Issues with Scala 2.13.7

[mfirry]

Hi all,
I think it's a bit weird that my code is not compiling with 2.13.7 but it does with 2.13.6.

Specifically with the following:

  val list = get(zero) {
    Ok(List(1, 2, 3).asJson)
  }

I now get

[error] /Users/pippo/personal/scala/web-frameworks-templates/finch/src/main/scala/WebServer.scala:22:17: overloaded method apply with alternatives:
[error]   (mapper: io.finch.internal.Mapper[cats.effect.IO,shapeless.HNil])io.finch.Endpoint[cats.effect.IO,mapper.Out] <and>
[error]   (input: io.finch.Input)io.finch.Endpoint.Result[cats.effect.IO,shapeless.HNil]
[error]  cannot be applied to (io.finch.Output[io.circe.Json])
[error]   val list = get(zero) {

(the complete code can be found here)

[rpless]

I'm not sure what changed in 2.13.7 to cause this, but this version seems to have broken a bunch of things in the ecosystem. For the endpoints that don't take arguments to their mapping function you can fix them by adding an empty argument block:

val list = get(zero) { () =>
  Ok(List(1, 2, 3).asJson)
}

However the sayHi endpoint is still broken and I'm not quite sure why.

[SethTisue]

This would have been caught ahead of time (before 2.13.7 was released) if Finch were in the Scala 2 community build.

I've opened a PR to add it, here: scala/community-build#1504

[gabro]

FWIW there seems to be something odd with

finch/core/src/main/scala/io/finch/internal/Mapper.scala

Lines 79 to 81 in 298889b

	implicit def mapperFromOutputValue[F[_], A](o: => Output[A])(implicit
	F: MonadError[F, Throwable]
	): Mapper.Aux[F, HNil, A] = instance(_.mapOutput(_ => o))

Given the error message, I would have expected the implicit conversion Output => Mapper to kick in, but it's clearly not happening

[mfirry]

I was wondering if there is any way to trace and log implicit resolution (I think there is) and the check it with 2.13.7 and 2.13.6

[joroKr21]

I was wondering if there is any way to trace and log implicit resolution (I think there is) and the check it with 2.13.7 and 2.13.6

You can enable the -Xlog-implicits compiler option for that

[SethTisue]

We're aiming to build 2.13.8 circa Dec 10 (I'll be announcing this soon on contributors.scala-lang.org), so time is running short for a minimized bug report if you were hoping this might get fixed in that release.

[rpless]

@SethTisue what would you consider ideal for a minimized bug report? Would extracting the failing finch examples from @mfirry's repo into a smaller self contained one be enough for a bug report on the Scala bug tracker? Or would it need to minimized further to exclude Finch entirely?

[SethTisue]

It’s strongly preferable to have a reproduction without an external JAR involved, so that we can turn it directly into a regression test.

[mfirry]

@rpless, happy to help. Let me know if this is useful => https://github.com/mfirry/finch-repro-1417

[rpless]

Thanks @mfirry. I'm hoping we can get a repro without dependencies, but we'll see.

Did a little digging this morning. One thing I noticed is that this still compiles:

  val sayHello: Endpoint[IO, Buf] = get("hello" :: param[String]("one") :: param[String]("two"))(
    (one: String, two: String) =>
      Ok(Buf.Utf8(s"Hello, $one. Hello $two")))

Which kinda tells me that any Mapper that needs to summon a FnToProduct.Aux from shapeless still works. I suspect that means the only ones that are affected are the instances that get a raw Output or are a single arity function.
I'm gonna try to reduce finch down to something that can demonstrate this, but I'm unsure I'll have anything before the next Scala 2.13 release.
I also think this won't be an issue in Scala 3, based on the fact that we can remove the magnet pattern in #1407, so thats something. @sergeykolbasov not sure if you have any thoughts on this.

[joroKr21]

Perhaps scala/scala#9691 surfaced a diverging implicit error

[joroKr21]

This would fix it:

--- a/src/compiler/scala/tools/nsc/typechecker/Infer.scala
+++ b/src/compiler/scala/tools/nsc/typechecker/Infer.scala
@@ -1101,10 +1101,10 @@ trait Infer extends Checkable {
                 enhanceBounds(adjusted.okParams, adjusted.okArgs, xs1)
                 xs1
             }
-          } else undetParams
+          } else if (context.implicitsEnabled) undetParams else Nil
         } catch ifNoInstance { msg =>
           NoMethodInstanceError(fn, args, msg)
-          undetParams
+          if (context.implicitsEnabled) undetParams else Nil
         }
       case x => throw new MatchError(x)
     }

Now all we need is a minimal reproduction and a ticket in scala/bug

[joroKr21]

What I missed in that PR was that if we are already typechecking an implicit view we can't recover with another implicit view.