Re: math constants

[Ed Smith-Rowland <3dw4rd at verizon dot net>]

On 02/06/2013 03:12 PM, Ulrich Drepper wrote:
> On Tue, Feb 5, 2013 at 9:32 AM, Ed Smith-Rowland <3dw4rd@verizon.net> wrote:
> > I just remembered I found it necessary to add math constants for the old TR1
> > math functions.
> >
> > If you look at include/tr1/special_function_util.h I had a
> > floating_point_constant template in analogy to the standard
> > integral_constant.
> > Then I added a bunch of math constants.  So we have some prior art in this.
> >
> > I would be in strong support of a fully standard solution along these lines.
> You really mean that the current form is what you prefer?  Why
> functions?  Why not constexpr?  You named numeric_constants after
> numeric_limits but it's implemented differently.
Ulrich,

Sorry, I like your constexpr solution.  I forgot to write that in my 
missive. These TR1 constants were pre-constexpr and are needed to work 
in pre-C++11 mode in any case.  I meant only that I had to move beyond 
the double-only constants in math.h.

Go with you idea and good luck.

Also, in a sort of related subject, I would like global functions that 
return numeric_limits things.  I want this to support, generically, 
numeric algorithms that might take, say, multi-precision reals and big 
integers that might have different behaviors and sizes depending on how 
they are constructed. Obviously these would refer to numeric_limits for 
primitive types. What do you all think?

Ed