adriangb commented on code in PR #20117: URL: https://github.com/apache/datafusion/pull/20117#discussion_r2760254821
########## datafusion/sqllogictest/test_files/push_down_filter.slt: ########## @@ -116,11 +116,12 @@ explain select * from (select column1, unnest(column2) as o from d) where o['a'] ---- physical_plan 01)ProjectionExec: expr=[column1@0 as column1, __unnest_placeholder(d.column2,depth=1)@1 as o] -02)--FilterExec: get_field(__unnest_placeholder(d.column2,depth=1)@1, a) = 1 +02)--FilterExec: __datafusion_extracted_1@0 = 1, projection=[column1@1, __unnest_placeholder(d.column2,depth=1)@2] 03)----RepartitionExec: partitioning=RoundRobinBatch(4), input_partitions=1 -04)------UnnestExec -05)--------ProjectionExec: expr=[column1@0 as column1, column2@1 as __unnest_placeholder(d.column2)] -06)----------DataSourceExec: partitions=1, partition_sizes=[1] +04)------ProjectionExec: expr=[get_field(__unnest_placeholder(d.column2,depth=1)@1, a) as __datafusion_extracted_1, column1@0 as column1, __unnest_placeholder(d.column2,depth=1)@1 as __unnest_placeholder(d.column2,depth=1)] +05)--------UnnestExec Review Comment: I think it would be quite complex to try to push the get_filter through the unnest, and ultimately I don't think Parquet would be able to optimize the scan (maybe I'm wrong about this?) so there would be little point. -- 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]
