If you might want another function to be selected instead, use enable_if If you just want an error message use concept checks.

Thanks a lot, it makes things far clearer.

Hello,

how can I build a concept which requires a given type X to either have an associated type named foo or an associated type named bar (silly names for example purpose). Checking for both is easy but i can't find how to check for OR another.

Maybe using SFINAE? template <class T> struct foo_or_bar { template <class U> static typename U::bar f(); template <class U> static typename U::foo f(); typedef BOOST_TYPEOF(f<T>()) type; }; foo_or_bar<something>::type will give you the type of the "foo" or "bar" nested in "something"... But it's an exclusive OR: if both are present, you have an ambiguity. Anyway I'm pretty sure someone else will find a better way... :-) Bruno

I would say don't. It's not friendly to generic code and it sounds

"or constraints" which were left out of C++0x for very good reasons. If you could be more specific about what you're trying to do I might be able to help find a better solution. My first guess is that the answer will be found in

like the use of traits. Basically I want to make a concept which represents class that supports the result_of protocol, ie exposes either a result_type or a result inner class. So my first idea was to see if a given type have either of this inner elements.

Thansk for the clues. I indeed turned to a traits based solution. I must admit that, if testing for the result_type typedef is easy, testing if a template inner class with an a priori unknown number of template parameters is harder. It's even harder cause I can't use a simple SFINAE on the result of boost::result_of cause it's defined even if neither result_of protocols are implemented. I'll try some exhaustive listing of result possibilities and see what it spawns.

on Mon Aug 11 2008, "joel falcou" wrote:

Forgot what I just said. I got this working :

http://codepad.org/mcY4K7Sf

Do anyone see any blatant misses in this ?

Doesn't work for struct me2 { template struct result { typedef void type; }; }; or struct me3 { template struct result { typedef void type; }; }; FWIW. -- Dave Abrahams BoostPro Computing http://www.boostpro.com

On Mon, Aug 11, 2008 at 9:05 AM, joel falcou wrote:

Yes, I thought that the boost::result_of protocol needed the result struct to be a one argument template only. I forgot about default argument and such.

Well, I used some preprocessor magic and got this : http://codepad.org/VrLTKmJO

Should do the tric IMHO. Comments on how to simplify this is of course welcomed. If it's of some interest, I can make a proper file to be included in the proper boost library (maybe next to result_of ?)

What is your use case? I found this sort of thing useful for finding out whether result_of would work for a specific set of arguments. Is your metafunction intended for testing whether result_of would work for _some_ (whichever) set of arguments? (it doesn't quite do that because it doesn't check that result has a nested type typedef). Stjepan

On Mon, Aug 11, 2008 at 9:33 AM, joel falcou wrote:

I was trying to use it as a simple traits to see if a given type fullfilled the result_of protocol whatever the arguments used. It works ok for the result_type variant but now I see it's still incomplete. The problem is that I can't see, for an arbitrary result structure, how I can see if it actually has at least one overload that has a proper nested type typedef.

I can't either :-)

Typical use is :

For F a given type, support_result_of_protocol<F>::type evaluates as a compile time boolean indicating if either F has a nested result structure or result_type typedef. It doesn't make any assumptions on wether or nor result is correcly implemented.

As for testing if a particuliar set of arguments produces a valid result_of, one can easily do :

template< class F , template<class> class T = boost::result_of<F>::type > struct valid_impl { typedef void type; };

template struct produce_valid_result_of : boost::false_type {};

template<class F> struct produce_valid_result_of : boost::true_type {};

or did I miss something in your question ?

I think you got my question, but that won't work. If you fix it to: template< class F , typename T = typename boost::result_of<F>::type > struct valid_impl { typedef void type; }; template struct produce_valid_result_of : boost::false_type {}; template<class F> struct produce_valid_result_of : boost::true_type {}; ...and then try: struct x {}; std::cout << produce_valid_result_of::type::value << std::endl; ... you will get an error. I get: /Development/boost/boost/utility/result_of.hpp: In instantiation of 'boost::detail::result_of_nested_result': /Development/boost/boost/utility/result_of.hpp:38: instantiated from 'boost::detail::result_of_impl' /Development/boost/boost/utility/result_of.hpp:31: instantiated from 'boost::result_of' ../../libs/patterns/example/video_gesture_recognition.cpp:224: instantiated from here /Development/boost/boost/utility/result_of.hpp:73: error: no class template named 'result' in 'struct x' With the way boost::result_of is implemented, I couldn't find a way to make an approach like the above work. The best I could do was the result_of_defined code. Kind regards, Stjepan

Sorry for adding noise. Here is the correct link : http://codepad.org/tp9rMX8E