feat(ccrd): open OCR originals in the HF Search Reader

app.py

@@ -16,7 +16,7 @@ from functools import lru_cache
 from contextlib import asynccontextmanager, contextmanager
 from pathlib import Path
 from typing import Literal, Optional
-from urllib.parse import parse_qs, quote, unquote
+from urllib.parse import parse_qs, quote, unquote, urlencode
 from fastapi import FastAPI, Query, Request
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.middleware.gzip import GZipMiddleware
@@ -27,6 +27,10 @@ from pydantic import BaseModel, Field
 import ocr_release
 BASE_DIR = Path(__file__).resolve().parent
 DATA_PATH = BASE_DIR / "data/search_data.json.gz"
+# Existing Hugging Face Search Reader; OCR previews deep-link to the original
+# document there instead of exposing the generated plain-text file.
+HF_SEARCH_READER = os.environ.get("CCRD_HF_SEARCH_READER",
+                                  "https://voiceofml-search.hf.space/static/reader.html")
 FOLDER_TREE_PATH = BASE_DIR / "data/folder_tree.json.gz"
 FOLDER_BROWSER_PATH = BASE_DIR / "data/folder_browser.json.gz"
 FULLTEXT_DIR = BASE_DIR / "data/fulltext"
@@ -103,6 +107,7 @@ FULLTEXT_DISK_SOURCES = frozenset(
 FULLTEXT_REFRESH_INTERVAL_SECONDS = max(1, int(os.environ.get("CCRD_FULLTEXT_REFRESH_INTERVAL_SECONDS", "5")))
 fulltext_refresh_stop = threading.Event()
 active_ocr_release = None
+ocr_reader_urls: dict[str, str] = {}
 
 def clear_metadata_match_cache() -> None:
     global metadata_match_cache_generation
@@ -1053,9 +1058,28 @@ def build_indexes() -> None:
         source_size_order[public_source_slug(records[idx])].append(idx)
     clear_metadata_match_cache()
 
+def ocr_reader_url(export: dict) -> str:
+    """Deep link to the existing HF Search Reader for an OCR origin document.
+
+    The Reader accepts only public ``VoiceOfML/<repo>`` dataset ``resolve/main``
+    sources; the commit-pinned revision remains provenance, not a Reader URL.
+    """
+    repo = export.get("origin_repo") or ""
+    path = export.get("origin_rel_path") or ""
+    if not repo.startswith("VoiceOfML/") or not path:
+        return ""
+    source = f"https://huggingface.co/datasets/{repo}/resolve/main/" + quote(path, safe="/")
+    filename = path.rsplit("/", 1)[-1]
+    extension = filename.rsplit(".", 1)[-1].lower() if "." in filename else ""
+    query = {"url": source}
+    if extension:
+        query["ext"] = extension
+    return f"{HF_SEARCH_READER}?{urlencode(query)}"
+
+
 def load_data() -> None:
     global records, record_map, record_map_index, sources, source_counts, folder_tree_data, folder_browser_data, active_ocr_release
-    global source_physical_map, physical_source_counts, ocr_public_browser
+    global source_physical_map, physical_source_counts, ocr_public_browser, ocr_reader_urls
     start = time.time()
     payload = load_json_gz(DATA_PATH)
     extra = None
@@ -1102,6 +1126,12 @@ def load_data() -> None:
         folder_browser_data.update(layout["browsers"])
         ocr_public_browser = layout["browsers"]
     active_ocr_release = extra
+    ocr_reader_urls = {}
+    if extra:
+        for export in extra.get("exports", []):
+            url = ocr_reader_url(export)
+            if url:
+                ocr_reader_urls[export["doc_id"]] = url
     build_indexes()
     print(f"loaded {len(records)} txt records in {time.time() - start:.2f}s")
 
@@ -1967,6 +1997,9 @@ def api_preview(doc_id: str):
     if rec.get("source") == "OCR":
         payload.update(text_quality=rec.get("text_quality", "unreviewed"),
                        text_url="/txt/OCR/" + quote(rec["display_rel_path"], safe="/"))
+        reader_url = ocr_reader_urls.get(doc_id)
+        if reader_url:
+            payload["reader_url"] = reader_url
     return JSONResponse(payload)
 @app.get("/api/snippet/{doc_id}")
 

static/app.js

@@ -1329,16 +1329,15 @@ async function openPreview(docId, keyword, options = {}) {
     } else if (keyword && rawText.length >= 500000) {
       showToast("文本较大，已跳过预览高亮");
     }
-    const textQuality = data.text_quality === "partially-reviewed" ? "部分校订" : data.text_quality === "unreviewed" ? "未校订 OCR" : "";
-    const plainTextLink = typeof data.text_url === "string" && data.text_url.startsWith("/txt/OCR/")
-      ? `<a href="${escapeHTML(data.text_url)}" target="_blank" rel="noopener">打开纯 TXT</a>` : "";
+    const readerLink = typeof data.reader_url === "string" && data.reader_url.startsWith("https://")
+      ? `<a href="${escapeHTML(data.reader_url)}" target="_blank" rel="noopener noreferrer">打开原文件</a>` : "";
     window.CCRDSpeech?.previewClosed();
     DOM.previewPanel.innerHTML = `
       <div class="preview-header">
         <div class="preview-heading">
           <div class="preview-title">${escapeHTML(data.title)}</div>
           <div class="preview-path">${escapeHTML(data.source)} / ${escapeHTML(data.path)}</div>
-          ${textQuality || plainTextLink ? `<div class="preview-path">${escapeHTML(textQuality)}${plainTextLink ? " · " + plainTextLink : ""}</div>` : ""}
+          ${readerLink ? `<div class="preview-path">${readerLink}</div>` : ""}
         </div>
         <div class="preview-tools">
           ${previewSpeechButtonHTML()}

tests/test_ocr_release.py

@@ -7,6 +7,7 @@ from pathlib import Path
 import tempfile
 import unittest
 from unittest.mock import patch
+from urllib.parse import quote, urlencode
 
 import app
 import build_fulltext_db as builder
@@ -278,6 +279,12 @@ class OcrReleaseTests(unittest.TestCase):
             self.assertEqual(client.get("/txt/OCR/missing.txt").status_code, 404)
             self.assertEqual(client.get("/api/preview/OCR:1").json()["text"], self.content)
             self.assertEqual(client.get("/api/preview/OCR:1").json()["text_quality"], "partially-reviewed")
+            preview = client.get("/api/preview/OCR:1").json()
+            expected_source = ("https://huggingface.co/datasets/VoiceOfML/Test/resolve/main/"
+                               + quote("原目录/book.pdf", safe="/"))
+            self.assertEqual(preview["reader_url"],
+                             "https://voiceofml-search.hf.space/static/reader.html?"
+                             + urlencode({"url": expected_source, "ext": "pdf"}))
             self.assertEqual(client.get("/api/preview/OCR:1").json()["path"], "Test/原目录/book.txt")
             self.assertEqual(client.get("/api/preview/OCR:1").json()["source"], "Test")
             self.assertEqual([item["slug"] for item in client.get("/api/sources").json()], ["CCRD", "Test"])