[
https://issues.apache.org/jira/browse/CALCITE-3541?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17114290#comment-17114290
]
Xiening Dai commented on CALCITE-3541:
--------------------------------------
hi [~zabetak], not sure what you mean by adding a predicate. I think we already
have a user predicate for ConverterRule constructor today, no?
{code:java}
public <R extends RelNode> ConverterRule(Class<R> clazz,
Predicate<? super R> predicate, RelTrait in, RelTrait out,
RelBuilderFactory relBuilderFactory, String descriptionPrefix) {
}
{code}
> Avoid transformations to Enumerable nodes for custom SqlOperators
> -----------------------------------------------------------------
>
> Key: CALCITE-3541
> URL: https://issues.apache.org/jira/browse/CALCITE-3541
> Project: Calcite
> Issue Type: Improvement
> Components: core
> Affects Versions: 1.21.0
> Reporter: Stamatis Zampetakis
> Priority: Major
>
> Most Enumerable converter rules apply a transformation as soon as the
> {{RelNode}} class and its convention match those defined by the rule.
> However, there are use-cases that we would like to restrict the matches even
> further to avoid generating unimplementable plans that will fail at runtime.
> The most prominent example comes from extending the standard operator set
> with new {{SqlOperator}} s that appear in filters and projections as part of
> a row expression ({{RexNode}}). If we use the default instance of the
> {{EnumerableCalcRule}} we might end-up with a plan that will fail at runtime
> since the new operator is not handled by the Enumerable convention. Most
> likely there is a {{RelNode}} in another convention that can handle this new
> operator.
> We could avoid such undesirable transformations by allowing instantiations of
> the Enumerable converter rules with user-defined predicates. This also means
> adding public constructors to the rules that currently they do not have one.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
