Re: Arrays of custom classes

[Stamatis Zampetakis <za...@gmail.com>]

Hi Matt,

Since you end-up with SqlTypeName.OTHER I guess when you create the type
you use RelDataTypeFactory#createJavaType [1].
Have you tried using RelDataTypeFactory#createStructType [2]?

There are already use-cases using combinations of arrays with structs and
Avatica seems to handle them without problem so it may be worth giving it a
try.

Best,
Stamatis

[1]
https://github.com/apache/calcite/blob/2088488ac8327b19512a76a122cae2961fc551c3/core/src/main/java/org/apache/calcite/rel/type/RelDataTypeFactory.java#L61
[2]
https://github.com/apache/calcite/blob/2088488ac8327b19512a76a122cae2961fc551c3/core/src/main/java/org/apache/calcite/rel/type/RelDataTypeFactory.java#L87

On Thu, Jul 30, 2020 at 11:22 PM Matt Burgess <ma...@apache.org> wrote:

> Hi all,
>
> Does anyone have an example of a Calcite table with a column with
> array type where the elements are "custom" Java classes (not in
> JavaToSqlTypeConversionRules)?
>
> I'm running into an issue in Apache NiFi where we have our own type
> "Record" and our table model can have a column of type Record and/or a
> column of type Array[Record]. For the simple Record case, the SQL type
> of Record comes back as OTHER, and since that's the only custom type
> in our system, we presume that OTHER means Record and the object we're
> looking at is a Record, and we proceed and things seem to work well
> for that case.
>
> I thought the same would go for Array[Record], but if I try to SELECT
> that column, I get an error:
>
> java.lang.IllegalStateException: Unhandled ARRAY component type:
> OBJECT, id: 1111
>
> Looking at the code, it appears that at some point
> RelDataTypeFactoryImpl tries to get a SqlTypeName for our Record
> class, and since it's not in the mapping in
> JavaToSqlTypeConversionRules, it returns SqlTypeName.OTHER. That maps
> to java.sql.Types.OTHER (1111), and the AbstractCursor.ArrayAccessor
> doesn't like it.
>
> I can wrap the Record.class JavaType in a MultisetType to get things
> like CARDINALITY to work, but that's a hack and I still can't SELECT
> the Array[Record] column.
>
> Should SqlTypeName.OTHER map to JAVA_OBJECT instead? Am I approaching
> this correctly? If so is there a workaround and if not, can you help
> me wrap my head around this? :) For reference, [1] is the Jira case
> I'm working on.
>
> Thank you,
> Matt
>
> [1] https://issues.apache.org/jira/browse/NIFI-7601
>