typeintersect bug with NTuple

[vtjnash]

julia> typeintersect((Tuple{Vararg{Union{},A}} where A), Tuple{Int64, Vararg{Int64, N} where N}) !== Union{}
false

This is bad since the intersection is non-empty when A is later given:

julia> typeintersect((Tuple{Vararg{Union{},3}} where A), Tuple{Int64, Vararg{Int64, N} where N}) !== Union{}
true

[Keno]

I thought we at some point had Tuple{Union{},Union{},Union{}} == Union{}. Did we change that?

[JeffBezanson]

No, I don't think we ever had that.

[vtjnash]

It turns out that we do have that, but the type system is inconsistent right now. For example, typeintersect(Tuple{Int32}, Tuple{Int64}) gives Union{} not Tuple{Union{}}, even though Tuple{Union{}} is a subtype of both of those. Fixing this by forbidding Tuple{Union{}} from being constructed would help eliminate some complicated inference cases from existing like this (I don't know exactly how this occurred, but somehow it does):

julia> typeintersect(Tuple{typeof(convert), Type{T}, T} where T, Tuple{typeof(convert), Type{<:Symbol}, Any})
Tuple{typeof(convert), Type{T}, T} where T<:Symbol

# ^ could be changed to Tuple{typeof(convert), Type{Symbol}, Symbol}, since Symbol is a leaf type