[
https://issues.apache.org/jira/browse/CALCITE-5703?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17723403#comment-17723403
]
Sergey Nuyanzin commented on CALCITE-5703:
------------------------------------------
I wouldn't say that the casting is redundant... at least so far
this is exactly the issue we faced in Flink with Janino 3.1.x (not an issue for
3.0.x)
[https://github.com/janino-compiler/janino/issues/188]
And explicit cast is a WA for that issue. I would wait for at least having it
fixed on Janino level.
Also +1 to [~julianhyde] and [~FrankZou] comments
> Reduce amount of generated runtime code
> ---------------------------------------
>
> Key: CALCITE-5703
> URL: https://issues.apache.org/jira/browse/CALCITE-5703
> Project: Calcite
> Issue Type: Improvement
> Components: core
> Affects Versions: 1.34.0
> Reporter: Evgeny Stanilovsky
> Assignee: Evgeny Stanilovsky
> Priority: Major
> Labels: patch-available
>
> In some cases runtime generates code like :
> {noformat}
> return case_when_value == null ? (String) null : some_oparation();
> or
> return input_value == null ? (Long) null : Long.valueOf(...;
> {noformat}
> this redundant casting probably not harmful, but there is another side -
> maximum method size, this size jdk[1], janino [2] throws : *Code grows beyond
> 64 KB* . This PR reduces code generated by calcite runtime thus more huge
> expressions can be executed.
> [1]
> https://github.com/openjdk/jdk/blob/d22bcc813eea719b817d3d541a843594675c0ca9/src/jdk.compiler/share/classes/com/sun/tools/javac/jvm/ClassFile.java#L101
> [2]
> https://github.com/janino-compiler/janino/blob/e69022f5aaabd36edc08a2074360d62514493a19/janino/src/main/java/org/codehaus/janino/CodeContext.java#L699
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
