On Tue, May 5, 2015 at 12:11 PM, Yichao Yu <[email protected]> wrote:
> Seems to be related to `typeintersect` and the error also disappears
> when the order of defining those functions are reversed
>
> Minimum working example:
>
> ```julia
> type A
> end
>
> type B
> end
Note that change A and/or B to abstract type doesn't change the
output. I use concrete type here just to make sure there's absolutely
no chance that the two can be a subtype of each other.
>
> f{T}(::Type{T}, ::T) = nothing
> f{T<:B}(::Type{A}, ::T) = nothing
>
> g{T<:B}(::Type{A}, ::T) = nothing
> g{T}(::Type{T}, ::T) = nothing
>
> t1 = f.env.defs.sig
> t2 = f.env.defs.next.sig
>
> println("Signatures: $t1, $t2")
> println("Intersect: $(typeintersect(t1, t2))")
> println("Intersect: $(typeintersect(t2, t1))")
> ```
>
> Output (on my branch that fixes the output formatting....)
>
> ```
> Warning: New definition
> f(Type{A}, T<:B) at /home/yuyichao/tmp/julia/type_intersect.jl:10
> is ambiguous with:
> f(Type{T}, T) at /home/yuyichao/tmp/julia/type_intersect.jl:9.
> To fix, define
> f(Type{A}, A)
> before the new definition.
> Signatures: Tuple{Type{T},T}, Tuple{Type{A},T<:B}
> Intersect: Union()
> Intersect: Tuple{Type{A},A}
> ```
>
> Should probably be reported as bug?
>
>
> On Tue, May 5, 2015 at 7:09 AM, Avik Sengupta <[email protected]> wrote:
>> Thanks Tovio
>>
>>> unless JString <: String (but I guess it isn't?).
>>
>> No, indeed, JString is not a subtype of a Julia string. It is simply a
>> wrapper around a pointer. It has no string-like behaviour in Julia, other
>> than being able to be converted back and forth from a Julia string.
>>
>>> it seems to me that your method can never be an identity conversion
>>
>> I could write a method definition such as
>>
>> convert(::Type{JString}, str::JString) = str
>>
>> I suppose that should be safe, but I don't understand why that should be
>> necessary, or what benefit it will have.
>>
>> On Tuesday, 5 May 2015 11:38:38 UTC+1, Toivo Henningsson wrote:
>>>
>>> This looks like a julia bug, unless JString <: String (but I guess it
>>> isn't?).
>>>
>>> The method from Base is the generic identity conversion, but it seems to
>>> me that your method can never be an identity conversion?
