Dandandan commented on pull request #9301: URL: https://github.com/apache/arrow/pull/9301#issuecomment-766160716
> @Dandandan , > > > Wouldn't it be possible to have a similar as previously implementation for `take_no_nulls` and `take_values_nulls`? I think those are the most common anyway. > > I am sorry, I did not understand: this PR does split the kernel in the 4 cases, and uses them accordingly depending on the null count of `values` and `indices`. Sorry, I'll try to be more clear. As far as I understand the undefined behavior comes from using the values from the null _indices_, so wouldn't it be possible to (mostly) recover the performance for the cases where there are either no null values at all, or only for the _values_? I think cases would occur more frequently (e.g. in DataFusion) than using the kernel with indices with nulls. ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: [email protected]
