Patch to fix PR9861

[TJ Laurenzo]

> By the way, considering the demangler is not an unimportant point as
> any ambiguity in demangling is an ambiguity that can be happen when
> linking C++ and Java code together.  For example, your proposed
> mangling for
>     java.lang.Math.acos(double)
> appears to be
>     _ZN4java4lang4Math4acosEdd
> which is the mangled name which will be used for the C++ function
>     java::lang::Math::acos(double, double)

I think your point is valid but this situation actually wouldn't
happen because the C++ mangler would recognize that java::lang::Math
is a Java type and would apply the Java mangling rules.    I think
that with this change, Java has reached the point where it needs to
diverge from the C++ ABI and it would probably be a good idea to have
a way to distinguish the Java form from the C++ form.  My only concern
is that introducing a new divergent form would break existing tools
that do demangling until they could catchup.

TJ