alamb commented on issue #9083: URL: https://github.com/apache/arrow-rs/issues/9083#issuecomment-3756356776
My thoughts from the 1. https://github.com/apache/arrow-rs/issues/9083#issuecomment-3710183483 2. https://github.com/apache/arrow-rs/issues/9083#issuecomment-3710188287 Basically, my personal opinion is that a parallel RowFormat for equality comparisons is likely going to be a lot of code to maintain and I think that for whatever usecase you have in mind, you would likely better off using a different approach. > row format is also useful when you need to copy a lot of columns around, for example when shuffling I don't understand why row format would be used here -- Arrow already has a format that is optimzied for copying values around (the IPC format) > grouping - grouping by a lot of columns does not need ordering requirements How about implementing a specialization for Structs, the way we have for other types: https://github.com/apache/datafusion/tree/main/datafusion/physical-plan/src/aggregates/group_values/multi_group_by That would likely be a lot faster than having to copy all the data once which a new row format would do - aggregation - using count(distinct <struct>) or array_agg with distinct, both of which do not need ordering requirement shuffling Similarly, I think if you implemented a specialized accumulator for structs you would be better off than using a RowFormat, right? -- 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]
