Inconsistent IA-64 name mangling between gcc and Clang for template functions with a local class template parameter

I recently happened along code for which gcc and Clang generate different IA-64 external names. A test case follows. The name in question is for the template specialization of 'foo' parametrized with local class 'Item' of 'X<int>::test<char>(char&) const'

$ cat test.cpp
template<typename T>
const T *foo(const T &t) {
       return &t;
}

template<typename T>
struct X {
       template<typename U>
       void test(U &) const {
           struct Item {} x;
           foo(x);
       }
};

int main() {
       X<int> x;
       char c;
       x.test(c);
}

$ g++ --version
g++ (GCC) 4.5.2
Copyright (C) 2010 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

$ g++ -std=c++0x -c test.cpp
$ nm test.o | grep foo
00000000 W _Z3fooIZNK1XIiE4testIcEEvRT_E4ItemEPKT_RS7_

$ clang --version
clang version 3.1 (branches/release_31)
Target: x86_64-unknown-linux-gnu
Thread model: posix

$ clang -std=c++0x -c test.cpp
$ nm test.o | grep foo
0000000000000020 t _Z3fooIZNK1XIiE4testIcEEvRT_E4ItemEPKS3_RS6_

For ease of comparison, that's:
_Z3fooIZNK1XIiE4testIcEEvRT_E4ItemEPKT_RS7_ (gcc)
_Z3fooIZNK1XIiE4testIcEEvRT_E4ItemEPKS3_RS6_ (clang)

It seems that Clang is implementing a substitution candidate that gcc is not. My reading of the IA-64 ABI specification (http://sourcery.mentor.com/public/cxx-abi/abi.html) leaves me unclear as to what the correct mangling should be.

Tom.

This has been brought up on cxx-abi-dev:

http://sourcerytools.com/pipermail/cxx-abi-dev/2012-June/002429.html