SI-8301 fix regression with refinement subtyping, wildcard type.

In the enclosed test case, the list of candidate implicit is
checked for eligibility with the view by:

    pt = Test.sym.type => ?{def asFreeType: ?}
    tp = (sym: Test.u.Symbol)Test.u.CompatibleSymbol
    matchesPt(tp, pt)

This ends up in `TypeComparers#specializesSym`, which compares:

   symLo = asFreeType: => Universe.this.TypeSymbol
   symHi = asFreeType: ?

Before the regression in fada1ef6, that method called `isStable`
on both of these to make sure that `symLo` was as least as stable
as `symHi`. Both returned `false`, and, they matched.

After that refactoring, two calls were made. `isStable` *only*
checked for stability, and gave the same results as before.
But, the new check for volatility believes that the low symbol
was more volatile than the high, because `WildCardType` is treated
as non-volatile.

Really, it should have the same volatility as whatever it is
being compared to, so that these sort of comparisons yield true.

This commit ammends `specializesSym` to special case `symHi: ?`,
and ignore the volatilty gap in that case.

1 files changed, 19 insertions, 0 deletions

diff --git a/test/files/pos/t8301.scala b/test/files/pos/t8301.scala
new file mode 100644
index 0000000000..2d10864c57
--- /dev/null
+++ b/test/files/pos/t8301.scala
@@ -0,0 +1,19 @@
+trait Universe {
+  type Symbol >: Null <: AnyRef with SymbolApi
+  trait SymbolApi
+
+  type TypeSymbol >: Null <: TypeSymbolApi with Symbol
+  trait TypeSymbolApi
+
+  implicit class CompatibleSymbol(sym: Symbol) {
+    def asFreeType: TypeSymbol = ???
+  }
+}
+
+object Test extends App {
+  val u: Universe = ???
+  import u._
+
+  val sym: Symbol = ???
+  sym.asFreeType
+}