[Dune] Dune on Windows MinGW

[Martin Nolte]

There is a key difference between static variables and methods: space for the
static variables has to be declared at exactly one point. In the case of types
like GMPField< precision >, we would have to know all useful template
arguments in advance. This is not so with static inline functions that are
instantiated if and only if they are used.

To store the expressions in static variables is, of course, still possible
with the static method approach. For GMPField< precision > I would simply use
a singleton class.

Best,

Martin


On 05/04/2011 11:33 AM, Jö Fahlke wrote:
> Am Wed,  4. May 2011, 09:46:15 +0200 schrieb Martin Nolte:
>> this is an interesting Detail. Maybe we should consider adding something
>> similar to numeric_limits to Dune, i.e., a specialized class containing
>> important mathematical constants. It could look somewhat like
>>
>> template<>
>> struct MathematicalConstants< double >
>> {
>>   static double e () { return exp( 1 ); }
>>   static double pi () { return 3.14...; }
>>   static double ln2 () { return log( 2 ); }
>> };
> 
> The glibc info documentation recommends "acos(-1.0)" for pi.
> 
> I would suggest to use static member variables instead of functions, so the
> mathematical functions don't have to be evaluated each time a constant is
> needed.  I believe Christian discovered at one point that exp() could be quite
> expensive, and that is probably true for the other mathematical functions as
> well.  (Or was it pow()?)
> 
> Therefore I suggest something more akin to this:
> 
>   template<class T>
>   struct MathematicalConstants {
>     static const T e;
>     static const T pi;
>     static const T ln2;
>     //...
>   };
>   template<class T> const T MathematicalConstants<T>::e = std::exp(T(1));
>   template<class T> const T MathematicalConstants<T>::pi = std::acos(T(-1));
>   template<class T> const T MathematicalConstants<T>::ln2 = std::log(T(2));
> 
> Of course, I'm all in favour of this idea.
> 
> Bye,
> Jö.
> 
> 
> 
> 
> _______________________________________________
> Dune mailing list
> Dune at dune-project.org
> http://lists.dune-project.org/mailman/listinfo/dune

-- 
Martin Nolte <nolte at mathematik.uni-freiburg.de>

Universität Freiburg                                   phone: +49-761-203-5630
Abteilung für angewandte Mathematik                    fax:   +49-761-203-5632
Hermann-Herder-Straße 10
79104 Freiburg, Germany