Problem with argument types (already fixed in 0.4)

[lmshk]

In 0.3.2, when defining

f{T}(x::T, ::Type{T}) = println("Yes: ", x)
f(x, ::Type) = println("No: ", x)

I would expect the first method to match iff x' type is a subtype of the second argument. It works for

f(1, Int) # prints "Yes: 1"
f(1.0, Int) # prints "No: 1.0"

but not for

f(1, Number) # prints "No: 1"

This is actually works in 0.4. @simonster said that this should probably also work in 0.3. Can it be backported? (I would be willing to do this myself if someone points me in the right direction.)

[jiahao]

I think this works in 0.4 because of the change to call constructors in #8712; the change in diagonal dispatch behavior came up in #8974 (comment). I don't think it makes sense to backport such a large change to the language.

[JeffBezanson]

Yes this was a deliberate change for 0.4. It changes method applicability and is definitely not a bug fix, so this can't be backported.