fix: control UnidentifiedImageError

private_gpt/components/readers/pptx2md/pptx2md_reader.py

@@ -11,8 +11,10 @@ from enum import Enum
 from pathlib import Path
 from typing import Any
 
+from PIL import UnidentifiedImageError
 from llama_index.core.ingestion import arun_transformations
 from llama_index.core.schema import BaseNode, Document
+from pptx.exc import PackageNotFoundError
 from pptx2md import convert  # ty:ignore[unresolved-import]
 
 from private_gpt.celery.notify import NotifyProtocol
@@ -177,11 +179,10 @@ class PPTX2MdReader(IngestionReader):
         output_md = temp_dir / "output.md"
         image_dir = temp_dir / "img"
 
-        convert(
+        convert_kwargs = dict(
             pptx_path=str(file_info.file_data.absolute()),
             output=str(output_md),
             image_dir=str(image_dir),
-            disable_image=False,
             disable_wmf=False,
             disable_notes=True,
             disable_escaping=True,
@@ -190,6 +191,24 @@ class PPTX2MdReader(IngestionReader):
             min_block_size=5,
         )
 
+        try:
+            convert(disable_image=False, **convert_kwargs)
+        except UnidentifiedImageError as e:
+            logger.warning(
+                "Image extraction failed for %s due to an unsupported/corrupt "
+                "embedded image (%s). Retrying conversion with images disabled.",
+                file_info.file_name,
+                e,
+            )
+            convert(disable_image=True, **convert_kwargs)
+        except PackageNotFoundError as e:
+            logger.error(
+                "File %s could not be opened as a valid PPTX package: %s",
+                file_info.file_name,
+                e,
+            )
+            raise
+
         with open(output_md, encoding="utf-8") as f:
             text = f.read()