fix: don't look for overshadow conflicts for symbols not in the scope

[kasiaMarek]

fixes: #6924

Before:

We search for overshadowed symbols before performing inline - names, which would refer to different symbol after inline, by looking at symbols available in the new scope.

Problem:

For symbol scala/Int#==() , conflicting scala/Any#==() was found.

Now:

Since symbols for methods and values on a class/type instance that is not an owner of the inlined method are not in the scope, we don't check them for conflicts.

[tgodzik]

What about:

class O {
  val t = new T()
  import t._
  val bb = a + a

  class Inner {
    val a = 123
    val cc = <<bb>>
  }
}

both a's have a type as parent. We should differentiate Select and Ident. I think the logic is correct on Select, but not on Ident

[kasiaMarek]

I found another example, where this logic fails but in a different way.

object O {
  def ==(o: O) = false
}
object P {
  def test() = {
   val isOne = O == O
    <<i>>sOne //claims it cannot inline
  }
}

Being concerned only with the first qualifier for select (this might need some special logic) should be a better approach. Ident was for sake of infix methods, I guess we should detect them and handle appropriately. (infix is also Select)

However, this will fail to cover overshadow connected to extension methods and implicit classes. But lets maybe make it into a followup issue.

[tgodzik]

However, this will fail to cover overshadow connected to extension methods and implicit classes. But lets maybe make it into a followup issue.

Yeah, we should be able to check that separately.

[kasiaMarek]

Also in your example reference to a gets turned into a Select in the typed tree.

[tgodzik]

That's not great then 🤔 Maybe we can check the ident span in Select(ident: Ident, sel) to see if it actually exist in the code ?

[kasiaMarek]

I changed the approach and I check if symbols are in scope before inline and look for conflicts on these symbols. Seems like the simplest solution.

[tgodzik]

LGTM!