Sorry for being vague about versions.  This was with Boost 1.49. Both Visual Studio 2012 and 2013 generates the warning.  Clang 3.5.0 and g++ 4.7.2 do not. They all compile fine at the end. 

Also I just tried with Boost 1.56.  There is no warning in either compiler!  Strangely, the result_type typedef in key extractor is still non-& in this version.  Perhaps Boost.Range changed... 

Any explanation of what may have changed and why not having a & in result_type works would be greatly appreciated. 

Thanks many,

On Thu, Mar 19, 2015 at 2:20 PM, Nick Stokes <randomaccessiterator@gmail.com> wrote:
... forgot to add, if I use a custom functor which does include the & in result_type, like:

struct FooName {  typedef const string& result_type;    result_type operator()(const Foo& f) const { return f.name;} };

in the same ranged for as above:

for( string& s : foos.get<1>() | boost::adaptors::transformed( FooName() ) ) ..

then there is no warning from the compiler.


On Thu, Mar 19, 2015 at 2:17 PM, Nick Stokes <randomaccessiterator@gmail.com> wrote:
There is a problem...   I noticed, when used as follows with boost::range,


for( string& s : foos.get<1>() | boost::adaptors::transformed( m.get<1>().key_extractor() ) )
   ...

Visual Studio 2012 generates an ugly warning " warning C4172: returning address of local variable or temporary", and compiles.  It also compiles on clang and gcc (sorry for not being too specific with versions, modern ones).   But this is suspicious enough so I poked around a little bit.   Sure enough, the key_from_value (the return type of key_extractor()) is a functor which is like

template< .. class Value  .. >
struct key_from_value {
     typedef  Value result_type;   // not Value&

     Value& operator()( Foo&.. ) const {  .. }

}; 

So isn't result_type being a non-reference type a problem here, given operator() is returning &? 

Regards,
Nick

On Thu, Mar 19, 2015 at 9:09 AM, Nick Stokes <randomaccessiterator@gmail.com> wrote:
Thanks Joaquín

On Wed, Mar 18, 2015 at 3:45 PM, Joaquin M Lopez Munoz <joaquin@tid.es> wrote:
Nick Stokes <randomaccessiterator <at> gmail.com> writes:

>
> Q1:  Is there a builtin in way in MI to iterate over keys rather
> than elements? I.e. the iterator's value type will be that of the
> key, not the element. 

No, there's none. Some iterator adaption is needed as you mention
below.

> Q2:  Is there a way to obtain the type of key extractors given a
> multi index container?  (it is related to Q1, as this sort of a
> thing could be passed to boost::transform iterator to obtain an
> alternative answer for Q1)

  auto beg=boost::make_transform_iterator(
             foos.get<1>().begin(),foos.get<1>().key_extractor()),
       end=boost::make_transform_iterator(
             foos.get<1>().end(),foos.get<1>().key_extractor());

  std::copy(
    beg,end,std::ostream_iterator<std::string>(std::cout," "));

Joaquín M López Muñoz
Telefónica
_______________________________________________
Boost-users mailing list
Boost-users@lists.boost.org
http://lists.boost.org/mailman/listinfo.cgi/boost-users