• via Digitalmars-d-bugs (27/28) Feb 25 2016 https://issues.dlang.org/show_bug.cgi?id=15722

via Digitalmars-d-bugs <digitalmars-d-bugs puremagic.com> writes:

https://issues.dlang.org/show_bug.cgi?id=15722

          Issue ID: 15722
           Summary: std.algorithm sum should favour speed
           Product: D
           Version: D2
          Hardware: x86_64
                OS: Windows
            Status: NEW
          Severity: enhancement
          Priority: P1
         Component: phobos
          Assignee: nobody puremagic.com
          Reporter: adamsibson hotmail.com

From the Learn forum thread

https://forum.dlang.org/post/vkiwojmfjrwhigbkenaa forum.dlang.org it became
apparent that sum is four times slower (on DMD and LDC) than the naive loop sum
method and similarly slower than reduce!((a, b) => a + b).

Would it not be better to have a fast, basic implementation in algorithm and a
slower, more accurate version in a sub-library? The performance gap is too
large to justify the current version being the standard. I would suggest that
most users just want a trivial sum function and the more technical users are
capable of finding and using the more accurate version, especially if we
properly document the library. The accuracy of the basic version is also not
terrible, it's usually going to be good enough. This is not playing fast and
loose with accuracy just for the sake of a few percent speed.

This kind of poor performance iceberg reflects poorly on D for new users. It's
not the right out of box experience.

--