Reproduces future value cast issue for scala 3

[mprevel]

Hi,

This PR is related to #906
I've added a test to reproduce the encountered issue.
It seems it may be related to derivation since adding the fields explicitly (commented code in the test) works, but I also encounter some strange behaviours while running the tests.

If I use +testQuick the new test fails for scala 3 on the 1st run, but it seems to pass on the next one (but I'm not sure it's running with scala 3, logs make me think it runs with scala 2.12)
If I use +testOnly test.issues.ticket_906.Issue906Spec*, it fails for scala 3 on each run.

[yanns]

Thanks for the reproduction!

The code generated for scala 2:

  sangria.schema.Field[test.issues.ticket_906.Issue906Context.MyRepository, test.issues.ticket_906.Issue906Context.SampleCaseClass, String, String](
    "myFutureString", 
    implicitly[sangria.macros.derive.GraphQLOutputTypeLookup[String]].graphqlType,
    scala.None,
    scala.collection.immutable.List(),
    ((c: sangria.schema.Context[test.issues.ticket_906.Issue906Context.MyRepository, test.issues.ticket_906.Issue906Context.SampleCaseClass]) => 
      c.value.graphqlMyFutureString),

The code generated for scala 3:

    sangria.schema.Field.apply[test.issues.ticket_906.Issue906Context.MyRepository, test.issues.ticket_906.Issue906Context.SampleCaseClass, scala.Any, scala.Any](
      "myFutureString",
      sangria.macros.derive.GraphQLOutputTypeLookup.outLookup[java.lang.String](sangria.schema.StringType).graphqlType,
      scala.Option.empty[java.lang.String],
      scala.Nil,
      ((`c₃`: sangria.schema.Context[test.issues.ticket_906.Issue906Context.MyRepository, test.issues.ticket_906.Issue906Context.SampleCaseClass]) =>
        ((`value₃`: scala.concurrent.Future[scala.Predef.String]) =>
          sangria.schema.Action.defaultAction[test.issues.ticket_906.Issue906Context.MyRepository, scala.concurrent.Future[scala.Predef.String]](`value₃`)).apply(`c₃`.value.graphqlMyFutureString)),

Instead of using the "simple"

((c: sangria.schema.Context[test.issues.ticket_906.Issue906Context.MyRepository, test.issues.ticket_906.Issue906Context.SampleCaseClass]) => 
      c.value.graphqlMyFutureString)

scala 3 is using something more complex:

      ((`c₃`: sangria.schema.Context[test.issues.ticket_906.Issue906Context.MyRepository, test.issues.ticket_906.Issue906Context.SampleCaseClass]) =>
        ((`value₃`: scala.concurrent.Future[scala.Predef.String]) =>
          sangria.schema.Action.defaultAction[test.issues.ticket_906.Issue906Context.MyRepository, scala.concurrent.Future[scala.Predef.String]](`value₃`)).apply(`c₃`.value.graphqlMyFutureString)),

It does not call graphqlMyFutureString anymore, but encapsulates the body of the function directly in the resolve?

[yanns]

for context, the PR that implemented the scala macros: #900

[yanns]

@jchyb pinging you if you have some quick ideas where to look at

[robberthcz]

I was looking at the code, and there is method collectFields in file DeriveObjectTypeMacro.scala. On line 124, the expression field.onType.memberType(field.method).widen.asType evaluates probably to Future[String] in case of method myFutureString.

Then on the line 132, the expression Expr.summon[t => Action[Ctx, t]] likely evaluates to something like Expr.summon[Future[String] => Action[Ctx, Future[String]]], which finds the implicit method Action.defaultAction and not the correct Action.futureAction.

It would correspond with the initial ClassCastException, where Sangria resolves the field probably as Future[Future[String]] and not the expected Future[String] at runtime.

Maybe I'm wrong, I don't really understand macros, so take it with a grain of salt. I will try to debug it more.

[yanns]

thanks for the investigation @robberthcz . It seems promising.
I also have no expertise with scala 3 macros, so I guess the only way to make progress is try and error (and debug)

[jchyb]

Apologies, it's been some time since I looked into the codebase here, so I cannot pinpoint anything yet. I'll try to see if this is something I can help with. I remember trying to match scala-2's behavior as much as possible, so any deviation here is probably not intended.

[jchyb]

I think I see the issue. In scala 2 the generated code does not have to type check, so we can rely on implicit conversions to do some of the job. So while we generate ((c: sangria.schema.Context[test.issues.ticket_906.Issue906Context.MyRepository, test.issues.ticket_906.Issue906Context.SampleCaseClass]) => c.value.graphqlMyFutureString) there, the c.value.graphqlMyFutureString result later gets wrapped in ReduceAction.futureAction() by the compiler:

implicit def futureAction[Ctx, Val](value: Future[Val]): ReduceAction[Ctx, Val] = FutureValue(
    value)

In scala 3 the generated code has to type check, so we summon it manually, but for some reason it summons defaultAction:

  implicit def defaultAction[Ctx, Val](value: Val): ReduceAction[Ctx, Val] = Value(value)

So it looks like the summon around line 330 might be wrong. I'll prepare a PR tomorrow

[yanns]

@jchyb you don't need to apologize, it's open source, and we're already so thankful to you for having done all the hard work for Scala 3! ❤️
And thank you again for your time here.

[yanns]

Can you all check that the new sangria release fixes your issue?

[robberthcz]

Big thanks. It fixed the issue for us.

[mprevel]

Can you all check that the new sangria release fixes your issue?

It's fixed, thanks.