perf(docx): memoize python-docx style lookups

docling/backend/msword_backend.py

@@ -3,6 +3,7 @@
 
 from __future__ import annotations
 
+import functools
 import logging
 import re
 import warnings
@@ -882,6 +883,12 @@ def __init__(
             self._default_paragraph_style = self.docx_obj.styles.default(
                 WD_STYLE_TYPE.PARAGRAPH
             )
+            # paragraph.style and run.style go through get_style(), which scans
+            # the whole styles part on every call; styles do not change during
+            # conversion, so memoize it on the loaded part.
+            self.current_part.get_style = functools.cache(  # type: ignore[method-assign]
+                self.current_part.get_style
+            )
             # Build comment mappings after loading document
             self._extract_comment_ranges()
 

tests/test_backend_msword_styles.py

@@ -0,0 +1,51 @@
+# SPDX-FileCopyrightText: The Docling Contributors
+# SPDX-License-Identifier: MIT
+
+"""Tests for Word style resolution.
+
+Kept separate from ``test_backend_msword.py`` so that file stays under the
+repository's per-file line limit.
+"""
+
+from docling_core.types.doc import DoclingDocument
+from docx import Document
+
+from docling.backend.msword_backend import MsWordDocumentBackend
+from docling.datamodel.base_models import InputFormat
+from docling.datamodel.document import InputDocument
+
+
+def _convert_docx(document, docx_path) -> DoclingDocument:
+    document.save(str(docx_path))
+    in_doc = InputDocument(
+        path_or_stream=docx_path,
+        format=InputFormat.DOCX,
+        backend=MsWordDocumentBackend,
+        filename=docx_path.name,
+    )
+    return MsWordDocumentBackend(in_doc=in_doc, path_or_stream=docx_path).convert()
+
+
+def test_style_lookups_do_not_leak_between_documents(tmp_path):
+    """Memoized style lookups must be scoped to the document they came from.
+
+    Both paragraphs resolve to the default paragraph style through the same
+    ``get_style(None, PARAGRAPH)`` lookup, but only the first document makes
+    that style bold. Converting them one after the other in one process must
+    report the bold of each document's own style.
+    """
+    bold_doc = Document()
+    bold_doc.styles["Normal"].font.bold = True
+    bold_doc.add_paragraph("bold by style")
+    plain_doc = Document()
+    plain_doc.add_paragraph("plain")
+
+    bold_texts = _convert_docx(bold_doc, tmp_path / "bold.docx").texts
+    plain_texts = _convert_docx(plain_doc, tmp_path / "plain.docx").texts
+
+    assert [t.text for t in bold_texts] == ["bold by style"]
+    assert bold_texts[0].formatting is not None
+    assert bold_texts[0].formatting.bold is True
+    assert [t.text for t in plain_texts] == ["plain"]
+    assert plain_texts[0].formatting is not None
+    assert plain_texts[0].formatting.bold is False