westonpace opened a new pull request, #34060: URL: https://github.com/apache/arrow/pull/34060
This PR introduces the concept of `ExecBatch:index` but does not yet do much with it. As a proof of concept this PR adds a fetch node which can be inserted anywhere in the plan (not just at the sink) to satisfy `LIMIT x OFFSET y` (Substrait calls this fetch and so I have also). This PR also introduces two sequencing accumulation queues which will be useful, I hope, for anyone implementing nodes that rely on ordered execution. This PR unfortunately introduces a new query option which is whether or not the sink node should pay the small performance hit required to sequence output. While considering how best to add this option I realized we will probably have more query options in the near future regarding "how much RAM to use" (e.g. spillover) and potentially more beyond that. So I have taken all the options and put them into `arrow::compute::QueryOptions` (this already existed but it was not user facing and I added more things to it). I added a new DeclarationToXyz overload that accepts QueryOptions. This has, unfortunately, led to a bit of overload explosion but I think this should be the last new addition to the overload set (and we can deprecate the older overloads at some point). This PR also includes a new `gen::Gen / gen::TestGen` facility for generating test tables for input. I'd like to eventually use this to simplify some of the existing exec plan tests as well. I'm willing to split this into a separate PR if that makes sense. -- 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]
