[PATCH] Implement std::experimental::sample

[Evgeniy Stepanov]

In http://reviews.llvm.org/D9044#161258, @EricWF wrote:

> In http://reviews.llvm.org/D9044#156827, @eugenis wrote:
>
> > In http://reviews.llvm.org/D9044#156813, @EricWF wrote:
> >
> > > You'll need to use reserved identifiers
> >
> >
> > Do you mean for local variable names? Is there a technical reason for this, or just a code style thing?
>
>
> For any name not defined in the standard you should use reserved identifiers. Assume that the user of the standard library defines every non-reserved, non-defined name as a macro. We use reserved identifiers so that conforming programs can define non-reserved names globally and still use the standard library.
>
> For example:
>
>   #define PopulationIterator char*
>   #define SampleIterator char*
>   #define UniformRandomNumberGenerator &rand
>   #include <experimental/algorithm>
>


Good point, will do.

> 

> 

> > > and guard the definition of `sample` inside a C++14 only block because adding a new definition of could break existing code.

> 

> > 

> 

> > 

> 

> > n4335 does not seem to require that this is only available in C++14. And aren't names under std:: already reserved, and an addition of a new std:: name could not break compliant code?

> 

> 

> I would prefer not to restrict these things to c++14 only, although there may be a reason that conforming implementations need to that isn't visible in n4335.

> 

> However because `sample` is a function it can be looked up by ADL. Since `sample` takes user-defined iterator types and RNG's it is possible that existing unqualified calls to `sample` in user code may now find `std::sample`.


That's a valid concern, but limiting this definition to C++14 will still break existing C++14 code. Is it really that much better than breaking C++11 and C++14 at the same time?
Also, that would be std::experimental::sample.


REPOSITORY
  rL LLVM

http://reviews.llvm.org/D9044

EMAIL PREFERENCES
  http://reviews.llvm.org/settings/panel/emailpreferences/