[jira] [Commented] (CALCITE-3629) Table-valued function implementor interface

["Rui Wang (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/CALCITE-3629?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17015640#comment-17015640 ] 

Rui Wang commented on CALCITE-3629:
-----------------------------------

Fixed in [b80bb1c|https://github.com/apache/calcite/commit/b80bb1cbceb11ed31b73e419916b5cc98610503e].


Sorry [~zabetak] I wasn't clear enough on this JIRA. This JIRA came from supporting TUMBLE as a table-valued function in which Danny wanted to implement an interface for table-valued function implementor. I don't know what is the better practice for this scenario: when working on a Jira, some new requirement appear. Should I create a separate Jira for that new requirement or just reuse the old jira for context? I chose to create a Jira but maybe the best way is to reuse the old Jira.



> Table-valued function implementor interface
> -------------------------------------------
>
>                 Key: CALCITE-3629
>                 URL: https://issues.apache.org/jira/browse/CALCITE-3629
>             Project: Calcite
>          Issue Type: Sub-task
>            Reporter: Rui Wang
>            Assignee: Rui Wang
>            Priority: Major
>             Fix For: 1.22.0
>
>
> The current implementation of TUMBLE is the following: 
> {code:java}
> if (isTumble()) {
>    // create expressions and blockstatement
> } else {
>  // other table function scan enumerable implementation.
> }
> {code}
> This style is less extensible. We will add more builtin table-valued function implementation and we shouldn't  add more if-else with complicated implementation.
> A better practice is in RexImpTable, each operator register its implementor, and then the code will be becomes:
> {code:java}
> if (implementor defined) {
>    return implementor.implement(...)
> }  else {
>  // default table function scan enumerable implementation for those that does not have implementor.
> } 
> {code}
> The current implementor[1] or [2] is not enough for table-valued function as table-valued function need both inputPhysType and outPhysType, and potential other params to implement itself.
> Another condition is we might need more table-valued function implementations to abstract a common interface. 
> [1]: https://github.com/apache/calcite/blob/master/core/src/main/java/org/apache/calcite/adapter/enumerable/CallImplementor.java#L38
> [2]: https://github.com/apache/calcite/blob/master/core/src/main/java/org/apache/calcite/adapter/enumerable/NotNullImplementor.java



--
This message was sent by Atlassian Jira
(v8.3.4#803005)