Rui Mo
Rui Mo
Try to fix this issue in https://github.com/facebookincubator/velox/pull/2386.
Hi Masha, thanks for your suggestion. > I suggest to unpack the row into 3 columns: count, sum, is_null. Looks like if we extract an extra column for is_null, the...
> Would you share the input data and query that produces the wrong result? This issue only happens when the input of final avg is the assembled row vector of...
> It sounds like the problem happens only if input to final aggregation is different from the output of the partial aggregation. Yes, they are different in our use case....
> It seem that we need a different solution which would preserve the null-ness of the intermediate results. Have you looked into a possibility of changing the Spark query plan?...
> Do you have a way to special case this some how in Gluten? Curious, how that trick works. For the temporary solution, currently just like the PR shows, we...
> @ChengjieLi28 I ran into the same problem and made a similar fix, but substrait tests started failing because checked in .json files need updating. > > I'm surprised that...
The length of this PR is 1.5k, but the json files occupies 800+ lines. And the changes from PR#1581 occupies 200 lines, which can be removed if PR#1581 gets merged.