std random

[Ed Smith-Rowland]

Ed Smith-Rowland wrote:
> Greetings,
>
> A while back I noted that the std random was different from the tr1 
> random facility.  My original plan was to simply rename the classes in 
> the std implementation.  Further investigation into the latest CD for 
> C++-0X revealed that the differences were far more pervasive and 
> deeper that simply renaming (even renaming involved more hacking of 
> TR1 than would make it worth while).
>
> The attached patch splits the random facility.  It also goes a long 
> way toward completing the std random facility.  The new analytic type 
> distributions were obtained by studying the gmp implementations.  The 
> sampling distributions are completely new as are two new engine 
> adaptors: independent_bits and shuffle_order.  The generate_canonical 
> and the seed_seq utilities are also new.  Even the engines and the 
> distributions that have direct analogs in tr1 have been substantially 
> reworked due to the changes in design in the later drafts.
>
> This patch is a tad more aggressive than one would want at this stage 
> ;-).  This would only go in for 4.5 I think.  That leaves the question 
> of what we should do for 4.4.  I think it may be misleading to keep 
> random in std.  I think it may be best to simply delete or disable 
> include/std/random for now.  What do all of you you think?
>
> The patch bootstraps and regtests with no new fails on 
> x86_64-unknown-linux-gnu.
>
> The one thing that I must figure out before release is that one of the 
> constructor overloads:
> template<typename SeedSeq>
>  some_engine(SeedSeq& sseq);
>
> eats a lot of calls that it shouldn't such as this:
>
> some_engine(unsigned long seed);
>
> and barfs because unsigned long doesn't have a generate() method.
>
> I need to kind of pseudo-concept for SeedSeq.  Any pointers on how 
> this mechanism works in gcc would be helpful.  I'm just starting to 
> look at this.  I need to request that a type support a 
> generate(Iter,Iter) method before being worthy of the first ctor.
>
> I'll make a list of other issues more specific to random that I and 
> others could help think about in another message.
>
> Thank you,
>
> Ed
>
Actually, the patch is mangled.  I'll put together another one.  Sorry.