Dify 1.17.1 fixes five bugs that corrupted indexed documents silently

Dify's 1.17.1 release fixes five bugs in how the platform reads documents before indexing them, and none of the five ever threw an error. They just changed the data.

CSV imports ran cells through pandas type inference first. A code or ID field like 00123 became 123.0. Empty cells became the literal string "nan", indexed and searchable as if a user had typed it. That's issue #41922.

Notion imports had their own failure. When a table cell mixed formatting, styled and plain text in the same cell, Dify emitted one Markdown column per formatting segment instead of one column per actual cell. Empty cells got skipped rather than preserved, which shifted every following cell left by one column. Rows didn't just lose data, they lost alignment (#41993).

Two more bugs hit rich_text and title properties in Notion, cut off at the first formatting change with everything after it lost. A fifth broke .xls rows whenever a cell contained an embedded double quote.

The release notes say indexed text is not retroactively repaired. Upgrading to 1.17.1 stops new corruption. It does nothing for documents already sitting in your vector store with truncated titles or shifted rows. Fixing those means re-importing every affected file by hand, and the notes don't say how to identify which ones are affected beyond knowing you used CSV, Notion, or .xls sources before this release.

Running Dify against Notion or spreadsheet sources means checking ingestion history against these five issue numbers before trusting a query result that looks plausible but wrong.