Skip to content

Build and run agents you can see, understand and trust.

fix(rag): normalize deferred PDF read errors (#2975)

SIcheng Han93b12e4bb3ae
diff --git a/src/agentscope/rag/_parser/_pdf.py b/src/agentscope/rag/_parser/_pdf.py
index 0a7d240c4..f3dbc87ec 100644
--- a/src/agentscope/rag/_parser/_pdf.py
+++ b/src/agentscope/rag/_parser/_pdf.py
@@ -74,19 +74,21 @@ class PDFParser(ParserBase):
try:
reader = PdfReader(io.BytesIO(file))
+ # pypdf reads page objects and content streams lazily, so read
+ # failures can also occur after the reader is constructed.
+ sections: list[Section] = []
+ for page_idx, page in enumerate(reader.pages, start=1):
+ text = page.extract_text() or ""
+ sections.append(
+ Section(
+ content=TextBlock(text=text),
+ source=filename,
+ metadata={"page": page_idx},
+ ),
+ )
except PdfReadError as e:
raise ValueError(
f"Failed to parse {filename!r} as PDF: {e}",
) from e
- sections: list[Section] = []
- for page_idx, page in enumerate(reader.pages, start=1):
- text = page.extract_text() or ""
- sections.append(
- Section(
- content=TextBlock(text=text),
- source=filename,
- metadata={"page": page_idx},
- ),
- )
return sections
diff --git a/tests/rag_parser_test.py b/tests/rag_parser_test.py
index 212d5ed6f..f8660e963 100644
--- a/tests/rag_parser_test.py
+++ b/tests/rag_parser_test.py
@@ -465,6 +465,74 @@ class PDFParserTest(IsolatedAsyncioTestCase):
with self.assertRaises(ValueError):
await parser.parse(b"not a pdf", "broken.pdf")
+ async def test_password_protected_pdf_raises_value_error(self) -> None:
+ """Errors deferred until page iteration retain filename context."""
+ from pypdf import PdfWriter
+ from pypdf.errors import FileNotDecryptedError
+
+ writer = PdfWriter()
+ writer.add_blank_page(width=72, height=72)
+ writer.encrypt("secret")
+ buffer = io.BytesIO()
+ writer.write(buffer)
+
+ with self.assertRaisesRegex(
+ ValueError,
+ r"Failed to parse 'locked\.pdf' as PDF:",
+ ) as context:
+ await PDFParser().parse(buffer.getvalue(), "locked.pdf")
+ self.assertIsInstance(
+ context.exception.__cause__,
+ FileNotDecryptedError,
+ )
+
+ async def test_empty_user_password_pdf_remains_readable(self) -> None:
+ """An encrypted PDF that opens without a password is still parsed."""
+ from pypdf import PdfWriter
+
+ writer = PdfWriter()
+ writer.add_blank_page(width=72, height=72)
+ writer.encrypt(user_password="", owner_password="owner")
+ buffer = io.BytesIO()
+ writer.write(buffer)
+
+ sections = await PDFParser().parse(buffer.getvalue(), "open.pdf")
+
+ self.assertEqual(
+ [section.model_dump() for section in sections],
+ [
+ {
+ "content": {
+ "type": "text",
+ "text": "",
+ "id": AnyString(),
+ "created_at": AnyString(),
+ "finished_at": None,
+ },
+ "source": "open.pdf",
+ "metadata": {"page": 1},
+ },
+ ],
+ )
+
+ async def test_text_extraction_read_error_raises_value_error(self) -> None:
+ """Read errors raised after page enumeration are wrapped as well."""
+ from unittest.mock import patch
+ from pypdf import PageObject
+ from pypdf.errors import PdfReadError
+
+ error = PdfReadError("broken content stream")
+ with patch.object(PageObject, "extract_text", side_effect=error):
+ with self.assertRaisesRegex(
+ ValueError,
+ r"Failed to parse 'broken-stream\.pdf' as PDF:",
+ ) as context:
+ await PDFParser().parse(
+ _make_pdf(["Hello"]),
+ "broken-stream.pdf",
+ )
+ self.assertIs(context.exception.__cause__, error)
+
async def test_supported_extensions(self) -> None:
"""``.pdf`` is the only extension exposed to the file picker."""
self.assertEqual(PDFParser.supported_extensions(), [".pdf"])