Inconsistent restriction when subtyping using typealias

[martinholters]

julia> abstract Foo{T}

julia> typealias Bar{T} Foo{T}
Foo{T}

julia> typealias Baz Foo
Foo{T}

julia> type A <: Foo; end

julia> type B <: Bar; end
ERROR: invalid subtyping in definition of B

julia> type C <: Baz; end

Arguably, having a supertype with unfixed parameters is not very useful (or am I missing something?), but if it works for A and C, why shouldn't it for B?

Somewhat more concrete:

julia> type UntypedVector <: AbstractVector; end
ERROR: invalid subtyping in definition of UntypedVector

julia> type UntypedVector <: AbstractArray{TypeVar(:T),1}; end # then try this instead

julia> supertype(UntypedVector) == AbstractVector
true

The root cause seems to be that in the cases that work, the given supertype is a DataType, while in those that don't, it is a TypeConstructor. Might make sense the way it is, but for a casual observer, above examples just look really inconsistent.

[JeffBezanson]

closed by #18457 - Now all of these cases are disallowed.

[tkelman]

@JeffBezanson you have or are adding test cases for each of these issues you're closing, right?