收口 MinerU OCR 任务链与上下文

2026-06-01 10:30:42 +08:00
parent 1481de2018
commit 610b23d3a5
13 changed files with 710 additions and 55 deletions
@@ -102,6 +102,17 @@ async function saveScreenshot(page, name) {
  return target;
 }

+function cleanupRoot(root) {
+  for (let attempt = 0; attempt < 5; attempt += 1) {
+    try {
+      fs.rmSync(root, { recursive: true, force: true, maxRetries: 3, retryDelay: 100 });
+      return;
+    } catch (error) {
+      if (attempt === 4) throw error;
+    }
+  }
+}
+
 async function main() {
  fs.mkdirSync(OUT_DIR, { recursive: true });
  const actorId = "mnote-e2e";
@@ -109,7 +120,9 @@ async function main() {
  const root = fs.mkdtempSync(path.join(os.tmpdir(), "mnote-task520-raw-target-"));
  const rootUri = fileUrl(root);
  const pagePath = "Page.md";
-  const rawPath = "Page/notes.txt";
+  const expectOcrContext = process.env.TASK520_OCR_CONTEXT === "1";
+  const rawPath = expectOcrContext ? "Page/photo.png" : "Page/notes.txt";
+  const ocrPath = "Page.ocr/photo.png.ocr.md";
  const documentId = localMdDocumentId(pagePath);
  const captured = [];
  let caughtError = null;
@@ -117,7 +130,41 @@ async function main() {
  fs.mkdirSync(path.join(root, "Page"), { recursive: true });
  writeWorkspaceManifest(root, actorId, workspaceId);
  fs.writeFileSync(path.join(root, pagePath), "# Page\n\nTask520 page\n", "utf8");
-  fs.writeFileSync(path.join(root, rawPath), "Task520 raw resource target\n", "utf8");
+  if (expectOcrContext) {
+    fs.writeFileSync(path.join(root, rawPath), Buffer.from([0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a]));
+    fs.mkdirSync(path.join(root, "Page.ocr"), { recursive: true });
+    fs.writeFileSync(
+      path.join(root, ocrPath),
+      "---\nmnote_ocr_version: 1\nprovider: mock\nmodel_version: vlm\nowner_document: ../Page.md\nsource_path: ./Page/photo.png\nsource_root_relative_path: Page/photo.png\nsource_size: 6\nsource_mtime_ms: 1\nstatus: done\ncreated_at: 1\nupdated_at: 1\n---\n\nTask520 OCR sidecar context text\n",
+      "utf8",
+    );
+    fs.writeFileSync(
+      path.join(root, ".mnote", "ocr-index.json"),
+      `${JSON.stringify({
+        version: 1,
+        entries: {
+          [rawPath]: {
+            jobId: "ocr_task520",
+            ownerDocumentId: documentId,
+            ownerDocumentPath: pagePath,
+            sourceRootRelativePath: rawPath,
+            ocrRootRelativePath: ocrPath,
+            provider: "mock",
+            modelVersion: "vlm",
+            status: "done",
+            sourceSize: 6,
+            sourceMtimeMs: 1,
+            createdAtMs: 1,
+            updatedAtMs: 1,
+            plainTextPreview: "Task520 OCR sidecar context text",
+          },
+        },
+      }, null, 2)}\n`,
+      "utf8",
+    );
+  } else {
+    fs.writeFileSync(path.join(root, rawPath), "Task520 raw resource target\n", "utf8");
+  }

  const browser = await chromium.launch({
    headless: process.env.HEADFUL !== "1",
@@ -243,7 +290,8 @@ async function main() {
    const targetChip = page.locator("[data-page-ai-target-chip]");
    await targetChip.waitFor({ state: "visible", timeout: UI_TIMEOUT_MS });
    const chipText = (await targetChip.innerText({ timeout: UI_TIMEOUT_MS })).trim();
-    assert(chipText.includes("notes.txt"), `raw resource 打开后 target chip 应指向 notes.txt: ${chipText}`);
+    const expectedName = expectOcrContext ? "photo.png" : "notes.txt";
+    assert(chipText.includes(expectedName), `raw resource 打开后 target chip 应指向 ${expectedName}: ${chipText}`);

    await page.locator("[data-page-ai-input]").fill("Task520 raw target", { timeout: UI_TIMEOUT_MS });
    await page.locator('[data-page-ai-action="send"]').click({ timeout: UI_TIMEOUT_MS });
@@ -268,19 +316,26 @@ async function main() {
    assert.strictEqual(runBody.targetPackage?.currentFile?.relativePath, rawPath, `targetPackage currentFile 应指向 raw resource: ${JSON.stringify(runBody.targetPackage)}`);
    assert.strictEqual(runBody.targetPackage?.currentFile?.objectIdentity, "resource:file:" + rootUri + ":" + rawPath, `currentFile objectIdentity 不应退化为 [object Object]: ${JSON.stringify(runBody.targetPackage?.currentFile)}`);
    assert(runBody.targetPackage?.allowedFiles?.includes(rawPath), `allowedFiles 应只包含 raw resource: ${JSON.stringify(runBody.targetPackage?.allowedFiles)}`);
+    if (expectOcrContext) {
+      assert.strictEqual(activeEditorRef.ocrContext?.source, "local_ocr_sidecar", `active_editor 应携带 OCR sidecar context: ${JSON.stringify(activeEditorRef)}`);
+      assert.strictEqual(activeEditorRef.ocrContext?.ocrRootRelativePath, ocrPath, `active_editor OCR path 不正确: ${JSON.stringify(activeEditorRef.ocrContext)}`);
+      assert(activeEditorRef.ocrContext?.plainTextPreview?.includes("Task520 OCR sidecar context text"), `active_editor OCR preview 缺失: ${JSON.stringify(activeEditorRef.ocrContext)}`);
+      assert.strictEqual(runBody.targetPackage?.ocrContext?.ocrRootRelativePath, ocrPath, `targetPackage 应携带 OCR sidecar context: ${JSON.stringify(runBody.targetPackage)}`);
+      assert.strictEqual(runBody.targetPackage?.currentFile?.ocrRootRelativePath, ocrPath, `currentFile 应携带 OCR sidecar path: ${JSON.stringify(runBody.targetPackage?.currentFile)}`);
+    }

    const screenshot = await saveScreenshot(page, "01-raw-resource-target");
-    const result = { ok: true, task: TASK, baseUrl: BASE_URL, root, rootUri, documentId, rawPath, screenshot, captured };
+    const result = { ok: true, task: TASK, baseUrl: BASE_URL, root, rootUri, documentId, rawPath, ocrPath: expectOcrContext ? ocrPath : "", screenshot, captured };
    fs.writeFileSync(RESULT_PATH, `${JSON.stringify(result, null, 2)}\n`, "utf8");
    console.log(JSON.stringify(result, null, 2));
  } catch (error) {
    caughtError = error;
    await saveScreenshot(page, "failure").catch(() => undefined);
  } finally {
-    fs.rmSync(root, { recursive: true, force: true });
    await page.close().catch(() => undefined);
    await context.close().catch(() => undefined);
    await browser.close().catch(() => undefined);
+    cleanupRoot(root);
  }

  if (caughtError) {