andygrove commented on PR #6457: URL: https://github.com/apache/datafusion-comet/pull/6457#issuecomment-5969837961
@parthchandra A correction to my earlier reply. `eb5c7af4f` did not move the signed-zero checks to `first_value`/`last_value` windows. It dropped the aggregate tie queries from the fixture, and `CometAggregateSuite` then checked each zero on its own, which passes under any tie rule. So the end-to-end check of the behavior this PR is about was gone. `40592a255` brings those checks back with the single-file guarantee you suggested. The fixture writes `mm_float` with `INSERT INTO ... SELECT /*+ COALESCE(1) */`, so both engines read the rows in id order in one task, and `CometAggregateSuite` writes with `repartition(1)` again. The ungrouped checks now run once per group, because over groups 2 and 3 together the first and the last zero have the same sign, so that query could not tell first-wins from last-wins. Making the ungrouped and then the grouped accumulator last-wins fails the matching query each time. I also merged main again in `a67baeb7e`, and `205a2c516` points the `array_remove` and `sort_array` tests from #6518 at the shared edge values in `float_semantics` instead of their own copy. -- 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]
