Tools
Dify 1.17.1 fixes five extraction bugs that were corrupting indexed documents without raising an error
CSV cells went through pandas type inference before indexing, so `00123` became `123.0` and empty cells became the literal string `nan` (#41922). Notion table cells with mixed formatting emitted one Markdown column per rich-text segment, corrupting rows, while empty cells were skipped and shifted rows left (#41993). Notion `rich_text` and `title` properties were truncated at the first formatting change, and `.xls` cells with embedded double quotes broke the row. The release explicitly says indexed text is not retroactively repaired and affected documents must be re-imported.
Source
↳ Follow the thread