[
https://issues.apache.org/jira/browse/CALCITE-3079?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16845842#comment-16845842
]
Feng Zhu commented on CALCITE-3079:
-----------------------------------
Hi, [~danny0405]. I have made a thorough investigation on *EnumerablWindow* and
fix it. However, when I add some unit tests, we find they fail caused by the
following Assert.
{code:java}
java.lang.AssertionError: RexInputRef index 5 out of range 0..4
at org.apache.calcite.util.Litmus$1.fail(Litmus.java:31)
at org.apache.calcite.rex.RexChecker.visitInputRef(RexChecker.java:121)
at org.apache.calcite.rex.RexChecker.visitInputRef(RexChecker.java:57)
at org.apache.calcite.rex.RexInputRef.accept(RexInputRef.java:112)
at org.apache.calcite.rex.RexChecker.visitCall(RexChecker.java:140)
at org.apache.calcite.rex.RexChecker.visitCall(RexChecker.java:57)
at org.apache.calcite.rex.RexCall.accept(RexCall.java:190)
at org.apache.calcite.rex.RexChecker.isValid(RexChecker.java:184)
at org.apache.calcite.rel.core.Window.isValid(Window.java:116)
at
org.apache.calcite.rel.AbstractRelNode.onRegister(AbstractRelNode.java:351)
at
org.apache.calcite.plan.volcano.VolcanoPlanner.registerImpl(VolcanoPlanner.java:1754)
at
org.apache.calcite.plan.volcano.VolcanoPlanner.register(VolcanoPlanner.java:923)
at
org.apache.calcite.plan.volcano.VolcanoPlanner.ensureRegistered(VolcanoPlanner.java:945)
at
org.apache.calcite.plan.volcano.VolcanoPlanner.ensureRegistered(VolcanoPlanner.java:102)
at
org.apache.calcite.rel.AbstractRelNode.onRegister(AbstractRelNode.java:334)
at
org.apache.calcite.plan.volcano.VolcanoPlanner.registerImpl(VolcanoPlanner.java:1754)
at
org.apache.calcite.plan.volcano.VolcanoPlanner.register(VolcanoPlanner.java:923)
at
org.apache.calcite.plan.volcano.VolcanoPlanner.ensureRegistered(VolcanoPlanner.java:945)
at
org.apache.calcite.plan.volcano.VolcanoPlanner.ensureRegistered(VolcanoPlanner.java:2030)
at
org.apache.calcite.plan.volcano.VolcanoRuleCall.transformTo(VolcanoRuleCall.java:138)
at
org.apache.calcite.plan.RelOptRuleCall.transformTo(RelOptRuleCall.java:234)
at
org.apache.calcite.rel.rules.ProjectToWindowRule$ProjectToLogicalProjectAndWindowRule.onMatch(ProjectToWindowRule.java:190)
{code}
It seems that the nested window is not the design intention. Consequently,
Calcite treats this LogicalWindow as invalid.
How about your idea?
> Successive dependent windows cannot be implemented in same expression level
> ---------------------------------------------------------------------------
>
> Key: CALCITE-3079
> URL: https://issues.apache.org/jira/browse/CALCITE-3079
> Project: Calcite
> Issue Type: Bug
> Components: core
> Affects Versions: 1.20.0
> Reporter: Feng Zhu
> Assignee: Feng Zhu
> Priority: Major
> Labels: pull-request-available
> Time Spent: 10m
> Remaining Estimate: 0h
>
> Recently, we encountered an IndexOutOfBoundsException when running a
> complicated query containing successive dependent windows.The issue can be
> reproduced by the following simple query on table *t1(a, b, c)*.
> {code:java}
> Q1:
> select sum(s) over (partition by aa) as ss " +
> from (
> select a as aa, sum(b) over (partition by a, c) as s
> from t1
> ) t2";{code}
> The exception is:
> {code:java}
> Exception in thread "main" java.sql.SQLException: Error while executing SQL
> "select sum(s) over (partition by aa) as ss from (select a as aa, sum(b) over
> (partition by a, c) as s from t1) t2": index (0) must be less than size (0)
> at org.apache.calcite.avatica.Helper.createException(Helper.java:56)
> at org.apache.calcite.avatica.Helper.createException(Helper.java:41)
> at
> org.apache.calcite.avatica.AvaticaStatement.executeInternal(AvaticaStatement.java:163)
> at
> org.apache.calcite.avatica.AvaticaStatement.executeQuery(AvaticaStatement.java:227)
> at org.apache.calcite.JDBCDemo.main(JDBCDemo.java:70)
> at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
> at
> sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
> at
> sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
> at java.lang.reflect.Method.invoke(Method.java:498)
> at com.intellij.rt.execution.application.AppMain.main(AppMain.java:147)
> Caused by: java.lang.IndexOutOfBoundsException: index (0) must be less than
> size (0)
> at
> com.google.common.base.Preconditions.checkElementIndex(Preconditions.java:310)
> at
> com.google.common.base.Preconditions.checkElementIndex(Preconditions.java:293)
> at
> com.google.common.collect.RegularImmutableList.get(RegularImmutableList.java:67)
> at
> org.apache.calcite.adapter.enumerable.EnumUtils$2.get(EnumUtils.java:115)
> at
> org.apache.calcite.adapter.enumerable.EnumUtils$2.get(EnumUtils.java:110)
> at
> org.apache.calcite.adapter.enumerable.EnumerableWindow.lambda$implement$0(EnumerableWindow.java:442)
> at
> org.apache.calcite.adapter.enumerable.EnumerableWindow$3.rexArguments(EnumerableWindow.java:854)
> .................................
> {code}
>
> However, the modified query below can be executed in a right way.
> {code:java}
> Q2:
> select sum(s) over (partition by aa) as ss " +
> from (
> select a as aa, sum(b) over (partition by a, c) + 0 as s
> from t1
> ) t2{code}
> This issue is caused by
> *_ProjectToWindowRule_*({color:#ff0000}CalcRelSplitter{color}). When
> splitting window expressions in Project node, the rule ignores to check
> whether a window and its input window are in the same level.Due to such
> beheavior, two successive window expressions are implemented in same level
> and the RelNode after transformation is:
> {code:java}
> LogicalProject($0=[$4])
> LogicalWindow(window#0=[window(partition {0, 2} order by [] range between
> UNBOUNDED PRECEDING and UNBOUNDED FOLLOWING aggs [SUM($1)])],
> window#1=[window(partition {0} order by [] range between UNBOUNDED PRECEDING
> and UNBOUNDED FOLLOWING aggs [SUM($3)])])
> EnumerableTableScan(subset=[rel#7:Subset#0.ENUMERABLE.[]], table=[[ttt,
> test]]){code}
> As for *Q2*, two window expressions are not "successive", an _*Add(+)*_
> operation results to implementing them in different levels. The RelNode after
> transformation is:
> {code:java}
> LogicalProject($0=[$2])
> LogicalWindow(window#0=[window(partition {0} order by [] range between
> UNBOUNDED PRECEDING and UNBOUNDED FOLLOWING aggs [SUM($1)])])
> LogicalProject(a=[$0], $1=[+($3, 0)])
> LogicalWindow(window#0=[window(partition {0, 2} order by [] range
> between UNBOUNDED PRECEDING and UNBOUNDED FOLLOWING aggs [SUM($1)])])
> EnumerableTableScan(subset=[rel#7:Subset#0.ENUMERABLE.[]],
> table=[[ttt, test]]){code}
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
