[
https://issues.apache.org/jira/browse/CALCITE-4616?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17351269#comment-17351269
]
Vladimir Ozerov commented on CALCITE-4616:
------------------------------------------
The proposed fix:
# Iterate over all inputs and check if they require aggregate pushdown.
# If no inputs require the pushdown, return.
# If at least one input requires a pushdown, do a pushdown for all other inputs
as well.
# This way we ensure that all inputs would have the same row type, and the
problem would go away. The {{AggregateRemoveRule}} may remove the unnecessary
aggregates afterward.
> AggregateUnionTransposeRule causes row type mismatch when some inputs have
> unique grouping key
> ----------------------------------------------------------------------------------------------
>
> Key: CALCITE-4616
> URL: https://issues.apache.org/jira/browse/CALCITE-4616
> Project: Calcite
> Issue Type: Bug
> Affects Versions: 1.26.0
> Reporter: Vladimir Ozerov
> Priority: Major
>
> Consider the following relational tree:
> {code}
> 1: Aggregate[group=a, SUM(b)]
> 2: Union
> 3: Input1[a, b, c]
> 4: Input2[a, b, c]
> {code}
> It may happen, that {{Input1}} has a unique key on {{[a]}}. In this case,
> Apache Calcite, will not install the {{Aggregate}} on top of it, which leads
> to the following tree:
> 1: Aggregate[group=a, SUM(b)]
> 2: Union
> 3: Input1[a, b, c]
> 4: Aggregate[group=a, SUM(b)]
> 5: Input2[a, b, c]
> {code}
> Obviously, the tree is incorrect, because {{Union}} inputs now have different
> row types.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
