As previously referred, in the ANTLR site there exists a grammar for C++, but, it is for C++ output, so, it isn't directly usable in Java programs. As always when I don't get exactly what I want, I googled a bit on this and found a question by Yolanda Jargudo in the ANTLR interest list about this. Well I contacted her and she had no luck [ :-( ], so, she made a version for java output and now, she will give it to me (I think) so that I won't need to make this work again [ :-)) ].
While I'm waiting for the grammar, I realized that there could be alternate paths for C++ parsing. There are several environments where this is needed also, like in IDEs that support C++. So I looked at the Eclipse CDT project and to the NetBeans cpplite project. cpplite development stopped and what exists supports very basic syntax highlighting, so, it isn't attractive. The CDT is active and it seams to be very well supported, with persons from IBM and QNX actively working on it.
Within CDT, the CDTCore contains almost every thing I could think off, in the functionality needed for the ArgoUML C++ module reverse engineering. Even better, it could be a new different way to look into how to solve many of the problems of the C++ module, since this project contains things such as refactoring support, makefile generation and import, multi-platform support, etc!
I found this resource on parsing C++ by Andrew Birkett.