Re: [Boost-users] [type traits / MPL] check whether a class has a certain member function

Subject: Re: [Boost-users] [type traits / MPL] check whether a class has a certain member function
From: Edward Diener (eldiener_at_[hidden])
Date: 2013-03-08 05:37:19

On 3/7/2013 3:53 PM, Stefan Strasser wrote:
> Am 07.03.2013 20:12, schrieb Jeffrey Lee Hellrung, Jr.:
>>             The macro you're looking for is
>> BOOST_TTI_HAS_MEMBER_FUNCTION.
>
>>     it requires you to know the exact signature of the member
>>     function. I'm trying to check whether it is ok to call c.f(x), which
>>     would be satisfied by each of the following 3 signatures (and more):
>>
>>     struct C{
>>          void f(X);
>>          void f(X &);
>>          void f(X const &);
>>          ...
>>     };
>>
>>     is there a way to check for that, or do I have to check for all
>>     (likely) signatures?
>
>> https://groups.google.com/group/comp.lang.c++.moderated/tree/browse_frm/thread/4f7c7a96f9afbe44/c95a7b4c645e449f?pli=1#doc_e5fbc9305539f699
>>
>>
>
>
> Thanks!
> it doesn't work for me, because it uses sizeof(void) if the function
> call is possible.
>
> However, it is enough to determine whether a member of a given name
> exists, without checking the signature. an implementation of has_member
> is here: http://www.rsdn.ru/forum/cpp/2720363
>
> Edward, can this be added to TTI, for example as BOOST_TTI_HAS_MEMBER?
> not only does this implementation look portable to me, but boost already
> uses it internally:
> http://www.boost.org/doc/libs/1_47_0/boost/thread/locks.hpp

I agree that having a TTI metafunction to check for a member function by 
name with any signature might be helpful. But even in your situation 
above, just knowing that some 'f' exists within 'C' does not tell you it 
is callable by passing 'X', 'X &', or 'X const &'.

Nonetheless I will look into the links presented by you and Jeffrey 
Hellrung, especially the last above.

First I need to finish my work and get TTI into Boost. I am very near 
the end of making coding changes as suggested by the review and largely 
have some more documentation and tests on some compilers ( Intel and Sun 
) to do. Gcc, clang, and VC++ are working fine so far with TTI and an 
occasional workaround ( VC++ and some slightly older versions of gcc ).

My major concern is how much time I should spend on trying to invent 
workarounds for compilers whose implementations of SFINAE are 
insufficient in certain situations need by TTI. But I think its most 
important just getting TTI into Boost, documenting what does not work on 
certain less SFINAE deficient compilers, and then look for possible 
workarounds later. The vast majority of C++ programmers probably use 
gcc, VC++, or clang nowadays. I have also delayed much too long in 
getting TTI fully ready for Boost and into the release branch and just 
need to get it done.