Created attachment 144788 [details]
Source files showing the regression

In the recent 1.6.1 valid documentation (at least I think it's correct, please
note if and why not) causes warnings. Doxygen 1.5.8 accepted this without a
problem. This is on Fedora 11.

I have attached a tarball with source files documenting the problem. Here the
verbal description:
- Class A is a pure virtual class with do_something(A &a), it is member of the
namespace N
- do_something(A &a) takes an instance of A as argument
- Class B inherits from A, but is *not* member of the namespace N, therefore it
must read: class B : public N::A {... do_something(N::A &a); ... }.

Now the documentation is rejected, because it seems not to properly map N::A
onto the class A in namespace N. Even if I remove the qualifier in B's
do_something (making a forward declaration in the namespace) gives an error if
I split the implementation into a .cpp file. In that case it says "no uniquely
matching class member found". In this case class_B.cpp had a "using namespace
N" clause, which seems to make doxygen again search for the qualified version
of do_something().

To me this looks like a bug and since it worked as expected 1.5.8 like a
regression. Or was that intended? What's the recommended code structure then?