[
https://issues.apache.org/jira/browse/CALCITE-5391?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17636938#comment-17636938
]
Scott Reynolds commented on CALCITE-5391:
-----------------------------------------
I am worried my application – which depends on field names – is going to break
in other ways because of this. I did some exploration and identified a few
other places where the field names change – some in meaningful ways. I found
Rules that rewrote all the names and ignoring all the aliases the user
provided. And I found – I believe mostly caused by ambiguous {{{}RexNode{}}}s –
field changes from {{Something}} to {{{}Something0{}}}.
How *should* applications code against the result set? Should applications keep
their own map of field names to ordinals instead of relying on the field names
from the root {{RelNode}} ?
h3. How I ran the Tests
I updated {{RelOptFixture#check()}} to compare the before and after plan's
{{RelDataType}}
{code:none}
diff --git a/testkit/src/main/java/org/apache/calcite/test/RelOptFixture.java
b/testkit/src/main/java/org/apache/calcite/test/RelOptFixture.java
index ecb474a9e..ed99437c7 100644
--- a/testkit/src/main/java/org/apache/calcite/test/RelOptFixture.java
+++ b/testkit/src/main/java/org/apache/calcite/test/RelOptFixture.java
@@ -400,6 +400,8 @@ private void checkPlanning(boolean unchanged) {
}
}
assertThat(relAfter, relIsValid());
+
+ assertThat(relAfter.getRowType(), is(relBefore.getRowType()));
}
{code}
And then ran {{./gradlew :core:test --tests
org.apache.calcite.test.RelOptRulesTest}}
h3. What I found
h4. Tests that rewrote all the column names
# testWithinDistinctCountDistinct
# testProjectToWindowRuleForMultipleWindows
# testReduceConstantsWindow
# testExpressionInWindowFunction
# testProjectWindowTransposeRuleWithConstants
# testProjectAggregateMergeSum0
h4. Column Names changed from {{Something}} to {{Something0}}
# testPushAggregateThroughOuterJoin2
# testPushAggregateThroughOuterJoin3
# testAggregateJoinRemove6
h4. Plans with expressions without an alias
There where a few plans that have expressions without a name so they get a name
change through the planning process. These seemed harmless.
> JoinOnUniqueToSemiJoinRule should preserve field names, if possible
> -------------------------------------------------------------------
>
> Key: CALCITE-5391
> URL: https://issues.apache.org/jira/browse/CALCITE-5391
> Project: Calcite
> Issue Type: Bug
> Affects Versions: 1.32.0
> Reporter: Scott Reynolds
> Assignee: Scott Reynolds
> Priority: Major
> Labels: pull-request-available
> Fix For: 1.33.0
>
> Time Spent: 10m
> Remaining Estimate: 0h
>
> The new rule pushes a copy of the [original project without the alias
> names|https://github.com/apache/calcite/pull/2848/files#r1026959819]
> {code:java}
> builder.project(project.getProjects());
> {code}
> This results in a new SQL plan without the name fields -- they become {{$fN}}
> fields in different plans.
> Small change is required to put the names from the {{RelDataType}} of the
> {{Project}} similar to the other {{Semijoin}} rule.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
