random numbers in bulk

[Ed Smith-Rowland]

On 08/25/2012 07:39 PM, Ulrich Drepper wrote:
> On Sat, Aug 25, 2012 at 7:37 PM, Jonathan Wakely <jwakely.gcc@gmail.com> wrote:
>> But iterators don't have to imply non-sequential storage. Using
>> iterators instead of pointers would allow you to store them in a
>> std::deque, for example, or in a std::vector using
>> std::back_insert_iterator.
> Yes, and this is already trivial to do with the operator() interface.
>
> The fill() interface is needed for performance, everything else is
> taken care by the operator() interface.
>
If you templatize the pointers as forward iterators there is nothing 
preventing you from *calling* with pointers on C arrays.  You then get 
the benefit that all containers can use this interface as well. Plus you 
have std::array which is a C array under th ehood.

I'm a little bit stuck on why vector would be slow if, say a caller used 
reserve.
Maybe the cost of default initialization and then setting the number later?

Anyway, like the idea.

How we juggle adding new members vs. standards compliance is another 
question.  We could add underscores and put them in the gnu extension 
namespace as we do with other extensions.  We would then have these 
lying around if (as I think they should) this gets standardized.  Or we 
could just put them in and risk being out of step if the LWG doesn't add 
them.  OTOH, if we added them it would create a precedent...  Politics ;-)

Ed