Build your personal knowledge base with Trilium Notes
TypeScript electronelectron-appknowledge-baseknowledge-graphknowledge-managementknowledge-management-graphlocal-firstnote-managmentnote-takernote-takingnotebooknotesnotes-apppersonal-knowledge-basepersonal-wikiscriptableself-hostedself-hostingwiki
fix(ocr): report failed PDF pages instead of calling scanned PDFs unsupported
Elian Doranaba6d56a02d3
diff --git a/apps/client/src/translations/en-GB/translation.json b/apps/client/src/translations/en-GB/translation.jsonindex be6b17048c..6b19a8d557 100644--- a/apps/client/src/translations/en-GB/translation.json+++ b/apps/client/src/translations/en-GB/translation.json@@ -248,5 +248,8 @@},"resource_download": {"license": "{{license}} licence"+ },+ "ocr": {+ "no_text_recognized": "OCR found no text to recognise."}}diff --git a/apps/client/src/translations/en/translation.json b/apps/client/src/translations/en/translation.jsonindex ee12d98525..af53c0b578 100644--- a/apps/client/src/translations/en/translation.json+++ b/apps/client/src/translations/en/translation.json@@ -3739,7 +3739,9 @@"processing_complete": "OCR processing complete.","processing_failed": "Failed to start OCR processing","text_filtered_low_confidence": "OCR detected text with {{confidence}}% confidence, but it was discarded because your minimum threshold is {{threshold}}%.",- "image_based_pdf_not_supported": "This PDF appears to be image-based (scanned). Text extraction from image-based PDFs is not yet supported.",+ "no_text_recognized": "OCR found no text to recognize.",+ "pages_failed_one": "OCR failed on {{count}} page, so its text is missing. The log has the reason.",+ "pages_failed_other": "OCR failed on {{count}} pages, so their text is missing. The log has the reason.","open_media_settings": "Open Settings","view_extracted_text": "View extracted text (OCR)"},diff --git a/apps/client/src/widgets/dialogs/ocr_text.spec.tsx b/apps/client/src/widgets/dialogs/ocr_text.spec.tsxindex 91d791297d..744effd64f 100644--- a/apps/client/src/widgets/dialogs/ocr_text.spec.tsx+++ b/apps/client/src/widgets/dialogs/ocr_text.spec.tsx@@ -10,6 +10,8 @@ import { beforeEach, describe, expect, it, vi } from "vitest";import Component from "../../components/component";import { renderInto } from "../../test/render";+import server from "../../services/server";+import toast from "../../services/toast";import { ParentComponent } from "../react/react_utils";import OcrTextDialog from "./ocr_text";@@ -17,6 +19,9 @@ const mocks = vi.hoisted(() => ({ standalone: false }));vi.mock("../../services/i18n", () => ({ t: (key: string) => key }));vi.mock("../../components/app_context", () => ({ default: { triggerCommand: vi.fn() } }));+vi.mock("../../services/toast", () => ({+ default: { showMessage: vi.fn(), showPersistent: vi.fn() }+}));// Answer the dialog's own request and nothing else: the modal's shortcut hints go through the same// service, and handing them this payload instead of the list they expect throws inside their fetch.vi.mock("../../services/server", () => ({@@ -54,6 +59,17 @@ async function open() {});}+/** Clicks "Process now" and lets the request and the toast it leads to settle. */+async function processNow(result: object) {+ vi.mocked(server.post).mockResolvedValue({ success: true, result, minConfidence: 0 });+ const button = [ ...document.querySelectorAll(".ocr-text-modal .modal-footer button") ]+ .find((b) => b.textContent?.trim() === "ocr.process_now");+ expect(button).toBeDefined();+ await act(async () => {+ button?.dispatchEvent(new MouseEvent("click", { bubbles: true }));+ });+}+function footerButtons(): string[] {return [ ...document.querySelectorAll(".ocr-text-modal .modal-footer button") ].map((button) => button.textContent?.trim() ?? "");@@ -63,6 +79,7 @@ describe("OcrTextDialog", () => {beforeEach(() => {mocks.standalone = false;document.body.innerHTML = "";+ vi.clearAllMocks();});it("shows the text, and offers to extract it again where an engine can", async () => {@@ -82,4 +99,24 @@ describe("OcrTextDialog", () => {// Copying is not extraction, and stays.expect(footerButtons()).toContain("info.copy_to_clipboard");});++ it("says no text was recognized when a PDF yields none", async () => {+ await open();+ await processNow({ text: "", confidence: 0, extractedAt: "", processingType: "pdf" });++ expect(toast.showMessage).toHaveBeenCalledWith(+ "ocr.no_text_recognized", expect.any(Number), expect.any(String));+ expect(toast.showPersistent).not.toHaveBeenCalled();+ });++ it("warns about the pages OCR failed on", async () => {+ await open();+ await processNow({+ text: "page two", confidence: 0.8, extractedAt: "", processingType: "pdf", failedPageCount: 1+ });++ expect(toast.showPersistent).toHaveBeenCalledWith(+ expect.objectContaining({ message: "ocr.pages_failed" }));+ expect(toast.showMessage).not.toHaveBeenCalled();+ });});diff --git a/apps/client/src/widgets/dialogs/ocr_text.tsx b/apps/client/src/widgets/dialogs/ocr_text.tsxindex 44ee6afedb..7e97b414f6 100644--- a/apps/client/src/widgets/dialogs/ocr_text.tsx+++ b/apps/client/src/widgets/dialogs/ocr_text.tsx@@ -87,12 +87,11 @@ function TextRepresentationModal({ textUrl, processUrl, onHidden }: TextRepresenconst result = response.result;const minConfidence = response.minConfidence ?? 0;- // Check if this is an image-based PDF (no text extracted)- if (result && !result.text && result.processingType === 'pdf') {+ if (result?.failedPageCount) {toast.showPersistent({- id: `ocr-pdf-unsupported-${randomString(8)}`,- icon: "bx bx-info-circle",- message: t("ocr.image_based_pdf_not_supported"),+ id: `ocr-pages-failed-${randomString(8)}`,+ icon: "bx bx-error",+ message: t("ocr.pages_failed", { count: result.failedPageCount }),timeout: 15000});// Check if text was filtered due to low confidence@@ -117,6 +116,8 @@ function TextRepresentationModal({ textUrl, processUrl, onHidden }: TextRepresen}}]});+ } else if (result && !result.text) {+ toast.showMessage(t("ocr.no_text_recognized"), 5000, "bx bx-info-circle");} else {toast.showMessage(t("ocr.processing_complete"));}diff --git a/apps/server/src/services/ocr/ocr_service.ts b/apps/server/src/services/ocr/ocr_service.tsindex ef1efbe49a..1c54c817f3 100644--- a/apps/server/src/services/ocr/ocr_service.ts+++ b/apps/server/src/services/ocr/ocr_service.ts@@ -14,6 +14,8 @@ export interface OCRResult {extractedAt: string;language?: string;pageCount?: number;+ /** Pages whose OCR threw, for a document whose other pages still produced a result. */+ failedPageCount?: number;processingType?: string;}diff --git a/apps/server/src/services/ocr/processors/pdf_processor.spec.ts b/apps/server/src/services/ocr/processors/pdf_processor.spec.tsindex 9730fd2ee3..ba36fa268c 100644--- a/apps/server/src/services/ocr/processors/pdf_processor.spec.ts+++ b/apps/server/src/services/ocr/processors/pdf_processor.spec.ts@@ -224,21 +224,45 @@ describe("PDFProcessor", () => {expect(result.text).toBe("");expect(result.confidence).toBe(0);+ expect(result.failedPageCount).toBeUndefined();});- it("tolerates a render failure on a page without aborting the document", async () => {+ it("keeps the other pages and counts the failure when OCR fails on one page", async () => {const processor = new PDFProcessor();- mockExtractText.mockResolvedValue({ totalPages: 1, text: [""] });- mockRenderPage.mockRejectedValue(new Error("broken page"));+ mockExtractText.mockResolvedValue({ totalPages: 2, text: [ "", "" ] });+ mockRenderPage+ .mockRejectedValueOnce(new Error("broken page"))+ .mockImplementationOnce(async () => renderedPage());+ mockRecognize.mockResolvedValue({ text: "second page", confidence: 0.8 });const result = await processor.extractText(buffer, { language: "eng" });- expect(result.text).toBe("");- expect(result.confidence).toBe(0);- expect(mockRecognize).not.toHaveBeenCalled();+ expect(result.text).toBe("second page");+ expect(result.failedPageCount).toBe(1);expect(mockLog.error).toHaveBeenCalledWith(expect.stringContaining("PDF OCR failed for page 1"));});+ it("fails the document when OCR fails on every scanned page", async () => {+ const processor = new PDFProcessor();+ mockExtractText.mockResolvedValue({ totalPages: 2, text: [ "", "" ] });+ mockRenderPage.mockRejectedValue(new Error("Cannot find module 'pdfium.wasm'"));++ await expect(processor.extractText(buffer, { language: "eng" })).rejects.toThrow(+ "OCR failed on all 2 scanned pages: Error: Cannot find module 'pdfium.wasm'");+ expect(mockRecognize).not.toHaveBeenCalled();+ });++ it("keeps the text pages but counts the failures when every scanned page fails", async () => {+ const processor = new PDFProcessor();+ mockExtractText.mockResolvedValue({ totalPages: 2, text: [ densePage("text page"), "" ] });+ mockRenderPage.mockRejectedValue(new Error("broken page"));++ const result = await processor.extractText(buffer, { language: "eng" });++ expect(result.text).toBe(densePage("text page").trim());+ expect(result.failedPageCount).toBe(1);+ });+it("caps the number of OCR pages and logs the pages it skipped", async () => {const processor = new PDFProcessor();const totalPages = 55;diff --git a/apps/server/src/services/ocr/processors/pdf_processor.ts b/apps/server/src/services/ocr/processors/pdf_processor.tsindex 9464629587..372c5fbdf9 100644--- a/apps/server/src/services/ocr/processors/pdf_processor.ts+++ b/apps/server/src/services/ocr/processors/pdf_processor.ts@@ -59,6 +59,8 @@ export class PDFProcessor extends FileProcessor {let renderBytes: Uint8Array | null = null;let ocrPagesProcessed = 0;let ocrPagesSkipped = 0;+ let ocrPagesFailed = 0;+ let firstOcrError: unknown;for (let pageNum = 1; pageNum <= totalPages; pageNum++) {const embedded = (pageTexts[pageNum - 1] ?? "").trim();@@ -82,7 +84,14 @@ export class PDFProcessor extends FileProcessor {ocrPagesProcessed++;renderBytes ??= new Uint8Array(buffer);const ocr = await this.ocrPage(renderBytes, pageNum, language);- if (ocr.text.length > 0) {+ if ("error" in ocr) {+ ocrPagesFailed++;+ firstOcrError ??= ocr.error;+ if (embedded.length > 0) {+ pageResults.push(embedded);+ pageConfidences.push(EMBEDDED_TEXT_CONFIDENCE);+ }+ } else if (ocr.text.length > 0) {pageResults.push(ocr.text);pageConfidences.push(ocr.confidence);} else if (embedded.length > 0) {@@ -97,6 +106,9 @@ export class PDFProcessor extends FileProcessor {}const text = pageResults.join("\n\n").trim();+ if (ocrPagesFailed > 0 && ocrPagesFailed === ocrPagesProcessed && text.length === 0) {+ throw new Error(`OCR failed on all ${ocrPagesFailed} scanned pages: ${firstOcrError}`);+ }const confidence = text.length === 0 || pageConfidences.length === 0? 0: pageConfidences.reduce((sum, c) => sum + c, 0) / pageConfidences.length;@@ -106,7 +118,8 @@ export class PDFProcessor extends FileProcessor {confidence,extractedAt: new Date().toISOString(),language,- pageCount: totalPages+ pageCount: totalPages,+ ...(ocrPagesFailed > 0 && { failedPageCount: ocrPagesFailed })};}@@ -115,16 +128,17 @@ export class PDFProcessor extends FileProcessor {}/**- * OCR a single page by rasterizing it and recognizing the result. Failures on an individual page- * are logged and treated as "no text" so one bad page never aborts the whole document.+ * OCR a single page by rasterizing it and recognizing the result. A failure is logged and+ * returned rather than thrown, so one bad page never aborts the whole document.*/- private async ocrPage(pdf: Uint8Array, pageNum: number, language: string): Promise<{ text: string; confidence: number }> {+ private async ocrPage(pdf: Uint8Array, pageNum: number, language: string):+ Promise<{ text: string; confidence: number } | { error: unknown }> {try {const page = await renderer.renderPage(pdf, pageNum);return await recognizer.recognize(await toPngBuffer(page), language);} catch (error) {getLog().error(`PDF OCR failed for page ${pageNum}: ${error}`);- return { text: "", confidence: 0 };+ return { error };}}}diff --git a/packages/commons/src/lib/server_api.ts b/packages/commons/src/lib/server_api.tsindex 10537c573e..da914f1c09 100644--- a/packages/commons/src/lib/server_api.ts+++ b/packages/commons/src/lib/server_api.ts@@ -984,6 +984,8 @@ export interface OCRProcessResponse {extractedAt: string;language?: string;pageCount?: number;+ /** Pages whose OCR threw, for a document whose other pages still produced a result. */+ failedPageCount?: number;processingType?: string;};/** The minimum confidence threshold that was applied (0-1 scale). */