Re: [Patch] DR 438

[Howard Hinnant <hhinnant at apple dot com>]

On Apr 27, 2007, at 3:59 PM, Paolo Carlini wrote:

> I noticed only today for the first time that the implementation is  
> allowed to do better than recognizing integers... (well, if ever  
> possible, I would suggest not spending too much time  reasoning  
> about those issues in the context of C++03 DRs, concepts will  
> chaneg everything here, right?)

<nod> :-)  Actually you can do a lot better than is_integral today.   
The code that inspired the issue had is_input_iterator which keyed on  
whether the type had a nested iterator_category which is convertible  
to std::input_iterator_tag (not 100% foolproof, but I think it is  
correct to 99.9999%).  And then the dispatching was done with  
enable_if on the user-level member templates, instead of dispatching  
internally after binding to the user-level member templates (much  
like concepts will do).  So implicit conversions on both f and l are  
"more correct" (with the concept model).  We might have to get  
explicit conversion operators to actually tell the difference  
(already blessed by EWG).

-Howard