namespace for 128-bit long double support

[Janis Johnson]

In February 2006, Jakub Jelinek added a patch to libstdc++ for 128-bit
long double support for {powerpc{,64},s90{,x},sparc,alpha}-linux.
It's likely that the answer to my question in somewhere in the long
thread about various versions of that patch, but if so I have not yet
found it.

The question: why are some libstdc++ functions for 128-bit long double
support now in namespace __gnu_cxx_ldbl128?  At least for powerpc{,64},
those functions were already in libstdc++ and the name mangling was
enough to differentiate them from the support for 64-bit long double.
Now mainline still defaults to 64-bit long double, for which the new
namespace is not used for 128-bit long double functions, but when GCC
is configured using --with-long-double-128, the new namespace is used.
This means that object files built with a compiler that defaults to
64-bit long doubles, but using -mlong-double-128, cannot be linked
using a driver and libraries from a compiler that defaults to 128-bit
long doubles.  Is this intended?

Is the use of this new namespace documented somewhere as part of a
modified C++ ABI?

Janis