Python tool for converting files and office documents to Markdown.
fix(docx): ignore malformed styles missing type (#2190)
S Mf9b0dcb4232e
diff --git a/packages/markitdown/src/markitdown/converter_utils/docx/pre_process.py b/packages/markitdown/src/markitdown/converter_utils/docx/pre_process.pyindex d6fa8db6..94c49de7 100644--- a/packages/markitdown/src/markitdown/converter_utils/docx/pre_process.py+++ b/packages/markitdown/src/markitdown/converter_utils/docx/pre_process.py@@ -115,6 +115,29 @@ def _pre_process_math(content: bytes) -> bytes:return str(soup).encode()+def _pre_process_styles(content: bytes) -> bytes:+ """+ Repairs DOCX style definitions that Mammoth cannot read.++ Mammoth indexes ``w:type`` and ``w:styleId`` directly, so a ``w:style``+ element missing either attribute causes conversion to fail with a+ ``KeyError`` before any document text can be extracted.++ ``w:type`` is optional in OOXML: when it is absent the style type defaults+ to ``paragraph``, so the attribute is filled in rather than dropping the+ style, which would discard its formatting (a heading would be emitted as+ plain body text). A style with no ``w:styleId`` cannot be referenced by the+ document body, so it is removed.+ """+ soup = BeautifulSoup(content, features="xml")+ for tag in soup.find_all("w:style"):+ if not tag.has_attr("w:styleId"):+ tag.decompose()+ elif not tag.has_attr("w:type"):+ tag["w:type"] = "paragraph"+ return str(soup).encode()++def pre_process_docx(input_docx: BinaryIO) -> BinaryIO:"""Pre-processes a DOCX file with provided steps.@@ -131,11 +154,12 @@ def pre_process_docx(input_docx: BinaryIO) -> BinaryIO:"""output_docx = BytesIO()# The files that need to be pre-processed from .docx- pre_process_enable_files = [- "word/document.xml",- "word/footnotes.xml",- "word/endnotes.xml",- ]+ pre_process_enable_files = {+ "word/document.xml": _pre_process_math,+ "word/footnotes.xml": _pre_process_math,+ "word/endnotes.xml": _pre_process_math,+ "word/styles.xml": _pre_process_styles,+ }with zipfile.ZipFile(input_docx, mode="r") as zip_input:files = {name: zip_input.read(name) for name in zip_input.namelist()}with zipfile.ZipFile(output_docx, mode="w") as zip_output:@@ -144,7 +168,7 @@ def pre_process_docx(input_docx: BinaryIO) -> BinaryIO:if name in pre_process_enable_files:try:# Pre-process the content- updated_content = _pre_process_math(content)+ updated_content = pre_process_enable_files[name](content)# In the future, if there are more pre-processing steps, they can be added herezip_output.writestr(name, updated_content)except Exception:diff --git a/packages/markitdown/tests/test_module_vectors.py b/packages/markitdown/tests/test_module_vectors.pyindex d6d9d680..3f51550a 100644--- a/packages/markitdown/tests/test_module_vectors.py+++ b/packages/markitdown/tests/test_module_vectors.py@@ -3,6 +3,8 @@ import osimport timeimport pytestimport base64+import re+import zipfilefrom pathlib import Path@@ -199,6 +201,37 @@ def test_convert_stream_keep_data_uris(test_vector):assert string not in result.markdown+def test_convert_docx_with_style_missing_type(tmp_path):+ """DOCX conversion should not fail when a style entry is missing w:type."""+ source_path = os.path.join(TEST_FILES_DIR, "test.docx")+ malformed_path = tmp_path / "missing_style_type.docx"++ with zipfile.ZipFile(source_path, mode="r") as zip_input:+ with zipfile.ZipFile(malformed_path, mode="w") as zip_output:+ for item in zip_input.infolist():+ content = zip_input.read(item.filename)+ if item.filename == "word/styles.xml":+ styles_xml = content.decode("utf-8")+ styles_xml, count = re.subn(+ r'<w:style\s+w:type="[^"]+"(\s+w:styleId="1")',+ r"<w:style\1",+ styles_xml,+ )+ # Guard against a future fixture change silently+ # neutralizing this test.+ assert count == 1+ content = styles_xml.encode("utf-8")+ zip_output.writestr(item, content)++ result = MarkItDown().convert(str(malformed_path))++ assert (+ "AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation"+ in result.markdown+ )+ assert "# Abstract" in result.markdown++if __name__ == "__main__":"""Runs this file's tests from the command line."""