[
https://issues.apache.org/jira/browse/FLINK-12298?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16824860#comment-16824860
]
sunjincheng commented on FLINK-12298:
-------------------------------------
Hi Dawid,
Thanks for mentioning the typesafe problem!
I think the typesafe problem does not only exist in the column functions. All
the TableAPIs with String and Expression as parameters have the same problem.
For example: `select("a, udf(b), ...")`, and the auxiliary operation APIs of
Window is also the same, the parameters are only String and Expression, such as
`Over.partitionBy(Expression...)`. Users can also write query like
`Over.partitionBy(1, 2) ` .
So this is a general problem in TableAPI, as Timo said, we need to be
considered from a Java DSL perspective. For example, `withColumns()` is a
function call, then it should be written as follows(from
[here|https://docs.google.com/document/d/1r3bfR9R6q5Km0wXKcnhfig2XQ4aMiLG5h2MTx960Fg8/edit?usp=sharing]
),
Select(call(withColumns, field("a"), range(...)))
When introducing Java DSL, we will introduce more types to users such as field,
call, range etc.
>From the points of my view, I suggest this change be done in conjunction with
>the Java DSL discussion, which we can discuss here
>https://docs.google.com/document/d/1r3bfR9R6q5Km0wXKcnhfig2XQ4aMiLG5h2MTx960Fg8/edit?usp=sharing
What do you think?
> Make column functions accept custom Range class rather than Expression
> ----------------------------------------------------------------------
>
> Key: FLINK-12298
> URL: https://issues.apache.org/jira/browse/FLINK-12298
> Project: Flink
> Issue Type: Improvement
> Components: Table SQL / API
> Reporter: Dawid Wysakowicz
> Priority: Major
>
> I would suggest to rework the column functions to a more typesafe approach
> with custom {{Range}} class. Right now {{withColumns}} accepts array of
> Expressions. We have
> {{org.apache.flink.table.api.scala.ImplicitExpressionOperations#to}} method,
> but also we have implicit conversion from {{scala.Range}} to Expression
> range. This already introduces ambiguity, as it is unclear what will be the
> end product of expression like {{1 to 9}}. This approach defers the checking
> of the types of expressions to expressions resolution phase.
> I would suggest to make
> {{org.apache.flink.table.api.scala.withColumns#apply}} always accept e.g.
> {{ColumnRange}} that could always accept only {{Integer}} or {{String}}
> instead of Expressions. Such class could look like (this is just a very rough
> outline):
> {code}
> class ColumnRange {
> IndexRange idx(Integer idx);
> ReferenceRange ref(String reference);
> }
> class IndexRange {
> IndexRange to(Integer idx);
> }
> class ReferenceRange {
> ReferenceRange to(String ref);
> }
> {code}
> We could also have implicit conversion from {{scala.Range}} to {{IndexRange}}.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
