langchain - 💡(How to fix) Fix ExperimentalMarkdownSyntaxTextSplitter silently drops all content after an unterminated code fence [1 pull requests]

Checked other resources

• This is a bug, not a usage question.
• I added a clear and descriptive title.
• I searched the LangChain issues and did not find a duplicate.
• This is reproducible on the latest langchain-text-splitters.

Package (Required)

• langchain-text-splitters

Description

ExperimentalMarkdownSyntaxTextSplitter._resolve_code_chunk returns "" when a markdown code fence is never closed, after consuming the rest of the input. As a result split_text silently discards the code block and every following section and its body — with no error or warning. This is easy to hit with truncated LLM output (cut at max_tokens mid code block), a truncated stream, or a malformed file, which RAG pipelines feed straight into the header splitter.

Example Code

from langchain_text_splitters.markdown import ExperimentalMarkdownSyntaxTextSplitter as S
md = "# Title\nintro\n\n```python\nx = 1\n\n## Important Section\ncritical content\n"
[d.page_content for d in S().split_text(md)]
# -> ['# Title\nintro\n\n']   (the code block AND "Important Section" are gone)

Error Message and Stack Trace (if applicable)

No error — the content is silently dropped.

Expected behavior

Content after the unterminated fence (the code block and the following ## Important Section) should be preserved, not silently discarded.

System Info

langchain-text-splitters (latest). A one-line fix plus a regression test is ready in #37964.