ObjectTpeJava is truly unique; bounds; isAnyTpe

[adriaanm]

Make sure the TypeRef is truly unique, even when embedded
in other types, such as a TypeBounds. Since we're dealing
with structural equality, TypeBounds(lo, hi) == TypeBounds(lo', hi')
if lo == lo' and hi == hi', if hi = ObjectTpe and hi' = ObjectTpe',
unique'ing (hashconsing) such a TypeBounds would indirectly
replace ObjectTpe with ObjectTpeJava or vice versa.

We can't use eq when trying to identify ObjectTpeJava in unique
due to cycles.

isAnyTpe returns true for ObjectTpeJava.

Further fixes by Lukas:

• Don't show ObjectTpeJava upper bound in TypeBounds.toString
• Upper bounds of wildcards should be ObjectTpeJava
• Fix wildcardExtrapolation for BoundedWildcardType with hi ObjectTpeJava

Reduce _ >: lo <: ObjectTpeJava to lo. This eliminates the
BoundedWildcardType in the expected type when type-checking the
lambda

(c: java.util.Collection[String]).removeIf(x => true)

where

boolean removeIf(Predicate<? super E> filter)

Follow up for #7966

Fix scala/bug#11525

[adriaanm]

Strawman fix. Not sure this is ok for performance, but the test reliably shows that if we don't do something like this, unique will collapse our two ObjectTpe variants.

ObjectTpeJava could still be replaced by ObjectTpe when uniqueing a Type that contains ObjectTpeJava, since e.g. TypeBounds(...ObjectTpeJava...).equalsTypeBounds(...ObjectTpe...) and thus uniqueing the TypeBounds will also replace our precious ObjectTpeJava by ObjectTpe.

[retronym]

This fix exposed another bug in my attempt at this: Optional<?> in Java code should be translated to Optional[_ <: Object[Java]]. Pushed an attempted fix.

[retronym]

Minimisation of the latest mole:

class Test {
  def foo(c: java.util.Collection[String]): Unit = {
    c.removeIf(x => true)
  }
}

[adriaanm]

Can you add an example for the importance of distinguishing AnyTpe and ObjectTpeJava here?

[retronym]

/synch

[retronym]

/synch