Spurious "cannot be checked at runtime" in latest 2.12.x (regression wrt. 2.12.8)

[sjrd]

Trying to build the latest 0.6.x branch of Scala.js with the latest 2.12.x, I get a warning that I do not get with 2.12.8.

Repro:

$ git clone [email protected]:scala-js/scala-js.git
$ cd scala-js
$ git checkout fa02c74
$ sbt
> set resolvers in Global += "scala-integration" at "https://scala-ci.typesafe.com/artifactory/scala-integration/"
> ++2.12.9-bin-ae27fe9
> compiler/compile
[...]
[warn] compiler/src/main/scala/org/scalajs/core/compiler/GenJSCode.scala:4421: The outer reference in this type test cannot be checked at run time.
[warn]         case JSUnaryOpMethodName(code) if argc == 0 =>
[warn]                                 ^

The line mentioned is here:
https://github.com/scala-js/scala-js/blob/fa02c745a64b869127b8cb68d594f5ca6c1b2d4e/compiler/src/main/scala/org/scalajs/core/compiler/GenJSCode.scala#L4421

The warning was not reported by 2.12.8.

Note that the next case with JSBinaryOpMethodName, which is extremely similar, does not emit a warning.

A similar problem exists on our master branch (commit scala-js/scala-js@9f4f59b), where this becomes an error because we have -Xfatal-warnings.

[eed3si9n]

It's 2 weeks before 2.12.9, but it doesn't look like there's an activity on this, so I am bumping this to 2.12.10.

[lrytz]

This is caused by one of the commits in scala/scala@8257b83...ae27fe9. With the parent merge commit (8257b83) the warning doesn't appear. @retronym

[Jasper-M]

Not a clue if that's the cause—or how it could possibly be the cause—but it seems that name.newName(str) no longer retains the "cachedness" of the Name.

[retronym]

Here's a smaller repro:

// scalac: -Xfatal-warnings -unchecked
object Test1 {
    val g: scala.tools.nsc.Global = ???
    import g._
    def test(sym: Symbol) = sym.name match {
        case _: TermName  =>
    }
}

ExplicitOuter opts not to emit an outer test anymore because although TermName has an outer accessor, it is now a final class. This is somehow tied up in the morass of #4440.

Ideally the compiler would be smart enough to avoid the runtime test in this case as we statically know that the prefixes are the same. But sym.name has an exotic type: sym.NameType an abstract type defined by >: Null <: Test1.g.Name. The code that tries to statically unify the prefixes doesn't deal with this correctly.