fix: handle backslash path separators in DOCX ZIP entries exported on…(#33129) (#33131)

Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>

api/dify_graph/nodes/document_extractor/node.py

@@ -4,6 +4,7 @@ import json
 import logging
 import os
 import tempfile
+import zipfile
 from collections.abc import Mapping, Sequence
 from typing import TYPE_CHECKING, Any
 
@@ -385,6 +386,32 @@ def parser_docx_part(block, doc: Document, content_items, i):
         content_items.append((i, "table", Table(block, doc)))
 
 
+def _normalize_docx_zip(file_content: bytes) -> bytes:
+    """
+    Some DOCX files (e.g. exported by Evernote on Windows) are malformed:
+    ZIP entry names use backslash (\\) as path separator instead of the forward
+    slash (/) required by both the ZIP spec and OOXML.  On Linux/Mac the entry
+    "word\\document.xml" is never found when python-docx looks for
+    "word/document.xml", which triggers a KeyError about a missing relationship.
+
+    This function rewrites the ZIP in-memory, normalizing all entry names to
+    use forward slashes without touching any actual document content.
+    """
+    try:
+        with zipfile.ZipFile(io.BytesIO(file_content), "r") as zin:
+            out_buf = io.BytesIO()
+            with zipfile.ZipFile(out_buf, "w", compression=zipfile.ZIP_DEFLATED) as zout:
+                for item in zin.infolist():
+                    data = zin.read(item.filename)
+                    # Normalize backslash path separators to forward slash
+                    item.filename = item.filename.replace("\\", "/")
+                    zout.writestr(item, data)
+            return out_buf.getvalue()
+    except zipfile.BadZipFile:
+        # Not a valid zip — return as-is and let python-docx report the real error
+        return file_content
+
+
 def _extract_text_from_docx(file_content: bytes) -> str:
     """
     Extract text from a DOCX file.
@@ -392,7 +419,15 @@ def _extract_text_from_docx(file_content: bytes) -> str:
     """
     try:
         doc_file = io.BytesIO(file_content)
-        doc = docx.Document(doc_file)
+        try:
+            doc = docx.Document(doc_file)
+        except Exception as e:
+            logger.warning("Failed to parse DOCX, attempting to normalize ZIP entry paths: %s", e)
+            # Some DOCX files exported by tools like Evernote on Windows use
+            # backslash path separators in ZIP entries and/or single-quoted XML
+            # attributes, both of which break python-docx on Linux. Normalize and retry.
+            file_content = _normalize_docx_zip(file_content)
+            doc = docx.Document(io.BytesIO(file_content))
         text = []
 
         # Keep track of paragraph and table positions

api/tests/unit_tests/core/workflow/nodes/test_document_extractor_node.py

@@ -16,6 +16,7 @@ from dify_graph.nodes.document_extractor.node import (
     _extract_text_from_excel,
     _extract_text_from_pdf,
     _extract_text_from_plain_text,
+    _normalize_docx_zip,
 )
 from dify_graph.variables import ArrayFileSegment
 from dify_graph.variables.segments import ArrayStringSegment
@@ -385,3 +386,58 @@ def test_extract_text_from_excel_numeric_type_column(mock_excel_file):
     expected_manual = "| 1.0 | 1.1 |\n| --- | --- |\n| Test | Test |\n\n"
 
     assert expected_manual == result
+
+
+def _make_docx_zip(use_backslash: bool) -> bytes:
+    """Helper to build a minimal in-memory DOCX zip.
+
+    When use_backslash=True the ZIP entry names use backslash separators
+    (as produced by Evernote on Windows), otherwise forward slashes are used.
+    """
+    import zipfile
+
+    sep = "\\" if use_backslash else "/"
+    buf = io.BytesIO()
+    with zipfile.ZipFile(buf, "w", compression=zipfile.ZIP_DEFLATED) as zf:
+        zf.writestr("[Content_Types].xml", b"<Types/>")
+        zf.writestr(f"_rels{sep}.rels", b"<Relationships/>")
+        zf.writestr(f"word{sep}document.xml", b"<w:document/>")
+        zf.writestr(f"word{sep}_rels{sep}document.xml.rels", b"<Relationships/>")
+    return buf.getvalue()
+
+
+def test_normalize_docx_zip_replaces_backslashes():
+    """ZIP entries with backslash separators must be rewritten to forward slashes."""
+    import zipfile
+
+    malformed = _make_docx_zip(use_backslash=True)
+    fixed = _normalize_docx_zip(malformed)
+
+    with zipfile.ZipFile(io.BytesIO(fixed)) as zf:
+        names = zf.namelist()
+
+    assert "word/document.xml" in names
+    assert "word/_rels/document.xml.rels" in names
+    # No entry should contain a backslash after normalization
+    assert all("\\" not in name for name in names)
+
+
+def test_normalize_docx_zip_leaves_forward_slash_unchanged():
+    """ZIP entries that already use forward slashes must not be modified."""
+    import zipfile
+
+    normal = _make_docx_zip(use_backslash=False)
+    fixed = _normalize_docx_zip(normal)
+
+    with zipfile.ZipFile(io.BytesIO(fixed)) as zf:
+        names = zf.namelist()
+
+    assert "word/document.xml" in names
+    assert "word/_rels/document.xml.rels" in names
+
+
+def test_normalize_docx_zip_returns_original_on_bad_zip():
+    """Non-zip bytes must be returned as-is without raising."""
+    garbage = b"not a zip file at all"
+    result = _normalize_docx_zip(garbage)
+    assert result == garbage