John Roesler created KAFKA-7267:
-----------------------------------
Summary: KafkaStreams Scala DSL process method should accept a
ProcessorSupplier
Key: KAFKA-7267
URL: https://issues.apache.org/jira/browse/KAFKA-7267
Project: Kafka
Issue Type: Improvement
Reporter: John Roesler
scala.KafkaStreams#process currently expects a ()=>Processor, which is
semantically equivalent to a ProcessorSupplier, but it's the only such method
to do so.
All the similar methods in the Scala DSL take a Supplier like their Java
counterparts.
Note that on Scala 2.12+, SAM conversion allows callers to pass either a
function or a supplier when the parameter is a ProcessorSupplier. (But if the
parameter is a function, you must pass a function)
But on scala 2.11-, callers will have to pass a function if the parameter is a
function and a supplier if the parameter is a supplier. This means that
currently, 2.11 users are confronted with an api that demands they construct
suppliers for all the methods *except* process, which demands a function.
Mitigating factor: we have some implicits available to convert a Function0 to a
supplier, and we could add an implicit from ProcessorSupplier to Function0 to
smooth over the API.
What to do about it?
We could just change the existing method to take a ProcessorSupplier instead.
* 2.12+ users would not notice a difference during compilation, as SAM
conversion would kick in. However, if they just swap in the new jar without
recompiling, I think they'd get a MethodDefNotFound error.
* 2.11- users would not be able to compile their existing code. They'd have to
swap their function out for a ProcessorSupplier or pull the implicit conversion
into scope.
* Note that we can delay this action until we drop 2.11 support, and we would
break no one.
We could deprecate the existing method and add a new one taking a
ProcessorSupplier.
* All scala users would be able to compile their existing code and also swap
in the new version at runtime.
* Anyone explicitly passing a function would get a deprecation warning,
though, regardless of SAM conversion or implicit conversion, since neither
conversion won't kick in if there's actually a method overload expecting a
function. This would drive everyone to explicitly create a supplier
(unnecessarily)
We could leave the existing method without deprecating it and add a new one
taking a ProcessorSupplier.
* All scala users would be able to compile their existing code and also swap
in the new version at runtime.
* There would be no unfortunate deprecation warnings.
* The interface would list two process methods, which is untidy.
* Once we drop 2.11 support, we would just drop the function variant.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
