typealiases with parameter no longer works as expected

[dlfivefifty]

The following returns true in 0.3 but false in 0.4

abstract Foo{T,S}
immutable B{T} end
typealias FooB{T} Foo{T,B}
immutable CFoo <: FooB{Float64}end
isa(CFoo(),FooB)

[ivarne]

I can confirm that the behaviour difference between 0.3 and 0.4.

I'm confused with such complicated types, so others will have to judge if this is a regression or a bugfix.

[stevengj]

Seems like a bug to me. If CFoo <: FooB{Float64} then we should always have CFoo <: FooB, but the latter returns false in Julia 0.4.

[elextr]

If CFoo <: FooB{Float64} implies the more general CFoo <: FooB does that not then imply then imply that CFoo <: FooB{Int64} or other more specific types, and is that the correct behaviour?

[stevengj]

@elextr, no it does not imply that, because FooB{Int64} <: FooB, not the other way around. A triangle is a polygon and a square is a polygon, but it does not follow that a triangle is a square.

[elextr]

D'oh, yeah FooB{Int64} <: FooB is down a different unrelated branch of the subtype tree from FooB{Float64} <: FooB.

[dlfivefifty]

To motivate it getting fixed: ApproxFun.jl won't work in 0.4 until it is