westonpace commented on issue #15280: URL: https://github.com/apache/arrow/issues/15280#issuecomment-1377946455
`libacero` will be easier than `libkernels` (there are still parts of the core lib that depend on compute kernels, I think) though both seem valid. The only gray area I can see is group-by. There is currently a standalone group-by implementation that runs outside of Acero, that is marked internal, that I am [trying to get rid of](https://github.com/apache/arrow/pull/14867). My vote would be that "group-by" is a part of Acero. Note, the `hash_` aggregate kernels themselves could remain as part of compute/kernels. Aggregations would still be usable outside of Acero. If that's the case then you could also move `arrow/compute/row` to Acero (and probably move it to `arrow/compute/exec/row` while you are at it). > I prefer to libacero because Gandiva uses libgandiva not libarrow_gandiva. ADBC also uses libadbc_* not libarrow_adbc_*. Both datasets and substrait will need to depend on libacero. It would be a bit weird to have `libarrow_dataset` -> `libacero` -> `libarrow_compute`. I'm ok with the weirdness though. -- 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]
