You don't need anything like this for C++

Posted Jul 9, 2012 10:46 UTC (Mon) by jwakely (subscriber, #60262)
In reply to: You don't need anything like this for C++ by cmccabe
Parent article: Calling for a new openSUSE development model

It's not really "very similar" to using namespace version2, an inline namespace is a lot more transparent so users never need know it exists (unless they look at the mangled symbols in their objects.)

A using-directive just makes names visible, but they are not treated as first-class members of the namespace containing the using-directive and will not be found by qualified name lookup if there are declarations of the same name in the namespace containing the using-directive. That's not the case for inline namespaces.

Templates defined in an inline namespace can be instantiated and specialized as though they are members of the enclosing namespace, not possible via a using-directive.

An inline namespace is an associated namespace of types in its enclosing namespace and vice versa, so it plays nicely with ADL.

Don't dismiss the feature because you don't understand it yet.

to post comments

You don't need anything like this for C++

Posted Jul 10, 2012 21:31 UTC (Tue) by cmccabe (guest, #60281) [Link]

It seems like the best practice is to put all of your C++ library declarations into one namespace-- for the sake of argument, call it foolib.

It seems like what you're recommending then is that you have something like this:

namespace foolib {
   namespace version1 {
       ...
   }
   inline namespace version2 {
       ...
   }
}

Given that use-case, I don't think we need to care about the scenario where there are declarations of the same name in the namespace containing the using-directive-- neither the library user nor the library implementor should be doing that.

Templates defined in an inline namespace can be instantiated and specialized as though they are members of the enclosing namespace, not possible via a using-directive.

That's fair. I suppose template specialization is where the "using directive" approach really breaks down.

The cost of this feature is that debugging becomes (even more) difficult, since you search the symbol table of your application for foolib::doit and you don't find it. Instead you find N versions and have to decide which one you're really using. But of course, none of us here writes bugs, so that shouldn't be a problem :)