Re: where are the implementations for <random>?

[Jonathan Wakely <jwakely dot gcc at gmail dot com>]

On 11 April 2017 at 21:17, Oliver Kullmann <o.kullmann@swansea.ac.uk> wrote:
>> Now only missing seeding for std::mt19937_64.
>>
>> This seems most complicated.
>> Apparently it is in
>> https://gcc.gnu.org/viewcvs/gcc/trunk/libstdc%2B%2B-v3/include/bits/random.tcc?revision=246542&view=markup
>>
>> Lines 346 - 389.
>>
>> I would guess that this functionality is most specialised, and most likely to be different
>> for different compilers.
>>
>> Really a pitty that this was not standardised.
>>
>
> The code at Lines 346 - 389 for the seed-member of the
> mersenne_twister_engine doesn't look too bad except of two aspects: What
> is
>
>  - __detail::_Shift ?
>  - __detail::__mod  ?
>
> It seems that apart from these functions, everything else is just
> arithmetic of bit-logic.
>
> And the creation of that sequence in Line 360:
>   __q.generate(__arr + 0, __arr + __n * __k);
> actually seems to be defined by the standard, so that can be used.
>
> ACTUALLY, I just realise that even if I can code the above myself, I
> couldn't use the Mersenne twister from the standard library, since
> apparently the only way of setting the internal state is by
> *constants*??
>
> All what there is seems to be
> http://www.cplusplus.com/reference/random/mersenne_twister_engine/

Use a better reference. http://en.cppreference.com is much better than
cplusplus.com

> template <class UIntType, size_t w, size_t n, size_t m, size_t r,
>           UIntType a, size_t u, UIntType d, size_t s,
>           UIntType b, size_t t,
>           UIntType c, size_t l, UIntType f>
>   class mersenne_twister_engine;
> and that has the state, which is set by the seed-member-function, only as template parameter.
>
> So I need to implement the Mersenne twister apparently myself.
> Or perhaps I don't understand that yet properly.
>
> Perhaps the seed function has nothing to do with those template parameters, which are the
> parameters of the engine, while the seed sets only the "internal state", and that perhaps
> can be done explicitly by calling some std-library facility?

I don't understand what you're trying to do or what the difficulty is.

The output of the mersenne_twister_engine is specified by the
standard, there should be no need to reimplement it. If you don't get
repeatable values for any implementation of mersenne_twister_engine
then it's a bug in that implementation.

The template arguments are just parameters for the internal algorithm,
so different parameters create different outputs. std::mt19937 is
simply a specialization of the mersenne_twister_engine with parameters
that are known to produce good pseudo-random numbers.

The behaviour of uniform_int_distribution is also specified by the
standard, but only the mathematical properties, not a specific
implementation. That means different implementations can give
different outputs. If you use your own uniform_int_distribution then
you ensure repeatable outputs across implementations. I repeat, there
should be no need to do that with the engines themselves.