pitrou commented on issue #50314: URL: https://github.com/apache/arrow/issues/50314#issuecomment-4890721463
> How do we choose the threshold for non-reasonable values? A conservative guesstimate is good enough IMHO. The underlying design of the encoding should suggest that a miniblock larger than 1024 values, for example, does not really make sense. > Maybe we can determine an upper bound from trusted information like the buffer size, rather than have to pick an arbitrary threshold. The problem is that a range of equal values can encode to exactly 0 bits. > Eg. looking at the Rust implementation, they don't resize `mini_block_bit_widths` until reading them [here](https://github.com/apache/arrow-rs/blob/8205c5421f2f08c22c74be909beaf8afb2774011/parquet/src/encodings/decoding.rs#L608-L609) We could mirror that in C++, though that's slightly more logic to write than simply disallowing extremely unlikely values. -- 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]
