Re: [boost] [foreach] cheap_copy crisis?

From: Eric Niebler (eric_at_[hidden])
Date: 2005-12-03 17:25:04

David Abrahams wrote:
> "Eric Niebler" <eric_at_[hidden]> writes:
> 
> 
>>David Abrahams wrote:
>>
>>>Eric Niebler <eric_at_[hidden]> writes:
>>>
>>>
>>>>Put the overload in an associated namespace of the type. Or for maximum 
>>>>portability, you could put the overload at global scope.
>>>
>>>
>>>Really?  If invoked from within a template, global scope overloads
>>>won't be found unless an argument is in the global namespace or the
>>>overloads happen to appear before the point of definition of the
>>>template.
>>
>>
>>That was the idea, yes.
>>
>>
>>
>>>If you want a way to make this portable, consider adding a dummy
>>>parameter that comes from namespace Boost and asking people to
>>>overload there (the Ramey trick).
>>
>>
>>I'm not sure how this helps on compilers that don't do ADL. This is
>>a macro. BOOST_FOREACH(foo, bar) will expand to something that makes
>>a non-qualified call to boost_foreach_has_cheap_copy *in the context
>>of the BOOST_FOREACH invocation*.  This could be any context, in any
>>namespace. 
> 
> 
> Right.  Including from within a template, which could foil lookup in
> the global namespace.
> 
> 
>>If I want this to work on compilers that don't do ADL,
>>requiring people to make their overloads visible at global scope
>>seems necessary, but I've likely overlooked something.
> 
> 
> namespace boost
> {
>   template <class T>
>   struct foreach_has_cheap_copy
>   {
>        char (& test(...)) [2];
>        char test(mpl::true_*);
>        typedef mpl::bool_<
>           sizeof(
>               test(
>                   boost_foreach_has_cheap_copy(
>                       (T*)0,
>                       boost::whatever()) ))
>        > type;
>   };
> }
> 
> Now all compilers will look in namespace boost, whether or not they
> support ADL.
> 

I thought of something like this shortly after I sent the email. Sadly, 
I cannot get VC6 to eat it. "error C2133: '$S38' : unknown size". Pbbth. 
I tried a few variations, and then gave up.

Templates don't "foil lookup in the global namespace." They merely 
require that the overload has been seen already. I have no problem 
requiring users of broken compilers to put the overload where 
BOOST_FOREACH can find it. Users of non-broken compilers can put the 
overload in an associated namespace and be happy.

-- 
Eric Niebler
Boost Consulting
www.boost-consulting.com