Regression in m2-oss/calypso - rhs of a given with type bounded param unable to resolve certain other givens

[jchyb]

Compiler version

3.4.0-RC1-bin-20240104-2746ee8-NIGHTLY (last successful build 3.4.0-RC1-bin-20231207-16f1680-NIGHTLY), via the community build: https://github.com/VirtusLab/community-build3/actions/runs/7427597656/job/20238015937

Minimized code

trait GeneratedEnum
trait Decoder[A]

object Decoder:
  given Decoder[Int] = ???

object GeneratedEnumDecoder:

  given [A <: GeneratedEnum]: Decoder[A] =
    summon[Decoder[Int]]
    ???

Output

[error] -- [E172] Type Error: /Users/jchyb/Documents/workspace/calypso/modules/scalapb/src/main/scala/ru/m2/calypso/scalapb/GeneratedEnumDecoder.scala:10:24 
[error] 10 |    summon[Decoder[Int]]
[error]    |                        ^
[error]    |No given instance of type Decoder[Int] was found for parameter x of method summon in object Predef.
[error]    |I found:
[error]    |
[error]    |    GeneratedEnumDecoder.given_Decoder_A[A]
[error]    |
[error]    |But given instance given_Decoder_A in object GeneratedEnumDecoder does not match type Decoder[Int].
[error] one error found

Expectation

no errors

[odersky]

What happens here is that the given in DeneratedDecoder itself is picked as the resolved instance. The whole problem is solved with #19392 but one has to compile with -source future.

It seems what happens is that previously the given in GeneratedEnumDecoder was rejected since Int does not conform to GeneratedEnum. But maybe now the bounds are not checked eagerly (I am only guessing). Can we get a bisect that shows where things go wrong?

[odersky]

It's actually #19392 that broke it. I.e. it now works with -source future, but not now. I'll look into it.