[Patch] Second half of tr1/complex

[Howard Hinnant]

On Jan 13, 2006, at 10:16 AM, Paolo Carlini wrote:

> Hi Howard,
>
>> I see one more issue that we might want to cover here.  The new
>> overloads look overly generic to me, which may break user code.  For
>> example, consider:
>
> I see, but then, if we want to make the library robust wrt this  
> kind of,
> well, perversion ;) (*), shouldn't we start sprinkling __enable_if
> everywhere?!? Or is there a cleaner solution?
>
> Paolo.
>
> (*) Is it even legal using the 'arg' name together with a using
> namespace like this?!?

<nod> It's legal.

Concepts in C++0X may provide a cleaner solution.

I think any time we add a function template with a parameter that  
isn't meant to work for user types, we need to be alert to this  
issue.  I admit I may be overly paranoid of this issue.  But when I  
first tried to fix up <cmath> with this logic and provided:

template <class T> double sqrt(T x);

to catch integral arguments, I broke real customer's code  
(unfortunately not detected until after shipping).  Thus was born the  
motivation for enable_if, at least for me (and which took a couple  
more years to materialize).

Yes, the std is chock full of overly generic code.  No, at this time  
I don't recommend that gcc retrofit all of it with enable_if.  But  
when we add new stuff, that could possibly take customers by  
surprise, I do recommend being cautious in this way.

-Howard