CAAC,EASG,CAAC,GAAS,EAAW,QAAU,IACjC,GAAK,KACH,CACE,QAAS,EACT,WAAY,EACZ,cAAe,EAAI,aACrB,EACA,4MACF,EAEF,IAAM,EAAa,EAAQ,EAAgB,EAAM,UAAU,EAAI,KACzD,EAAU,EAAa,EAAmB,EAAY,EAAO,UAAU,EAAI,KACjF,GAAI,EAAS,CACX,IAAM,EAAS,EAAQ,OAAS,EAC1B,EAAiB,IAAI,EAC3B,EAAe,SACb,GACA,gBACA,6BAA6B,EAAS,EAAQ,MAAQ,EAAI,EAAQ,MAAA,CAAO,QAAQ,CAAC,EAAE,IAAI,EAAS,SAAW,QAAQ,0EAA0E,GAAO,IAAM,UAAU,WAAW,EAAQ,WAAW,qBAAqB,EAAQ,WAAW,8FAC3R,oBACA,CACE,oBAAqB,GAAO,IAAM,KAClC,eAAgB,EAAQ,WACxB,cAAe,EAAQ,WACvB,MAAO,EAAQ,KACjB,CACF,EACA,EAAkB,EAA2B,EAAO,SAAU,EAAe,SAAS,CAAC,EACvF,GAAK,KACH,CAAE,oBAAqB,GAAO,GAAI,MAAO,EAAQ,KAAM,EACvD,wFACF,CACF,CACF,OAAS,EAAK,CAIZ,GAAK,KAAK,CAAE,KAAI,EAAG,gEAAgE,CACrF,CAiDF,GA9CA,MAAM,EAAW,OAAO,EAAa,CACnC,OAAQ,YACR,WAAY,IAAI,KAChB,OAAQ,CACN,SAAU,EAAO,SACjB,cAAe,EAAO,cACtB,WAAY,EAAO,WACnB,YAAa,EAAO,YACpB,WAAY,EAAO,WACnB,iBAAkB,EAAO,iBAIzB,eAAgB,EAAO,eACvB,gBAAiB,EAAO,eAC1B,EAOA,aAAc,KAAK,MAAM,KAAK,UAAU,EAAO,MAAM,CAAC,EAMtD,eAAgB,KAAK,MAAM,KAAK,UAAU,CAAe,CAAC,EAG1D,WAAY,KAAK,MAAM,KAAK,UAAU,EAAO,UAAU,CAAC,CAC1D,CAAC,EAGD,EAA0B,EAAa,CACrC,SAAU,EAAO,SACjB,cAAe,EAAO,cACtB,WAAY,EAAO,UACrB,CAAC,EAOG,GAAY,EAAO,MAAM,OAAS,EACpC,GAAI,CACF,MAAM,EAAS,EAAO,MAAO,CAC3B,cACA,SACA,GAAI,IAAQ,IAAA,IAAa,CAAE,OAAQ,CAAI,CACzC,CAAC,EAGD,GAAK,KAAK,CAAE,cAAa,MAAO,EAAO,MAAM,MAAO,EAAG,2BAA2B,CACpF,OAAS,EAAK,CACZ,GAAK,KACH,CAAE,cAAa,MAAO,EAAO,MAAM,OAAQ,KAAI,EAK/C,yFACF,CACF,CAGF,GAAK,KACH,CACE,SAAU,EAAO,SACjB,cAAe,EAAO,cACtB,WAAY,EAAO,WACnB,YAAa,EAAO,YACpB,QAAS,EAAO,gBAClB,EACA,sBACF,CACF,CAAC,CACL,CAGA,eAAe,EAAW,EAA6B,EAAY,EAA+B,CAChG,GAAI,CACF,MAAM,EAAW,OAAO,EAAI,CAC1B,OAAQ,SACR,WAAY,IAAI,KAChB,aAAc,CAAE,MAAO,CAAO,CAChC,CAAC,EAMD,EAAuB,EAAI,CAAM,CACnC,MAAQ,CAGR,CACF,CA6CA,SAAgB,EACd,EACuB,CACvB,GAAM,CAAE,aAAY,UAAW,EAE/B,MAAO,CACL,UAAW,KAAO,IAA4C,CACxD,EAAM,UAAY,eAEtB,MAAM,EAAS,SAAY,CACzB,IAAM,EAAM,GAAQ,MAAM,CACxB,MAAO,mBACP,MAAO,EAAM,MACb,KAAM,aACR,CAAC,EAYK,EAAQ,EAAW,SAAS,EAK5B,GAAM,MAJU,EAAW,SAAS,CACxC,MAAO,EAAG,EAAM,WAAY,EAAM,KAAK,EACvC,MAAO,CACT,CAAC,EAAA,CACmB,GACpB,GAAI,CAAC,EAAK,CAKR,GAAK,KAAK,uDAAuD,EACjE,MACF,CAMA,GAAI,EAAI,SAAW,WAAa,EAAI,SAAW,UAAW,CACxD,GAAK,KACH,CAAE,YAAa,EAAI,GAAI,OAAQ,EAAI,MAAO,EAC1C,8EACF,EACA,MACF,CAQI,EAAI,SAAW,WACjB,GAAK,KACH,CAAE,YAAa,EAAI,GAAI,SAAU,EAAM,QAAS,EAChD,gGACF,EAGF,MAAM,EAAW,EAAY,EAAI,GAAI,EAAM,KAAK,EAChD,GAAK,KACH,CAAE,YAAa,EAAI,GAAI,SAAU,EAAM,SAAU,YAAa,EAAI,MAAO,EACzE,2CACF,CACF,CAAC,CACH,CACF,CACF"}

package/dist/{xlsx-reader-Ccesszbc.mjs → xlsx-reader-Dsq_ERiB.mjs}

@@ -1,2 +1,2 @@
-import{l as e}from"./runner-C-s6SD7t.mjs";import{open as t}from"node:fs/promises";var n=class extends Error{headerRowNumber;constructor(e){super(`xlsx header row ${e} has no readable columns — the file may be corrupt, or its shared strings failed to load. Refusing rather than importing empty rows.`),this.name=`XlsxEmptyHeaderError`,this.headerRowNumber=e}},r=class extends Error{constructor(e){super(`xlsx path "${e}" is not a regular file`),this.name=`XlsxNotAFileError`}};function i(e){if(e==null)return``;if(typeof e==`string`)return e;if(typeof e==`number`||typeof e==`boolean`)return String(e);if(e instanceof Date)return e.toISOString();if(`richText`in e&&Array.isArray(e.richText))return e.richText.map(e=>e.text).join(``);if(`text`in e&&typeof e.text==`string`)return e.text;if(`result`in e&&e.result!==void 0&&e.result!==null){let t=e.result;return t instanceof Date?t.toISOString():typeof t==`object`?``:String(t)}return``}function a(e){let t=[];if(Array.isArray(e))for(let n=1;n<e.length;n+=1)t[n-1]=i(e[n]);else for(let[n,r]of Object.entries(e)){let e=Number.parseInt(n,10);!Number.isNaN(e)&&e>=1&&(t[e-1]=i(r))}for(let e=0;e<t.length;e+=1)t[e]===void 0&&(t[e]=``);return t}async function*o(i){let{filePath:o,hasHeader:s=!0,headerRowIndex:c=0,sheet:l}=i,u=await import(`exceljs`),d;if(l!==void 0&&(d=Number.parseInt(l,10),Number.isNaN(d)||d<1))throw Error(`Invalid xlsx sheet "${l}" — expected a 1-based worksheet index`);let f=await t(o,`r`),p;try{if(!(await f.stat()).isFile())throw new r(o);p=f.createReadStream({autoClose:!1});let t=new u.stream.xlsx.WorkbookReader(p,{worksheets:`emit`,sharedStrings:`cache`,styles:`ignore`,hyperlinks:`ignore`,entries:`ignore`});t.model={sheets:[]};let i=0,l=0,m=!1,h=t[Symbol.asyncIterator](),g=!1;try{for(;;){let t=await h.next();if(t.done){g=!0;break}let r=t.value;if(l+=1,!(d===void 0?l===1:l===d)){for await(let e of r);continue}m=!0;let o=0,u=null;for await(let t of r){o+=1;let r=a(t.values);if(s&&u===null){if(o<c+1)continue;if(u=r.map(e=>e.trim()),u.every(e=>e===``))throw new n(o);e(u);continue}i+=1;let l={};if(u)for(let e=0;e<u.length;e+=1){let t=u[e];t&&(l[t]=r[e]??``)}else for(let e=0;e<r.length;e+=1)l[String(e)]=r[e]??``;yield{rowNumber:i,row:l}}break}}finally{try{for(;!g;){let e=await h.next();if(e.done)break;for await(let t of e.value);}}catch{}}if(d!==void 0&&!m)throw Error(`xlsx worksheet ${d} not found`)}finally{p?.destroy(),await f.close().catch(()=>{})}}export{n as XlsxEmptyHeaderError,r as XlsxNotAFileError,o as streamXlsx};
-//# sourceMappingURL=xlsx-reader-Ccesszbc.mjs.map
+import{g as e}from"./runner-wPnCZlB4.mjs";import{open as t}from"node:fs/promises";var n=class extends Error{headerRowNumber;constructor(e){super(`xlsx header row ${e} has no readable columns — the file may be corrupt, or its shared strings failed to load. Refusing rather than importing empty rows.`),this.name=`XlsxEmptyHeaderError`,this.headerRowNumber=e}},r=class extends Error{constructor(e){super(`xlsx path "${e}" is not a regular file`),this.name=`XlsxNotAFileError`}};function i(e){if(e==null)return``;if(typeof e==`string`)return e;if(typeof e==`number`||typeof e==`boolean`)return String(e);if(e instanceof Date)return e.toISOString();if(`richText`in e&&Array.isArray(e.richText))return e.richText.map(e=>e.text).join(``);if(`text`in e&&typeof e.text==`string`)return e.text;if(`result`in e&&e.result!==void 0&&e.result!==null){let t=e.result;return t instanceof Date?t.toISOString():typeof t==`object`?``:String(t)}return``}function a(e){let t=[];if(Array.isArray(e))for(let n=1;n<e.length;n+=1)t[n-1]=i(e[n]);else for(let[n,r]of Object.entries(e)){let e=Number.parseInt(n,10);!Number.isNaN(e)&&e>=1&&(t[e-1]=i(r))}for(let e=0;e<t.length;e+=1)t[e]===void 0&&(t[e]=``);return t}async function*o(i){let{filePath:o,hasHeader:s=!0,headerRowIndex:c=0,sheet:l}=i,u=await import(`exceljs`),d;if(l!==void 0&&(d=Number.parseInt(l,10),Number.isNaN(d)||d<1))throw Error(`Invalid xlsx sheet "${l}" — expected a 1-based worksheet index`);let f=await t(o,`r`),p;try{if(!(await f.stat()).isFile())throw new r(o);p=f.createReadStream({autoClose:!1});let t=new u.stream.xlsx.WorkbookReader(p,{worksheets:`emit`,sharedStrings:`cache`,styles:`ignore`,hyperlinks:`ignore`,entries:`ignore`});t.model={sheets:[]};let i=0,l=0,m=!1,h=t[Symbol.asyncIterator](),g=!1;try{for(;;){let t=await h.next();if(t.done){g=!0;break}let r=t.value;if(l+=1,!(d===void 0?l===1:l===d)){for await(let e of r);continue}m=!0;let o=0,u=null;for await(let t of r){o+=1;let r=a(t.values);if(s&&u===null){if(o<c+1)continue;if(u=r.map(e=>e.trim()),u.every(e=>e===``))throw new n(o);e(u);continue}i+=1;let l={};if(u)for(let e=0;e<u.length;e+=1){let t=u[e];t&&(l[t]=r[e]??``)}else for(let e=0;e<r.length;e+=1)l[String(e)]=r[e]??``;yield{rowNumber:i,row:l}}break}}finally{try{for(;!g;){let e=await h.next();if(e.done)break;for await(let t of e.value);}}catch{}}if(d!==void 0&&!m)throw Error(`xlsx worksheet ${d} not found`)}finally{p?.destroy(),await f.close().catch(()=>{})}}export{n as XlsxEmptyHeaderError,r as XlsxNotAFileError,o as streamXlsx};
+//# sourceMappingURL=xlsx-reader-Dsq_ERiB.mjs.map

package/dist/{xlsx-reader-Ccesszbc.mjs.map → xlsx-reader-Dsq_ERiB.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"xlsx-reader-Ccesszbc.mjs","names":[],"sources":["../src/xlsx-reader.ts"],"sourcesContent":["/**\n * Streaming XLSX reader (commerce-demo phase 02, D005) — the exceljs branch of\n * the feed reader. Yields the SAME `AsyncIterable<{ rowNumber, row }>` contract as\n * the csv-parse `streamFeed`, so the runner / transform layer downstream is\n * format-agnostic.\n *\n * **exceljs over SheetJS** (R02): we parse operator-uploaded spreadsheets — the\n * exact threat model of the unfixed npm `xlsx` prototype-pollution CVE\n * (CVE-2023-30533). exceljs is MIT, ships its own types, has a real streaming\n * `WorkbookReader`, and carries no equivalent advisory.\n *\n * **exceljs is `await import()`-ed**, never statically imported, so it stays out\n * of this package's static module graph — heavy + Node-only, it must not be\n * inlinable into any `*-ui` server-pages bundle (CLAUDE.md bundle budgets). The\n * `read-feed` dispatcher additionally only loads THIS module for `format: 'xlsx'`.\n */\n\nimport { open } from 'node:fs/promises'\nimport type { CellValue } from 'exceljs'\nimport { assertUniqueHeaders, type StreamFeedRow } from './streaming.js'\n\n/**\n * Thrown by {@link streamXlsx} when the configured header row parses to zero\n * readable columns — the signature of the silent-corruption defect documented\n * on the guard below.\n *\n * A distinct CLASS rather than a bare `Error` because the retroactive audit\n * (`auditEmptyHeaderRuns`, #842) has to tell \"this file's header is\n * unreadable\" apart from every other way reading a spreadsheet can fail\n * (missing object, unreadable zip, wrong sheet index). Matching on the message\n * would make that classification a regex over prose that anyone is free to\n * reword — which is exactly the kind of detector that silently stops detecting.\n * The message text is unchanged from the original throw.\n */\nexport class XlsxEmptyHeaderError extends Error {\n  /** 1-based worksheet row the empty header was read from. */\n  readonly headerRowNumber: number\n\n  constructor(headerRowNumber: number) {\n    super(\n      `xlsx header row ${headerRowNumber} has no readable columns — the file may be corrupt, ` +\n        'or its shared strings failed to load. Refusing rather than importing empty rows.',\n    )\n    this.name = 'XlsxEmptyHeaderError'\n    this.headerRowNumber = headerRowNumber\n  }\n}\n\n/**\n * Thrown by {@link streamXlsx} when the path resolves to something that is not a\n * regular file — a directory, most often. `open()` SUCCEEDS on a directory on\n * Linux (POSIX allows `O_RDONLY` on one) and only the read fails, so this is a\n * check the reader has to make itself.\n *\n * Typed for the same reason {@link XlsxEmptyHeaderError} is: the #842 audit has\n * to tell \"the evidence is unreadable\" from \"the evidence is readable and bad\",\n * and matching prose is how a classifier quietly stops classifying.\n */\nexport class XlsxNotAFileError extends Error {\n  constructor(filePath: string) {\n    super(`xlsx path \"${filePath}\" is not a regular file`)\n    this.name = 'XlsxNotAFileError'\n  }\n}\n\nexport interface XlsxReadOptions {\n  /** Path to the .xlsx file on local disk. */\n  filePath: string\n  /** First row is the header (column names). `false` → positional `{ \"0\": … }` keys. Default `true`. */\n  hasHeader?: boolean\n  /** 0-based index of the header row (rows before it are skipped). Default `0`. */\n  headerRowIndex?: number\n  /** Worksheet to read, as a 1-based index string. Omit → the first worksheet. */\n  sheet?: string\n}\n\n/** Convert one exceljs cell value to a plain string (feeds are text/number cells). */\nfunction cellToString(v: CellValue): string {\n  if (v === null || v === undefined) return ''\n  if (typeof v === 'string') return v\n  if (typeof v === 'number' || typeof v === 'boolean') return String(v)\n  if (v instanceof Date) return v.toISOString()\n  // Object cell values: rich text, hyperlink, formula result, error.\n  if ('richText' in v && Array.isArray(v.richText)) {\n    return v.richText.map((r) => r.text).join('')\n  }\n  if ('text' in v && typeof v.text === 'string') return v.text\n  if ('result' in v && v.result !== undefined && v.result !== null) {\n    const r = v.result\n    if (r instanceof Date) return r.toISOString()\n    // A nested error object (formula error) surfaces as empty; primitives stringify.\n    return typeof r === 'object' ? '' : String(r)\n  }\n  return ''\n}\n\n/** Normalise an exceljs `row.values` (1-indexed array OR object form) to a 0-indexed string[]. */\nfunction rowCells(values: CellValue[] | { [key: string]: CellValue }): string[] {\n  const out: string[] = []\n  if (Array.isArray(values)) {\n    // exceljs row arrays are 1-indexed (`values[0]` is empty); shift to 0-based.\n    for (let i = 1; i < values.length; i += 1) out[i - 1] = cellToString(values[i])\n  } else {\n    for (const [k, v] of Object.entries(values)) {\n      const idx = Number.parseInt(k, 10)\n      if (!Number.isNaN(idx) && idx >= 1) out[idx - 1] = cellToString(v)\n    }\n  }\n  // Fill any holes (sparse rows) with empty strings — the contract is \"missing\n  // cell = ''\", matching streamFeed.\n  for (let i = 0; i < out.length; i += 1) if (out[i] === undefined) out[i] = ''\n  return out\n}\n\n/**\n * Async-iterable over the data rows of an .xlsx file. `rowNumber` is the 1-based\n * DATA row index (header excluded). Cell values are coerced to strings so the\n * mapping engine's per-column transforms (which expect raw string cells) apply\n * uniformly across csv/tsv/xlsx.\n */\nexport async function* streamXlsx(options: XlsxReadOptions): AsyncIterable<StreamFeedRow> {\n  const { filePath, hasHeader = true, headerRowIndex = 0, sheet } = options\n  const ExcelJS = await import('exceljs')\n\n  // `WorksheetReader` doesn't expose its sheet id, so count emitted worksheets\n  // (1-based) to match the configured `sheet`. Default: the first worksheet.\n  // A provided-but-invalid sheet FAILS FAST rather than silently reading sheet 1\n  // — and before we touch the filesystem, so a config typo costs no I/O.\n  let targetSheetIndex: number | undefined\n  if (sheet !== undefined) {\n    targetSheetIndex = Number.parseInt(sheet, 10)\n    if (Number.isNaN(targetSheetIndex) || targetSheetIndex < 1) {\n      throw new Error(`Invalid xlsx sheet \"${sheet}\" — expected a 1-based worksheet index`)\n    }\n  }\n\n  // OPEN THE FILE OURSELVES rather than handing exceljs a path, because a path\n  // makes every filesystem failure an UNCAUGHT EXCEPTION that kills the process\n  // instead of a rejection the caller can handle.\n  //\n  // `_getStream` does `fs.createReadStream(input)` for a string input and then\n  // `stream.pipe(zip)` (`workbook-reader.js:38,81`). `pipe()` does not forward\n  // SOURCE errors, and exceljs registers no `error` listener on that stream — so\n  // an ENOENT / EACCES / EISDIR surfaces as `Unhandled 'error' event`. Verified\n  // by execution, not inference: all three crash the process outright. In the\n  // queue worker that kills the job mid-flight instead of routing through\n  // `markFailed`; in the #842 audit one unreadable file discards the whole sweep.\n  //\n  // An fd fixes it at the root rather than patching the symptom:\n  //   - `open()` rejects normally on ENOENT and EACCES.\n  //   - a DIRECTORY still opens on Linux (checked — O_RDONLY on a dir succeeds\n  //     and only the read gives EISDIR), so the `isFile()` check below is\n  //     load-bearing. It runs against the fd, so it cannot be raced.\n  //   - once the fd is held, the inode survives `unlink` (checked: the read\n  //     still returns the bytes). That closes the window against the\n  //     `@every 1h` retention sweep, which deletes exactly these files and which\n  //     the audit is documented as racing.\n  const handle = await open(filePath, 'r')\n  // Held so the outer `finally` can destroy it BEFORE the fd goes away. With\n  // `autoClose: false` nothing tears this stream down except reaching the end,\n  // and the drain below deliberately swallows errors — so a malformed trailing\n  // worksheet can leave the source mid-read while `handle.close()` invalidates\n  // the descriptor underneath it. exceljs registers no `error` listener on the\n  // stream it pipes (the whole reason this function opens the file itself), so\n  // that EBADF would arrive as an uncaught event: the exact failure mode the fd\n  // was introduced to eliminate, re-entering through the back door.\n  let source: ReturnType<typeof handle.createReadStream> | undefined\n  try {\n    const stats = await handle.stat()\n    if (!stats.isFile()) {\n      throw new XlsxNotAFileError(filePath)\n    }\n\n    source = handle.createReadStream({ autoClose: false })\n    const reader = new ExcelJS.stream.xlsx.WorkbookReader(source, {\n      worksheets: 'emit',\n      sharedStrings: 'cache',\n      styles: 'ignore',\n      hyperlinks: 'ignore',\n      entries: 'ignore',\n    })\n\n    // Seed `model` so exceljs cannot dereference it before it has parsed\n    // `xl/workbook.xml` — an UPSTREAM defect, and the cause of the xlsx-reader\n    // \"flake\" that failed CI intermittently for months (#619, #835).\n    //\n    // Its streaming reader parses a worksheet INLINE, rather than deferring it to\n    // a temp file, when `this.sharedStrings && this.workbookRels` are both set as\n    // the entry arrives (`workbook-reader.js:109`). Neither implies\n    // `xl/workbook.xml` has been read. The inline path then does\n    // `matchingRel && (this.model.sheets || []).find(…)` (`:303`) — guarding\n    // `this.workbookRels` on the line above and `this.model.sheets` not at all —\n    // so any workbook whose zip orders `sharedStrings.xml` before the worksheet,\n    // with `workbook.xml` after it, throws `Cannot read properties of undefined\n    // (reading 'sheets')`.\n    //\n    // That is a PRODUCTION hazard, not a test artifact: this function reads\n    // operator-uploaded spreadsheets and the writer chooses the entry order. It\n    // presented as flakiness only because the fixtures our own tests write happen\n    // to sit on the safe side of the condition.\n    //\n    // An empty `sheets` makes the lookup miss, so the worksheet keeps its default\n    // id/name/state until `_parseWorkbook` overwrites `model` with the real one a\n    // few entries later — which is exactly what the deferred path already does.\n    // Preferred over a `pnpm patch` (exceljs 4.4.0 is the latest release, so there\n    // is no upstream fix to take, but a patch is a standing maintenance cost and\n    // this is one assignment that stays correct even if upstream fixes it).\n    //\n    // `sheets: []` rather than `{}` because `(this.model.sheets || [])` is what is\n    // read.\n    //\n    // The double assertion is NOT laziness and a single `as typeof reader.model`\n    // does not compile — checked, not assumed:\n    //\n    //   TS2352: Conversion of type '{ sheets: never[]; }' to type 'WorkbookModel'\n    //   may be a mistake because neither type sufficiently overlaps with the\n    //   other. … missing the following properties: creator, lastModifiedBy,\n    //   lastPrinted, created, and 17 more.\n    //\n    // A deliberately PARTIAL seed is the whole point — exceljs overwrites it with\n    // the real model a few entries later, and filling 21 fields with invented\n    // values to satisfy the compiler would be worse than the assertion: it would\n    // put fabricated workbook metadata one `_parseWorkbook` failure away from\n    // being read as real.\n    // toolkit-smell-allow: partial seed for a 21-field type; single assertion rejected by TS2352 (see above)\n    reader.model = { sheets: [] } as unknown as typeof reader.model\n\n    let dataRowNumber = 0\n    let sheetIndex = 0\n    let matchedTarget = false\n\n    // Drive the worksheet iterator BY HAND rather than with `for await`, so the\n    // `finally` below can resume it. Abandoning exceljs's `parse()` generator —\n    // which `return`/`break` out of a `for await` does — LEAKS a temp file and an\n    // fd per deferred worksheet, because `tempFileCleanupCallback()` sits after\n    // the `yield*` with no `finally` of its own (`workbook-reader.js:140-149`).\n    //\n    // Measured, with a negative control: 25 early-returning reads of a two-sheet\n    // deferred-order workbook leaked 50 fds and 50 temp files; the draining shape\n    // below leaks 0 of each. `tmp`'s graceful-cleanup hook reclaims the files at\n    // NORMAL process exit only, which is no help to a long-lived queue worker and\n    // no help at all against the fd ceiling: at the #842 audit's 5000-run cap\n    // this is an EMFILE, and an EMFILE arrives as the uncaught stream error the\n    // `open()` above exists to prevent.\n    //\n    // THE COST IS PROPORTIONAL TO THE WORKSHEETS AFTER THE TARGET, which is zero\n    // for the single-sheet feeds this reader mostly sees and 4.7× on a workbook\n    // whose target is sheet 1 and whose sheet 2 holds 40k rows. Advancing the\n    // parent iterator WITHOUT consuming each trailing worksheet is far cheaper\n    // and does release the temp files — but it strands exceljs's\n    // `createReadStream` on each one, i.e. it trades the disk leak for the fd\n    // leak, and the fd leak is the one that ends in EMFILE. So: full drain.\n    const worksheets = reader[Symbol.asyncIterator]()\n    let readerExhausted = false\n    try {\n      while (true) {\n        const next = await worksheets.next()\n        if (next.done) {\n          readerExhausted = true\n          break\n        }\n        const worksheet = next.value\n        sheetIndex += 1\n        const isTarget =\n          targetSheetIndex === undefined ? sheetIndex === 1 : sheetIndex === targetSheetIndex\n\n        if (!isTarget) {\n          // Drain a non-target worksheet so the underlying zip stream advances to the\n          // next; exceljs's streaming reader stalls if a worksheet is left unconsumed.\n          for await (const _skip of worksheet) {\n            void _skip\n          }\n          continue\n        }\n        matchedTarget = true\n\n        let excelRowNum = 0\n        let header: string[] | null = null\n        for await (const row of worksheet) {\n          excelRowNum += 1\n          const cells = rowCells(row.values)\n\n          if (hasHeader && header === null) {\n            // Skip rows before the configured header row, then capture it (trimmed).\n            if (excelRowNum < headerRowIndex + 1) continue\n            header = cells.map((c) => c.trim())\n            // FAIL LOUDLY on an empty header, because the alternative is SILENT DATA\n            // CORRUPTION and we have watched it happen.\n            //\n            // exceljs assigns `this.sharedStrings = []` at the TOP of\n            // `_parseSharedStrings`, before parsing a single entry — and an empty\n            // array is truthy, which is what the inline-worksheet guard tests. So a\n            // shared-strings entry that yields nothing (a truncated or starved read;\n            // observed in CI under load) leaves every string cell resolving to\n            // `sharedStrings[index]` → `undefined`. The header parses to zero\n            // columns, every data row maps to `{}`, and the import reports SUCCESS\n            // with every value dropped. Reproduced deterministically by rebuilding a\n            // workbook with an empty `sst`: two rows read, no values, no error.\n            //\n            // A header row with no usable column is never a legitimate import, so\n            // there is no case to preserve. Refusing costs an operator a clear error\n            // and saves them a silently empty dataset they would have to notice\n            // themselves — the same posture the rest of this package takes toward\n            // config mistakes.\n            if (header.every((c) => c === '')) {\n              throw new XlsxEmptyHeaderError(excelRowNum)\n            }\n            assertUniqueHeaders(header)\n            continue\n          }\n\n          dataRowNumber += 1\n          const rec: Record<string, string> = {}\n          if (header) {\n            for (let i = 0; i < header.length; i += 1) {\n              const key = header[i]\n              if (key) rec[key] = cells[i] ?? ''\n            }\n          } else {\n            for (let i = 0; i < cells.length; i += 1) rec[String(i)] = cells[i] ?? ''\n          }\n          yield { rowNumber: dataRowNumber, row: rec }\n        }\n\n        // Target worksheet fully read. Stop yielding — the `finally` walks the\n        // remaining worksheets so exceljs reaches its cleanup.\n        break\n      }\n    } finally {\n      // Runs on every exit: target read, an empty-header throw, or the CONSUMER\n      // abandoning us mid-stream (a `break` in the caller's `for await` calls\n      // `.return()`, which lands here). Whatever the reason, exceljs still owns\n      // temp files it only releases as the consumer pulls past each worksheet.\n      //\n      // The whole body is CONTAINED, for the same reason `handle.close()` below\n      // is: a throw from a `finally` REPLACES whatever the try block was\n      // propagating. Trailing worksheets are parsed here, so a malformed one\n      // would overwrite an `XlsxEmptyHeaderError` with an XML error — turning\n      // the #842 audit's `corrupt` verdict into a meaningless `error` and\n      // failing an import whose target sheet read perfectly. Measured: two of\n      // four malformed-trailing-sheet variants masked the refusal before this\n      // catch existed. Cleanup is best-effort by definition; the error the\n      // caller came for is not.\n      try {\n        while (!readerExhausted) {\n          const rest = await worksheets.next()\n          if (rest.done) break\n          for await (const _skip of rest.value) {\n            void _skip\n          }\n        }\n      } catch {\n        // Deliberately swallowed — see above. A trailing worksheet we only read\n        // to satisfy exceljs's cleanup contract cannot be allowed to speak for\n        // the read the caller actually asked for. Worst case we leak the temp\n        // files this drain existed to release, which is what `main` did on\n        // EVERY read.\n      }\n    }\n\n    // Reached only when a specific sheet was requested but never matched — the\n    // first-sheet default always matches on a non-empty workbook.\n    if (targetSheetIndex !== undefined && !matchedTarget) {\n      throw new Error(`xlsx worksheet ${targetSheetIndex} not found`)\n    }\n  } finally {\n    // Order matters: destroy the reader of the fd before closing the fd.\n    source?.destroy()\n    await handle.close().catch(() => {\n      // Swallowed deliberately, and biome's `noUnsafeFinally` agrees: a throw\n      // here would REPLACE whatever the try block was propagating, and what it\n      // propagates is the point of this reader — an empty-header refusal has to\n      // reach the caller intact. Closing a read-only fd we are finished with can\n      // fail only in ways that are not actionable, and the fd is reclaimed at\n      // process exit regardless.\n    })\n  }\n}\n"],"mappings":"kFAkCA,IAAa,EAAb,cAA0C,KAAM,CAE9C,gBAEA,YAAY,EAAyB,CACnC,MACE,mBAAmB,EAAgB,qIAErC,EACA,KAAK,KAAO,uBACZ,KAAK,gBAAkB,CACzB,CACF,EAYa,EAAb,cAAuC,KAAM,CAC3C,YAAY,EAAkB,CAC5B,MAAM,cAAc,EAAS,wBAAwB,EACrD,KAAK,KAAO,mBACd,CACF,EAcA,SAAS,EAAa,EAAsB,CAC1C,GAAI,GAAM,KAAyB,MAAO,GAC1C,GAAI,OAAO,GAAM,SAAU,OAAO,EAClC,GAAI,OAAO,GAAM,UAAY,OAAO,GAAM,UAAW,OAAO,OAAO,CAAC,EACpE,GAAI,aAAa,KAAM,OAAO,EAAE,YAAY,EAE5C,GAAI,aAAc,GAAK,MAAM,QAAQ,EAAE,QAAQ,EAC7C,OAAO,EAAE,SAAS,IAAK,GAAM,EAAE,IAAI,CAAC,CAAC,KAAK,EAAE,EAE9C,GAAI,SAAU,GAAK,OAAO,EAAE,MAAS,SAAU,OAAO,EAAE,KACxD,GAAI,WAAY,GAAK,EAAE,SAAW,IAAA,IAAa,EAAE,SAAW,KAAM,CAChE,IAAM,EAAI,EAAE,OAGZ,OAFI,aAAa,KAAa,EAAE,YAAY,EAErC,OAAO,GAAM,SAAW,GAAK,OAAO,CAAC,CAC9C,CACA,MAAO,EACT,CAGA,SAAS,EAAS,EAA8D,CAC9E,IAAM,EAAgB,CAAC,EACvB,GAAI,MAAM,QAAQ,CAAM,EAEtB,IAAK,IAAI,EAAI,EAAG,EAAI,EAAO,OAAQ,GAAK,EAAG,EAAI,EAAI,GAAK,EAAa,EAAO,EAAE,OAE9E,IAAK,GAAM,CAAC,EAAG,KAAM,OAAO,QAAQ,CAAM,EAAG,CAC3C,IAAM,EAAM,OAAO,SAAS,EAAG,EAAE,EAC7B,CAAC,OAAO,MAAM,CAAG,GAAK,GAAO,IAAG,EAAI,EAAM,GAAK,EAAa,CAAC,EACnE,CAIF,IAAK,IAAI,EAAI,EAAG,EAAI,EAAI,OAAQ,GAAK,EAAO,EAAI,KAAO,IAAA,KAAW,EAAI,GAAK,IAC3E,OAAO,CACT,CAQA,eAAuB,EAAW,EAAwD,CACxF,GAAM,CAAE,WAAU,YAAY,GAAM,iBAAiB,EAAG,SAAU,EAC5D,EAAU,MAAM,OAAO,WAMzB,EACJ,GAAI,IAAU,IAAA,KACZ,EAAmB,OAAO,SAAS,EAAO,EAAE,EACxC,OAAO,MAAM,CAAgB,GAAK,EAAmB,GACvD,MAAU,MAAM,uBAAuB,EAAM,uCAAuC,EAyBxF,IAAM,EAAS,MAAM,EAAK,EAAU,GAAG,EASnC,EACJ,GAAI,CAEF,GAAI,EAAC,MADe,EAAO,KAAK,EAAA,CACrB,OAAO,EAChB,MAAM,IAAI,EAAkB,CAAQ,EAGtC,EAAS,EAAO,iBAAiB,CAAE,UAAW,EAAM,CAAC,EACrD,IAAM,EAAS,IAAI,EAAQ,OAAO,KAAK,eAAe,EAAQ,CAC5D,WAAY,OACZ,cAAe,QACf,OAAQ,SACR,WAAY,SACZ,QAAS,QACX,CAAC,EA6CD,EAAO,MAAQ,CAAE,OAAQ,CAAC,CAAE,EAE5B,IAAI,EAAgB,EAChB,EAAa,EACb,EAAgB,GAuBd,EAAa,EAAO,OAAO,cAAc,CAAC,EAC5C,EAAkB,GACtB,GAAI,CACF,OAAa,CACX,IAAM,EAAO,MAAM,EAAW,KAAK,EACnC,GAAI,EAAK,KAAM,CACb,EAAkB,GAClB,KACF,CACA,IAAM,EAAY,EAAK,MAKvB,GAJA,GAAc,EAIV,EAFF,IAAqB,IAAA,GAAY,IAAe,EAAI,IAAe,GAEtD,CAGb,UAAW,IAAM,KAAS,GAG1B,QACF,CACA,EAAgB,GAEhB,IAAI,EAAc,EACd,EAA0B,KAC9B,UAAW,IAAM,KAAO,EAAW,CACjC,GAAe,EACf,IAAM,EAAQ,EAAS,EAAI,MAAM,EAEjC,GAAI,GAAa,IAAW,KAAM,CAEhC,GAAI,EAAc,EAAiB,EAAG,SAoBtC,GAnBA,EAAS,EAAM,IAAK,GAAM,EAAE,KAAK,CAAC,EAmB9B,EAAO,MAAO,GAAM,IAAM,EAAE,EAC9B,MAAM,IAAI,EAAqB,CAAW,EAE5C,EAAoB,CAAM,EAC1B,QACF,CAEA,GAAiB,EACjB,IAAM,EAA8B,CAAC,EACrC,GAAI,EACF,IAAK,IAAI,EAAI,EAAG,EAAI,EAAO,OAAQ,GAAK,EAAG,CACzC,IAAM,EAAM,EAAO,GACf,IAAK,EAAI,GAAO,EAAM,IAAM,GAClC,MAEA,IAAK,IAAI,EAAI,EAAG,EAAI,EAAM,OAAQ,GAAK,EAAG,EAAI,OAAO,CAAC,GAAK,EAAM,IAAM,GAEzE,KAAM,CAAE,UAAW,EAAe,IAAK,CAAI,CAC7C,CAIA,KACF,CACF,QAAU,CAeR,GAAI,CACF,KAAO,CAAC,GAAiB,CACvB,IAAM,EAAO,MAAM,EAAW,KAAK,EACnC,GAAI,EAAK,KAAM,MACf,UAAW,IAAM,KAAS,EAAK,OAGjC,CACF,MAAQ,CAMR,CACF,CAIA,GAAI,IAAqB,IAAA,IAAa,CAAC,EACrC,MAAU,MAAM,kBAAkB,EAAiB,WAAW,CAElE,QAAU,CAER,GAAQ,QAAQ,EAChB,MAAM,EAAO,MAAM,CAAC,CAAC,UAAY,CAOjC,CAAC,CACH,CACF"}
+{"version":3,"file":"xlsx-reader-Dsq_ERiB.mjs","names":[],"sources":["../src/xlsx-reader.ts"],"sourcesContent":["/**\n * Streaming XLSX reader (commerce-demo phase 02, D005) — the exceljs branch of\n * the feed reader. Yields the SAME `AsyncIterable<{ rowNumber, row }>` contract as\n * the csv-parse `streamFeed`, so the runner / transform layer downstream is\n * format-agnostic.\n *\n * **exceljs over SheetJS** (R02): we parse operator-uploaded spreadsheets — the\n * exact threat model of the unfixed npm `xlsx` prototype-pollution CVE\n * (CVE-2023-30533). exceljs is MIT, ships its own types, has a real streaming\n * `WorkbookReader`, and carries no equivalent advisory.\n *\n * **exceljs is `await import()`-ed**, never statically imported, so it stays out\n * of this package's static module graph — heavy + Node-only, it must not be\n * inlinable into any `*-ui` server-pages bundle (CLAUDE.md bundle budgets). The\n * `read-feed` dispatcher additionally only loads THIS module for `format: 'xlsx'`.\n */\n\nimport { open } from 'node:fs/promises'\nimport type { CellValue } from 'exceljs'\nimport { assertUniqueHeaders, type StreamFeedRow } from './streaming.js'\n\n/**\n * Thrown by {@link streamXlsx} when the configured header row parses to zero\n * readable columns — the signature of the silent-corruption defect documented\n * on the guard below.\n *\n * A distinct CLASS rather than a bare `Error` because the retroactive audit\n * (`auditEmptyHeaderRuns`, #842) has to tell \"this file's header is\n * unreadable\" apart from every other way reading a spreadsheet can fail\n * (missing object, unreadable zip, wrong sheet index). Matching on the message\n * would make that classification a regex over prose that anyone is free to\n * reword — which is exactly the kind of detector that silently stops detecting.\n * The message text is unchanged from the original throw.\n */\nexport class XlsxEmptyHeaderError extends Error {\n  /** 1-based worksheet row the empty header was read from. */\n  readonly headerRowNumber: number\n\n  constructor(headerRowNumber: number) {\n    super(\n      `xlsx header row ${headerRowNumber} has no readable columns — the file may be corrupt, ` +\n        'or its shared strings failed to load. Refusing rather than importing empty rows.',\n    )\n    this.name = 'XlsxEmptyHeaderError'\n    this.headerRowNumber = headerRowNumber\n  }\n}\n\n/**\n * Thrown by {@link streamXlsx} when the path resolves to something that is not a\n * regular file — a directory, most often. `open()` SUCCEEDS on a directory on\n * Linux (POSIX allows `O_RDONLY` on one) and only the read fails, so this is a\n * check the reader has to make itself.\n *\n * Typed for the same reason {@link XlsxEmptyHeaderError} is: the #842 audit has\n * to tell \"the evidence is unreadable\" from \"the evidence is readable and bad\",\n * and matching prose is how a classifier quietly stops classifying.\n */\nexport class XlsxNotAFileError extends Error {\n  constructor(filePath: string) {\n    super(`xlsx path \"${filePath}\" is not a regular file`)\n    this.name = 'XlsxNotAFileError'\n  }\n}\n\nexport interface XlsxReadOptions {\n  /** Path to the .xlsx file on local disk. */\n  filePath: string\n  /** First row is the header (column names). `false` → positional `{ \"0\": … }` keys. Default `true`. */\n  hasHeader?: boolean\n  /** 0-based index of the header row (rows before it are skipped). Default `0`. */\n  headerRowIndex?: number\n  /** Worksheet to read, as a 1-based index string. Omit → the first worksheet. */\n  sheet?: string\n}\n\n/** Convert one exceljs cell value to a plain string (feeds are text/number cells). */\nfunction cellToString(v: CellValue): string {\n  if (v === null || v === undefined) return ''\n  if (typeof v === 'string') return v\n  if (typeof v === 'number' || typeof v === 'boolean') return String(v)\n  if (v instanceof Date) return v.toISOString()\n  // Object cell values: rich text, hyperlink, formula result, error.\n  if ('richText' in v && Array.isArray(v.richText)) {\n    return v.richText.map((r) => r.text).join('')\n  }\n  if ('text' in v && typeof v.text === 'string') return v.text\n  if ('result' in v && v.result !== undefined && v.result !== null) {\n    const r = v.result\n    if (r instanceof Date) return r.toISOString()\n    // A nested error object (formula error) surfaces as empty; primitives stringify.\n    return typeof r === 'object' ? '' : String(r)\n  }\n  return ''\n}\n\n/** Normalise an exceljs `row.values` (1-indexed array OR object form) to a 0-indexed string[]. */\nfunction rowCells(values: CellValue[] | { [key: string]: CellValue }): string[] {\n  const out: string[] = []\n  if (Array.isArray(values)) {\n    // exceljs row arrays are 1-indexed (`values[0]` is empty); shift to 0-based.\n    for (let i = 1; i < values.length; i += 1) out[i - 1] = cellToString(values[i])\n  } else {\n    for (const [k, v] of Object.entries(values)) {\n      const idx = Number.parseInt(k, 10)\n      if (!Number.isNaN(idx) && idx >= 1) out[idx - 1] = cellToString(v)\n    }\n  }\n  // Fill any holes (sparse rows) with empty strings — the contract is \"missing\n  // cell = ''\", matching streamFeed.\n  for (let i = 0; i < out.length; i += 1) if (out[i] === undefined) out[i] = ''\n  return out\n}\n\n/**\n * Async-iterable over the data rows of an .xlsx file. `rowNumber` is the 1-based\n * DATA row index (header excluded). Cell values are coerced to strings so the\n * mapping engine's per-column transforms (which expect raw string cells) apply\n * uniformly across csv/tsv/xlsx.\n */\nexport async function* streamXlsx(options: XlsxReadOptions): AsyncIterable<StreamFeedRow> {\n  const { filePath, hasHeader = true, headerRowIndex = 0, sheet } = options\n  const ExcelJS = await import('exceljs')\n\n  // `WorksheetReader` doesn't expose its sheet id, so count emitted worksheets\n  // (1-based) to match the configured `sheet`. Default: the first worksheet.\n  // A provided-but-invalid sheet FAILS FAST rather than silently reading sheet 1\n  // — and before we touch the filesystem, so a config typo costs no I/O.\n  let targetSheetIndex: number | undefined\n  if (sheet !== undefined) {\n    targetSheetIndex = Number.parseInt(sheet, 10)\n    if (Number.isNaN(targetSheetIndex) || targetSheetIndex < 1) {\n      throw new Error(`Invalid xlsx sheet \"${sheet}\" — expected a 1-based worksheet index`)\n    }\n  }\n\n  // OPEN THE FILE OURSELVES rather than handing exceljs a path, because a path\n  // makes every filesystem failure an UNCAUGHT EXCEPTION that kills the process\n  // instead of a rejection the caller can handle.\n  //\n  // `_getStream` does `fs.createReadStream(input)` for a string input and then\n  // `stream.pipe(zip)` (`workbook-reader.js:38,81`). `pipe()` does not forward\n  // SOURCE errors, and exceljs registers no `error` listener on that stream — so\n  // an ENOENT / EACCES / EISDIR surfaces as `Unhandled 'error' event`. Verified\n  // by execution, not inference: all three crash the process outright. In the\n  // queue worker that kills the job mid-flight instead of routing through\n  // `markFailed`; in the #842 audit one unreadable file discards the whole sweep.\n  //\n  // An fd fixes it at the root rather than patching the symptom:\n  //   - `open()` rejects normally on ENOENT and EACCES.\n  //   - a DIRECTORY still opens on Linux (checked — O_RDONLY on a dir succeeds\n  //     and only the read gives EISDIR), so the `isFile()` check below is\n  //     load-bearing. It runs against the fd, so it cannot be raced.\n  //   - once the fd is held, the inode survives `unlink` (checked: the read\n  //     still returns the bytes). That closes the window against the\n  //     `@every 1h` retention sweep, which deletes exactly these files and which\n  //     the audit is documented as racing.\n  const handle = await open(filePath, 'r')\n  // Held so the outer `finally` can destroy it BEFORE the fd goes away. With\n  // `autoClose: false` nothing tears this stream down except reaching the end,\n  // and the drain below deliberately swallows errors — so a malformed trailing\n  // worksheet can leave the source mid-read while `handle.close()` invalidates\n  // the descriptor underneath it. exceljs registers no `error` listener on the\n  // stream it pipes (the whole reason this function opens the file itself), so\n  // that EBADF would arrive as an uncaught event: the exact failure mode the fd\n  // was introduced to eliminate, re-entering through the back door.\n  let source: ReturnType<typeof handle.createReadStream> | undefined\n  try {\n    const stats = await handle.stat()\n    if (!stats.isFile()) {\n      throw new XlsxNotAFileError(filePath)\n    }\n\n    source = handle.createReadStream({ autoClose: false })\n    const reader = new ExcelJS.stream.xlsx.WorkbookReader(source, {\n      worksheets: 'emit',\n      sharedStrings: 'cache',\n      styles: 'ignore',\n      hyperlinks: 'ignore',\n      entries: 'ignore',\n    })\n\n    // Seed `model` so exceljs cannot dereference it before it has parsed\n    // `xl/workbook.xml` — an UPSTREAM defect, and the cause of the xlsx-reader\n    // \"flake\" that failed CI intermittently for months (#619, #835).\n    //\n    // Its streaming reader parses a worksheet INLINE, rather than deferring it to\n    // a temp file, when `this.sharedStrings && this.workbookRels` are both set as\n    // the entry arrives (`workbook-reader.js:109`). Neither implies\n    // `xl/workbook.xml` has been read. The inline path then does\n    // `matchingRel && (this.model.sheets || []).find(…)` (`:303`) — guarding\n    // `this.workbookRels` on the line above and `this.model.sheets` not at all —\n    // so any workbook whose zip orders `sharedStrings.xml` before the worksheet,\n    // with `workbook.xml` after it, throws `Cannot read properties of undefined\n    // (reading 'sheets')`.\n    //\n    // That is a PRODUCTION hazard, not a test artifact: this function reads\n    // operator-uploaded spreadsheets and the writer chooses the entry order. It\n    // presented as flakiness only because the fixtures our own tests write happen\n    // to sit on the safe side of the condition.\n    //\n    // An empty `sheets` makes the lookup miss, so the worksheet keeps its default\n    // id/name/state until `_parseWorkbook` overwrites `model` with the real one a\n    // few entries later — which is exactly what the deferred path already does.\n    // Preferred over a `pnpm patch` (exceljs 4.4.0 is the latest release, so there\n    // is no upstream fix to take, but a patch is a standing maintenance cost and\n    // this is one assignment that stays correct even if upstream fixes it).\n    //\n    // `sheets: []` rather than `{}` because `(this.model.sheets || [])` is what is\n    // read.\n    //\n    // The double assertion is NOT laziness and a single `as typeof reader.model`\n    // does not compile — checked, not assumed:\n    //\n    //   TS2352: Conversion of type '{ sheets: never[]; }' to type 'WorkbookModel'\n    //   may be a mistake because neither type sufficiently overlaps with the\n    //   other. … missing the following properties: creator, lastModifiedBy,\n    //   lastPrinted, created, and 17 more.\n    //\n    // A deliberately PARTIAL seed is the whole point — exceljs overwrites it with\n    // the real model a few entries later, and filling 21 fields with invented\n    // values to satisfy the compiler would be worse than the assertion: it would\n    // put fabricated workbook metadata one `_parseWorkbook` failure away from\n    // being read as real.\n    // toolkit-smell-allow: partial seed for a 21-field type; single assertion rejected by TS2352 (see above)\n    reader.model = { sheets: [] } as unknown as typeof reader.model\n\n    let dataRowNumber = 0\n    let sheetIndex = 0\n    let matchedTarget = false\n\n    // Drive the worksheet iterator BY HAND rather than with `for await`, so the\n    // `finally` below can resume it. Abandoning exceljs's `parse()` generator —\n    // which `return`/`break` out of a `for await` does — LEAKS a temp file and an\n    // fd per deferred worksheet, because `tempFileCleanupCallback()` sits after\n    // the `yield*` with no `finally` of its own (`workbook-reader.js:140-149`).\n    //\n    // Measured, with a negative control: 25 early-returning reads of a two-sheet\n    // deferred-order workbook leaked 50 fds and 50 temp files; the draining shape\n    // below leaks 0 of each. `tmp`'s graceful-cleanup hook reclaims the files at\n    // NORMAL process exit only, which is no help to a long-lived queue worker and\n    // no help at all against the fd ceiling: at the #842 audit's 5000-run cap\n    // this is an EMFILE, and an EMFILE arrives as the uncaught stream error the\n    // `open()` above exists to prevent.\n    //\n    // THE COST IS PROPORTIONAL TO THE WORKSHEETS AFTER THE TARGET, which is zero\n    // for the single-sheet feeds this reader mostly sees and 4.7× on a workbook\n    // whose target is sheet 1 and whose sheet 2 holds 40k rows. Advancing the\n    // parent iterator WITHOUT consuming each trailing worksheet is far cheaper\n    // and does release the temp files — but it strands exceljs's\n    // `createReadStream` on each one, i.e. it trades the disk leak for the fd\n    // leak, and the fd leak is the one that ends in EMFILE. So: full drain.\n    const worksheets = reader[Symbol.asyncIterator]()\n    let readerExhausted = false\n    try {\n      while (true) {\n        const next = await worksheets.next()\n        if (next.done) {\n          readerExhausted = true\n          break\n        }\n        const worksheet = next.value\n        sheetIndex += 1\n        const isTarget =\n          targetSheetIndex === undefined ? sheetIndex === 1 : sheetIndex === targetSheetIndex\n\n        if (!isTarget) {\n          // Drain a non-target worksheet so the underlying zip stream advances to the\n          // next; exceljs's streaming reader stalls if a worksheet is left unconsumed.\n          for await (const _skip of worksheet) {\n            void _skip\n          }\n          continue\n        }\n        matchedTarget = true\n\n        let excelRowNum = 0\n        let header: string[] | null = null\n        for await (const row of worksheet) {\n          excelRowNum += 1\n          const cells = rowCells(row.values)\n\n          if (hasHeader && header === null) {\n            // Skip rows before the configured header row, then capture it (trimmed).\n            if (excelRowNum < headerRowIndex + 1) continue\n            header = cells.map((c) => c.trim())\n            // FAIL LOUDLY on an empty header, because the alternative is SILENT DATA\n            // CORRUPTION and we have watched it happen.\n            //\n            // exceljs assigns `this.sharedStrings = []` at the TOP of\n            // `_parseSharedStrings`, before parsing a single entry — and an empty\n            // array is truthy, which is what the inline-worksheet guard tests. So a\n            // shared-strings entry that yields nothing (a truncated or starved read;\n            // observed in CI under load) leaves every string cell resolving to\n            // `sharedStrings[index]` → `undefined`. The header parses to zero\n            // columns, every data row maps to `{}`, and the import reports SUCCESS\n            // with every value dropped. Reproduced deterministically by rebuilding a\n            // workbook with an empty `sst`: two rows read, no values, no error.\n            //\n            // A header row with no usable column is never a legitimate import, so\n            // there is no case to preserve. Refusing costs an operator a clear error\n            // and saves them a silently empty dataset they would have to notice\n            // themselves — the same posture the rest of this package takes toward\n            // config mistakes.\n            if (header.every((c) => c === '')) {\n              throw new XlsxEmptyHeaderError(excelRowNum)\n            }\n            assertUniqueHeaders(header)\n            continue\n          }\n\n          dataRowNumber += 1\n          const rec: Record<string, string> = {}\n          if (header) {\n            for (let i = 0; i < header.length; i += 1) {\n              const key = header[i]\n              if (key) rec[key] = cells[i] ?? ''\n            }\n          } else {\n            for (let i = 0; i < cells.length; i += 1) rec[String(i)] = cells[i] ?? ''\n          }\n          yield { rowNumber: dataRowNumber, row: rec }\n        }\n\n        // Target worksheet fully read. Stop yielding — the `finally` walks the\n        // remaining worksheets so exceljs reaches its cleanup.\n        break\n      }\n    } finally {\n      // Runs on every exit: target read, an empty-header throw, or the CONSUMER\n      // abandoning us mid-stream (a `break` in the caller's `for await` calls\n      // `.return()`, which lands here). Whatever the reason, exceljs still owns\n      // temp files it only releases as the consumer pulls past each worksheet.\n      //\n      // The whole body is CONTAINED, for the same reason `handle.close()` below\n      // is: a throw from a `finally` REPLACES whatever the try block was\n      // propagating. Trailing worksheets are parsed here, so a malformed one\n      // would overwrite an `XlsxEmptyHeaderError` with an XML error — turning\n      // the #842 audit's `corrupt` verdict into a meaningless `error` and\n      // failing an import whose target sheet read perfectly. Measured: two of\n      // four malformed-trailing-sheet variants masked the refusal before this\n      // catch existed. Cleanup is best-effort by definition; the error the\n      // caller came for is not.\n      try {\n        while (!readerExhausted) {\n          const rest = await worksheets.next()\n          if (rest.done) break\n          for await (const _skip of rest.value) {\n            void _skip\n          }\n        }\n      } catch {\n        // Deliberately swallowed — see above. A trailing worksheet we only read\n        // to satisfy exceljs's cleanup contract cannot be allowed to speak for\n        // the read the caller actually asked for. Worst case we leak the temp\n        // files this drain existed to release, which is what `main` did on\n        // EVERY read.\n      }\n    }\n\n    // Reached only when a specific sheet was requested but never matched — the\n    // first-sheet default always matches on a non-empty workbook.\n    if (targetSheetIndex !== undefined && !matchedTarget) {\n      throw new Error(`xlsx worksheet ${targetSheetIndex} not found`)\n    }\n  } finally {\n    // Order matters: destroy the reader of the fd before closing the fd.\n    source?.destroy()\n    await handle.close().catch(() => {\n      // Swallowed deliberately, and biome's `noUnsafeFinally` agrees: a throw\n      // here would REPLACE whatever the try block was propagating, and what it\n      // propagates is the point of this reader — an empty-header refusal has to\n      // reach the caller intact. Closing a read-only fd we are finished with can\n      // fail only in ways that are not actionable, and the fd is reclaimed at\n      // process exit regardless.\n    })\n  }\n}\n"],"mappings":"kFAkCA,IAAa,EAAb,cAA0C,KAAM,CAE9C,gBAEA,YAAY,EAAyB,CACnC,MACE,mBAAmB,EAAgB,qIAErC,EACA,KAAK,KAAO,uBACZ,KAAK,gBAAkB,CACzB,CACF,EAYa,EAAb,cAAuC,KAAM,CAC3C,YAAY,EAAkB,CAC5B,MAAM,cAAc,EAAS,wBAAwB,EACrD,KAAK,KAAO,mBACd,CACF,EAcA,SAAS,EAAa,EAAsB,CAC1C,GAAI,GAAM,KAAyB,MAAO,GAC1C,GAAI,OAAO,GAAM,SAAU,OAAO,EAClC,GAAI,OAAO,GAAM,UAAY,OAAO,GAAM,UAAW,OAAO,OAAO,CAAC,EACpE,GAAI,aAAa,KAAM,OAAO,EAAE,YAAY,EAE5C,GAAI,aAAc,GAAK,MAAM,QAAQ,EAAE,QAAQ,EAC7C,OAAO,EAAE,SAAS,IAAK,GAAM,EAAE,IAAI,CAAC,CAAC,KAAK,EAAE,EAE9C,GAAI,SAAU,GAAK,OAAO,EAAE,MAAS,SAAU,OAAO,EAAE,KACxD,GAAI,WAAY,GAAK,EAAE,SAAW,IAAA,IAAa,EAAE,SAAW,KAAM,CAChE,IAAM,EAAI,EAAE,OAGZ,OAFI,aAAa,KAAa,EAAE,YAAY,EAErC,OAAO,GAAM,SAAW,GAAK,OAAO,CAAC,CAC9C,CACA,MAAO,EACT,CAGA,SAAS,EAAS,EAA8D,CAC9E,IAAM,EAAgB,CAAC,EACvB,GAAI,MAAM,QAAQ,CAAM,EAEtB,IAAK,IAAI,EAAI,EAAG,EAAI,EAAO,OAAQ,GAAK,EAAG,EAAI,EAAI,GAAK,EAAa,EAAO,EAAE,OAE9E,IAAK,GAAM,CAAC,EAAG,KAAM,OAAO,QAAQ,CAAM,EAAG,CAC3C,IAAM,EAAM,OAAO,SAAS,EAAG,EAAE,EAC7B,CAAC,OAAO,MAAM,CAAG,GAAK,GAAO,IAAG,EAAI,EAAM,GAAK,EAAa,CAAC,EACnE,CAIF,IAAK,IAAI,EAAI,EAAG,EAAI,EAAI,OAAQ,GAAK,EAAO,EAAI,KAAO,IAAA,KAAW,EAAI,GAAK,IAC3E,OAAO,CACT,CAQA,eAAuB,EAAW,EAAwD,CACxF,GAAM,CAAE,WAAU,YAAY,GAAM,iBAAiB,EAAG,SAAU,EAC5D,EAAU,MAAM,OAAO,WAMzB,EACJ,GAAI,IAAU,IAAA,KACZ,EAAmB,OAAO,SAAS,EAAO,EAAE,EACxC,OAAO,MAAM,CAAgB,GAAK,EAAmB,GACvD,MAAU,MAAM,uBAAuB,EAAM,uCAAuC,EAyBxF,IAAM,EAAS,MAAM,EAAK,EAAU,GAAG,EASnC,EACJ,GAAI,CAEF,GAAI,EAAC,MADe,EAAO,KAAK,EAAA,CACrB,OAAO,EAChB,MAAM,IAAI,EAAkB,CAAQ,EAGtC,EAAS,EAAO,iBAAiB,CAAE,UAAW,EAAM,CAAC,EACrD,IAAM,EAAS,IAAI,EAAQ,OAAO,KAAK,eAAe,EAAQ,CAC5D,WAAY,OACZ,cAAe,QACf,OAAQ,SACR,WAAY,SACZ,QAAS,QACX,CAAC,EA6CD,EAAO,MAAQ,CAAE,OAAQ,CAAC,CAAE,EAE5B,IAAI,EAAgB,EAChB,EAAa,EACb,EAAgB,GAuBd,EAAa,EAAO,OAAO,cAAc,CAAC,EAC5C,EAAkB,GACtB,GAAI,CACF,OAAa,CACX,IAAM,EAAO,MAAM,EAAW,KAAK,EACnC,GAAI,EAAK,KAAM,CACb,EAAkB,GAClB,KACF,CACA,IAAM,EAAY,EAAK,MAKvB,GAJA,GAAc,EAIV,EAFF,IAAqB,IAAA,GAAY,IAAe,EAAI,IAAe,GAEtD,CAGb,UAAW,IAAM,KAAS,GAG1B,QACF,CACA,EAAgB,GAEhB,IAAI,EAAc,EACd,EAA0B,KAC9B,UAAW,IAAM,KAAO,EAAW,CACjC,GAAe,EACf,IAAM,EAAQ,EAAS,EAAI,MAAM,EAEjC,GAAI,GAAa,IAAW,KAAM,CAEhC,GAAI,EAAc,EAAiB,EAAG,SAoBtC,GAnBA,EAAS,EAAM,IAAK,GAAM,EAAE,KAAK,CAAC,EAmB9B,EAAO,MAAO,GAAM,IAAM,EAAE,EAC9B,MAAM,IAAI,EAAqB,CAAW,EAE5C,EAAoB,CAAM,EAC1B,QACF,CAEA,GAAiB,EACjB,IAAM,EAA8B,CAAC,EACrC,GAAI,EACF,IAAK,IAAI,EAAI,EAAG,EAAI,EAAO,OAAQ,GAAK,EAAG,CACzC,IAAM,EAAM,EAAO,GACf,IAAK,EAAI,GAAO,EAAM,IAAM,GAClC,MAEA,IAAK,IAAI,EAAI,EAAG,EAAI,EAAM,OAAQ,GAAK,EAAG,EAAI,OAAO,CAAC,GAAK,EAAM,IAAM,GAEzE,KAAM,CAAE,UAAW,EAAe,IAAK,CAAI,CAC7C,CAIA,KACF,CACF,QAAU,CAeR,GAAI,CACF,KAAO,CAAC,GAAiB,CACvB,IAAM,EAAO,MAAM,EAAW,KAAK,EACnC,GAAI,EAAK,KAAM,MACf,UAAW,IAAM,KAAS,EAAK,OAGjC,CACF,MAAQ,CAMR,CACF,CAIA,GAAI,IAAqB,IAAA,IAAa,CAAC,EACrC,MAAU,MAAM,kBAAkB,EAAiB,WAAW,CAElE,QAAU,CAER,GAAQ,QAAQ,EAChB,MAAM,EAAO,MAAM,CAAC,CAAC,UAAY,CAOjC,CAAC,CACH,CACF"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@murumets-ee/imports",
-  "version": "0.49.0",
+  "version": "0.51.0",
   "license": "Elastic-2.0",
   "type": "module",
   "exports": {
@@ -45,13 +45,13 @@
     "drizzle-orm": "^0.45.2",
     "exceljs": "^4.4.0",
     "zod": "^3.24.1",
-    "@murumets-ee/core": "0.49.0",
-    "@murumets-ee/db": "0.49.0",
-    "@murumets-ee/entity": "0.49.0",
-    "@murumets-ee/logging": "0.49.0",
-    "@murumets-ee/queue": "0.49.0",
-    "@murumets-ee/search-elasticsearch": "0.49.0",
-    "@murumets-ee/storage": "0.49.0"
+    "@murumets-ee/core": "0.51.0",
+    "@murumets-ee/db": "0.51.0",
+    "@murumets-ee/entity": "0.51.0",
+    "@murumets-ee/logging": "0.51.0",
+    "@murumets-ee/queue": "0.51.0",
+    "@murumets-ee/storage": "0.51.0",
+    "@murumets-ee/search-elasticsearch": "0.51.0"
   },
   "devDependencies": {
     "@types/node": "^20.19.42",

package/dist/carmaker-feed-CxhZBrkO.mjs

@@ -1,2 +0,0 @@
-import{n as e,r as t,s as n}from"./parts-schema-D0gMhNWj.mjs";const r=`commerce:carmaker-feed`,i=e,a=n({targetSchema:`parts`,numberFormat:{decimal:`,`,thousands:`.`},columnMap:[{source:`ArticleName`,target:`code`,transforms:[{op:`trim`}]},{source:`NetPrice/Discount`,target:`net_price`,transforms:[{op:`europeanNumber`}]},{source:`GrossPrice`,target:`gross_price`,transforms:[{op:`europeanNumber`}]},{source:`CurrencyCd`,target:`currency`,transforms:[{op:`trim`}]},{source:`Barcode`,target:`barcode`,transforms:[{op:`trim`}]},{source:`Description1`,target:`description1`,transforms:[{op:`trim`}]},{source:`Description2`,target:`description2`,transforms:[{op:`trim`}]},{source:`Description_DE`,target:`name_de`,transforms:[{op:`trim`}]},{source:`Description_EN`,target:`name_en`,transforms:[{op:`trim`}]},{source:`Description_ES`,target:`name_es`,transforms:[{op:`trim`}]},{source:`Description_FR`,target:`name_fr`,transforms:[{op:`trim`}]},{source:`Description_IT`,target:`name_it`,transforms:[{op:`trim`}]},{source:`Description_NL`,target:`name_nl`,transforms:[{op:`trim`}]},{source:`Description_PT`,target:`name_pt`,transforms:[{op:`trim`}]},{source:`Description_RO`,target:`name_ro`,transforms:[{op:`trim`}]}]},t);export{i as n,a as r,r as t};
-//# sourceMappingURL=carmaker-feed-CxhZBrkO.mjs.map

package/dist/parts-schema-x8_GRYWc.d.mts.map

@@ -1 +0,0 @@
-{"version":3,"file":"parts-schema-x8_GRYWc.d.mts","names":[],"sources":["../src/mapping/number-format.ts","../src/mapping/column-transforms.ts","../src/mapping/target-schema.ts","../src/mapping/parts-schema.ts"],"mappings":";;;;;;;;;;AAqBA;;;;AAIW;AAsBX;;;;;;;;AAA6D;AA+B7D;AAAA,UAzDiB,YAAA;;EAEf,OAAA;EAuDwC;EArDxC,SAAS;AAAA;;;AA+EkC;;;;AC9E7C;;;;;;;;;;iBDqBgB,WAAA,CAAY,GAAA,UAAa,MAAA,EAAQ,YAAY;;;;;;;;iBA+B7C,YAAA,CAAa,KAAa;;;;;;;;;;;;;;;;;;;;;;;iBA0B1B,mBAAA,CAAoB,CAAS;;;;;AAzDgB;AA+B7D;;cCpDa,qBAAA,EAAqB,CAAA,CAAA,qBAAA,QAAA,CAAA,CAAA,SAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;KAYtB,eAAA,GAAkB,CAAA,CAAE,KAAK,QAAQ,qBAAA;;KAGjC,iBAAA,GAAoB,eAAe;;UAG9B,2BAAA;EACf,EAAA,EAAI,iBAAiB;EACrB,KAAA;;EAEA,QAAA;AAAA;;;;;;cAQW,oBAAA,WAA+B,2BAA2B;;KAY3D,SAAA;;;;;;;;iBASI,qBAAA,CACd,GAAA,UACA,UAAA,WAAqB,eAAA,IACrB,MAAA,EAAQ,YAAA,GACP,SAAA;;;;UC/Dc,WAAA;EF6B4C;EE3B3D,GAAA;EF0Dc;EExDd,KAAA;;EAEA,QAAA;EFsDwC;EEpDxC,IAAA;EF8EiC;;;AAAU;;;;AC9E7C;;;;;;;;ECgBE,OAAA;AAAA;;KAIU,SAAA,GAAY,MAAM,SAAS,SAAA;;UAGtB,eAAA;EACf,WAAA;EACA,SAAA;;;;;EAKA,MAAA,EAAQ,MAAM;AAAA;;UAIC,YAAA;;EAEf,IAAA;;EAEA,MAAA,WAAiB,WAAA;;EAEjB,QAAA,CAAS,MAAA,EAAQ,SAAA,EAAW,GAAA,EAAK,eAAA,GAAkB,SAAA,CAAU,IAAA;AAAA;;;;;;iBAY/C,UAAA,CAAW,CAAwB,EAArB,SAAS;;KAO3B,aAAA;EACN,IAAA;EAAe,KAAA;AAAA;EACf,IAAA;AAAA;EACA,IAAA;AAAA;;;;;;iBAOU,UAAA,CAAW,CAAA,EAAG,SAAA,eAAwB,aAAa;;iBASnD,QAAA,CAAS,SAAA,UAAmB,OAAA,UAAiB,KAAA;;;;;;;;;;AD9E7D;;;;;;AAAA,cEyBa,iBAAA,EAAiB,CAAA,CAAA,SAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;KAqBlB,WAAA,GAAc,CAAA,CAAE,KAAK,QAAQ,iBAAA;;;;;;;;iBAsMzB,aAAA,CAAc,CAAS;;;;;;cAuF1B,WAAA,EAAa,YAAY,CAAC,aAAA"}

package/dist/runner-C-s6SD7t.mjs

@@ -1,2 +0,0 @@
-import{z as e}from"zod";import{createReadStream as t}from"node:fs";import{parse as n}from"csv-parse";import{bulkPartialUpdate as r,bulkUpsert as i}from"@murumets-ee/search-elasticsearch";function a(e){let t=new Set;for(let n of e)if(n!==``){if(t.has(n))throw Error(`Duplicate header "${n}" after trimming — column names must be unique`);t.add(n)}}async function o(e,n={}){let{hasHeader:r=!0}=n,i=t(e),a=0,o=0;for await(let e of i){for(let t=0;t<e.length;t+=1)e[t]===10&&(a+=1);e.length>0&&(o=e[e.length-1]??0)}return o!==0&&o!==10&&(a+=1),r?Math.max(0,a-1):a}async function*s(e){let{filePath:r,delimiter:i=`	`,hasHeader:o=!0,columns:s,relaxColumnCount:c=!1,quote:l=!1}=e,u=s?Array.from(s):o?e=>{let t=e.map(e=>e.trim());return a(t),t}:!1,d=t(r),f=d.pipe(n({delimiter:i,columns:u,bom:!0,skip_empty_lines:!0,relax_column_count:c,quote:l})),p=0;try{for await(let e of f){p+=1;let t={};if(Array.isArray(e)){let n=e;for(let e=0;e<n.length;e+=1)t[String(e)]=n[e]??``}else for(let[n,r]of Object.entries(e))t[n]=r??``;yield{rowNumber:p,row:t}}}finally{d.destroy()}}const c=e.object({format:e.enum([`csv`,`tsv`,`xlsx`]).optional(),delimiter:e.string().min(1).max(4).optional(),hasHeader:e.boolean().optional(),headerRowIndex:e.number().int().min(0).optional(),sheet:e.string().min(1).max(128).optional()}).strict();function l(e){let t=e&&typeof e==`object`&&!Array.isArray(e)?e:{},n=t.feed===void 0||t.feed===null?{}:t.feed;return c.safeParse(n)}function u(e){let t=e.delimiter??(e.format===`csv`?`,`:`	`);return{filePath:e.filePath,delimiter:t,...e.hasHeader!==void 0&&{hasHeader:e.hasHeader},...e.columns!==void 0&&{columns:e.columns},...e.relaxColumnCount!==void 0&&{relaxColumnCount:e.relaxColumnCount},...e.quote!==void 0&&{quote:e.quote}}}async function*d(e){if(e.format===`xlsx`){let{streamXlsx:t}=await import(`./xlsx-reader-Ccesszbc.mjs`);yield*t({filePath:e.filePath,...e.hasHeader!==void 0&&{hasHeader:e.hasHeader},...e.headerRowIndex!==void 0&&{headerRowIndex:e.headerRowIndex},...e.sheet!==void 0&&{sheet:e.sheet}});return}yield*s(u(e))}const f=50,p=5;var m=class{patterns=new Map;maxPatterns;maxSamplesPerPattern;droppedSignatures=0;constructor(e={}){this.maxPatterns=e.maxPatterns??50,this.maxSamplesPerPattern=e.maxSamplesPerPattern??5}addError(e,t,n,r,i){let a=`${t}:${r??`GENERAL`}:${n}`,o=this.patterns.get(a);if(!o){if(this.patterns.size>=this.maxPatterns){this.droppedSignatures+=1;return}o={errorType:t,field:r??null,message:n,count:0,firstOccurrence:e,lastOccurrence:e,samples:[]},this.patterns.set(a,o)}o.count+=1,o.lastOccurrence=e,o.samples.length<this.maxSamplesPerPattern&&o.samples.push({rowNumber:e,rowData:i})}getTotalErrorCount(){let e=0;for(let t of this.patterns.values())e+=t.count;return e}getDistinctPatternCount(){return this.patterns.size}getDroppedSignatureCount(){return this.droppedSignatures}getTopPatterns(){let e=Array.from(this.patterns.values()).sort((e,t)=>t.count-e.count),t=e.reduce((e,t)=>e+t.count,0);return e.map(e=>({errorType:e.errorType,field:e.field,message:e.message,count:e.count,firstOccurrence:e.firstOccurrence,lastOccurrence:e.lastOccurrence,samples:e.samples.slice(),percentage:t>0?e.count/t*100:0}))}snapshot(){return{totalErrors:this.getTotalErrorCount(),distinctPatterns:this.getDistinctPatternCount(),droppedSignatures:this.droppedSignatures,patterns:this.getTopPatterns()}}};const h=1e3;async function g(e){let{importRunId:t,runLabel:n,params:a,transform:o,feed:s,esClient:c,esIndex:l,batchSize:u=h,onProgress:f,rowLimit:p,signal:g,errorTracker:_=new m,warningTracker:v=new m,logger:y}=e;if(u<1)throw Error(`batchSize must be >= 1 (got ${u})`);let b=Date.now();y?.info({importRunId:t,runLabel:n,batchSize:u,esIndex:l},`imports.runner: started`);let x=0,S=0,C=0,w=0,T=0,E=0,D=[],O=[],k=[],A=async()=>{if(k.length===0)return;let e=k;k=[];let n=Date.now(),r;try{r=await i(c,{index:l,docs:e.map(({id:e,doc:t})=>({id:e,doc:t})),...g!==void 0&&{signal:g}})}catch(t){let n=g?.aborted??(t instanceof Error&&(t.name===`AbortError`||/abort/i.test(t.message)))?`aborted`:`bulk_request_failed`,r=t instanceof Error?t.message:String(t);for(let{rowNumber:t}of e)_.addError(t,n,r,void 0,null);throw C+=e.length,t}S+=r.succeeded,C+=r.failures.length,E+=1;let a=Date.now()-n,o=(Date.now()-b)/1e3;if(y?.info({importRunId:t,batch:E,flushMs:a,submitted:e.length,succeeded:r.succeeded,failed:r.failures.length,rowsRead:x,rowsPerSecond:o>0?Math.round(x/o):0},`imports.runner: batch flushed`),r.failures.length>0){let t=new Map(e.map(e=>[e.id,e.rowNumber]));for(let e of r.failures){let n=t.get(e.id)??-1;_.addError(n,e.type,e.reason,void 0,{id:e.id})}}let s=new Set(r.failures.map(e=>e.id));for(let t of e)if(!(!t.warnings||t.warnings.length===0)&&!s.has(t.id)){T+=1;for(let e of t.warnings)v.addError(t.rowNumber,e.warningType,e.message,e.field,t.rowData)}for(let t of e)!t.pairs||t.pairs.length===0||s.has(t.id)||D.push(...t.pairs);if(f){let e=(Date.now()-b)/1e3;f({rowsRead:x,rowsSucceeded:S,rowsFailed:C,rowsSkipped:w,rowsWarned:T,batchesCompleted:E,elapsedSeconds:e,rowsPerSecond:e>0?x/e:0,distinctErrorPatterns:_.getDistinctPatternCount()})}},j=u,M=new Map;for await(let{rowNumber:e,row:r}of d(s)){if(g?.aborted||p!==void 0&&x>=p)break;if(x+=1,x===1&&y?.info({importRunId:t,rowNumber:e,sampleKeys:Object.keys(r)},`imports.runner: STREAM_FIRST_ROW`),x%j===0){let e=(Date.now()-b)/1e3;y?.info({importRunId:t,rowsRead:x,rowsSucceeded:S,rowsFailed:C,rowsSkipped:w,pending:k.length,elapsedSec:Math.round(e*10)/10,rowsPerSecond:e>0?Math.round(x/e):0,distinctFirstErrorTypes:M.size},`imports.runner: STREAM_TICK`)}let i={importRunId:t,params:a,runLabel:n,rowNumber:e},s;try{s=await o(r,i)}catch(n){let i=n instanceof Error?n.message:String(n);_.addError(e,`transform_threw`,i,void 0,r),M.has(`transform_threw`)||(M.set(`transform_threw`,{rowNumber:e,message:i}),y?.warn({importRunId:t,rowNumber:e,errorType:`transform_threw`,message:i},`imports.runner: FIRST_ERROR transform_threw`)),C+=1;continue}if(s.kind===`skip`){w+=1,s.patches&&s.patches.length>0&&O.push(...s.patches);continue}if(s.kind===`error`){_.addError(e,s.error.errorType,s.error.message,s.error.field,r),M.has(s.error.errorType)||(M.set(s.error.errorType,{rowNumber:e,message:s.error.message}),y?.warn({importRunId:t,rowNumber:e,errorType:s.error.errorType,field:s.error.field,message:s.error.message},`imports.runner: FIRST_ERROR ${s.error.errorType}`)),C+=1;continue}k.push({id:s.id,doc:s.doc,rowNumber:e,rowData:r,...s.warnings&&s.warnings.length>0&&{warnings:s.warnings},...s.pairs&&s.pairs.length>0&&{pairs:s.pairs}}),k.length>=u&&await A()}await A();let N=0,P=0;for(let e=0;e<O.length;e+=u){if(g?.aborted)throw Error(`imports: run aborted during core-charge attach (phase 2)`);let t=O.slice(e,e+u),n;try{n=await r(c,{index:l,updates:t.map(({id:e,doc:t})=>({id:e,doc:t})),...g!==void 0&&{signal:g}})}catch(e){if(g?.aborted??(e instanceof Error&&(e.name===`AbortError`||/abort/i.test(e.message))))throw e;let n=e instanceof Error?e.message:String(e);for(let e of t)v.addError(-1,`patch_attach_failed`,n,e.orphanWarning?.field,{target:e.id});continue}if(N+=n.succeeded,n.failures.length>0){let e=new Map(t.map(e=>[e.id,e]));for(let t of n.failures){let n=e.get(t.id)?.orphanWarning;t.status===404||t.type===`document_missing_exception`?(P+=1,v.addError(-1,n?.warningType??`orphan_patch`,n?.message??t.reason,n?.field,{target:t.id})):v.addError(-1,`patch_attach_failed`,t.reason,n?.field,{target:t.id})}}}let F=(Date.now()-b)/1e3;return y?.info({importRunId:t,rowsRead:x,rowsSucceeded:S,rowsFailed:C,rowsSkipped:w,batchesCompleted:E,patchesApplied:N,patchesOrphaned:P,totalSec:Math.round(F*10)/10,rowsPerSecond:F>0?Math.round(x/F):0},`imports.runner: finished`),{rowsRead:x,rowsSucceeded:S,rowsFailed:C,rowsSkipped:w,rowsWarned:T,batchesCompleted:E,errors:_.snapshot(),warnings:v.snapshot(),pairs:D,patchesApplied:N,patchesOrphaned:P}}export{m as a,l as c,s as d,p as i,a as l,g as n,c as o,f as r,d as s,h as t,o as u};
-//# sourceMappingURL=runner-C-s6SD7t.mjs.map

package/dist/runner-C-s6SD7t.mjs.map

@@ -1 +0,0 @@
-{"version":3,"file":"runner-C-s6SD7t.mjs","names":[],"sources":["../src/streaming.ts","../src/read-feed.ts","../src/error-tracker.ts","../src/runner.ts"],"sourcesContent":["/**\n * Tab-delimited / CSV streaming reader. Generic over delimiter so the\n * same path handles `.txt` (tab), `.csv` (comma), and the rare `;`\n * European export dialect.\n *\n * Built on `csv-parse` per giga-test precedent — node-stream-based, low\n * memory, handles UTF-8 BOM (the carmaker feeds are Windows-exported\n * and ship with a BOM that breaks naive split-on-tab parsers).\n *\n * The reader yields `{ rowNumber, row }` pairs where:\n * - `rowNumber` is 1-based and counts the header as row 0.\n * - `row` is `Record<string, string>` keyed by header name. Empty cells\n *   are the empty string, NOT `undefined` — feed transforms test with\n *   `value === ''` consistently.\n *\n * Why this lives in a streaming reader and not inside the transform\n * itself: the transform sees one already-parsed row at a time, never\n * the file. That keeps transform implementations free of I/O concerns\n * and makes them trivially unit-testable with a fixture row map.\n */\n\nimport { createReadStream } from 'node:fs'\nimport { parse } from 'csv-parse'\n\n/**\n * Throw if two (non-empty) header names collide. Headers are the keys the mapping\n * engine matches a `source` against; after trimming, two columns differing only in\n * whitespace would collapse to one key and silently overwrite each other's data.\n * Fail fast instead — the operator fixes the feed or the mapping. Shared by the\n * csv/tsv reader (below) and the xlsx reader. Blank columns are ignored (unmapped).\n */\nexport function assertUniqueHeaders(header: readonly string[]): void {\n  const seen = new Set<string>()\n  for (const h of header) {\n    if (h === '') continue\n    if (seen.has(h)) {\n      throw new Error(`Duplicate header \"${h}\" after trimming — column names must be unique`)\n    }\n    seen.add(h)\n  }\n}\n\n/**\n * Pre-pass: count data rows in a delimited file (excluding the header\n * row when `hasHeader` is true; default `true` to match `streamFeed`).\n *\n * Implementation: stream the file as raw bytes and tally `\\n`. NOT a\n * full parse — escaped newlines inside quoted fields would over-count.\n * That's acceptable for the supplier-feed use case because we default\n * `quote: false` (every `\\n` IS a row boundary). For genuine RFC-4180\n * CSV inputs with embedded newlines, treat the result as an upper\n * bound; the actual parsed row count from `streamFeed` is authoritative.\n *\n * Why bytes-only and not csv-parse: a 16 MB feed counts in ~50 ms with\n * `\\n` tally vs ~500 ms with csv-parse parsing. For 500 MB the gap\n * matters — pre-pass needs to be a small fraction of the actual import\n * or the operator perceives the run as \"stuck on pre-pass\" with no UI.\n *\n * Returns the number of DATA rows (header excluded when present).\n * Empty trailing lines (no `\\n` after the final row, or a single\n * trailing `\\n`) are handled identically to how csv-parse treats them\n * with `skip_empty_lines: true`.\n */\nexport async function countDataRows(\n  filePath: string,\n  options: { hasHeader?: boolean } = {},\n): Promise<number> {\n  const { hasHeader = true } = options\n  const stream = createReadStream(filePath)\n  let count = 0\n  let lastByte = 0\n  for await (const chunk of stream as AsyncIterable<Buffer>) {\n    for (let i = 0; i < chunk.length; i += 1) {\n      if (chunk[i] === 0x0a /* \\n */) count += 1\n    }\n    if (chunk.length > 0) lastByte = chunk[chunk.length - 1] ?? 0\n  }\n  // If the file doesn't end with a newline, the final row didn't get\n  // counted by the `\\n` tally — add one. Skipping when the file is\n  // entirely empty (count=0 and lastByte=0).\n  if (lastByte !== 0 && lastByte !== 0x0a) count += 1\n  return hasHeader ? Math.max(0, count - 1) : count\n}\n\nexport interface StreamFeedOptions {\n  /** Path to the file on disk. The PoC uploads land on local disk; S3-keyed reads come later. */\n  filePath: string\n  /**\n   * Single-character field delimiter. Default `\\t` (the carmaker feed\n   * format). Pass `,` for CSV, `;` for some European dialects.\n   */\n  delimiter?: string\n  /**\n   * `true` (default): the first row is the header and column names come\n   * from it. `false`: rows are emitted as positional `{ \"0\": ..., \"1\": ... }`\n   * and the transform reads by index — useful for headerless feeds that\n   * commit to a documented column order.\n   */\n  hasHeader?: boolean\n  /**\n   * Optional explicit column-name list. When provided, takes precedence\n   * over `hasHeader` (header row, if present, is skipped but its values\n   * are ignored). Useful when the upstream header is unstable but the\n   * positional shape isn't.\n   */\n  columns?: ReadonlyArray<string>\n  /**\n   * Forward to `csv-parse` `relax_column_count`. Default `false` —\n   * a row whose column count doesn't match the header surfaces as a\n   * parser error so the transform isn't silently fed truncated data.\n   */\n  relaxColumnCount?: boolean\n  /**\n   * Forward to `csv-parse` `quote`. The character (or `false`) that\n   * encloses fields containing the delimiter. Default `false` — most\n   * supplier feeds in this package's target use case are tab-delimited\n   * exports that contain literal `\"` characters in description fields\n   * (e.g. `LIVRE \"INVERSION D'IMAGE\"` from the Mercedes carmaker feed)\n   * with NO quote-as-field-wrapper convention. csv-parse's library\n   * default `'\"'` then misreads the literal `\"` as an opening quote\n   * and rejects the row with `Invalid Opening Quote`.\n   *\n   * Pass `'\"'` (or any character) to opt back into RFC-4180-style\n   * CSV parsing for genuine quoted-CSV inputs. Pass `'\\''` for\n   * single-quoted dialects.\n   */\n  quote?: string | false\n}\n\nexport interface StreamFeedRow {\n  /** 1-based row number. Header (when present) is row 0; first data row is row 1. */\n  rowNumber: number\n  /**\n   * Cell values keyed by column name (or string-position when\n   * `hasHeader: false` AND no `columns`).\n   *\n   * **Cell-value invariants:**\n   * - Empty cells (`A\\t\\tC`) → `''` (empty string).\n   * - Missing TRAILING cells in `relaxColumnCount: true` mode → the\n   *   key is **absent** from the object, not present-with-`''`. csv-parse\n   *   does not emit keys for short rows. Transforms reading those\n   *   columns get `undefined` from `row['col']` and must handle it\n   *   (`row['col'] ?? ''` is the canonical idiom).\n   * - With the default `relaxColumnCount: false`, short rows reject at\n   *   the parser, so this case never reaches the transform.\n   */\n  row: Record<string, string>\n}\n\n/**\n * Async-iterable over the parsed rows of a delimited file. Use with\n * `for await (const { rowNumber, row } of streamFeed({ filePath, ... }))`.\n *\n * The iterator owns its file descriptor — the `for await` loop closes\n * the underlying stream when it returns or breaks. Aborting mid-stream\n * (`break`, `throw`, signal) is safe; csv-parse propagates the close.\n */\nexport async function* streamFeed(options: StreamFeedOptions): AsyncIterable<StreamFeedRow> {\n  const {\n    filePath,\n    delimiter = '\\t',\n    hasHeader = true,\n    columns,\n    relaxColumnCount = false,\n    quote = false,\n  } = options\n\n  // When the caller provides explicit `columns`, prefer them. When the file\n  // has a header but no explicit `columns`, derive the column names from the\n  // first row via a function so we can TRIM them — real supplier feeds ship\n  // leading/trailing-space headers (R03 Samples A/C: `\" JAEHIND KM-TA\"`), and\n  // an untrimmed key wouldn't match a mapping's `source`. The xlsx reader trims\n  // headers the same way. When neither is true, rows get string-position keys.\n  //\n  // csv-parse's `columns` accepts `ColumnOption[] | boolean | ((record) =>\n  // ColumnOption[])`; `string[]` satisfies `ColumnOption[]`.\n  const columnConfig: string[] | boolean | ((header: string[]) => string[]) = columns\n    ? Array.from(columns)\n    : hasHeader\n      ? (header: string[]) => {\n          const trimmed = header.map((h) => h.trim())\n          assertUniqueHeaders(trimmed)\n          return trimmed\n        }\n      : false\n\n  const stream = createReadStream(filePath)\n  const parser = stream.pipe(\n    parse({\n      delimiter,\n      columns: columnConfig,\n      bom: true,\n      skip_empty_lines: true,\n      // csv-parse defaults to strict column count; opt-in relaxation only.\n      relax_column_count: relaxColumnCount,\n      // Default `quote: false` — see StreamFeedOptions.quote JSDoc.\n      // Carmaker / parts-feed exports contain literal `\"` characters\n      // in description fields and would otherwise reject as\n      // `Invalid Opening Quote`. Genuine RFC-4180 CSV consumers opt\n      // back in by passing `quote: '\"'`.\n      quote,\n    }),\n  )\n\n  let rowNumber = 0\n  try {\n    for await (const rawRow of parser as AsyncIterable<\n      ReadonlyArray<string> | Record<string, string | undefined>\n    >) {\n      rowNumber += 1\n      // csv-parse emits records with string keys when `columns` is\n      // truthy, otherwise an Array. Normalise both shapes to\n      // `Record<string, string>` so transforms can rely on `value === ''`\n      // for missing cells (csv-parse leaves trailing missing cells as\n      // `undefined` when `relax_column_count: true`; this collapses\n      // them to `''` to keep the contract uniform).\n      const row: Record<string, string> = {}\n      if (Array.isArray(rawRow)) {\n        const arr = rawRow as ReadonlyArray<string | undefined>\n        for (let i = 0; i < arr.length; i += 1) {\n          row[String(i)] = arr[i] ?? ''\n        }\n      } else {\n        for (const [k, v] of Object.entries(rawRow)) {\n          row[k] = v ?? ''\n        }\n      }\n      yield { rowNumber, row }\n    }\n  } finally {\n    // Safety: ensure the underlying file descriptor closes even if the\n    // consumer breaks mid-iteration. Node closes streams on\n    // garbage-collection but this makes it deterministic under tests.\n    stream.destroy()\n  }\n}\n","/**\n * Feed reader dispatcher (commerce-demo phase 02, D005). Selects the reader by\n * `format` and yields the uniform `AsyncIterable<{ rowNumber, row }>` the runner\n * consumes — csv/tsv via the existing `streamFeed` (csv-parse), xlsx via the\n * exceljs-backed `streamXlsx`.\n *\n * **Back-compat**: with no `format` (the carmaker worker passes only `filePath`),\n * this defaults to tab-delimited — the original `streamFeed` behaviour — so the\n * existing import path is unchanged.\n *\n * **Bundle safety**: the xlsx reader (and its heavy, Node-only exceljs dep) is\n * `await import()`-ed ONLY for `format: 'xlsx'`, so the csv/tsv path — and anything\n * that statically imports the runner — never pulls exceljs into its module graph.\n */\n\nimport { z } from 'zod'\nimport { type StreamFeedOptions, type StreamFeedRow, streamFeed } from './streaming.js'\n\n/**\n * The reader-config subset safe to persist in `import_run.params.feed` and rehydrate\n * in the worker (commerce-demo PR3). It is exactly the {@link ReadFeedOptions} fields\n * that describe HOW to parse a feed — `filePath` is supplied by the worker after it\n * materialises the upload, never carried in params. Validated at BOTH ends: the\n * commerce run route builds it from a saved `import_mapping`, the worker re-parses it\n * before handing it to {@link readFeed} (defense-in-depth against a tampered row).\n */\nexport const feedReaderConfigSchema = z\n  .object({\n    format: z.enum(['csv', 'tsv', 'xlsx']).optional(),\n    delimiter: z.string().min(1).max(4).optional(),\n    hasHeader: z.boolean().optional(),\n    headerRowIndex: z.number().int().min(0).optional(),\n    sheet: z.string().min(1).max(128).optional(),\n  })\n  .strict()\nexport type FeedReaderConfig = z.infer<typeof feedReaderConfigSchema>\n\n/**\n * Rehydrate the reader config a run persisted in `import_run.params.feed`.\n *\n * ONE implementation, because two would drift and drift here is silent: the\n * worker uses it to decide how to READ a feed, and the #842 audit uses it to\n * re-read the same file the same way. An audit that rehydrates differently from\n * the run it is auditing invents corruption, or misses it.\n *\n * ABSENT means \"the carmaker default\", i.e. tab-delimited, which is what an\n * empty config parses to — so a non-object `params`, or `params` with no\n * `feed`, is a real answer rather than a gap. A PRESENT `feed` is validated\n * whatever its shape: a scalar or an array there is a malformed persisted blob,\n * and reading it as \"no config\" would silently import an xlsx run as tsv.\n */\nexport function readPersistedFeedConfig(\n  params: unknown,\n): ReturnType<typeof feedReaderConfigSchema.safeParse> {\n  const record =\n    params && typeof params === 'object' && !Array.isArray(params)\n      ? (params as Record<string, unknown>)\n      : {}\n  // `null` is how \"absent\" round-trips through JSONB, so it defaults with\n  // `undefined`. Everything else present goes to the schema and is rejected on\n  // its merits rather than being flattened into the default.\n  const feed = record.feed === undefined || record.feed === null ? {} : record.feed\n  return feedReaderConfigSchema.safeParse(feed)\n}\n\nexport interface ReadFeedOptions {\n  /** Path to the file on local disk. */\n  filePath: string\n  /** Feed format. Omitted → tab-delimited (tsv), matching the legacy carmaker reader. */\n  format?: 'csv' | 'tsv' | 'xlsx'\n  /** Explicit delimiter override (csv/tsv). Defaults from `format` (`csv` → `,`, else `\\t`). */\n  delimiter?: string\n  /** First row is the header. Default `true`. */\n  hasHeader?: boolean\n  /** 0-based header row index (xlsx). */\n  headerRowIndex?: number\n  /** xlsx worksheet as a 1-based index string. Omit → first sheet. */\n  sheet?: string\n  /** Explicit column-name list (csv/tsv) — see `streamFeed`. */\n  columns?: ReadonlyArray<string>\n  /** Forward to csv-parse `relax_column_count` (csv/tsv). */\n  relaxColumnCount?: boolean\n  /** Forward to csv-parse `quote` (csv/tsv). */\n  quote?: string | false\n}\n\nfunction toStreamFeedOptions(o: ReadFeedOptions): StreamFeedOptions {\n  // No format → tab (the carmaker feed default). `csv` → comma; everything else → tab.\n  const delimiter = o.delimiter ?? (o.format === 'csv' ? ',' : '\\t')\n  return {\n    filePath: o.filePath,\n    delimiter,\n    ...(o.hasHeader !== undefined && { hasHeader: o.hasHeader }),\n    ...(o.columns !== undefined && { columns: o.columns }),\n    ...(o.relaxColumnCount !== undefined && { relaxColumnCount: o.relaxColumnCount }),\n    ...(o.quote !== undefined && { quote: o.quote }),\n  }\n}\n\n/** Stream a feed file's data rows, dispatching on `format`. */\nexport async function* readFeed(options: ReadFeedOptions): AsyncIterable<StreamFeedRow> {\n  if (options.format === 'xlsx') {\n    const { streamXlsx } = await import('./xlsx-reader.js')\n    yield* streamXlsx({\n      filePath: options.filePath,\n      ...(options.hasHeader !== undefined && { hasHeader: options.hasHeader }),\n      ...(options.headerRowIndex !== undefined && { headerRowIndex: options.headerRowIndex }),\n      ...(options.sheet !== undefined && { sheet: options.sheet }),\n    })\n    return\n  }\n  yield* streamFeed(toStreamFeedOptions(options))\n}\n","/**\n * Aggregates per-row errors into top-N pattern buckets so a feed of 1M\n * malformed rows surfaces as a handful of actionable signatures rather\n * than a million identical strings.\n *\n * Lifted from giga-test (`backend/src/workers/csv-importer.ts`) and\n * generalized:\n * - Configurable caps so the importer can tune memory bounds per run.\n * - Pure data — no logging side effects, no I/O.\n * - JSON-serialisable output via {@link ErrorTracker.snapshot} for the\n *   `import_run.errorSummary` column.\n *\n * Pattern signature shape: `${errorType}:${field || 'GENERAL'}:${message}`.\n * Same `errorType + field + message` collapses to one bucket; differing\n * messages stay separate. This is intentional: a parser error on column\n * `NetPrice/Discount` (\"invalid number 'NA,5'\") and the same on column\n * `GrossPrice` are operationally distinct even if the parser is the same.\n *\n * Memory bounds: the patterns map is capped at `maxPatterns`. Once full,\n * additional NEW signatures are dropped — known patterns keep accumulating\n * counts. This is the \"top-N most common\" model: rare-but-novel errors\n * past the cap are invisible, but the cap protects against a runaway\n * adversarial feed exploding the map. Sample arrays are independently\n * capped at `maxSamplesPerPattern`.\n */\n\n/**\n * Recursively-defined JSON-serialisable value. Mirrors the shape of\n * `JsonValue` in `@murumets-ee/entity` without taking a dependency on\n * that package — error-tracker is otherwise standalone, and the\n * `ImportRun.errorSummary` JSONB column accepts anything in this shape.\n */\nexport type ImportJsonValue =\n  | string\n  | number\n  | boolean\n  | null\n  | ImportJsonValue[]\n  | { [key: string]: ImportJsonValue }\n\n/** One sample row attached to a pattern. `rowData` is the originally-parsed row. */\nexport interface ErrorSample {\n  rowNumber: number\n  rowData: ImportJsonValue\n}\n\n/** Public shape of an aggregated pattern as returned by {@link ErrorTracker.getTopPatterns}. */\nexport interface ErrorPattern {\n  errorType: string\n  field: string | null\n  message: string\n  count: number\n  firstOccurrence: number\n  lastOccurrence: number\n  samples: ReadonlyArray<ErrorSample>\n  /** `count / totalErrors`, scaled 0..100. `0` when there are zero errors total. */\n  percentage: number\n}\n\nexport interface ErrorTrackerConfig {\n  /** Hard cap on distinct signatures. Defaults to 50. New signatures past the cap are dropped. */\n  maxPatterns?: number\n  /** Hard cap on samples retained per pattern. Defaults to 5. Excess samples are dropped. */\n  maxSamplesPerPattern?: number\n}\n\n/** Default caps — match giga-test for compatibility. */\nexport const DEFAULT_MAX_PATTERNS = 50\nexport const DEFAULT_MAX_SAMPLES_PER_PATTERN = 5\n\ninterface InternalPattern {\n  errorType: string\n  field: string | null\n  message: string\n  count: number\n  firstOccurrence: number\n  lastOccurrence: number\n  samples: ErrorSample[]\n}\n\nexport class ErrorTracker {\n  private readonly patterns = new Map<string, InternalPattern>()\n  private readonly maxPatterns: number\n  private readonly maxSamplesPerPattern: number\n  private droppedSignatures = 0\n\n  constructor(config: ErrorTrackerConfig = {}) {\n    this.maxPatterns = config.maxPatterns ?? DEFAULT_MAX_PATTERNS\n    this.maxSamplesPerPattern = config.maxSamplesPerPattern ?? DEFAULT_MAX_SAMPLES_PER_PATTERN\n  }\n\n  /**\n   * Record one error. Same `(errorType, field, message)` triple bumps the\n   * existing bucket; a new triple opens a new one (subject to {@link maxPatterns}).\n   *\n   * `field` is optional — pass `undefined` for errors not tied to a single\n   * column (e.g. parse errors at row level). Internally normalised to the\n   * literal string `'GENERAL'` so it shares a bucket with other genericised\n   * errors of the same type+message.\n   */\n  addError(\n    rowNumber: number,\n    errorType: string,\n    message: string,\n    field: string | undefined,\n    rowData: ImportJsonValue,\n  ): void {\n    const fieldKey = field ?? 'GENERAL'\n    const signature = `${errorType}:${fieldKey}:${message}`\n\n    let pattern = this.patterns.get(signature)\n    if (!pattern) {\n      if (this.patterns.size >= this.maxPatterns) {\n        this.droppedSignatures += 1\n        return\n      }\n      pattern = {\n        errorType,\n        field: field ?? null,\n        message,\n        count: 0,\n        firstOccurrence: rowNumber,\n        lastOccurrence: rowNumber,\n        samples: [],\n      }\n      this.patterns.set(signature, pattern)\n    }\n\n    pattern.count += 1\n    pattern.lastOccurrence = rowNumber\n    if (pattern.samples.length < this.maxSamplesPerPattern) {\n      pattern.samples.push({ rowNumber, rowData })\n    }\n  }\n\n  /** Total count across every pattern. Counts errors, not patterns. */\n  getTotalErrorCount(): number {\n    let sum = 0\n    for (const p of this.patterns.values()) sum += p.count\n    return sum\n  }\n\n  /** Number of distinct signatures retained in the map (≤ `maxPatterns`). */\n  getDistinctPatternCount(): number {\n    return this.patterns.size\n  }\n\n  /**\n   * Number of NEW signatures dropped because the map was already at\n   * capacity. Surfacing this in the import_run summary tells the operator\n   * \"the top-N was saturated — there's a long tail you're not seeing\".\n   */\n  getDroppedSignatureCount(): number {\n    return this.droppedSignatures\n  }\n\n  /**\n   * Top patterns sorted by descending count, capped at `maxPatterns`.\n   * Stable secondary order is insertion order (Map iteration order is\n   * insertion order; Array.sort is stable in V8).\n   */\n  getTopPatterns(): ErrorPattern[] {\n    const all = Array.from(this.patterns.values()).sort((a, b) => b.count - a.count)\n    const totalErrors = all.reduce((sum, p) => sum + p.count, 0)\n    return all.map((p) => ({\n      errorType: p.errorType,\n      field: p.field,\n      message: p.message,\n      count: p.count,\n      firstOccurrence: p.firstOccurrence,\n      lastOccurrence: p.lastOccurrence,\n      samples: p.samples.slice(),\n      percentage: totalErrors > 0 ? (p.count / totalErrors) * 100 : 0,\n    }))\n  }\n\n  /**\n   * Compact JSON-serialisable snapshot for `import_run.errorSummary`.\n   * Aside from the patterns array, includes the totals so a reader of\n   * just this column doesn't have to re-derive them.\n   */\n  snapshot(): ErrorTrackerSnapshot {\n    return {\n      totalErrors: this.getTotalErrorCount(),\n      distinctPatterns: this.getDistinctPatternCount(),\n      droppedSignatures: this.droppedSignatures,\n      patterns: this.getTopPatterns(),\n    }\n  }\n}\n\nexport interface ErrorTrackerSnapshot {\n  totalErrors: number\n  distinctPatterns: number\n  droppedSignatures: number\n  patterns: ErrorPattern[]\n}\n","/**\n * One-shot importer: stream rows → transform → batched bulk-write to ES,\n * accumulating per-row errors into `ErrorTracker` and reporting progress\n * to the queue every batch.\n *\n * Per PLAN-ECOMMERCE.md PR 7 (PoC scope):\n * - **Batch size 1000.** Matches giga-test precedent. Configurable for\n *   integration tests that don't want a 1k row floor.\n * - **No resumability and no automatic retries.** The `imports:run` queue\n *   job is registered with `defaultRetries: 0` (see `worker.ts`) so a\n *   failed handler does NOT re-enqueue itself — re-running a multi-batch\n *   import against the same `import_run.id` while the previous attempt\n *   may still be writing is a footgun (duplicate batches, double-counted\n *   progress). Operator retries by creating a NEW `import_run` row.\n * - **No per-supplier transform plugin.** The runner takes a single\n *   `RowTransform<TDoc>` from the registry and applies it to every row;\n *   PR 8 may diverge but only by registering a different transform name.\n * - **Direct `bulkUpsert` into the live aliased index** (D6 alias is\n *   set up by PR 4's `ensureAliasedIndex`; the importer doesn't reindex).\n *\n * Per D21 (sanctioned bulk path): this runner intentionally bypasses\n * AdminClient and entity hooks. Per-batch audit / observability lives\n * on the surrounding `import_run` row + queue progress, NOT per-row.\n */\n\nimport {\n  type BulkIndexResult,\n  bulkPartialUpdate,\n  bulkUpsert,\n  type EsClientLike,\n} from '@murumets-ee/search-elasticsearch'\nimport { ErrorTracker } from './error-tracker.js'\nimport { readFeed, type ReadFeedOptions } from './read-feed.js'\nimport type {\n  DirectionalPair,\n  DocPatch,\n  RowTransform,\n  RowWarning,\n  TransformContext,\n} from './transform.js'\n\n/** Soft default; chosen to match giga-test. ES bulk requests over ~5MB get split server-side anyway. */\nexport const DEFAULT_BATCH_SIZE = 1000\n\nexport interface RunImportOptions<TDoc> {\n  /** UUID of the `import_run` row driving this run. Forwarded to every transform invocation. */\n  importRunId: string\n  /** Operator-supplied label for the run. Forwarded to the transform context. */\n  runLabel: string\n  /** Opaque per-run params copied from `import_run.params`. */\n  params: Record<string, unknown>\n  /** Transform applied to every parsed row. */\n  transform: RowTransform<TDoc>\n  /** Feed reader options — file path + format (csv/tsv/xlsx) + delimiter/header config. */\n  feed: ReadFeedOptions\n  /** ES client (low-level shape from `@murumets-ee/search-elasticsearch`). */\n  esClient: EsClientLike\n  /** Index alias to write to. Per D6, callers always pass an alias, never a physical index. */\n  esIndex: string\n  /** Rows per `bulkUpsert` call. Default {@link DEFAULT_BATCH_SIZE}. */\n  batchSize?: number\n  /**\n   * Callback invoked after every batch. The handler in `worker.ts`\n   * forwards this to `ctx.updateProgress` for the queue UI; tests\n   * inspect it directly. Synchronous + cheap so a slow callback can't\n   * back-pressure the importer.\n   */\n  onProgress?: (progress: ImportRunProgress) => void\n  /** Optional: stop processing after this many rows. Tests use it; production passes `undefined`. */\n  rowLimit?: number\n  /** Abort signal threaded into the underlying ES client request — cooperative cancel. */\n  signal?: AbortSignal\n  /** Optional ErrorTracker config (caps). Default: top-50 patterns × 5 samples. */\n  errorTracker?: ErrorTracker\n  /**\n   * Optional separate tracker for NON-FATAL row warnings (commerce-demo PR4 —\n   * `no_purchase_price`). Distinct from {@link errorTracker} so warnings never\n   * inflate `totals.failed` / the error summary: a warned row is a SUCCESS whose\n   * doc is indexed. Default: a fresh tracker with the same caps.\n   */\n  warningTracker?: ErrorTracker\n  /**\n   * Optional structural logger for per-batch telemetry. Receives\n   * `{ batch, ms, succeeded, failed, rowsRead, rowsPerSecond }` after\n   * every flush, plus a one-shot `started` and `finished` event. The\n   * worker forwards `app.logger.child({ pkg: 'imports' })`; tests\n   * leave it undefined.\n   *\n   * Why: without per-batch visibility, a stalled run looks identical\n   * to \"first batch in progress\" forever — the only feedback is the\n   * queue's `progress` jsonb (which writes after each batch) and the\n   * eventual `succeeded`/`failed` job event. Adding this lets the\n   * operator tail the worker log and immediately see whether ES is\n   * rate-limiting, the transform is rejecting every row, or the\n   * stream is actually progressing.\n   */\n  logger?: {\n    info: (data: Record<string, unknown>, msg: string) => void\n    warn: (data: Record<string, unknown>, msg: string) => void\n  }\n}\n\n/**\n * Progress payload written to `toolkit_jobs.progress` after every batch.\n * Caps + flush rules live on the queue's `updateProgress` debounce —\n * callers don't need to throttle.\n */\nexport interface ImportRunProgress {\n  rowsRead: number\n  rowsSucceeded: number\n  rowsFailed: number\n  rowsSkipped: number\n  /** Rows indexed WITH a non-fatal warning (commerce-demo PR4 — `no_purchase_price`). */\n  rowsWarned: number\n  batchesCompleted: number\n  /** Wall-clock seconds since the runner started. */\n  elapsedSeconds: number\n  /** Rows / second, computed at every batch. */\n  rowsPerSecond: number\n  /** Distinct error patterns currently held by the tracker. Saturates at the cap. */\n  distinctErrorPatterns: number\n  /**\n   * Total data-row count from a pre-pass scan, when the worker computed one.\n   * Lets the UI render `processed / total` percent. Absent when the worker\n   * skipped the count (file too large for the count budget, count failed,\n   * etc.) — UI must treat it as optional.\n   */\n  totalRows?: number\n}\n\n/**\n * Final result returned by {@link runImport}. The handler writes these\n * onto the `import_run` row alongside the ErrorTracker snapshot.\n */\nexport interface RunImportResult {\n  /** Total rows read from the file (excludes skipped empty lines). */\n  rowsRead: number\n  /** Rows the transform turned into a successful doc AND the ES cluster acknowledged. */\n  rowsSucceeded: number\n  /**\n   * Rows that the transform rejected (`{ kind: 'error' }`) OR that ES\n   * rejected on bulk-write (per-doc failure). Both are aggregated by\n   * `errorTracker` for the import_run summary.\n   */\n  rowsFailed: number\n  /** Rows that the transform skipped (`{ kind: 'skip' }`) — header noise, blank lines, intentional drop. */\n  rowsSkipped: number\n  /**\n   * Rows that succeeded (doc indexed) but carried ≥1 non-fatal warning —\n   * counted SEPARATELY from `rowsSucceeded`/`rowsFailed` (a warned row is also\n   * counted in `rowsSucceeded`). commerce-demo PR4: `no_purchase_price`.\n   */\n  rowsWarned: number\n  /** Number of `bulkUpsert` calls made. */\n  batchesCompleted: number\n  /** Final value of {@link ErrorTracker.snapshot}. */\n  errors: ReturnType<ErrorTracker['snapshot']>\n  /** Aggregated NON-FATAL warnings (commerce-demo PR4). Empty when no row warned. */\n  warnings: ReturnType<ErrorTracker['snapshot']>\n  /**\n   * Directional supersession pairs harvested as a side-output (commerce-demo PR6,\n   * SD007). Buffered in memory from ES-ACCEPTED rows only (same accounting as\n   * warnings — a row whose doc was rejected by ES contributes no pair). Empty unless\n   * a schema emits them (the `parts` schema, when a row's `superseding_code` is set).\n   * The worker hands these to an injected `pairSink` AFTER the run for the commerce\n   * replacement-merge engine; the runner itself never writes the consumer's tables.\n   */\n  pairs: readonly DirectionalPair[]\n  /**\n   * Doc-patches applied by the phase-2 pass (commerce-demo PR7): a `<code>*S`\n   * core-fee row is not a saleable doc — it emits a {@link DocPatch} that merges its\n   * deposit onto the PARENT part's `core_charge_minor`. The runner buffers these\n   * during the stream (sparse) and applies them via `bulkPartialUpdate` AFTER the ES\n   * pass. `patchesApplied` = targets the cluster merged; `patchesOrphaned` = patches\n   * whose parent genuinely wasn't in the index (a `document_missing`/404 → an\n   * `orphan_core_charge` warning). An INFRA failure (whole-batch transport error, or a\n   * non-404 per-doc failure) is NOT counted as orphaned — it lands in a distinct\n   * `patch_attach_failed` warning bucket so it can't masquerade as a data orphan. Zero\n   * unless a schema emits patches.\n   */\n  patchesApplied: number\n  patchesOrphaned: number\n}\n\n/**\n * Apply the runner against a feed file. Stops on rowLimit OR end-of-file\n * OR if `signal` aborts. Throws if the streaming reader / ES client\n * throws — caller (the queue handler) catches that and writes\n * `import_run.status = 'failed'` with the error message in\n * `errorSummary.fatal`.\n */\nexport async function runImport<TDoc>(options: RunImportOptions<TDoc>): Promise<RunImportResult> {\n  const {\n    importRunId,\n    runLabel,\n    params,\n    transform,\n    feed,\n    esClient,\n    esIndex,\n    batchSize = DEFAULT_BATCH_SIZE,\n    onProgress,\n    rowLimit,\n    signal,\n    errorTracker = new ErrorTracker(),\n    warningTracker = new ErrorTracker(),\n    logger,\n  } = options\n\n  if (batchSize < 1) {\n    throw new Error(`batchSize must be >= 1 (got ${batchSize})`)\n  }\n\n  const startedAt = Date.now()\n  logger?.info({ importRunId, runLabel, batchSize, esIndex }, 'imports.runner: started')\n  let rowsRead = 0\n  let rowsSucceeded = 0\n  let rowsFailed = 0\n  let rowsSkipped = 0\n  let rowsWarned = 0\n  let batchesCompleted = 0\n\n  // Side-output buffer (commerce-demo PR6, SD007): directional supersession pairs\n  // accumulated from ES-ACCEPTED rows across the whole run. An array append per\n  // sparse superseded row — it never blocks the `bulkUpsert` loop. Handed to the\n  // injected `pairSink` by the worker AFTER the run (read-once / merge-after).\n  const collectedPairs: DirectionalPair[] = []\n\n  // Side-output buffer (commerce-demo PR7): doc-patches from `*S` core-fee rows —\n  // each attaches a core charge onto a DIFFERENT (parent) doc. Collected off the\n  // `skip` result (a `*S` row is never a saleable doc), sparse, applied in a phase-2\n  // `bulkPartialUpdate` after the ES pass (below) — order-independent because ES\n  // `update` is realtime by `_id`.\n  const collectedPatches: DocPatch<TDoc>[] = []\n\n  // Warnings + pairs ride on the pending item (with the raw row for the warning\n  // sample) so they're accounted at FLUSH time — only for rows ES actually accepted.\n  // A warned row is by contract a SUCCESSFUL row, so counting it at push time (before\n  // the bulk response) would wrongly keep a later-ES-rejected row in rowsWarned /\n  // warnings; pairs follow the same accepted-rows-only rule.\n  let pending: Array<{\n    id: string\n    doc: TDoc\n    rowNumber: number\n    rowData: Record<string, string>\n    warnings?: readonly RowWarning[]\n    pairs?: readonly DirectionalPair[]\n  }> = []\n\n  const flush = async (): Promise<void> => {\n    if (pending.length === 0) return\n    const batch = pending\n    pending = []\n    const flushStartedAt = Date.now()\n    let result: BulkIndexResult\n    try {\n      result = await bulkUpsert<TDoc>(esClient, {\n        index: esIndex,\n        docs: batch.map(({ id, doc }) => ({ id, doc })),\n        ...(signal !== undefined && { signal }),\n      })\n    } catch (err) {\n      // Cluster- or transport-level failure — the whole batch is\n      // unaccounted for. Distinguish abort (operator-driven cancel) from\n      // a real cluster failure so the errorSummary doesn't mislabel a\n      // cancelled run as broken cluster connectivity.\n      const isAbort =\n        signal?.aborted ??\n        (err instanceof Error && (err.name === 'AbortError' || /abort/i.test(err.message)))\n      const errorType = isAbort ? 'aborted' : 'bulk_request_failed'\n      const reason = err instanceof Error ? err.message : String(err)\n      for (const { rowNumber } of batch) {\n        errorTracker.addError(rowNumber, errorType, reason, undefined, null)\n      }\n      rowsFailed += batch.length\n      throw err\n    }\n\n    rowsSucceeded += result.succeeded\n    rowsFailed += result.failures.length\n    batchesCompleted += 1\n\n    const flushMs = Date.now() - flushStartedAt\n    const elapsedSec = (Date.now() - startedAt) / 1000\n    logger?.info(\n      {\n        importRunId,\n        batch: batchesCompleted,\n        flushMs,\n        submitted: batch.length,\n        succeeded: result.succeeded,\n        failed: result.failures.length,\n        rowsRead,\n        rowsPerSecond: elapsedSec > 0 ? Math.round(rowsRead / elapsedSec) : 0,\n      },\n      'imports.runner: batch flushed',\n    )\n\n    if (result.failures.length > 0) {\n      // Map each ES failure back to its source row via `id`. The bulk\n      // response order matches the request order, but ES doesn't promise\n      // that; matching by `id` is the safe path. PoC volume is small\n      // enough that the O(failures × batch) cost is irrelevant.\n      const byId = new Map(batch.map((b) => [b.id, b.rowNumber]))\n      for (const fail of result.failures) {\n        const rowNumber = byId.get(fail.id) ?? -1\n        errorTracker.addError(rowNumber, fail.type, fail.reason, undefined, { id: fail.id })\n      }\n    }\n\n    // Account non-fatal warnings ONLY for rows ES accepted (commerce-demo PR4) — a\n    // warned row is by contract a SUCCESSFUL row, so a row that warned AND then failed\n    // the bulk write must NOT land in rowsWarned/warnings (it's already in rowsFailed).\n    const failedIds = new Set(result.failures.map((f) => f.id))\n    for (const item of batch) {\n      if (!item.warnings || item.warnings.length === 0) continue\n      if (failedIds.has(item.id)) continue\n      rowsWarned += 1\n      for (const w of item.warnings) {\n        warningTracker.addError(item.rowNumber, w.warningType, w.message, w.field, item.rowData)\n      }\n    }\n\n    // Buffer directional pairs for rows ES accepted (commerce-demo PR6, SD007) — same\n    // accepted-rows-only rule as warnings, so a pair from a doc ES rejected is dropped.\n    for (const item of batch) {\n      if (!item.pairs || item.pairs.length === 0) continue\n      if (failedIds.has(item.id)) continue\n      collectedPairs.push(...item.pairs)\n    }\n\n    if (onProgress) {\n      const elapsedSeconds = (Date.now() - startedAt) / 1000\n      onProgress({\n        rowsRead,\n        rowsSucceeded,\n        rowsFailed,\n        rowsSkipped,\n        rowsWarned,\n        batchesCompleted,\n        elapsedSeconds,\n        rowsPerSecond: elapsedSeconds > 0 ? rowsRead / elapsedSeconds : 0,\n        distinctErrorPatterns: errorTracker.getDistinctPatternCount(),\n      })\n    }\n  }\n\n  // Streaming heartbeat — sample every Nth row so the operator sees the\n  // streamer is alive even when a transform-rejected file would prevent\n  // any flush from firing. Without this, a feed that's 100% rejected\n  // looks identical to \"stuck\" until end-of-file. Sample frequency is\n  // batchSize so the cadence matches the natural batch rhythm.\n  const streamSampleEvery = batchSize\n\n  // First-error capture per error type — surface what's actually rejecting\n  // the rows when 100% of a feed is rejected. Without this, the only signal\n  // is `errorTracker.snapshot()` which only writes to import_run.errorSummary\n  // at the END of the run. Mid-run we want to know NOW.\n  const firstErrorByType = new Map<string, { rowNumber: number; message: string }>()\n\n  for await (const { rowNumber, row } of readFeed(feed)) {\n    if (signal?.aborted) break\n    if (rowLimit !== undefined && rowsRead >= rowLimit) break\n    rowsRead += 1\n\n    if (rowsRead === 1) {\n      logger?.info(\n        { importRunId, rowNumber, sampleKeys: Object.keys(row) },\n        'imports.runner: STREAM_FIRST_ROW',\n      )\n    }\n    if (rowsRead % streamSampleEvery === 0) {\n      const elapsedSec = (Date.now() - startedAt) / 1000\n      // Telemetry only — no row content, no error sample text.\n      // PR #262 review (M-5): the prior `firstErrors:\n      // Object.fromEntries(firstErrorByType)` payload included the\n      // raw error message captured from the transform, which for the\n      // carmaker feed embeds supplier codes + price fragments. The\n      // detailed first-error breakdown is already persisted on\n      // `import_run.errorSummary` via `errorTracker.snapshot()` —\n      // operators inspect it there, not in the worker log.\n      logger?.info(\n        {\n          importRunId,\n          rowsRead,\n          rowsSucceeded,\n          rowsFailed,\n          rowsSkipped,\n          pending: pending.length,\n          elapsedSec: Math.round(elapsedSec * 10) / 10,\n          rowsPerSecond: elapsedSec > 0 ? Math.round(rowsRead / elapsedSec) : 0,\n          distinctFirstErrorTypes: firstErrorByType.size,\n        },\n        'imports.runner: STREAM_TICK',\n      )\n    }\n\n    const ctx: TransformContext = { importRunId, params, runLabel, rowNumber }\n    let result: Awaited<ReturnType<typeof transform>>\n    try {\n      result = await transform(row, ctx)\n    } catch (err) {\n      // A throw from the transform is a programmer error — surface it as\n      // a row-level error so the run can continue. (If the bug is\n      // catastrophic, the operator sees the same message repeated and\n      // can stop the run.)\n      const reason = err instanceof Error ? err.message : String(err)\n      errorTracker.addError(rowNumber, 'transform_threw', reason, undefined, row)\n      if (!firstErrorByType.has('transform_threw')) {\n        firstErrorByType.set('transform_threw', { rowNumber, message: reason })\n        // PR #262 review (M-5): drop the `row` payload from the\n        // log line — supplier feeds carry pricing + internal codes.\n        // The full row is on `import_run.errorSummary.samples` for\n        // post-mortem review by operators with view permission.\n        logger?.warn(\n          { importRunId, rowNumber, errorType: 'transform_threw', message: reason },\n          'imports.runner: FIRST_ERROR transform_threw',\n        )\n      }\n      rowsFailed += 1\n      continue\n    }\n\n    if (result.kind === 'skip') {\n      rowsSkipped += 1\n      // A `*S` core-fee row skips the saleable index but may carry a doc-patch\n      // (commerce-demo PR7) — buffer it for the phase-2 attach. Unlike pairs, a\n      // patch has no ES write of its own to gate on, so it's collected here\n      // unconditionally; its success is decided by the phase-2 update.\n      if (result.patches && result.patches.length > 0) {\n        collectedPatches.push(...result.patches)\n      }\n      continue\n    }\n    if (result.kind === 'error') {\n      errorTracker.addError(\n        rowNumber,\n        result.error.errorType,\n        result.error.message,\n        result.error.field,\n        row,\n      )\n      if (!firstErrorByType.has(result.error.errorType)) {\n        firstErrorByType.set(result.error.errorType, {\n          rowNumber,\n          message: result.error.message,\n        })\n        logger?.warn(\n          {\n            importRunId,\n            rowNumber,\n            errorType: result.error.errorType,\n            field: result.error.field,\n            message: result.error.message,\n          },\n          `imports.runner: FIRST_ERROR ${result.error.errorType}`,\n        )\n      }\n      rowsFailed += 1\n      continue\n    }\n\n    // Non-fatal warnings ride on a SUCCESS — the doc is still indexed\n    // (commerce-demo PR4: `no_purchase_price`). Carry them (+ the raw row for the\n    // sample) on the pending item; the flush aggregates them into the separate\n    // `warningTracker` AFTER the bulk response confirms the row was accepted, so a\n    // warned-then-ES-rejected row never inflates rowsWarned / the warning summary.\n    pending.push({\n      id: result.id,\n      doc: result.doc,\n      rowNumber,\n      rowData: row,\n      ...(result.warnings && result.warnings.length > 0 && { warnings: result.warnings }),\n      ...(result.pairs && result.pairs.length > 0 && { pairs: result.pairs }),\n    })\n    if (pending.length >= batchSize) {\n      await flush()\n    }\n  }\n\n  await flush()\n\n  // Phase 2 (commerce-demo PR7): attach the buffered `*S` core-charge patches onto\n  // their PARENT docs. Runs AFTER the whole stream so every parent that exists in\n  // this run is already indexed; ES `update` is realtime by `_id`, so the merge\n  // lands regardless of whether the parent streamed before or after its `*S` row\n  // (order-independent). Patches are sparse (only reman parts carry `*S`), but batch\n  // anyway for a giant catalogue. A patch whose parent isn't in the index comes back\n  // as a per-doc failure (`document_missing_exception`) → an `orphan_core_charge`\n  // warning, never a phantom doc, never a silent drop (D004). A cluster failure of\n  // the whole batch is logged as a warning too — the saleable docs already indexed;\n  // only the deferred core-charge attach didn't happen — so we never fail the run\n  // over it.\n  let patchesApplied = 0\n  let patchesOrphaned = 0\n  for (let i = 0; i < collectedPatches.length; i += batchSize) {\n    // Operator-driven cancel BETWEEN batches: THROW rather than break, so a run\n    // cancelled after phase 1 is not marked `succeeded` with core charges left\n    // unattached. Mirrors phase-1's flush, which also re-throws on abort. The queue\n    // handler catches this and marks the run failed.\n    if (signal?.aborted) {\n      throw new Error('imports: run aborted during core-charge attach (phase 2)')\n    }\n    const slice = collectedPatches.slice(i, i + batchSize)\n    let updateResult: BulkIndexResult\n    try {\n      updateResult = await bulkPartialUpdate<TDoc>(esClient, {\n        index: esIndex,\n        updates: slice.map(({ id, doc }) => ({ id, doc })),\n        ...(signal !== undefined && { signal }),\n      })\n    } catch (err) {\n      // An abort surfacing as an AbortError from the ES client is a cancel, NOT an\n      // attach failure — re-throw so the run fails (same as phase 1), never a silent\n      // \"succeeded with patches unapplied\".\n      const isAbort =\n        signal?.aborted ??\n        (err instanceof Error && (err.name === 'AbortError' || /abort/i.test(err.message)))\n      if (isAbort) throw err\n      // Whole-batch transport/cluster failure (ES down, timeout). The parents may\n      // well exist — the attach simply didn't happen — so this is an INFRA failure,\n      // NOT a data orphan: it gets the distinct `patch_attach_failed` bucket and is\n      // NOT counted in patchesOrphaned. Best-effort: the saleable docs already\n      // indexed, and a re-import re-attaches, so we warn + continue rather than fail.\n      const reason = err instanceof Error ? err.message : String(err)\n      for (const patch of slice) {\n        warningTracker.addError(-1, 'patch_attach_failed', reason, patch.orphanWarning?.field, {\n          target: patch.id,\n        })\n      }\n      continue\n    }\n    patchesApplied += updateResult.succeeded\n    if (updateResult.failures.length > 0) {\n      const byId = new Map(slice.map((p) => [p.id, p]))\n      for (const fail of updateResult.failures) {\n        const w = byId.get(fail.id)?.orphanWarning\n        // A missing target (`document_missing_exception` / 404) is a genuine data\n        // orphan — the referenced parent isn't in this import (D004: warn, never a\n        // phantom doc). A stable message (from the patch's orphanWarning) aggregates\n        // orphans; the missing parent id rides in the sample. Any OTHER per-doc\n        // failure (mapper error, etc.) is an infra/attach failure kept in a distinct\n        // bucket so it can't masquerade as a data orphan.\n        const isOrphan = fail.status === 404 || fail.type === 'document_missing_exception'\n        if (isOrphan) {\n          patchesOrphaned += 1\n          warningTracker.addError(-1, w?.warningType ?? 'orphan_patch', w?.message ?? fail.reason, w?.field, {\n            target: fail.id,\n          })\n        } else {\n          warningTracker.addError(-1, 'patch_attach_failed', fail.reason, w?.field, {\n            target: fail.id,\n          })\n        }\n      }\n    }\n  }\n\n  const totalSec = (Date.now() - startedAt) / 1000\n  logger?.info(\n    {\n      importRunId,\n      rowsRead,\n      rowsSucceeded,\n      rowsFailed,\n      rowsSkipped,\n      batchesCompleted,\n      patchesApplied,\n      patchesOrphaned,\n      totalSec: Math.round(totalSec * 10) / 10,\n      rowsPerSecond: totalSec > 0 ? Math.round(rowsRead / totalSec) : 0,\n    },\n    'imports.runner: finished',\n  )\n\n  return {\n    rowsRead,\n    rowsSucceeded,\n    rowsFailed,\n    rowsSkipped,\n    rowsWarned,\n    batchesCompleted,\n    errors: errorTracker.snapshot(),\n    warnings: warningTracker.snapshot(),\n    pairs: collectedPairs,\n    patchesApplied,\n    patchesOrphaned,\n  }\n}\n"],"mappings":"2LA+BA,SAAgB,EAAoB,EAAiC,CACnE,IAAM,EAAO,IAAI,IACjB,IAAK,IAAM,KAAK,EACV,OAAM,GACV,IAAI,EAAK,IAAI,CAAC,EACZ,MAAU,MAAM,qBAAqB,EAAE,+CAA+C,EAExF,EAAK,IAAI,CAAC,CAF8E,CAI5F,CAuBA,eAAsB,EACpB,EACA,EAAmC,CAAC,EACnB,CACjB,GAAM,CAAE,YAAY,IAAS,EACvB,EAAS,EAAiB,CAAQ,EACpC,EAAQ,EACR,EAAW,EACf,UAAW,IAAM,KAAS,EAAiC,CACzD,IAAK,IAAI,EAAI,EAAG,EAAI,EAAM,OAAQ,GAAK,EACjC,EAAM,KAAO,KAAe,GAAS,GAEvC,EAAM,OAAS,IAAG,EAAW,EAAM,EAAM,OAAS,IAAM,EAC9D,CAKA,OADI,IAAa,GAAK,IAAa,KAAM,GAAS,GAC3C,EAAY,KAAK,IAAI,EAAG,EAAQ,CAAC,EAAI,CAC9C,CA2EA,eAAuB,EAAW,EAA0D,CAC1F,GAAM,CACJ,WACA,YAAY,IACZ,YAAY,GACZ,UACA,mBAAmB,GACnB,QAAQ,IACN,EAWE,EAAsE,EACxE,MAAM,KAAK,CAAO,EAClB,EACG,GAAqB,CACpB,IAAM,EAAU,EAAO,IAAK,GAAM,EAAE,KAAK,CAAC,EAE1C,OADA,EAAoB,CAAO,EACpB,CACT,EACA,GAEA,EAAS,EAAiB,CAAQ,EAClC,EAAS,EAAO,KACpB,EAAM,CACJ,YACA,QAAS,EACT,IAAK,GACL,iBAAkB,GAElB,mBAAoB,EAMpB,OACF,CAAC,CACH,EAEI,EAAY,EAChB,GAAI,CACF,UAAW,IAAM,KAAU,EAExB,CACD,GAAa,EAOb,IAAM,EAA8B,CAAC,EACrC,GAAI,MAAM,QAAQ,CAAM,EAAG,CACzB,IAAM,EAAM,EACZ,IAAK,IAAI,EAAI,EAAG,EAAI,EAAI,OAAQ,GAAK,EACnC,EAAI,OAAO,CAAC,GAAK,EAAI,IAAM,EAE/B,MACE,IAAK,GAAM,CAAC,EAAG,KAAM,OAAO,QAAQ,CAAM,EACxC,EAAI,GAAK,GAAK,GAGlB,KAAM,CAAE,YAAW,KAAI,CACzB,CACF,QAAU,CAIR,EAAO,QAAQ,CACjB,CACF,CCjNA,MAAa,EAAyB,EACnC,OAAO,CACN,OAAQ,EAAE,KAAK,CAAC,MAAO,MAAO,MAAM,CAAC,CAAC,CAAC,SAAS,EAChD,UAAW,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,EAC7C,UAAW,EAAE,QAAQ,CAAC,CAAC,SAAS,EAChC,eAAgB,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,EACjD,MAAO,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,GAAG,CAAC,CAAC,SAAS,CAC7C,CAAC,CAAC,CACD,OAAO,EAiBV,SAAgB,EACd,EACqD,CACrD,IAAM,EACJ,GAAU,OAAO,GAAW,UAAY,CAAC,MAAM,QAAQ,CAAM,EACxD,EACD,CAAC,EAID,EAAO,EAAO,OAAS,IAAA,IAAa,EAAO,OAAS,KAAO,CAAC,EAAI,EAAO,KAC7E,OAAO,EAAuB,UAAU,CAAI,CAC9C,CAuBA,SAAS,EAAoB,EAAuC,CAElE,IAAM,EAAY,EAAE,YAAc,EAAE,SAAW,MAAQ,IAAM,KAC7D,MAAO,CACL,SAAU,EAAE,SACZ,YACA,GAAI,EAAE,YAAc,IAAA,IAAa,CAAE,UAAW,EAAE,SAAU,EAC1D,GAAI,EAAE,UAAY,IAAA,IAAa,CAAE,QAAS,EAAE,OAAQ,EACpD,GAAI,EAAE,mBAAqB,IAAA,IAAa,CAAE,iBAAkB,EAAE,gBAAiB,EAC/E,GAAI,EAAE,QAAU,IAAA,IAAa,CAAE,MAAO,EAAE,KAAM,CAChD,CACF,CAGA,eAAuB,EAAS,EAAwD,CACtF,GAAI,EAAQ,SAAW,OAAQ,CAC7B,GAAM,CAAE,cAAe,MAAM,OAAO,8BACpC,MAAO,EAAW,CAChB,SAAU,EAAQ,SAClB,GAAI,EAAQ,YAAc,IAAA,IAAa,CAAE,UAAW,EAAQ,SAAU,EACtE,GAAI,EAAQ,iBAAmB,IAAA,IAAa,CAAE,eAAgB,EAAQ,cAAe,EACrF,GAAI,EAAQ,QAAU,IAAA,IAAa,CAAE,MAAO,EAAQ,KAAM,CAC5D,CAAC,EACD,MACF,CACA,MAAO,EAAW,EAAoB,CAAO,CAAC,CAChD,CC7CA,MAAa,EAAuB,GACvB,EAAkC,EAY/C,IAAa,EAAb,KAA0B,CACxB,SAA4B,IAAI,IAChC,YACA,qBACA,kBAA4B,EAE5B,YAAY,EAA6B,CAAC,EAAG,CAC3C,KAAK,YAAc,EAAO,aAAA,GAC1B,KAAK,qBAAuB,EAAO,sBAAA,CACrC,CAWA,SACE,EACA,EACA,EACA,EACA,EACM,CAEN,IAAM,EAAY,GAAG,EAAU,GADd,GAAS,UACiB,GAAG,IAE1C,EAAU,KAAK,SAAS,IAAI,CAAS,EACzC,GAAI,CAAC,EAAS,CACZ,GAAI,KAAK,SAAS,MAAQ,KAAK,YAAa,CAC1C,KAAK,mBAAqB,EAC1B,MACF,CACA,EAAU,CACR,YACA,MAAO,GAAS,KAChB,UACA,MAAO,EACP,gBAAiB,EACjB,eAAgB,EAChB,QAAS,CAAC,CACZ,EACA,KAAK,SAAS,IAAI,EAAW,CAAO,CACtC,CAEA,EAAQ,OAAS,EACjB,EAAQ,eAAiB,EACrB,EAAQ,QAAQ,OAAS,KAAK,sBAChC,EAAQ,QAAQ,KAAK,CAAE,YAAW,SAAQ,CAAC,CAE/C,CAGA,oBAA6B,CAC3B,IAAI,EAAM,EACV,IAAK,IAAM,KAAK,KAAK,SAAS,OAAO,EAAG,GAAO,EAAE,MACjD,OAAO,CACT,CAGA,yBAAkC,CAChC,OAAO,KAAK,SAAS,IACvB,CAOA,0BAAmC,CACjC,OAAO,KAAK,iBACd,CAOA,gBAAiC,CAC/B,IAAM,EAAM,MAAM,KAAK,KAAK,SAAS,OAAO,CAAC,CAAC,CAAC,MAAM,EAAG,IAAM,EAAE,MAAQ,EAAE,KAAK,EACzE,EAAc,EAAI,QAAQ,EAAK,IAAM,EAAM,EAAE,MAAO,CAAC,EAC3D,OAAO,EAAI,IAAK,IAAO,CACrB,UAAW,EAAE,UACb,MAAO,EAAE,MACT,QAAS,EAAE,QACX,MAAO,EAAE,MACT,gBAAiB,EAAE,gBACnB,eAAgB,EAAE,eAClB,QAAS,EAAE,QAAQ,MAAM,EACzB,WAAY,EAAc,EAAK,EAAE,MAAQ,EAAe,IAAM,CAChE,EAAE,CACJ,CAOA,UAAiC,CAC/B,MAAO,CACL,YAAa,KAAK,mBAAmB,EACrC,iBAAkB,KAAK,wBAAwB,EAC/C,kBAAmB,KAAK,kBACxB,SAAU,KAAK,eAAe,CAChC,CACF,CACF,ECnJA,MAAa,EAAqB,IAqJlC,eAAsB,EAAgB,EAA2D,CAC/F,GAAM,CACJ,cACA,WACA,SACA,YACA,OACA,WACA,UACA,YAAY,EACZ,aACA,WACA,SACA,eAAe,IAAI,EACnB,iBAAiB,IAAI,EACrB,UACE,EAEJ,GAAI,EAAY,EACd,MAAU,MAAM,+BAA+B,EAAU,EAAE,EAG7D,IAAM,EAAY,KAAK,IAAI,EAC3B,GAAQ,KAAK,CAAE,cAAa,WAAU,YAAW,SAAQ,EAAG,yBAAyB,EACrF,IAAI,EAAW,EACX,EAAgB,EAChB,EAAa,EACb,EAAc,EACd,EAAa,EACb,EAAmB,EAMjB,EAAoC,CAAC,EAOrC,EAAqC,CAAC,EAOxC,EAOC,CAAC,EAEA,EAAQ,SAA2B,CACvC,GAAI,EAAQ,SAAW,EAAG,OAC1B,IAAM,EAAQ,EACd,EAAU,CAAC,EACX,IAAM,EAAiB,KAAK,IAAI,EAC5B,EACJ,GAAI,CACF,EAAS,MAAM,EAAiB,EAAU,CACxC,MAAO,EACP,KAAM,EAAM,KAAK,CAAE,KAAI,UAAW,CAAE,KAAI,KAAI,EAAE,EAC9C,GAAI,IAAW,IAAA,IAAa,CAAE,QAAO,CACvC,CAAC,CACH,OAAS,EAAK,CAQZ,IAAM,EAFJ,GAAQ,UACP,aAAe,QAAU,EAAI,OAAS,cAAgB,SAAS,KAAK,EAAI,OAAO,IACtD,UAAY,sBAClC,EAAS,aAAe,MAAQ,EAAI,QAAU,OAAO,CAAG,EAC9D,IAAK,GAAM,CAAE,eAAe,EAC1B,EAAa,SAAS,EAAW,EAAW,EAAQ,IAAA,GAAW,IAAI,EAGrE,KADA,IAAc,EAAM,OACd,CACR,CAEA,GAAiB,EAAO,UACxB,GAAc,EAAO,SAAS,OAC9B,GAAoB,EAEpB,IAAM,EAAU,KAAK,IAAI,EAAI,EACvB,GAAc,KAAK,IAAI,EAAI,GAAa,IAe9C,GAdA,GAAQ,KACN,CACE,cACA,MAAO,EACP,UACA,UAAW,EAAM,OACjB,UAAW,EAAO,UAClB,OAAQ,EAAO,SAAS,OACxB,WACA,cAAe,EAAa,EAAI,KAAK,MAAM,EAAW,CAAU,EAAI,CACtE,EACA,+BACF,EAEI,EAAO,SAAS,OAAS,EAAG,CAK9B,IAAM,EAAO,IAAI,IAAI,EAAM,IAAK,GAAM,CAAC,EAAE,GAAI,EAAE,SAAS,CAAC,CAAC,EAC1D,IAAK,IAAM,KAAQ,EAAO,SAAU,CAClC,IAAM,EAAY,EAAK,IAAI,EAAK,EAAE,GAAK,GACvC,EAAa,SAAS,EAAW,EAAK,KAAM,EAAK,OAAQ,IAAA,GAAW,CAAE,GAAI,EAAK,EAAG,CAAC,CACrF,CACF,CAKA,IAAM,EAAY,IAAI,IAAI,EAAO,SAAS,IAAK,GAAM,EAAE,EAAE,CAAC,EAC1D,IAAK,IAAM,KAAQ,EACb,MAAC,EAAK,UAAY,EAAK,SAAS,SAAW,IAC3C,GAAU,IAAI,EAAK,EAAE,EACzB,IAAc,EACd,IAAK,IAAM,KAAK,EAAK,SACnB,EAAe,SAAS,EAAK,UAAW,EAAE,YAAa,EAAE,QAAS,EAAE,MAAO,EAAK,OAAO,CAF3E,CAQhB,IAAK,IAAM,KAAQ,EACb,CAAC,EAAK,OAAS,EAAK,MAAM,SAAW,GACrC,EAAU,IAAI,EAAK,EAAE,GACzB,EAAe,KAAK,GAAG,EAAK,KAAK,EAGnC,GAAI,EAAY,CACd,IAAM,GAAkB,KAAK,IAAI,EAAI,GAAa,IAClD,EAAW,CACT,WACA,gBACA,aACA,cACA,aACA,mBACA,iBACA,cAAe,EAAiB,EAAI,EAAW,EAAiB,EAChE,sBAAuB,EAAa,wBAAwB,CAC9D,CAAC,CACH,CACF,EAOM,EAAoB,EAMpB,EAAmB,IAAI,IAE7B,UAAW,GAAM,CAAE,YAAW,SAAS,EAAS,CAAI,EAAG,CAErD,GADI,GAAQ,SACR,IAAa,IAAA,IAAa,GAAY,EAAU,MASpD,GARA,GAAY,EAER,IAAa,GACf,GAAQ,KACN,CAAE,cAAa,YAAW,WAAY,OAAO,KAAK,CAAG,CAAE,EACvD,kCACF,EAEE,EAAW,IAAsB,EAAG,CACtC,IAAM,GAAc,KAAK,IAAI,EAAI,GAAa,IAS9C,GAAQ,KACN,CACE,cACA,WACA,gBACA,aACA,cACA,QAAS,EAAQ,OACjB,WAAY,KAAK,MAAM,EAAa,EAAE,EAAI,GAC1C,cAAe,EAAa,EAAI,KAAK,MAAM,EAAW,CAAU,EAAI,EACpE,wBAAyB,EAAiB,IAC5C,EACA,6BACF,CACF,CAEA,IAAM,EAAwB,CAAE,cAAa,SAAQ,WAAU,WAAU,EACrE,EACJ,GAAI,CACF,EAAS,MAAM,EAAU,EAAK,CAAG,CACnC,OAAS,EAAK,CAKZ,IAAM,EAAS,aAAe,MAAQ,EAAI,QAAU,OAAO,CAAG,EAC9D,EAAa,SAAS,EAAW,kBAAmB,EAAQ,IAAA,GAAW,CAAG,EACrE,EAAiB,IAAI,iBAAiB,IACzC,EAAiB,IAAI,kBAAmB,CAAE,YAAW,QAAS,CAAO,CAAC,EAKtE,GAAQ,KACN,CAAE,cAAa,YAAW,UAAW,kBAAmB,QAAS,CAAO,EACxE,6CACF,GAEF,GAAc,EACd,QACF,CAEA,GAAI,EAAO,OAAS,OAAQ,CAC1B,GAAe,EAKX,EAAO,SAAW,EAAO,QAAQ,OAAS,GAC5C,EAAiB,KAAK,GAAG,EAAO,OAAO,EAEzC,QACF,CACA,GAAI,EAAO,OAAS,QAAS,CAC3B,EAAa,SACX,EACA,EAAO,MAAM,UACb,EAAO,MAAM,QACb,EAAO,MAAM,MACb,CACF,EACK,EAAiB,IAAI,EAAO,MAAM,SAAS,IAC9C,EAAiB,IAAI,EAAO,MAAM,UAAW,CAC3C,YACA,QAAS,EAAO,MAAM,OACxB,CAAC,EACD,GAAQ,KACN,CACE,cACA,YACA,UAAW,EAAO,MAAM,UACxB,MAAO,EAAO,MAAM,MACpB,QAAS,EAAO,MAAM,OACxB,EACA,+BAA+B,EAAO,MAAM,WAC9C,GAEF,GAAc,EACd,QACF,CAOA,EAAQ,KAAK,CACX,GAAI,EAAO,GACX,IAAK,EAAO,IACZ,YACA,QAAS,EACT,GAAI,EAAO,UAAY,EAAO,SAAS,OAAS,GAAK,CAAE,SAAU,EAAO,QAAS,EACjF,GAAI,EAAO,OAAS,EAAO,MAAM,OAAS,GAAK,CAAE,MAAO,EAAO,KAAM,CACvE,CAAC,EACG,EAAQ,QAAU,GACpB,MAAM,EAAM,CAEhB,CAEA,MAAM,EAAM,EAaZ,IAAI,EAAiB,EACjB,EAAkB,EACtB,IAAK,IAAI,EAAI,EAAG,EAAI,EAAiB,OAAQ,GAAK,EAAW,CAK3D,GAAI,GAAQ,QACV,MAAU,MAAM,0DAA0D,EAE5E,IAAM,EAAQ,EAAiB,MAAM,EAAG,EAAI,CAAS,EACjD,EACJ,GAAI,CACF,EAAe,MAAM,EAAwB,EAAU,CACrD,MAAO,EACP,QAAS,EAAM,KAAK,CAAE,KAAI,UAAW,CAAE,KAAI,KAAI,EAAE,EACjD,GAAI,IAAW,IAAA,IAAa,CAAE,QAAO,CACvC,CAAC,CACH,OAAS,EAAK,CAOZ,GAFE,GAAQ,UACP,aAAe,QAAU,EAAI,OAAS,cAAgB,SAAS,KAAK,EAAI,OAAO,IACrE,MAAM,EAMnB,IAAM,EAAS,aAAe,MAAQ,EAAI,QAAU,OAAO,CAAG,EAC9D,IAAK,IAAM,KAAS,EAClB,EAAe,SAAS,GAAI,sBAAuB,EAAQ,EAAM,eAAe,MAAO,CACrF,OAAQ,EAAM,EAChB,CAAC,EAEH,QACF,CAEA,GADA,GAAkB,EAAa,UAC3B,EAAa,SAAS,OAAS,EAAG,CACpC,IAAM,EAAO,IAAI,IAAI,EAAM,IAAK,GAAM,CAAC,EAAE,GAAI,CAAC,CAAC,CAAC,EAChD,IAAK,IAAM,KAAQ,EAAa,SAAU,CACxC,IAAM,EAAI,EAAK,IAAI,EAAK,EAAE,CAAC,EAAE,cAOZ,EAAK,SAAW,KAAO,EAAK,OAAS,8BAEpD,GAAmB,EACnB,EAAe,SAAS,GAAI,GAAG,aAAe,eAAgB,GAAG,SAAW,EAAK,OAAQ,GAAG,MAAO,CACjG,OAAQ,EAAK,EACf,CAAC,GAED,EAAe,SAAS,GAAI,sBAAuB,EAAK,OAAQ,GAAG,MAAO,CACxE,OAAQ,EAAK,EACf,CAAC,CAEL,CACF,CACF,CAEA,IAAM,GAAY,KAAK,IAAI,EAAI,GAAa,IAiB5C,OAhBA,GAAQ,KACN,CACE,cACA,WACA,gBACA,aACA,cACA,mBACA,iBACA,kBACA,SAAU,KAAK,MAAM,EAAW,EAAE,EAAI,GACtC,cAAe,EAAW,EAAI,KAAK,MAAM,EAAW,CAAQ,EAAI,CAClE,EACA,0BACF,EAEO,CACL,WACA,gBACA,aACA,cACA,aACA,mBACA,OAAQ,EAAa,SAAS,EAC9B,SAAU,EAAe,SAAS,EAClC,MAAO,EACP,iBACA,iBACF,CACF"}

package/dist/transform-CABBaYmF.d.mts.map

@@ -1 +0,0 @@
-{"version":3,"file":"transform-CABBaYmF.d.mts","names":[],"sources":["../src/transform.ts"],"mappings":";;AA0BA;;;;AAAyB;AAGzB;;;;;;;;;;AAQW;AAoBX;;;;AAII;AAgBJ;;;KAnDY,aAAA;;UAGK,gBAAA;EA6DW;EA3D1B,WAAA;EA8CwB;EA5CxB,MAAA,EAAQ,MAAM;EAgDd;EA9CA,QAAA;EA8Ca;EA5Cb,SAAA;AAAA;;AAqD0B;AAmC5B;;;;;;;;;;;;;;;UApEiB,eAAA;EAyEX;EAvEJ,IAAA;EAwEI;EAtEJ,EAAE;AAAA;;;;;;;;;AAyEgC;AAGpC;;;;UA5DiB,QAAA;EA8Df;EA5DA,EAAA;EA6DK;EA3DL,GAAA,EAAK,OAAA,CAAQ,IAAA;EAoEE;;;;;;;;EA3Df,aAAA,GAAgB,UAAA;AAAA;;;;;;;;;;;;;;;;;;;AAyES;AAoB3B;;;;;;;;;;;;;KA1DY,SAAA;EAEN,IAAA;EACA,EAAA;EACA,GAAA,EAAK,IAAA;EACL,QAAA,YAAoB,UAAA;EACpB,KAAA,YAAiB,eAAA;AAAA;EAEjB,IAAA;EAAc,MAAA;EAAgB,OAAA,YAAmB,QAAA,CAAS,IAAA;AAAA;EAC1D,IAAA;EAAe,KAAA,EAAO,QAAA;AAAA;;UAGX,QAAA;EACf,SAAA;EACA,OAAA;EACA,KAAA;AAAA;;;;;;;UASe,UAAA;EACf,WAAA;EACA,OAAA;EACA,KAAA;AAAA;;;AAwGuD;AAiCzD;;KAjIY,YAAA,UACV,MAAA,EAAQ,MAAA,kBACR,GAAA,EAAK,gBAAA,KACF,OAAA,CAAQ,SAAA,CAAU,IAAA;;;;;;;;;;;;;AAgIQ;;;;;;cA5GlB,iBAAA;EAAA,iBACM,OAAA;;;;;;;EAQjB,QAAA,MAAA,CAAe,IAAA,EAAM,aAAA,EAAe,SAAA,EAAW,YAAA,CAAa,IAAA;;;;;;;;;;;;EAoB5D,GAAA,MAAA,CAAU,IAAA,EAAM,aAAA,EAAe,SAAA,EAAW,YAAA,CAAa,IAAA;;EAKvD,GAAA,CAAI,IAAA,EAAM,aAAA,GAAgB,YAAA;;EAK1B,GAAA,CAAI,IAAA,EAAM,aAAA;;EAKV,IAAA,CAAA,GAAQ,aAAA;;EAKR,KAAA,CAAA;AAAA;;iBAwBc,oBAAA,CAAA,GAAwB,iBAAiB;;;;;;;;;;;;;;;;;;;;;;;;iBAiCzC,uBAAA,MAAA,CACd,IAAA,EAAM,aAAA,EACN,SAAA,EAAW,YAAA,CAAa,IAAA"}

package/dist/transform-DIzWppgJ.mjs

@@ -1,2 +0,0 @@
-import{t as e}from"./rolldown-runtime-DK3Fl9T5.mjs";import{c as t,u as n}from"./parts-schema-D0gMhNWj.mjs";import{n as r,r as i}from"./schema-registry-C_lQje-K.mjs";import{behavior as a,defineEntity as o,field as s,searchable as c}from"@murumets-ee/entity";import{z as l}from"zod";import{sql as u}from"drizzle-orm";var d=e({IMPORT_RUN_STATUSES:()=>f,ImportRun:()=>p});const f=[`pending`,`running`,`succeeded`,`failed`,`cancelled`],p=o({name:`import_run`,fields:{label:s.text({required:!0,maxLength:255}),status:s.select({options:f,default:`pending`,indexed:!0}),filePath:s.text({required:!0,maxLength:2048}),transformName:s.text({required:!0,maxLength:128,indexed:!0}),params:s.json(),totals:s.json(),errorSummary:s.json(),warningSummary:s.json(),startedAt:s.date(),finishedAt:s.date(),queueJobId:s.text({maxLength:64,indexed:!0})},behaviors:[a.auditable(),c({fields:[`label`,`transformName`,`queueJobId`,`status`,`filePath`],projection:e=>({id:e.id,label:e.label,description:`${e.transformName} · ${e.status}`})})],scope:`global`,admin:{group:`imports`,label:`Import runs`,labelSingular:`Import run`,icon:`upload`,hideFromMenu:!0}}),m={decimal:`.`,thousands:``},h=l.array(t).max(512),g=l.strictObject({columns:l.array(l.string().min(1).max(256)).min(1).max(256),rows:l.array(l.record(l.string().max(500))).max(5)});function _(e){if(e.headerSnapshot!==void 0&&e.headerSnapshot!==null){let t=g.safeParse(e.headerSnapshot);if(!t.success)throw Error(`Invalid headerSnapshot: ${t.error.issues[0]?.message??`bad shape`}`);let n=JSON.stringify(e.headerSnapshot).length;if(n>1e5)throw Error(`headerSnapshot too large (${n} > 100000 serialized chars) — trim sample rows/cells`)}let t=typeof e.targetSchema==`string`?e.targetSchema:``,a=r(t);if(!a)throw Error(`Unknown target schema "${t}" (registered: ${i().join(`, `)})`);let o=n.safeParse(e.numberFormat);if(!o.success)throw Error(`Invalid numberFormat: ${o.error.issues[0]?.message??`bad shape`}`);let s=h.safeParse(e.columnMap);if(!s.success)throw Error(`Invalid columnMap: ${s.error.issues[0]?.message??`bad shape`}`);let c=JSON.stringify(e.columnMap).length;if(c>1e5)throw Error(`columnMap too large (${c} > 100000 serialized chars)`);let l=new Set(a.fields.map(e=>e.key)),u=new Set,d=new Set;for(let e of s.data){if(!l.has(e.target))throw Error(`Mapping target "${e.target}" is not a field of schema "${a.name}" (allowed: ${[...l].join(`, `)})`);u.has(e.target)&&d.add(e.target),u.add(e.target)}if(d.size>0)throw Error(`Mapping for schema "${a.name}" maps a target more than once: ${[...d].join(`, `)}`);let f=new Set(s.data.map(e=>e.target)),p=a.fields.filter(e=>e.required&&!f.has(e.key)).map(e=>e.key);if(p.length>0)throw Error(`Mapping for schema "${a.name}" is missing required target(s): ${p.join(`, `)}`)}const v={name:`imports:import-mapping-validation`,hooks:{beforeCreate:async e=>(_({targetSchema:e.targetSchema,numberFormat:e.numberFormat??m,columnMap:e.columnMap??[],headerSnapshot:e.headerSnapshot}),e),beforeUpdate:async(e,t,n)=>{let r=await n?.loadCurrent?.()??{};return _({targetSchema:`targetSchema`in t?t.targetSchema:r.targetSchema,numberFormat:`numberFormat`in t?t.numberFormat:r.numberFormat??m,columnMap:`columnMap`in t?t.columnMap:r.columnMap??[],headerSnapshot:`headerSnapshot`in t?t.headerSnapshot:void 0}),t}}},y=[`parts`,`discount_table`],b=[`csv`,`tsv`,`xlsx`],x=o({name:`import_mapping`,scope:`global`,fields:{supplierId:s.reference({entity:`supplier`,required:!0,indexed:!0}),brandId:s.reference({entity:`brand`,indexed:!0}),targetSchema:s.select({options:y,required:!0,indexed:!0}),name:s.text({required:!0,maxLength:255}),format:s.select({options:b,required:!0}),delimiter:s.text({maxLength:4}),hasHeader:s.boolean({default:!0}),headerRowIndex:s.number({integer:!0,min:0,default:0}),sheet:s.text({maxLength:128}),numberFormat:s.json({default:{decimal:`.`,thousands:``}}),columnMap:s.json({default:[]}),headerSnapshot:s.json(),active:s.boolean({default:!0,indexed:!0})},behaviors:[a.auditable(),c({fields:[`name`,`targetSchema`,`format`],projection:e=>({id:e.id,label:e.name,description:`${e.targetSchema} · ${e.format}`})}),{name:`imports:import-mapping-unique`,constraints:(e,t)=>{if(t!==`main`)return[];let n=e.supplierId,r=e.brandId,i=e.targetSchema;return!n||!r||!i?[]:[{kind:`uniqueIndex`,name:`uq_import_mapping_supplier_brand_schema`,on:[n,r,i],where:u`${r} IS NOT NULL`},{kind:`uniqueIndex`,name:`uq_import_mapping_supplier_schema_default`,on:[n,i],where:u`${r} IS NULL`}]}},v],admin:{group:`imports`,label:`Import mappings`,labelSingular:`Import mapping`,icon:`table`,hideFromMenu:!0}});var S=class{entries=new Map;register(e,t){if(this.entries.has(e))throw Error(`Transform "${e}" is already registered — two plugins cannot register the same transform name.`);this.entries.set(e,t)}set(e,t){this.entries.set(e,t)}get(e){return this.entries.get(e)}has(e){return this.entries.has(e)}list(){return Array.from(this.entries.keys()).sort()}clear(){this.entries.clear()}};const C=Symbol.for(`@murumets-ee/imports:transforms`);function w(){let e=globalThis,t=e[C];return t||(t=new S,e[C]=t),t}function T(e,t){w().set(e,t)}export{y as a,v as c,d,b as i,f as l,w as n,x as o,T as r,_ as s,S as t,p as u};
-//# sourceMappingURL=transform-DIzWppgJ.mjs.map