Hello all!

Python's fsum is ready for D :
https://github.com/D-Programming-Language/phobos/pull/2991

Destroy!

On 2/26/15 5:34 AM, Ilya Yaroshenko wrote:
> Hello all!
>
> Python's fsum is ready for D :
> https://github.com/D-Programming-Language/phobos/pull/2991
>
> Destroy!

Unfortunate overlap in functionality is unfortunate.

$ git grep -i kahan
std/algorithm/iteration.d:$(D sum) uses the $(WEB en.wikipedia.org/wiki/Kahan_summation,
std/algorithm/iteration.d:Kahan summation) algorithm.)
std/algorithm/iteration.d:            return sumKahan!E(seed, r);
std/algorithm/iteration.d:// Kahan algo http://en.wikipedia.org/wiki/Kahan_summation_algorithm
std/algorithm/iteration.d:private auto sumKahan(Result, R)(Result result, R r)
$ _


Andrei

> Unfortunate overlap in functionality is unfortunate.
>
> $ git grep -i kahan
> std/algorithm/iteration.d:$(D sum) uses the $(WEB en.wikipedia.org/wiki/Kahan_summation,
> std/algorithm/iteration.d:Kahan summation) algorithm.)
> std/algorithm/iteration.d:            return sumKahan!E(seed, r);
> std/algorithm/iteration.d:// Kahan algo http://en.wikipedia.org/wiki/Kahan_summation_algorithm
> std/algorithm/iteration.d:private auto sumKahan(Result, R)(Result result, R r)
> $ _
>
>
> Andrei

Would one of the following solutions be fortunate ?

I. Use alias of fsum!(Summation. Appropriate) for floating version of std.algorithm.sum.

    Appropriate algorithm performs D Pairwise summation for random access ranges.
    Otherwise performs D KBN summation of floating point and complex numbers and
    Kahan summation of user defined types.

or

II. Move this submodule to std.algorithm.iteration.

or

III. Remove Kahan (including KBN and KB2) and Pairwise algorithms from fsum, and use only Precise algorithm for std.numeric.

On Friday, 27 February 2015 at 09:27:49 UTC, Ilya Yaroshenko
wrote:
>> Unfortunate overlap in functionality is unfortunate.
>>
>> $ git grep -i kahan
>> std/algorithm/iteration.d:$(D sum) uses the $(WEB en.wikipedia.org/wiki/Kahan_summation,
>> std/algorithm/iteration.d:Kahan summation) algorithm.)
>> std/algorithm/iteration.d:            return sumKahan!E(seed, r);
>> std/algorithm/iteration.d:// Kahan algo http://en.wikipedia.org/wiki/Kahan_summation_algorithm
>> std/algorithm/iteration.d:private auto sumKahan(Result, R)(Result result, R r)
>> $ _
>>
>>
>> Andrei
>
> Would one of the following solutions be fortunate ?
>
> I. Use alias of fsum!(Summation. Appropriate) for floating version of std.algorithm.sum.
>
>     Appropriate algorithm performs D Pairwise summation for random access ranges.
>     Otherwise performs D KBN summation of floating point and complex numbers and
>     Kahan summation of user defined types.
>
> or
>
> II. Move this submodule to std.algorithm.iteration.
>
> or
>
> III. Remove Kahan (including KBN and KB2) and Pairwise algorithms from fsum, and use only Precise algorithm for std.numeric.

Probably, I have found good solution. Summator struct can be
represented in std.numeric and std.algorithm.sum can be extended
by the new functionality. So there will be no overlap in
functionality.