Fully qualified std::swap

[Morwenn]

I read the algorithms' code and it appears that you are using a fully qualified std::swap everywhere. It would be interesting to find swap with ADL instead to benefit from user-defined swap functions, be it for correctness or performance improvement.

[fjtapia]

Hi Morwen,

about your question. In the generic algorithms, don't know the data type to
use. Perhaps a single number, perhaps a very complex data structure.

The data type to use must accept the move semantic. Perhaps don't have
assignation operator , but can be moved. The assignation operator only is
required for the iterators.

The std::swap use the move semantic. The code of the std::swap is ( more
or less)

template

void swap (T &A T&B)

{ T Aux(A);

A = std::move ( B) ;

B = std::move ( Aux) ;

};

In an old project, I had a very complex data structure, and I designed a
specific swap function

struct myclass

{

 void swap ( myclass & m){ …….}

};

But after this, I wrote an overload of the std::swap

std::swap (myclass & m1, myclass &m2) { m1.swap(m2);};

Without this overload, perhaps you have problems with many of the standard
algorithms and data structures, because many of them, internally use the
std::swap

Francisco

2016-03-23 12:33 GMT+01:00 Morwenn notifications@github.com:

I read the algorithms' code and it appears that you are using a fully
qualified std::swap everywhere. It would be interesting to find swap with
ADL instead to benefit from user-defined swap functions, be it for
correctness or performance improvement.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub
#2

[Morwenn]

As far as I know, the standard library algorithms and data structures are required to use using std::swap; swap(*it1, *it2) (or std::iter_swap(it1, it2)) whenever they need to swap two values, so no matter whether you defined a swap function in the data structure namespace or whether you overloaded std::swap for your data structure, the right swap function should be found, and thus move semantics should always work.

[biddisco]

We had the same problem with the HPX version of the code. Using std::iter_swap(it1, it2) fixed a bunch of compilation problems that were triggered when using zip iterators to sort using a different key.