Ranges and random numbers -- again

[Joseph Rushton Wakeling]

On 06/18/2013 08:35 PM, Jonathan M Davis wrote:
> Regardless, I think that we should be sure of what we want to do with the 
> module as a whole before we start making tweaks like that, since if we decide 
> that we really do want std.random2, there's not much point in tweaking 
> std.random, and std.random2.rndGen could return what we want. If we do stick 
> with std.random though, I think that we should come up with a better name than 
> rndGenRef (maybe rndRange or genRnd or simply random).

If you're having a std.random2 then I don't see a reason to change the name --
keep it as rndGen.  In _most_ cases of downstream code it won't break, so it
smooths the upgrade path.  I presume using std.random and std.random2 together
would be considered an error, so is there really any risk of a clash?

Also, assuming we do go with std.random2 rather than an upgrade to std.random,
it may be better to go with structs-carrying-a-pointer-to-a-payload rather than
with classes; that in turn might allow for less breakage in users' code.

For example, if the Random alias is switched from a struct to a class, then
users would have to go through their code tweaking stuff like this:

     - auto gen = Random(unpredictableSeed);
     + auto gen = new Random(unpredictableSeed);

Whereas if RNGs are still structs on the outside (with reference to payload
inside) then the original line should still work.