"Kai Nacke" <kai@redstar.de> writes:
>> Maybe I can help?  I can pull in longdouble2 branch and see what it fixes for me and see if anything new breaks.
>
> You're welcome! I rebased the PR after the 0.15.1 release so it should be almost up to date. The failing test(s) in std.math should be easy to spot (just look at the Travis output). The other failing unit test is the CTFE test which grumble on the numerical precising of the results.
>
> Regards,
> Kai

Kai, I got your longdouble2 branch merged into my local repo with iOS support.  I did have to make a few additions to the ldc::longdouble class, probably due to my build env with different headers (MacOS SDK bundled with Xcode).  I'll share the updates once the dust settles.

Good news it that longdouble2 works and I get same unittest results as my exiting kludge to hardcode real to double (6 module unittests fail due to real value miscompare on my iPad mini).  I think it is time for me to look closer at the failures: at least one is even documented that it only works for 80-bit reals.  -- Dan

Dan Olson <zans.is.for.cans@yahoo.com> writes:
>
> Kai, I got your longdouble2 branch merged into my local repo with iOS support.  I did have to make a few additions to the ldc::longdouble class, probably due to my build env with different headers (MacOS SDK bundled with Xcode).  I'll share the updates once the dust settles.
>
> Good news it that longdouble2 works and I get same unittest results as my exiting kludge to hardcode real to double (6 module unittests fail due to real value miscompare on my iPad mini).  I think it is time for me to look closer at the failures: at least one is even documented that it only works for 80-bit reals.

It appears that some others unittests fail because ARM Cortex uses flush-to-zero mode by default instead of subnormal floats.  Seems many unittests are on the boundary of normal and subnormal.  I'll stomp these out one at a time by either disabling or changing the unittest for ARM.

Hi Dan!

On Saturday, 24 January 2015 at 23:20:00 UTC, Dan Olson wrote:
> Kai, I got your longdouble2 branch merged into my local repo with iOS
> support.  I did have to make a few additions to the ldc::longdouble
> class, probably due to my build env with different headers (MacOS SDK
> bundled with Xcode).  I'll share the updates once the dust settles.

Really nice!

> Good news it that longdouble2 works and I get same unittest results as
> my exiting kludge to hardcode real to double (6 module unittests fail
> due to real value miscompare on my iPad mini).  I think it is time for
> me to look closer at the failures: at least one is even documented that
> it only works for 80-bit reals.  -- Dan

Yes, not all tests are implement for all floating point types.

Regards,
Kai