Minimal example program: #include <iostream> #include <algorithm> template< typename T > struct A; template< typename T, typename U > struct A< std::pair< T, U > > { static void f() { std::cout << __PRETTY_FUNCTION__ << std::endl; } }; int main() { A< std::pair< int, int > >::f(); } Output of __PRETTY_FUNCTION__ from this program: static void A<std::pair<_T1, _T2> >::f() [with T = int, U = int] Problem: _T1 and _T2 are printed as the pair's parameter names, but it then say T=int and U=int. Additional information: This only seems to occur with partial specialization.
It uses the names as in the std::pair template declaration, so you could argue this is correct.
I understand where the names come from, but that doesn't make the message correct. Consider the specialization in the example program to use <_T2,_T1> (note the reversed order) and an instantiation with <int,double>, the resulting message is: static void A<std::pair<_T1, _T2> >::f() [with _T2 = int, _T1 = double] which is *really* misleading. Since it's actually vice versa. Additionally I just found out that the same problem also occurs within error and warning messages from the compiler, so it's not limited to __PRETTY_FUNCTION__. I triggered a warning for an unused variable, which reads: foo.cpp: In static member function ‘static void A<std::pair<_T1, _T2> >::f() [with T = int, U = int]’: foo.cpp:17: instantiated from here foo.cpp:10: warning: unused variable ‘i’ Regards, Daniel
Confirmed. This is a diagnostics bug. (And perhaps a duplicate). Thanks for the report. If you would like to contribute a patch, please read http://gcc.gnu.org/contribute.html
More information: I checked out the trunk and checked it again. The bug is still present, but while testing it, I noticed that the problem does not occur with my own types, only with types from the standard library. Here's the code I used: #include <algorithm> template< typename X, typename Y > struct P {}; template< typename > struct A; template< typename T, typename U > struct A< P< T, U > > { static void f() { int i; } }; template< typename V, typename W > struct A< std::pair< V, W > > { static void f() { int i; } }; int main() { A< P< int, double > >::f(); A< std::pair< int, double > >::f(); } and the output is: frey@fiasko:~/work/test/foo$ /home/frey/work/svn/gcc/host-i686-pc-linux-gnu/gcc/g++ -B/home/frey/work/svn/gcc/host-i686-pc-linux-gnu/gcc/ -ansi -pedantic -Wall -Wextra -I/home/frey/work/svn/gcc/i686-pc-linux-gnu/libstdc++-v3/include/i686-pc-linux-gnu -I/home/frey/work/svn/gcc/i686-pc-linux-gnu/libstdc++-v3/include -I/home/frey/work/svn/gcc/libstdc++-v3/libsupc++ foo.cpp -c -o foo foo.cpp: In static member function ‘static void A<P<T, U> >::f() [with T = int, U = double]’: foo.cpp:12: instantiated from here foo.cpp:7: warning: unused variable ‘i’ foo.cpp: In static member function ‘static void A<std::pair<_T1, _T2> >::f() [with V = int, W = double]’: foo.cpp:13: instantiated from here foo.cpp:8: warning: unused variable ‘i’ frey@fiasko:~/work/test/foo$ /home/frey/work/svn/gcc/host-i686-pc-linux-gnu/gcc/g++ --version g++ (GCC) 4.4.0 20080227 (experimental) [snip]
This is most likely a dup of bug 99.
I looked at bug #99, but I am unsure whether this bug is really a dup of it. #99 is about overloads and occurs with user types, while this bug is about partial template specializations and occurs only with certain types from the STL.
No, bug 99 has specialization too: template<unsigned int n, unsigned int m> double ch(dummy<n>, dummy<m>); template<unsigned int n> double ch(dummy<n>, dummy<0>); template<unsigned int m> double ch(dummy<0>, dummy<m>); Those are all specialization of the first one, not overloaded functions. See also comment #11. *** This bug has been marked as a duplicate of 99 ***