Thanks for filing the issue, it's clearer the way you've written it.
Yes, I agree that one shouldn't expect Julia to be faster for untyped
arrays than Python. I just didn't expect Julia to be more than three times
slower than Python.
On Monday, January 12, 2015 at 5:19:00 PM UTC+1, Steven G. Johnson wrote:
>
>
>
> On Monday, January 12, 2015 at 8:36:28 AM UTC-5, Andras Niedermayer wrote:
>>
>> It's still not entirely satisfactory that sorting arrays of tuples is so
>> much slower in Julia than in Python,
>>
>
> (Note that for sorting an untyped array of tuples, Julia may never have
> much if any advantage over Python. Python is already using C code for the
> sort implementation and C code to unbox the type and call the comparison
> operation. But I agree that for typed arrays of tuples Julia could, in
> principle, be faster, especially if you specify similar sort-stability
> requirements in both languages.)
>
>
>> @StefanKarpinski Stable vs unstable sort explains the performance
>> difference between `type Pair` and `type Pair <: Number`, but it would
>> suggest the opposite of what one sees for sorting `Vector{(Any,Any)}`
>> (faster) vs `Vector{(Float64,Int64)}` (slower)
>>>
>>>
> I agree that adding type information should not slow things down. I've
> filed an issue https://github.com/JuliaLang/julia/issues/9741
>
