Re: Synopsis Question

Stefan Seefeld <[email protected]> writes:

> David Abrahams wrote:
>
>>>Huh ? Isn't an anonymous (or 'unnamed' as per the spec 7.3.1.1) namespace
>>> specifically provided to avoid accidental ODR violations, 
>> Well, if you use it in source (.cpp) files, yeah.  But then the
>> symbol
>> is hidden from other TUs because, well, it's not *present* in those
>> TUs.  
>
> It could, without the anonymous namespace, still be referenced ('forward declared')
> as 'extern', and the linker would resolve it. With anonymous namespaces the
> linker will complain.

Yes.  I can't envision how you'd create that secenario in a useful
program, though.

>>>i.e. to make symbols local to the file containing that anonymous
>>>namespace ?
>> Yeah, but that doesn't prevent ODR violations by itself.  If you use
>> the unnamed namespace in a header it's as likely as not to *cause* ODR
>> violations, indirectly, when a template uses the symbols.
>
> You are right. I'v never used it in headers, and in fact, I don't quite
> understand what the purpose would be to put it into a header (see
> http://synopsis.fresco.org/issues/bug32 for example).

If you want to make an object available at namespace scope in a header
without order-of-initialization issues, you can use this technique

  namespace mine {

  template <class T>
  struct singleton
  {
     T& instance() { static T x; return x; }
  };

  namespace 
  {
    Whatever& x = singleton<Whatever>::instance();
  }

  }

HTH,
-- 
Dave Abrahams
Boost Consulting
www.boost-consulting.com