fix(markdown): keep fenced code out of the document outline

pageindex/page_index_md.py

@@ -32,26 +32,34 @@ async def generate_summaries_for_structure_md(structure, summary_token_threshold
 def extract_nodes_from_markdown(markdown_content):
     header_pattern = r'^(#{1,6})\s+(.+)$'
     bold_heading_pattern = r'^\*\*(.+?)\*\*\s*$'
-    code_block_pattern = r'^```'
+    code_block_pattern = r'^ {0,3}(`{3,}|~{3,})(.*)$'
     node_list = []
 
     lines = markdown_content.split('\n')
-    in_code_block = False
+    fence = None
 
     for line_num, line in enumerate(lines, 1):
         stripped_line = line.strip()
 
-        # Check for code block delimiters (triple backticks)
-        if re.match(code_block_pattern, stripped_line):
-            in_code_block = not in_code_block
+        delimiter = re.match(code_block_pattern, line)
+        if fence is not None:
+            if (delimiter and delimiter.group(1)[0] == fence[0]
+                    and len(delimiter.group(1)) >= len(fence)
+                    and not delimiter.group(2).strip()):
+                fence = None
             continue
-
+        if delimiter:
+            marker, info = delimiter.groups()
+            if marker[0] != '`' or '`' not in info:
+                fence = marker
+                continue
+
         # Skip empty lines
         if not stripped_line:
             continue
 
         # Only look for headers when not inside a code block
-        if not in_code_block:
+        if fence is None:
             match = re.match(header_pattern, stripped_line)
             if match:
                 title = match.group(2).strip()

tests/test_page_index_md.py

@@ -4,6 +4,24 @@
 
 
 class ExtractNodesFromMarkdownTest(unittest.TestCase):
+    def test_fenced_examples_do_not_become_document_headings(self):
+        cases = [
+            "~~~python\n# example\n**also code**\n~~~",
+            "````markdown\n```python\n# example\n```\n````",
+            "```python\n~~~\n# example\n~~~\n```",
+            "```python\n``` not a closing fence\n# example\n```",
+        ]
+        for code in cases:
+            with self.subTest(code=code):
+                text = "# Before\n" + code + "\n## After"
+                nodes, lines = extract_nodes_from_markdown(text)
+                self.assertEqual([n["node_title"] for n in nodes], ["Before", "After"])
+                self.assertEqual(nodes[-1]["line_num"], len(lines))
+
+    def test_unclosed_tilde_fence_hides_remaining_headings(self):
+        nodes, _ = extract_nodes_from_markdown("# Before\n~~~\n## Example")
+        self.assertEqual([n["node_title"] for n in nodes], ["Before"])
+
     def test_skips_bold_heading_with_only_whitespace(self):
         nodes, _ = extract_nodes_from_markdown("**   **\n**Valid heading**")