koriyoshi2041 commented on PR #26788: URL: https://github.com/apache/datafusion/pull/26788#issuecomment-6873780655
Yes. I traced the `single_stream.rs` pattern, and it fits both blocking operators better than the explicit `Poll` state added here. For `WindowAggStream`, the generator can consume `input.next().await`, drop the input, compute the materialized result once, and emit `result.slice(offset, batch_size)` from a local loop. For grouped TopK, it can consume and intern inputs in the same way, then emit the final materialized batch. That removes `done`, `output`, and `output_offset` from the structs while preserving the `4/4/2` and `2/2/1` regression expectations. Wrapping each generator in `RecordBatchStreamAdapter`/`ObservedStream` also keeps output metrics at the emitted-slice boundary. I will revise the branch in that form and rerun the two focused suites. -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
