From 322cd37de10b01aa44a1ce4abe463788311cf512 Mon Sep 17 00:00:00 2001
From: Jiaquan Yi <2018298805@qq.com>
Date: Mon, 9 Mar 2026 10:49:42 +0800
Subject: [PATCH] =?UTF-8?q?fix:=20handle=20backslash=20path=20separators?=
=?UTF-8?q?=20in=20DOCX=20ZIP=20entries=20exported=20on=E2=80=A6(#33129)?=
=?UTF-8?q?=20(#33131)?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>
---
.../nodes/document_extractor/node.py | 37 +++++++++++-
.../nodes/test_document_extractor_node.py | 56 +++++++++++++++++++
2 files changed, 92 insertions(+), 1 deletion(-)
diff --git a/api/dify_graph/nodes/document_extractor/node.py b/api/dify_graph/nodes/document_extractor/node.py
index 5945e57926..e1ae6f0199 100644
--- a/api/dify_graph/nodes/document_extractor/node.py
+++ b/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
diff --git a/api/tests/unit_tests/core/workflow/nodes/test_document_extractor_node.py b/api/tests/unit_tests/core/workflow/nodes/test_document_extractor_node.py
index 5e20b1e12f..abdf0954c4 100644
--- a/api/tests/unit_tests/core/workflow/nodes/test_document_extractor_node.py
+++ b/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"")
+ zf.writestr(f"_rels{sep}.rels", b"")
+ zf.writestr(f"word{sep}document.xml", b"")
+ zf.writestr(f"word{sep}_rels{sep}document.xml.rels", b"")
+ 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