Is it worth it to have separate namespaces for interfaces and implementations?

Question

Is it worth it to have separate namespaces for interfaces and implementations?

Stroustrup's advice in his C++ book (fourth edition) is that we should use separate namespaces for interfaces and implementations. Can the more experienced folks say something on this one? I mean it sounds nice, but is it really practical, does it make sense in real world projects?

Answer 1

A namespace tells you something about who the definition belongs to. Of course it makes sense for the interface to belong to a different group than the implementation; that's the whole point of interfaces, separation of concerns.

Answer 2

In code that is heavy on implementation (say, some meta-programming monstrosity inside Boost), it can be useful to spot at a glance which code you are expected to be able to use directly, and which code you can safely ignore. Code in a library's detail namespace is deemed to be "internal" code, so you don't need to spend time hunting through documentation when you see a detail symbol in a stack trace.

I wouldn't say there's a huge benefit to it and certainly not so in the general case, but since it doesn't do any harm you might as well keep things tidy and segregated.