Double eta-expansion when ascribing contextual types

[biboudis]

When we use a contextual type as a return type

class Test  {
   def f(): given Int => Boolean = true
}

it η-expands once ->

def f(): ImplicitFunction1[Int, Boolean] =  {
   def $anonfun(implicit evidence$1: Int): Boolean = true
   closure($anonfun)
}

However, if we provide an explicit ascription:

class Test  {
   def f(): given Int => Boolean = true : (given Int => Boolean)
}

it expands unnecessarily twice:

def f(): ImplicitFunction1[Int, Boolean] =  {
   def $anonfun(implicit evidence$1: Int): Boolean =  (
      {
         def $anonfun(implicit evidence$2: Int): Boolean = true
         closure($anonfun)
      }:ImplicitFunction1[Int, Boolean]).apply(evidence$1)
   closure($anonfun)
}

[nicolasstucki]

Note that this affects all inline methods that return a contextual function.

[anatoliykmetyuk]

This one's moved to a new milestone for the third time already...

[biboudis]

Will get back to it.

[anatoliykmetyuk]

I'll remove it from the milestones FTTB

[odersky]

This is hard to avoid, since there are a number of ways how we could close over a context parameter twice. I think it's easiest to detect and drop these patterns later in a miniphase.