Context bound desugared at the wrong place in extension method

[rjolly]

Compiler version

3.0.0-RC1

Minimized code

trait Conv[T]

extension [T: Conv](x: T)
  def combine (y: T): T = ???

Compiler output with -Xprint:typer

    extension [T >: Nothing <: Any](x: T) def combine(y: T)(
      implicit evidence$1: test.Conv[T]
    ): T = ???

Expectation

    extension [T >: Nothing <: Any](x: T)(using x$2: test.Conv[T]) def combine(
      y: T
    ): T = ???

, obtained with "using" syntax:

extension [T](x: T)(using Conv[T])
  def combine (y: T): T = ???

[odersky]

The current definition and implementation looks more useful since it allows to pass a Conv[T] explicitly.

[prolativ]

@odersky IMHO the suggested desugaring makes more sense and it still makes it possible to pass Conv[T] explicitly although the order of arguments would be different

"abc".combine(using given_Conv_String)("def")

[neko-kai]

@odersky
Useful or not, I believe it's unintuitive, irregular and without precedent.

The precedent that def x[T: U](a: A) is interchangeable with def x[T](a: A)(using U[T]) is broken here in a way that's very hard to explain when writing, teaching or even performing code reviews.

How would I explain why I wrote using U[T] in once place where [T: U] would obviously be shorter to someone reviewing my changes? That would require writing a whole paragraph explaining the difference in type inference, and why the context bound would not work just for this particular case even though usually it means the same thing. This is just for a minor and strange syntactic glitch, it would be very hard to justify why there's a difference in this particular in an otherwise completely intiuitive syntax.

[rjolly]

@odersky To pass it explicitely you also have the raw form combine("abc")(using given_Conv_String)("def").

[rjolly]

@odersky I understand it if you don' have time for this, but can we keep the issue open so that others can take over. Thanks!