SI-6493 Fix existential type used to hide local classes

In:

    def foo = { object O { class C }; new O.C }

The inferred return type of can't refer to the local
symbols; they must not leak out of the compilation unit.

Instead, `packSymbols` is used to choose a less precise
existential type, `(AnyRef { type C <: AnyRef })#C)`.

This is implemented as a `TypeMap`, which is supposed to
takes care of rebinding `C` to something valid in the new
prefix (`(AnyRef { type C <: AnyRef })`).

If `C` was orginally a type alias, and the original
prefix was a refinement type, this was handled in
`AliasTypeRef#coevolveSym`, which looks for a type
in the new prefix with the name `C`. But for other
type refs (e.g. a class type ref, as in this case),
a no-op `coevolveSym` was used, deferring the rebinding
of abstract classes until `typeRef`.

But our case falls between the cracks, and we end up
propagating a type ref in which the prefix does not
contain the member.

With the help of `-uniqid`, this is clear:

    <method> def foo#7445(): scala#21.this.AnyRef#2222{type Bar#12153 <: scala#21.this.AnyRef#2222}#Bar#12125

Notice the reference to symbol `#12125`, rather than `#12153`.

This commit moves the `coevolveSym` logic up to `TypeRef`,
generalizing it to work for any `pre1`. This example answered
the question in that code:

> // TODO: is there another way a typeref's symbol can refer to a symbol defined in its pre?

Author: retronym

src/reflect/scala/reflect/internal/Types.scala

@@ -2177,19 +2177,6 @@ trait Types
     // appliedType(sym.info, typeArgs).asSeenFrom(pre, sym.owner)
     override def betaReduce = transform(sym.info.resultType)
 
-    // #3731: return sym1 for which holds: pre bound sym.name to sym and
-    // pre1 now binds sym.name to sym1, conceptually exactly the same
-    // symbol as sym.  The selection of sym on pre must be updated to the
-    // selection of sym1 on pre1, since sym's info was probably updated
-    // by the TypeMap to yield a new symbol, sym1 with transformed info.
-    // @returns sym1
-    override def coevolveSym(pre1: Type): Symbol =
-      if (pre eq pre1) sym else (pre, pre1) match {
-        // don't look at parents -- it would be an error to override alias types anyway
-        case (RefinedType(_, _), RefinedType(_, decls1)) => decls1 lookup sym.name
-        // TODO: is there another way a typeref's symbol can refer to a symbol defined in its pre?
-        case _                                           => sym
-      }
     override def kind = "AliasTypeRef"
   }
 
@@ -2313,7 +2300,20 @@ trait Types
 
     // only need to rebind type aliases, as typeRef already handles abstract types
     // (they are allowed to be rebound more liberally)
-    def coevolveSym(pre1: Type): Symbol = sym
+    // SI-6493 Wrong! we also have to do this when existentially abstracting from `Foo.Bar` to `{ type Bar }#Bar`
+
+    // #3731: return sym1 for which holds: pre bound sym.name to sym and
+    // pre1 now binds sym.name to sym1, conceptually exactly the same
+    // symbol as sym.  The selection of sym on pre must be updated to the
+    // selection of sym1 on pre1, since sym's info was probably updated
+    // by the TypeMap to yield a new symbol, sym1 with transformed info.
+    // @returns sym1
+    final def coevolveSym(pre1: Type): Symbol =
+      if (pre eq pre1) sym else pre1 match {
+        // don't look at parents -- it would be an error to override alias types anyway
+        case RefinedType(_, decls1) => decls1 lookup sym.name
+        case _                      => sym
+      }
 
     //@M! use appliedType on the polytype that represents the bounds (or if aliastype, the rhs)
     def transformInfo(tp: Type): Type = appliedType(asSeenFromOwner(tp), args)

test/files/run/t6493/A_1.scala

@@ -0,0 +1,3 @@
+object one {
+  def foo() = { object Foo { class Bar } ; new Foo.Bar }
+}

test/files/run/t6493/B_2.scala

@@ -0,0 +1,3 @@
+object Test {
+  def main(args: Array[String]): Unit = one.foo
+}