raghavyadav01 commented on issue #26514: URL: https://github.com/apache/datafusion/issues/26514#issuecomment-6695956169
Found where this comes from. arrow-csv hands writing to the `csv-core` crate, and its `quote()` only escapes the quote character, never the escape character itself. So with `double_quote = false`, a `\` in a value is written as-is and the reader treats it as an escape. DataFusion can't fix the round trip just by choosing writer options. Suggested fix, in two steps: 1. **Now, in DataFusion:** reject `format.escape` when writing, unless `double_quote` is `false` or the escape is the quote character. The issue's repro then gives a clear error instead of a file that can't be read back. I have this ready with sqllogictest cases. The trade-off: a write with `escape` set and no backslashes in the data would now error too, even though it works today. 2. **Upstream:** open an issue in rust-csv so the writer also escapes the escape character when `double_quote` is false. Once that's in, DataFusion can apply `escape` on write properly and drop the error from step 1. Does that sound right? If so, I'll open the PR for step 1 and file the rust-csv issue. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
