From 14565b289af86145f7bbd56f42c9f2e85d5021a0 Mon Sep 17 00:00:00 2001
From: minion1227 <romantymkiv1999@gmail.com>
Date: Tue, 23 Jun 2026 22:16:16 -0700
Subject: [PATCH] Fix: docx parsing raises ValueError on 'Heading' styles
 (#16284)

---
 rag/nlp/__init__.py                           |  7 +-
 .../unit_test/rag/test_docx_question_level.py | 88 +++++++++++++++++++
 2 files changed, 94 insertions(+), 1 deletion(-)
 create mode 100644 test/unit_test/rag/test_docx_question_level.py

diff --git a/rag/nlp/__init__.py b/rag/nlp/__init__.py
index 9e7f332dbd..6e994ac4cc 100644
--- a/rag/nlp/__init__.py
+++ b/rag/nlp/__init__.py
@@ -1201,7 +1201,12 @@ def naive_merge_with_images(texts, images, chunk_token_num=128, delimiter="\n。
 def docx_question_level(p, bull=-1):
     txt = re.sub(r"\u3000", " ", p.text).strip()
     if hasattr(p.style, 'name') and p.style.name and p.style.name.startswith('Heading'):
-        return int(p.style.name.split(' ')[-1]), txt
+        # Heading styles are usually "Heading N", but the base "Heading" style,
+        # custom "Heading"-prefixed styles, or "HeadingN" (no space) have no
+        # space-separated trailing integer. Extract the level digits safely and
+        # fall back to the top heading level instead of raising ValueError (#16163).
+        m = re.search(r"\d+", p.style.name)
+        return (int(m.group()) if m else 1), txt
     else:
         if bull < 0:
             return 0, txt
diff --git a/test/unit_test/rag/test_docx_question_level.py b/test/unit_test/rag/test_docx_question_level.py
new file mode 100644
index 0000000000..7272d15ea2
--- /dev/null
+++ b/test/unit_test/rag/test_docx_question_level.py
@@ -0,0 +1,88 @@
+#
+#  Copyright 2025 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+
+import sys
+import types
+
+import pytest
+
+
+def _stub(name, **attrs):
+    mod = types.ModuleType(name)
+    for key, value in attrs.items():
+        setattr(mod, key, value)
+    sys.modules.setdefault(name, mod)
+    return mod
+
+
+# Stub heavy module-level imports so rag.nlp can be imported in isolation.
+_stub("common.token_utils", num_tokens_from_string=lambda *a, **k: 0)
+_stub("roman_numbers")
+_stub("word2number", w2n=types.SimpleNamespace())
+_stub("cn2an", cn2an=lambda *a, **k: 0)
+_pil = _stub("PIL")
+_pil.Image = _stub("PIL.Image")
+_stub("chardet")
+
+from rag.nlp import docx_question_level
+
+
+class _Style:
+    def __init__(self, name):
+        self.name = name
+
+
+class _Paragraph:
+    def __init__(self, style_name, text="Some title"):
+        self.style = _Style(style_name)
+        self.text = text
+
+
+@pytest.mark.p2
+@pytest.mark.parametrize(
+    "style_name, expected_level",
+    [
+        ("Heading 1", 1),
+        ("Heading 2", 2),
+        ("Heading 9", 9),
+        ("Heading 10", 10),
+        ("Heading1", 1),  # no space
+        ("Heading", 1),  # base style, no number -> top level
+        ("HeadingTitle", 1),  # custom prefix, no number -> top level
+        ("Heading Title", 1),  # custom prefix with space, no number -> top level
+    ],
+)
+def test_docx_question_level_heading_styles(style_name, expected_level):
+    level, text = docx_question_level(_Paragraph(style_name))
+    assert level == expected_level
+    assert text == "Some title"
+
+
+@pytest.mark.p2
+def test_docx_question_level_no_number_does_not_raise():
+    # Regression for #16163: a "Heading"-prefixed style without a parseable
+    # number used to raise ValueError: invalid literal for int().
+    for name in ("Heading", "HeadingTitle", "Heading Title"):
+        level, _ = docx_question_level(_Paragraph(name))
+        assert level == 1
+
+
+@pytest.mark.p2
+def test_docx_question_level_non_heading_default_bull():
+    # Non-heading paragraph with the default bull=-1 returns level 0 (body text).
+    level, text = docx_question_level(_Paragraph("Normal", text="just a body line"))
+    assert level == 0
+    assert text == "just a body line"