Why doesn't iter_swap use swap?

[Daniel Frey]

chris jefferson wrote:
> Daniel Frey wrote:
> 
>> I'm currently discussing the same problem (swap, iter_swap and DR 187) 
>> with some Boost folks. While it looks promising at first that 
>> iter_swap uses ADL to find specializations, it is also intrusive for 
>> the user as he can no longer protect calls from ADL by qualifying 
>> them. Also, if a qualified call to iter_swap uses ADL internally, why 
>> doesn't a qualified call to swap uses ADL, too? Technically it's 
>> possible, Howard Hinnard already tried it out in the Metrowerks STL. 
>> We are still undecided whether or not this is a good idea.
>>
> However, at least in stdlibc++, almost all uses of iter_swap are written 
> std::iter_swap. Therefore at the moment there is no way to user-defined 
> overloads, as users can't (or really, really shouldn't) put things in 
> std::, and iter_swap doesn't call swap. I think this is a worse situation

Maybe I should separate issues more clearly. This first issue is, how 
std::iter_swap should be implemented. The text of DR 187 suggests that 
is makes an unqualified call to swap. I think this is intrusive to the 
user as it activates ADL. I think this should be avoided as we don't 
have evidence of such behaviour elsewhere in the standard (AFAIK). IMHO 
the DR should be changed to make clear that iter_swap it meant to make a 
qualified call to std::swap.

The second, independent issue is, whether a qualified call to std::swap 
should activate ADL internally, thus finding user-defined swap functions 
as if the call would be made through "using std::swap; swap( a, b );". 
For iter_swap, the second issue is probably void.

Regards, Daniel
	
-- 
Daniel Frey

aixigo AG - financial solutions & technology
Schloß-Rahe-Straße 15, 52072 Aachen, Germany
fon: +49 (0)241 936737-42, fax: +49 (0)241 936737-99
eMail: daniel.frey@aixigo.de, web: http://www.aixigo.de