Re: [boost] [Numeric/Interval] Mingw-3.4.1 failures

From: Guillaume Melquiond (guillaume.melquiond_at_[hidden])
Date: 2004-09-22 04:27:19

Le mar 21/09/2004 à 23:46, Aaron W. LaFramboise a écrit :
> Bronek Kozicki wrote:
> 
> > Aaron W. LaFramboise wrote:
> > 
> >>If you are willing to contribute a public domain implementation of the
> >>real arc-hyperbolic trig functions, I am sure MinGW would not reject them.
> > 
> > If Boost license is compatible with requirements of MinGW, then maybe 
> > following will be good enough?
> 
> Unfortunately, it isn't.  For various reasons, the MinGW runtime is
> public domain, meaning that authors have disclaimed any copyright
> interest that they might have.  The author of special_functions could do
> this, of course, but it would undermine his legal ability to enforce the
> Boost license for this code, if this was something he was interested in.
> 
> For Boost's purposes, isn't it best just to use Boost's implementation
> of these functions on MinGW?

Unfortunately no. It has already been discussed on this mailing-list,
and so I will just give a quick summary here. The Boost implementation
of these functions is not correctly rounded. When used with the interval
library, it would systematically lead to wrong results.

If you think the same hold true for standard libm, you are right. But at
least the user has the possibility of using a precise libm as a
replacement. And it is not just theoretical, such libraries exist (the
crlibm project for example).

Regards,

Guillaume