zabetak commented on a change in pull request #2690:
URL: https://github.com/apache/calcite/pull/2690#discussion_r805648149
##########
File path: core/src/main/java/org/apache/calcite/rel/core/RepeatUnion.java
##########
@@ -61,12 +64,20 @@
*/
public final int iterationLimit;
+ /**
+ * Transient table where repeat union's intermediate results will be stored.
+ */
+ protected final RelOptTable transientTable;
+
Review comment:
Do we need to update the class javadoc based on this change to explain
how is this table supposed to be used? Also can this be null (optional) in case
some concrete implementations do not want to make use of it?
##########
File path:
core/src/test/java/org/apache/calcite/test/enumerable/EnumerableRepeatUnionTest.java
##########
@@ -233,4 +245,83 @@
.returnsOrdered("i=1", "i=2", "i=null", "i=3", "i=2", "i=3", "i=3");
}
+ /** Test case for
+ * <a
href="https://issues.apache.org/jira/browse/CALCITE-4054";>[CALCITE-4054]
+ * RepeatUnion containing a Correlate with a transientScan on its RHS causes
NPE</a>. */
+ @Test void testRepeatUnionWithCorrelateWithTransientScanOnItsRight() {
+ CalciteAssert.that()
+ .with(CalciteConnectionProperty.LEX, Lex.JAVA)
+ .with(CalciteConnectionProperty.FORCE_DECORRELATE, false)
+ .withSchema("s", new ReflectiveSchema(new HierarchySchema()))
+ .query("?")
+ .withHook(Hook.PLANNER, (Consumer<RelOptPlanner>) planner -> {
+ planner.addRule(JoinToCorrelateRule.Config.DEFAULT.toRule());
+ planner.removeRule(JoinCommuteRule.Config.DEFAULT.toRule());
+ planner.removeRule(EnumerableRules.ENUMERABLE_MERGE_JOIN_RULE);
+ planner.removeRule(EnumerableRules.ENUMERABLE_JOIN_RULE);
+ })
+ .withRel(builder -> {
+ builder
+ // WITH RECURSIVE delta(empid, name) as (
+ // SELECT empid, name FROM emps WHERE empid = 2
+ // UNION ALL
+ // SELECT e.empid, e.name FROM delta d
+ // JOIN hierarchies h ON d.empid =
h.managerid
+ // JOIN emps e ON
h.subordinateid = e.empid
+ // )
+ // SELECT empid, name FROM delta
+ .scan("s", "emps")
+ .filter(
+ builder.equals(
+ builder.field("empid"),
+ builder.literal(2)))
+ .project(
+ builder.field("emps", "empid"),
+ builder.field("emps", "name"))
+
+ .transientScan("#DELTA#");
+ RelNode transientScan = builder.build(); // pop the transientScan to
use it later
+
+ builder
+ .scan("s", "hierarchies")
+ .push(transientScan) // use the transientScan as right input of
the join
+ .join(
+ JoinRelType.INNER,
+ builder.equals(
+ builder.field(2, "#DELTA#", "empid"),
+ builder.field(2, "hierarchies", "managerid")))
+
+ .scan("s", "emps")
+ .join(
+ JoinRelType.INNER,
+ builder.equals(
+ builder.field(2, "hierarchies", "subordinateid"),
+ builder.field(2, "emps", "empid")))
+ .project(
+ builder.field("emps", "empid"),
+ builder.field("emps", "name"))
+ .repeatUnion("#DELTA#", true);
+ return builder.build();
+ })
+ .explainHookMatches(""
+ + "EnumerableRepeatUnion(all=[true])\n"
+ + " EnumerableTableSpool(readType=[LAZY], writeType=[LAZY],
table=[[#DELTA#]])\n"
+ + " EnumerableCalc(expr#0..4=[{inputs}], expr#5=[2],
expr#6=[=($t0, $t5)], empid=[$t0], name=[$t2], $condition=[$t6])\n"
+ + " EnumerableTableScan(table=[[s, emps]])\n"
+ + " EnumerableTableSpool(readType=[LAZY], writeType=[LAZY],
table=[[#DELTA#]])\n"
+ + " EnumerableCalc(expr#0..8=[{inputs}], empid=[$t4],
name=[$t6])\n"
+ + " EnumerableCorrelate(correlation=[$cor1],
joinType=[inner], requiredColumns=[{1}])\n"
+ // It is important to have EnumerableCorrelate + #DELTA# table
scan on its right
Review comment:
An explanation of **why** is it important would be helpful.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
