[jira] [Commented] (CALCITE-557) Remove AbstractConverter

["Vladimir Sitnikov (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/CALCITE-557?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14278570#comment-14278570 ] 

Vladimir Sitnikov commented on CALCITE-557:
-------------------------------------------

[~julianhyde], you've just updated {{RuleMatchImportanceComparator}} to compare by rule class name first.
Why is that?
If that is intentional, I suppose some comment should be added / javadoc updated.

> Remove AbstractConverter
> ------------------------
>
>                 Key: CALCITE-557
>                 URL: https://issues.apache.org/jira/browse/CALCITE-557
>             Project: Calcite
>          Issue Type: Bug
>            Reporter: Julian Hyde
>            Assignee: Julian Hyde
>             Fix For: 1.0.0-incubating
>
>
> Currently each subset contains an AbstractConverter to each subset in the same set. If there are a lot of active traits, there are many, many AbstractConverters that do not add anything to the planning.
> This task would remove those AbstractConverters. In fact AbstractConverter would never be created.
> Also, make costing of joins more deterministic when both inputs have the same number of rows. Currently we use the RelNode.id of the inputs, but this is unstable if we change what rules are activated. So, we should order the inputs based on the number of columns first, and on id only if that fails.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)