Does anyone know if avro-codegen supports Value Classes in any way?
@ebiggs: IIRC the message trait has a type param so that a
GeneratedMessageCompanion[A] can require that A conform to [A <: GeneratedMessage with Message[A]], this lets it use the toMutable method defined on Message[A] as such:def toMutable(msg: A): org.apache.avro.generic.GenericRecord = msg.toMutable
@zczapran: avro-codegen does not support value classes. In the future I'd like to add support for custom types via something similar to ScalaPB's TypeMapper: https://trueaccord.github.io/ScalaPB/customizations.html
@ebiggs is currently working on a custom schema parser. When that's done we will be able to augment the avro schema specification to include an optional
scalaType that will function similarly to the protobuf annotation used by ScalaPB: [(scalapb.field).type = "mydomain.Seconds"].
@pkinsky and I will work on this tomorrow and get this sorted out
(1) make sure that the codegen part uses the new partial schema parser
(2) have it work with both .avsc and .avsp
(3) update the end-to-end tests to include partial schemas, verify that they are parsed correctly
(4) update the .travis.yml to include "sbt test"
when we fix, we'll release
0.3.1
hey so does anybody else feel like it might be nice to move the Av* stuff into runtime, and put the string literals as partials into a central file and then replace the long canonical strings in each generated class file with a call to AvModule?
that way there's no repetition in the string literals.
Regarding your commit message about namespaces, I want to explain what I ran into and maybe we can come up with a better solution... The avro specification makes a distinction between no namespace and the null namespace. The null namespace is referred to by an empty string, And specifying no namespace means to inherit from the parent context. However, if there is no parent context, then use the null namespace. Unfortunately there's no way to distinguish between these two different semantics using the fully qualified name, but the Av* AST does make a distinction. So if you specified namesakes == None, it infers the null namespace and a reference that refers to None will not work.. I think there's probably a way to clean this up I just didn't take the time to think it through and write the tests etc..
wrt to the runtime stuff @ebiggs : I think that could work. It is generated code, though, so I don't know how much of a burden this change will lift off our collective shoulders... You're advocating this just to reduce repetition?
wrt namespaces: ah, I didn't realize that distinction in the spec
with that, I think what I had earlier was a test exercising the null namespace...but without actually using null.<name_of_schema>
so that e2e test case was actually probably wrong to begin with...
With this new info, I'm not sure if this namespace thing is as important. IMO everyone should put their schemas into namespaces. Maybe we could have an option to fail the codegen if we encounter a schema w/o a namespace?
this does seem like a more minor issue, right? so we could open it up and maybe flesh out some details there
Yeah it's minor, I mean it doesn't bother me, but it might help make the software more approachable to newbies
but essentially want I want to be able to do is have it codegen case class that take a type parameter, which when specified will create different Avro schema depending on the parameter you call it with
oh interesting...
I'll catch up on messages in a little while