[
https://issues.apache.org/jira/browse/FLINK-5256?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15826132#comment-15826132
]
ASF GitHub Bot commented on FLINK-5256:
---------------------------------------
Github user fhueske commented on a diff in the pull request:
https://github.com/apache/flink/pull/3033#discussion_r96419377
--- Diff:
flink-libraries/flink-table/src/main/scala/org/apache/flink/table/plan/nodes/dataset/DataSetSingleRowJoin.scala
---
@@ -144,21 +150,46 @@ class DataSetSingleRowJoin(
val condition = codeGenerator.generateExpression(joinCondition)
- val joinMethodBody = s"""
- |${condition.code}
- |if (${condition.resultTerm}) {
- | ${conversion.code}
- |
${codeGenerator.collectorTerm}.collect(${conversion.resultTerm});
- |}
- |""".stripMargin
+ val joinMethodBody =
+ if (joinType == JoinRelType.INNER) {
+ s"""
+ |${condition.code}
+ |if (${condition.resultTerm}) {
+ | ${conversion.code}
+ |
${codeGenerator.collectorTerm}.collect(${conversion.resultTerm});
+ |}
+ |""".stripMargin
+ } else {
+ val singleNode =
+ if (leftIsSingle) {
+ leftNode
+ }
+ else {
+ rightNode
+ }
+ val notSuitedToCondition = singleNode
+ .getRowType
+ .getFieldList
+ .map(field => getRowType.getFieldNames.indexOf(field.getName))
+ .map(i => s"${conversion.resultTerm}.setField($i,null);")
--- End diff --
This only works if the result term is a `Row`.
It think a better approach is to generate a dedicated result expression for
the null case based on `CodeGenerator.generateResultExpression()`, where the
field expressions are the outer fields and the null expressions. Then we could
use the current result expression if the condition is met and the result
expression with null values if that is not the case.
We would need to extend the `CodeGenerator` with methods for
`generateFieldAccesses(boolean firstInput)` (see
`generateConverterResultExpression()`) and `generateNullValues(boolean
firstInput)` which generates expressions for null values.
> Extend DataSetSingleRowJoin to support Left and Right joins
> -----------------------------------------------------------
>
> Key: FLINK-5256
> URL: https://issues.apache.org/jira/browse/FLINK-5256
> Project: Flink
> Issue Type: Improvement
> Components: Table API & SQL
> Affects Versions: 1.2.0
> Reporter: Fabian Hueske
> Assignee: Anton Mushin
>
> The {{DataSetSingleRowJoin}} is a broadcast-map join that supports arbitrary
> inner joins where one input is a single row.
> I found that Calcite translates certain subqueries into non-equi left and
> right joins with single input. These cases can be handled if the
> {{DataSetSingleRowJoin}} is extended to support outer joins on the
> non-single-row input, i.e., left joins if the right side is single input and
> vice versa.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
