Merge pull request #5064 from myhloli/dev

mineru/cli/common.py

@@ -23,7 +23,10 @@ from mineru.backend.vlm.vlm_analyze import aio_doc_analyze as aio_vlm_doc_analyz
 from mineru.backend.office.pptx_analyze import office_pptx_analyze
 from mineru.backend.office.xlsx_analyze import office_xlsx_analyze
 from mineru.backend.office.docx_analyze import office_docx_analyze
-from mineru.utils.pdfium_guard import rewrite_pdf_bytes_with_pdfium
+from mineru.utils.pdfium_guard import (
+    get_loadable_pdfium_page_indices,
+    rewrite_pdf_bytes_with_pdfium,
+)
 
 os.environ["TORCH_CUDNN_V8_API_DISABLED"] = "1"
 if os.getenv("MINERU_LMDEPLOY_DEVICE", "") == "maca":
@@ -190,12 +193,50 @@ def convert_pdf_bytes_to_bytes(pdf_bytes, start_page_id=0, end_page_id=None):
         )
         if rebuilt_pdf_bytes:
             return rebuilt_pdf_bytes
-        logger.warning("PDFium rewrite returned empty bytes, using original PDF bytes.")
+        logger.warning(
+            "PDFium rewrite returned empty bytes, trying to skip broken pages."
+        )
     except Exception as fallback_error:
         logger.warning(
             f"Error in converting PDF bytes with pdfium: {fallback_error}, "
+            "trying to skip broken pages."
+        )
+
+    try:
+        loadable_page_indices, broken_page_indices = get_loadable_pdfium_page_indices(
+            pdf_bytes,
+            start_page_id=start_page_id,
+            end_page_id=end_page_id,
+        )
+        if broken_page_indices:
+            skipped_pages = [page_index + 1 for page_index in broken_page_indices]
+            logger.warning(
+                f"Skipped broken PDF pages during PDFium rewrite: {skipped_pages}"
+            )
+        if not loadable_page_indices:
+            logger.warning(
+                "PDFium skip-broken-page rewrite found no loadable pages, "
+                "using original PDF bytes."
+            )
+            return pdf_bytes
+
+        rebuilt_pdf_bytes = rewrite_pdf_bytes_with_pdfium(
+            pdf_bytes,
+            start_page_id=start_page_id,
+            end_page_id=end_page_id,
+            page_indices=loadable_page_indices,
+        )
+        if rebuilt_pdf_bytes:
+            return rebuilt_pdf_bytes
+        logger.warning(
+            "PDFium skip-broken-page rewrite returned empty bytes, "
             "using original PDF bytes."
         )
+    except Exception as fallback_error:
+        logger.warning(
+            "Error in converting PDF bytes with skip-broken-page fallback: "
+            f"{fallback_error}, using original PDF bytes."
+        )
     return pdf_bytes
 
 

mineru/utils/pdfium_guard.py

@@ -62,6 +62,49 @@ def close_pdfium_objects_safely(*pdfium_objs, owner: str = "pdfium cleanup") ->
             logger.warning(f"Failed to close PDFium object during {owner}: {exc}")
 
 
+def get_loadable_pdfium_page_indices(
+    src_pdf_bytes: bytes,
+    start_page_id: int = 0,
+    end_page_id: int | None = None,
+) -> tuple[list[int], list[int]]:
+    """逐页探测 PDFium 可加载页面，返回可保留页和损坏页的 0-based 索引。"""
+    import pypdfium2 as pdfium
+
+    loadable_page_indices = []
+    broken_page_indices = []
+    pdf_doc = None
+
+    try:
+        with pdfium_guard():
+            pdf_doc = pdfium.PdfDocument(src_pdf_bytes)
+            total_page_count = len(pdf_doc)
+            if total_page_count == 0:
+                return [], []
+
+            normalized_start_page_id = max(0, start_page_id)
+            normalized_end_page_id = get_end_page_id(end_page_id, total_page_count)
+            if normalized_start_page_id > normalized_end_page_id:
+                return [], []
+
+            for page_index in range(
+                normalized_start_page_id,
+                normalized_end_page_id + 1,
+            ):
+                page = None
+                try:
+                    page = pdf_doc[page_index]
+                    page.get_size()
+                    loadable_page_indices.append(page_index)
+                except Exception:
+                    broken_page_indices.append(page_index)
+                finally:
+                    close_pdfium_child(page)
+    finally:
+        close_pdfium_document(pdf_doc)
+
+    return loadable_page_indices, broken_page_indices
+
+
 def rewrite_pdf_bytes_with_pdfium(
     src_pdf_bytes: bytes,
     start_page_id: int = 0,