No need to implement it. http://dlang.org/phobos/std_algorithm.html#.sum
It isn't accurate. Python's fsum is around 100 lines of c-code and AFAIK based on this algorithm:
http://www.cs.berkeley.edu/~jrs/papers/robustr.pdf more: https://hg.python.org/cpython/file/7c183c782401/Modules/mathmodule.c#l1063 http://www-pequan.lip6.fr/~graillat/papers/rr2005-03.pdf But I think an integer version is faster on a large dataset.
