[boost] Re: octonion_test and quaternion_test failures on Intel 8.1/Windows

From: Hubert Holin (Hubert.Holin_at_[hidden])
Date: 2005-05-31 05:20:01

Somewhere in the E.U., le 31/05/2005

   Bonjour

In article <6.0.3.0.2.20050524220323.037d0c20_at_[hidden]>,
 Beman Dawes <bdawes_at_[hidden]> wrote:

> Both octonion_test.cpp and quaternion_test.cpp, at about line 94, contain 
> the following:

[SNIP pb & fix]

> I've committed this change to CVS for both source files on the assumption 
> that this suits the library maintainer. If not, or there is something wrong 
> with the fix, please feel free to revert the changes.
> 
> --Beman

      Thanks for the fix! This of course suits me. I did further modify 
it in the hope (apparently well-founded) that it would also fix the CW 9 
on Intel problem.

      The remaining outstanding problems with quaternions, octonions and 
special functions all revolve around "long double" problems which I 
believe are due to the libstdc++ of these platforms which seem to have 
some deficiency w.r.t. 64 bits computing (the Beman/Win32/gcc result is 
odd, though, and will need further investigation). Jonathan Wakely is 
investigating the matter and in the end, if necessary, I can (but would 
rather not) disable the test for "long double".

      A few scattered problems remain as well, COMSOFT apparently fails 
to define the "float atan(float)" signature (and provides useless output 
for the quaternion and special functions tests), the FFT-icc 8 results 
seem irrelevant and SunPro 5 gets tripped by "for_each" (which seems to 
pulled in somehow...). At worst these can be marked as expected failures.

   Merci

Hubert

P.S.: sorry for the rather asinch replies I make to various threads, but 
things are a bit hectic lately, and I have not even managed to read all 
I would like to here; should be smoother next month, and definitely 
better in August :-)  (not to mention July, where I'll be almost one 
month off-line, on vacation :-))   ).