Should inline matches strip type ascription?

[nicolasstucki]

Compiler version

3.0.0.M3

Minimized example

inline def meth =
  inline ("a": Any) match
    case _: String => "ok"

def test = meth

Output

-- Error: Foo.scala:6:11 -------------------------------------------------------
6 |def test = meth
  |           ^^^^
  |           cannot reduce inline match with
  |            scrutinee:  "a":Any : Any
  |            patterns :  case _:String
  | This location contains code that was inlined from Foo.scala:3

Expectation

Should this code compile? It looks like it should. Not sure if there are use-cases where we would want to avoid this.

[nicolasstucki]

@odersky @OlivierBlanvillain WDYT?

[abgruszecki]

I think at one point we discussed that inline matches should try to re-type their scrutinees as accurately as possible before matching. Stripping type ascriptions seems to be a part of that.

[nicolasstucki]

We do retype the scrutinee. But as the ascription is not removed/ignored the type will not get more precise.

[odersky]

Not sure. Is there a reason why one would need to add : Any to the scrutinee?

[abgruszecki]

A type ascription could prevent a case from matching, so that's one usage. At the same time, one should be able to use .asInstanceOf to accomplish the same. I'd say that the usefulness of stripping type ascriptions is more important than being able to prevent a case from matching. For instance, one may add a type ascription to the value returned by one inline, and I wouldn't expect that to alter how a different inline reduces.