std.math unittests - accuracy of floating point results

[Dan Olson via digitalmars-d-ldc]

"Johan Engelen" <j at j.nl> writes:

> Hi all,
>  I am working on making more of the std.math unittests pass (I'm new
> to the project, and it is a nice niche thing to tinker on, learning
> the codebase, workflow, etc.).
> I've hit on a problem that I do not know how to handle: floating point
> comparison.

Hi Johan.  I think you are running into similar problems I ran into with
64-bit reals on ARM.  Hopefully there is a common solution for all
64-bit real architectures.

>
> (Incidentally, an inaccuracy of 1 ulp also haunts a std.csv unittest,
> but I do not yet know why exactly)

I encountered this too.  I found that std.conv.parse!double with 64-bit
real is often off by 1 ulp compared to strod().  For 80-bit reals, they
match.  I think the fix for this problem is a change to std.conv.parse.

> How should I go about fixing these unittests for us?

You could look in https://github.com/smolt/phobos/blob/ios/std/math.d
and see if any changes I made for iOS may help you.  I have all
remaining problems marked with D versions that are prefixed with "WIP_",
like WIP_FloatPrecIssue.

Here are the phobos files possibly of interest:
 std/csv.d
 std/internal/math/errorfunction.d
 std/internal/math/gammafunction.d
 std/math.d

Hope this helps,
Dan