On Tuesday, 18 June 2013 at 19:21:23 UTC, Andrei Alexandrescu
wrote:
On 6/18/13 2:11 PM, Jonathan M Davis wrote:
On Tuesday, June 18, 2013 10:39:38 Andrei Alexandrescu wrote:
On 6/18/13 10:16 AM, H. S. Teoh wrote:
I say again that RNGs being passed by value is a major BUG.
The above
situation is a prime example of this problem. We *need* to
make RNGs
passed by reference. For situations where you *want* to
duplicate a
pseudo random sequence, an explicit method should be
provided to clone
the RNG. Duplication of RNGs should never be implicit.
Occasionally copying a RNG's state is useful, but I agree
most of the
time you want to just take a reference to it.
I think a good way toward a solution is
http://d.puremagic.com/issues/show_bug.cgi?id=10404.
That may be a good way to go about it, but the default such as
what rndGen
returns really should be a reference type, and only ranges
whose state can
actually be copied such that continuing to iterate over them
gives the same
values should be forward ranges IMHO. Using Class!T might be
able to make it
so that we can avoid needing a std.random2, but we'll probably
still need some
minor code breakage in there somewhere.
- Jonathan M Davis
How about defining rndGenRef and putting rndGen on the
deprecation path?
Andrei
Even with 10404, I still think we'd need a new module. An adaptor
helps, but these ranges really need to be ref by default (amongst
other changes). We'd probably have to end up with MersenneRef, as
well as pretty much everything else...
Also: Class!PRNG.save would not actually save, it would merely
return a copy of its PRNG payload, and not a new class instance.
This could be problematic (for starters, it wouldn't qualify for
forward range...)
