typealias bug?

[StefanKarpinski]

julia> typealias SelfDict{T} Dict{T,T}
Dict{T,T}

julia> isself(d::Dict) = false
# methods for generic function isself
isself(d::Dict{K,V}) at none:1

julia> isself(d::SelfDict) = true
# methods for generic function isself
isself(d::Dict{T,T}) at none:1

julia> isself([1=>2])
true

julia> isself([1=>"foo"])
true

[JeffBezanson]

Yeah this isn't implemented yet. We need to convert it to the normal form isself{T}(d::SelfDict{T}) which is a bit tricky.

[mbauman]

Is this the same root cause as the following method ordering bug?

julia> type A{T<:Integer}; end

julia> f{T}(::A{T}) = 1      # Julia doesn't consider that T must be <: Integer here
       f(::A, ::Any...) = 2  # But implicitly does so here
f (generic function with 2 methods)

julia> f(A{Int}())           # So method 2 is more specific
2

julia> g{T<:Integer}(::A{T}) = 1 # But if I manually tell Julia about the restriction
       g(::A, ::Any...) = 2
g (generic function with 2 methods)

julia> g(A{Int}())  # Then dispatch works as expected
1

[yuyichao]

@mbauman Feels like a variant of #11597 ?

[mbauman]

Yes, that's much more applicable. I came here from #7453, which also seemed similar but was closed as a duplicate of this guy. It's not really a typealias problem though — it's a restricted type parameter. So I wasn't sure if the same restricting/unrestricting problem was at the root cause here.

[yuyichao]

Actually thanks for pinging this, since I actually notice exactly this a few days ago.

[vtjnash]

removing the milestone tag since this is being tracked through #8974

[JeffBezanson]

closed by #18457