[
https://issues.apache.org/jira/browse/CALCITE-987?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15025085#comment-15025085
]
Julian Hyde commented on CALCITE-987:
-------------------------------------
Regarding metadata validation.
Whether the validation is done by the metadata provider or consumer or by the
framework is a matter of taste. But clearly someone has to do it!
validateResult is the framework trying to apply a light sanity check. But
consumers should also do some validation: I think it would be foolish for a
consumer to divide by a value before they've checked that it is not null and
greater than zero.
It would seem to me that if a method has very few providers and a lot of
consumers then it would be better to validate in the providers; and vice versa.
I don't feel strongly about this. But if you decide to change the policy just
make sure do it throughout calcite and change the documentation.
> Push limit 0 will result in an infinite loop
> --------------------------------------------
>
> Key: CALCITE-987
> URL: https://issues.apache.org/jira/browse/CALCITE-987
> Project: Calcite
> Issue Type: Bug
> Reporter: Pengcheng Xiong
> Assignee: Pengcheng Xiong
> Attachments: CALCITE-987.01.patch
>
>
> We use "checkInputForCollationAndLimit" in RelMdUtil.java to check the input
> #rows. However, it calls RelMetadataQuery.getRowCount which will validate the
> #rows. The validation will change #row=0 to #row=1. This will result in an
> infinite loop to push limit. The affected rules include
> SortUnionTransposeRule and any Sort***TransposeRules that call
> checkInputForCollationAndLimit.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
