@semiont/jobs 0.5.30 → 0.5.32

package/dist/worker-main.js

@@ -1,5 +1,5 @@
-import { replyChannelsFor, createTomlConfigLoader, didToAgent, baseUrl, STARTUP_FETCH_RETRY, retryWithBackoff, isTransientFetchError, isObject, isString, isNumber, busRequest, isArray, resourceId, getPrimaryMediaType, isGenerationJobParams, capabilitiesOf, assembleAnnotation, findClaimSpan, textExtractionOf, GENERATABLE_MEDIA_TYPES, locate, createFragmentSelector, deriveViews, estimateTokens, chunkText, reconcileSelector, getLocaleEnglishName } from '@semiont/core';
-import { archivistContentReads, extractPdfTextLayer, EXTRACTORS, calculateChecksum, withinByteBudget, MAX_PDF_BYTES } from '@semiont/content';
+import { replyChannelsFor, createTomlConfigLoader, didToAgent, baseUrl, STARTUP_FETCH_RETRY, retryWithBackoff, isTransientFetchError, isObject, isString, isNumber, busRequest, isArray, resourceId, getPrimaryMediaType, isGenerationJobParams, capabilitiesOf, assembleAnnotation, findClaimSpan, textSourceOf, yieldsGeometryOf, decodeRepresentation, GENERATABLE_MEDIA_TYPES, locate, createFragmentSelector, deriveViews, estimateTokens, chunkText, reconcileSelector, getLocaleEnglishName } from '@semiont/core';
+import { archivistContentReads, extractPdfTextLayer, withinByteBudget, MAX_PDF_BYTES } from '@semiont/content';
 import { withSpan, SpanKind, recordJobOutcome, recordAnchorOutcome, recordDetectionCall } from '@semiont/observability';
 import { createInferenceClient, StructuredReadError } from '@semiont/inference';
 import { execFileSync } from 'child_process';
@@ -11216,28 +11216,43 @@ async function processGenerationJob(inferenceClient, params, onProgress, logger2
 }
 
 // src/workers/detection/prepare-detection.ts
-async function prepareDetection(mediaType, content, resourceId, userId, generator, store) {
-  const extractor = EXTRACTORS[textExtractionOf(mediaType)];
-  if (!extractor) return { declined: "no-extractor" };
-  const { data } = await content.getBinary(resourceId);
-  const bytes = Buffer.from(data);
-  const extracted = await extractor.extract(bytes, mediaType, {
-    key: calculateChecksum(bytes),
-    store
-  });
-  if (extracted.kind === "declined") return extracted;
-  if (!extracted.text.trim()) return { declined: "empty" };
-  const items = extracted.items;
-  if (items && items.length > 0) {
-    const anchored = { text: extracted.text, items };
-    return {
-      text: extracted.text,
-      buildAnnotation: (motivation, match, body) => buildPdfAnnotation(anchored, resourceId, userId, generator, motivation, match, body)
-    };
+async function prepareDetection(mediaType, content, resourceId, userId, generator, consult) {
+  if (textSourceOf(mediaType) === "none") return { declined: "no-extractor" };
+  if (yieldsGeometryOf(mediaType)) {
+    const answer = await consult(resourceId);
+    switch (answer.kind) {
+      case "extracted": {
+        if (!answer.text.trim()) return { declined: "empty" };
+        const anchored = { text: answer.text, items: answer.items ?? [] };
+        return {
+          text: answer.text,
+          buildAnnotation: (motivation, match, body) => buildPdfAnnotation(anchored, resourceId, userId, generator, motivation, match, body)
+        };
+      }
+      // The Smelter's own decline (encrypted, corrupt) — passed through by name.
+      case "declined":
+        return { declined: answer.declined };
+      // Named absences: `not-yet` retries, the other two are terminal.
+      case "not-yet":
+        return { declined: "not-yet" };
+      case "no-map":
+        return { declined: "no-map" };
+      case "unknown":
+        return { declined: "unknown" };
+      default: {
+        const unhandled = answer;
+        throw new DeterministicJobError(
+          `Unhandled anchored-text answer for resource ${resourceId}: ${JSON.stringify(unhandled)}`
+        );
+      }
+    }
   }
+  const { data } = await content.getBinary(resourceId);
+  const text = decodeRepresentation(Buffer.from(data), mediaType);
+  if (!text.trim()) return { declined: "empty" };
   return {
-    text: extracted.text,
-    buildAnnotation: (motivation, match, body) => buildTextAnnotation(extracted.text, resourceId, userId, generator, motivation, match, body)
+    text,
+    buildAnnotation: (motivation, match, body) => buildTextAnnotation(text, resourceId, userId, generator, motivation, match, body)
   };
 }
 function referenceIdOf(job) {
@@ -11372,13 +11387,19 @@ async function handleJobInner(adapter, config, job, completedUnitsByJob, signal)
     const mediaType = getPrimaryMediaType(descriptor);
     const source = await withSpan(
       "detection:prepare",
-      () => prepareDetection(mediaType ?? "", config.contentReads, resourceId$1, userId, generator, config.anchoredTextStore),
+      () => prepareDetection(mediaType ?? "", config.contentReads, resourceId$1, userId, generator, (rid) => session.client.browse.resourceAnchoredText(rid)),
       { attrs: { "resource.id": resourceId$1, "media.type": mediaType ?? "unknown" } }
     );
     if ("declined" in source) {
+      if (source.declined === "not-yet") {
+        throw new Error(`Anchored text not yet derived for resource ${resourceId$1} \u2014 Smelter has not settled; retrying`);
+      }
       if (source.declined === "no-extractor") {
         throw new DeterministicJobError(`Cannot run ${jobType} on resource ${resourceId$1}: media type '${mediaType ?? "unknown"}' has no extractable text to analyze`);
       }
+      if (source.declined === "no-map" || source.declined === "unknown") {
+        throw new DeterministicJobError(`Cannot run ${jobType} on resource ${resourceId$1}: anchored-text consult returned '${source.declined}'`);
+      }
       await emitEvent(session, "job:complete", {
         ...lifecycleBase,
         result: {
@@ -11611,29 +11632,6 @@ async function handleJobInner(adapter, config, job, completedUnitsByJob, signal)
   }
 }
 
-// src/anchored-text-over-bus.ts
-function anchoredTextOverBus(client, logger2) {
-  return {
-    async read(key) {
-      try {
-        return await client.browse.anchoredTextByChecksum(key);
-      } catch (error) {
-        logger2?.debug("Anchored-text consult failed \u2014 treating as miss", {
-          key,
-          reason: error instanceof Error ? error.message : String(error)
-        });
-        return null;
-      }
-    },
-    async write(key) {
-      logger2?.debug("Anchored-text write skipped \u2014 the Smelter is the sole writer", { key });
-    },
-    async list() {
-      return [];
-    }
-  };
-}
-
 // src/worker-runtime.ts
 var import_rxjs2 = __toESM(require_cjs());
 function buildHealthPayload(workers) {
@@ -11676,7 +11674,12 @@ function startStallWatchdog(opts) {
 var WORKER_AWAITED_OPERATIONS = [
   "job:claim",
   "browse:resource-requested",
-  "browse:anchored-text-by-checksum-requested",
+  // Canonical geometry for a geometry-bearing detection: the consult behind
+  // `ConsultAnchoredText` (SMELTER-OWNS-OCR P2). Its omission broke every
+  // PDF detection job at the transport probe
+  // (.plans/WORKER-ANCHORED-TEXT-CHANNEL.md); the census below now fails the
+  // BUILD when this list and the declared awaits drift.
+  "browse:anchored-text-requested",
   // Durability acknowledgement for a unit's annotations (JOB-RESTART-SAFETY
   // P6). The worker AWAITS this one — a unit may not advance until its
   // annotations are in the event log — so its replies must be in the narrow
@@ -11788,11 +11791,10 @@ async function startAgentWorker(opts) {
     jobTypes: group.jobTypes,
     inferenceClient: group.client,
     generator,
-    // The extraction seam's cache, consulted over the bus and never written
-    // (ANCHORED-TEXT-TO-SMELTER D2): the Smelter owns this store, the
-    // Archivist answers the checksum-addressed read, and a worker that
-    // misses extracts locally and discards.
-    anchoredTextStore: anchoredTextOverBus(client, logger2),
+    // Byte reads for decode-path media only. A geometry-bearing type's text
+    // never comes from bytes here — it is CONSULTED from the Smelter's
+    // canonical anchored text over the bus (SMELTER-OWNS-OCR P2), and this
+    // worker cannot derive even by mistake (READ-VS-EXTRACT P2).
     contentReads: contentReads2,
     logger: logger2
   });