[Dune] Dune on Windows MinGW

[Martin Nolte]

Hi Jö,

I don't have any problem with a static variable -- as long as it is really
faster. If Carstens suggestion is correct, the static variable variant might
be slower due to an additional if-condition whether the variable has already
been initialized.

With respect to the specialization: You are right in that this is possible. On
the other hand, your approach allows asking for MC< int >::e, which might give
a pretty stupid error message. Such points are exactly what needs discussion.

But I think there are more important points to be discussed: The names, static
methods vs. static variables. These have an impact on code actually asking for
these values. For example, I guess we want to change the use of Pi in the DGF
parser (and the other places indicated by Bernd) soon. It would be great, if
we agreed on an interface before these changes.

You are, of course, right that templatized variables generate weak symbols --
I indeed overlooked this. Does anyone know why numeric_limits uses static
methods over static member variables? For me, this was the inspiration to
suggest static methods.

Best,

Martin

On 05/04/2011 01:31 PM, Jö Fahlke wrote:
> Am Wed,  4. May 2011, 12:46:38 +0200 schrieb Martin Nolte:
>> Ok, I just added a first implementation of such a class to dune-common.
>> Improvements are, of course, welcome. I also think that naming can still be
>> discussed.
>>
>> The idea of the file math.hh is to have a "replacement" for the include of
>> <cmath>, which is included by math.hh. Any objections, suggestions, etc.?
> 
> Yes.  Instead of
> 
>   static double e () { return std::exp( 1.0 ); }
> 
> use
> 
>   static const double &e () {
>     static const double val = std::exp( 1.0 );
>     return val;
>   }
> 
> This should should also avoid recomputation of the value on every call.
> 
> Also, I don't see why this has to be specialized for every data type.  One
> templated class should be good for most if not all relevant types.  Or are
> there no overloads of std::exp() and std::acos() for GMPField?  Even so, at
> least long double would benefit from a general template, and it would make
> maintanance easier.
> 
> Thanks,
> Jö.
> 

-- 
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