Skip to content

Open-source Python library for AI browser agents, plus Browser Use Cloud: stealth browsers and a hosted agent API.

fix(dom): clamp final chunk char_end to len(content) for trailing newline

Diwak4r2bef1f0cacd2
diff --git a/browser_use/dom/markdown_extractor.py b/browser_use/dom/markdown_extractor.py
index 668ba7d99..fad209207 100644
--- a/browser_use/dom/markdown_extractor.py
+++ b/browser_use/dom/markdown_extractor.py
@@ -378,8 +378,9 @@ def _parse_atomic_blocks(content: str) -> list[_AtomicBlock]:
)
offset = para_end
- # Fix last block char_end: content may not end with \n
- if blocks and content and not content.endswith('\n'):
+ # Fix last block char_end: the parser counts +1 per line for the newline it
+ # split on, so content ending in \n (or a final blank line) overshoots len(content).
+ if blocks and content and blocks[-1].char_end > len(content):
blocks[-1] = _AtomicBlock(
block_type=blocks[-1].block_type,
lines=blocks[-1].lines,
diff --git a/tests/ci/test_markdown_chunking.py b/tests/ci/test_markdown_chunking.py
index b4616bd1e..a7ba7d386 100644
--- a/tests/ci/test_markdown_chunking.py
+++ b/tests/ci/test_markdown_chunking.py
@@ -43,6 +43,21 @@ class TestChunkMarkdownBasic:
# Last chunk ends at content length
assert chunks[-1].char_offset_end == len(content)
+ def test_last_chunk_offset_equals_len_for_trailing_newline(self):
+ """Content ending in a newline must report char_offset_end == len(content).
+
+ Regression: the parser adds +1 per line for the newline it split on, so a
+ trailing \\n overshot the final offset by one. That offset feeds next_start_char
+ for pagination, and chunk_markdown_by_structure returns [] for start_from_char
+ >= len(content), so newline-terminated markdown silently broke continuation.
+ """
+ for content in ['# Header\n\nParagraph one.\n\n# Header 2\n\nParagraph two.\n', 'Hello\n', 'Hello\n\nWorld\n']:
+ chunks = chunk_markdown_by_structure(content, max_chunk_chars=20)
+ assert chunks[-1].char_offset_end == len(content), (
+ f'Expected last chunk to end at {len(content)}, got {chunks[-1].char_offset_end} '
+ f'for content {content!r}'
+ )
+
class TestChunkMarkdownHeaders:
"""Header boundary splitting."""