[
https://issues.apache.org/jira/browse/CALCITE-3228?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16899594#comment-16899594
]
jin xing commented on CALCITE-3228:
-----------------------------------
[^CALCITE-3228.patch]
Hi [~Chunwei Lei]
I try to dig and learn from this issue;
*Project::getMapping* try to return an inverse-surjection mapping.
*SurjectionWithInverse* is defined as below:
{code:java}
/**
* A surjection with inverse has precisely one source for each target.
* (Whereas a general surjection has at least one source for each target.)
* Every source has at most one target.
*/
{code}
That's why the test you provided throws
{code:java}
...
Caused by: java.lang.IllegalArgumentException: source #0 is already mapped to
target #1
at
org.apache.calcite.util.mapping.Mappings$SurjectionWithInverse.set(Mappings.java:1326)
at org.apache.calcite.rel.core.Project.getMapping(Project.java:285)
at org.apache.calcite.rel.core.Project.getMapping(Project.java:256)
at
org.apache.calcite.rel.rules.ProjectTableScanRule.apply(ProjectTableScanRule.java:107)
at
org.apache.calcite.rel.rules.ProjectTableScanRule$2.onMatch(ProjectTableScanRule.java:83)
at
org.apache.calcite.plan.volcano.VolcanoRuleCall.onMatch(VolcanoRuleCall.java:208)
...
{code}
I have two ideas and it's great if you can confirm and rectify when you have
time:
1. *ProjectTableScanRule* should not be limited by the Project::getMapping. I
mean ProjectTableScanRule can handle to create a
BindableTableScan as long as *project.getProjects().stream().allMatch(p -> p
instanceof RexInputRef) == true,* and no need to be limited by a
inverse-surjection mapping.
2. The definition of *Project::getMapping* is as below
{code:java}
/**
* Returns a mapping of a set of project expressions.
*
* <p>The mapping is an inverse surjection.
* Every target has a source field, but
* a source field may appear as zero, one, or more target fields.
* Thus you can safely call
* {@link
org.apache.calcite.util.mapping.Mappings.TargetMapping#getTarget(int)}.
*
{code}
But in my understanding, it's not safe to call *getTarget(int)* but safe to
call *getSource(int);*
I try to file a fix for this issue and illustrate my idea. It's great if you
can shepherd and tell if it's the right direction.
If you haven't started a PR on this issue, I can work on it under your shepherd.
But since you have the expertise, if you have already working on, I will learn
and track your PR.
> Error while applying rule ProjectScanRule: interpreter
> ------------------------------------------------------
>
> Key: CALCITE-3228
> URL: https://issues.apache.org/jira/browse/CALCITE-3228
> Project: Calcite
> Issue Type: Bug
> Components: core
> Reporter: Chunwei Lei
> Priority: Critical
> Fix For: 1.21.0
>
> Attachments: CALCITE-3228.patch
>
>
> The following test can reproduce the issue.
>
> {code:java}
> // FrameworksTest.java
> @Test public void testMinMax() throws Exception {
> Table table = new TableImpl();
> final SchemaPlus rootSchema = Frameworks.createRootSchema(true);
> SchemaPlus schema = rootSchema.add("x", new AbstractSchema());
> schema.add("MYTABLE", table);
> List<RelTraitDef> traitDefs = new ArrayList<>();
> traitDefs.add(ConventionTraitDef.INSTANCE);
> traitDefs.add(RelDistributionTraitDef.INSTANCE);
> SqlParser.Config parserConfig =
> SqlParser.configBuilder(SqlParser.Config.DEFAULT)
> .setCaseSensitive(false)
> .build();
> final FrameworkConfig config = Frameworks.newConfigBuilder()
> .parserConfig(parserConfig)
> .defaultSchema(schema)
> .traitDefs(traitDefs)
> // define the rules you want to apply
> .ruleSets(
> RuleSets.ofList(AbstractConverter.ExpandConversionRule.INSTANCE,
> ProjectTableScanRule.INSTANCE))
> .programs(Programs.ofRules(Programs.RULE_SET))
> .build();
> executeQuery(config, " select min(id) as mi, max(id) as ma from mytable where
> id=1 group by id",
> CalciteSystemProperty.DEBUG.value());
> }
> {code}
>
--
This message was sent by Atlassian JIRA
(v7.6.14#76016)
