Re: Suggestion: Change the specification for grand in glib

[Sebastian Wilhelmi <wilhelmi ira uka de>]

Hi Sverre,

> If the application for g_rand_double is g_rand_double_range or
> g_rand_int_range you are correct. But in simulations where we test on
> really low probabilities it is important that g_rand_double works near
> 0.0.  But in that case we do not need all 53 bits of accuracy.
> 
> If you whant to spare one test then the following could be used:
> 
> gdouble
> g_rand_double (GRand* rand)
> {
>   gdouble transform = G_RAND_DOUBLE_TRANSFORM;
>   guint32 r1 = g_rand_int (rand);
>   gdouble r2 = g_rand_int (rand);
> 
>   if (r1 < 1<<20)
>     {
>       while (r1 == 0)
>         {
>           transform *= G_RAND_DOUBLE_TRANSFORM;
>           r1 = g_rand_int (rand);
>         }
>       r2 += g_rand_int (rand) * G_RAND_DOUBLE_TRANSFORM;
>     }
> 
>   return (r2 * G_RAND_DOUBLE_TRANSFORM + r1) * transform;
> }
> 
> The following version should be sufficient:
> 
> gdouble
> g_rand_double (GRand* rand)
> {
>   gdouble transform = G_RAND_DOUBLE_TRANSFORM;
>   guint32 r1 = g_rand_int (rand);
> 
>   while (r1 == 0)
>     {
>       transform *= G_RAND_DOUBLE_TRANSFORM;
>       r1 = g_rand_int (rand);
>     }
> 
>   return (g_rand_int (rand) * G_RAND_DOUBLE_TRANSFORM + r1) * transform;
> }
> 
> Or even simpler without the loop:
> 
> gdouble
> g_rand_double (GRand* rand)
> {
>   guint32 r1 = g_rand_int (rand);
> 
>   if (r1 == 0)
>     return (g_rand_int (rand) * G_RAND_DOUBLE_TRANSFORM + g_rand_int (rand))
>       * (G_RAND_DOUBLE_TRANSFORM * G_RAND_DOUBLE_TRANSFORM)
>   else
>     return (g_rand_int (rand) * G_RAND_DOUBLE_TRANSFORM + r1)
>       * G_RAND_DOUBLE_TRANSFORM;
> }

I'm sorry, but I'm not going to include that. It happens in .00000002% of all
cases and is IMHO of dubious usefulness.

God Jul og Hej då,
Sebastian
-- 
Sebastian Wilhelmi
mailto:wilhelmi ira uka de
http://goethe.ira.uka.de/~wilhelmi