Mir Random and Dlang Ranges [Example]

[Joseph Rushton Wakeling via Digitalmars-d]

On Friday, 25 November 2016 at 16:08:15 UTC, Ilya Yaroshenko 
wrote:
> On Friday, 25 November 2016 at 16:04:07 UTC, Andrei 
> Alexandrescu wrote:
>> I'd say we can bury the hatchet and move on with Ilya's API. 
>> -- Andrei
>
> Yes, this was clear. There are also others who may disagree. 
> This is the reason for this thread. --Ilya

Without wanting to re-open the debate now, it seems to me that 
it's worth distinguishing carefully between:

   * the changes necessary to support Ilya's goals in terms of 
generating really
     high quality random variates (where I'm not sure the switch 
from ranges to
     functors really matters that much, except as a preference), 
and

   * the changes necessary to avoid the known issues in terms of 
accidental
     copying-by-value of RNG state (where the functor-based 
proposal is
     interesting, but not necessarily the best or only solution).

On the second point, I would suggest maintaining an open mind 
both towards Ilya's proposal but also to the possibility that 
there might be better alternatives.

(Minor remark: the copying-by-value of RNG state is not limited 
to random engines [i.e. pseudo-random algorithms]; it also 
applies to e.g. a random device which caches a buffer of bits 
read from a 'true' random source.)