[ 
https://issues.apache.org/jira/browse/FLINK-3849?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15927290#comment-15927290
 ] 

ASF GitHub Bot commented on FLINK-3849:
---------------------------------------

Github user KurtYoung commented on a diff in the pull request:

    https://github.com/apache/flink/pull/3520#discussion_r106322568
  
    --- Diff: 
flink-libraries/flink-table/src/main/scala/org/apache/flink/table/plan/nodes/dataset/DataSetCalc.scala
 ---
    @@ -42,63 +41,40 @@ class DataSetCalc(
         traitSet: RelTraitSet,
         input: RelNode,
         rowRelDataType: RelDataType,
    -    private[flink] val calcProgram: RexProgram, // for tests
    +    calcProgram: RexProgram,
         ruleDescription: String)
    -  extends SingleRel(cluster, traitSet, input)
    +  extends Calc(cluster, traitSet, input, calcProgram)
    --- End diff --
    
    This is because i want to unify the PushFilterIntoScan rule's code for both 
batch and stream mode. During executing the rule, we may need to create a new 
copy of the DataSetCalc or DataStreamCalc. It make things more easier to let 
these two classes inherit from `Calc`, and use `Calc.copy` to create a new 
copied instance. 
    
    I do encountered some problem after i changed the hierarchy, some unit 
tests failed because of the plan changed. But it's because we don't calculate 
the cost for Calc right. I added some logic to `CommanCalc.computeSelfCost`, 
and everything works fine.


> Add FilterableTableSource interface and translation rule
> --------------------------------------------------------
>
>                 Key: FLINK-3849
>                 URL: https://issues.apache.org/jira/browse/FLINK-3849
>             Project: Flink
>          Issue Type: New Feature
>          Components: Table API & SQL
>            Reporter: Fabian Hueske
>            Assignee: Kurt Young
>
> Add a {{FilterableTableSource}} interface for {{TableSource}} implementations 
> which support filter push-down.
> The interface could look as follows
> {code}
> def trait FilterableTableSource {
>   // returns unsupported predicate expression
>   def setPredicate(predicate: Expression): Expression
> }
> {code}
> In addition we need Calcite rules to push a predicate (or parts of it) into a 
> TableScan that refers to a {{FilterableTableSource}}. We might need to tweak 
> the cost model as well to push the optimizer in the right direction.



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

Reply via email to