Without this, the unrelated type specified on the LHS of a property reference
literal was considered to be an extension receiver of the candidate, and the
resolution was erroneously successul. This is only reproducible for properties,
because if we're trying to resolve an extension, we consider all properties
from the scope, even non-extensions, because there may be a property of an
extension-functional type (T.() -> R). (We don't do this for functions.)
#KT-7430 Fixed
#KT-7945 Fixed
For example,
KMemberFunction2<T, P1, P2, R> : KMemberFunction<T, R>
So for this inheritance the heuristic that was present ("take the last K type
parameters of the subclass, and substitute for K parameters of the superclass")
was wrong. The new heuristic for this case is: take type parameters with the
same names.
Also don't store "parameters" in a lazy value, since it's easy to compute and
it's computed practically every time anyway
Metadata for KFunction classes is now longer serialized along with built-in
classes. This effectively means that it's no longer possible to find KFunction
classes via dependency on built-ins. There should be a kotlin-runtime library
in the specified classpath for reflection types to be resolvable.
A lot of tests were moved and changed, because tests on callable references
require stdlib in classpath from now on
It's not clear in which order should the two receiver arguments be passed to
them, and there's no corresponding K*FunctionN class for now
#KT-1183 In Progress
Alexander Udalov