Hi Jonathan
Thanks very much for your reply. I'm sorry that I didn't provide the necessary information.
> Are you using the same linker version in both cases? That might be as
> significant as the compiler version.
gcc 4.6.3 is running with ld 2.22 <==== Errors
gcc 4.4.7 is running with ld 2.20 <==== No errors
> How is EVD.o compiled?
g++ -c -Wall -m32 -I/usr/include/python2.7 -I/usr/include/boost -O3 EVD.cpp -o _gnuRelease/EVD.o
> At a guess, maybe because the function is declared as something like
> Matrix<T>::eye(unsigned, double = 0.0) i.e. just because you pass the function
> an int doesn't mean that's what the function is declared to take.
You are absolutely correct, of course, eye() is defined in Matrix.h as follows:
template<class T> class Matrix : public MatrixBase
{
public:
<snip>
static Matrix<T> eye( unsigned a_size, T a_scalingFactor=1.0 ) DONT_INLINE;
<snip>
}
Does that point to the root cause of the problem?
Best regards
David
> -----Original Message-----
> From: Jonathan Wakely [mailto:jwakely.gcc@xxxxxxxxx]
> Sent: 04 November 2013 15:19
> To: David Aldrich
> Cc: gcc-help@xxxxxxxxxxx
> Subject: Re: Template linker errors with gcc 4.6.3
>
> On 4 November 2013 15:07, David Aldrich wrote:
> > Hi
> >
> > We have a mature project that builds fine with gcc 4.4, but when we build it
> with gcc 4.6.3 on Ubuntu 12.04 LTS we get linker errors that are related to
> templates.
>
> Are you using the same linker version in both cases? That might be as
> significant as the compiler version.
>
>
> > I am finding it difficult to understand the errors. As a general question, was
> template processing significantly changed between gcc 4.4 and 4.6 in a way
> that might cause linker errors?
>
> Not that I can think of.
>
> > Here is an example of an error that I am getting:
> >
> > + make
> > g++ -o _gnuRelease/myapp -Wl,-whole-archive,-export-dynamic
> ../Kernel/_gnuRelease/libKernel.a -lboost_python-mt-py27 -lpython2.7 -Wl,--
> no-whole-archive -ldl
> > ../Kernel/_gnuRelease/libKernel.a(EVD.o): In function
> `eigSymmetric(Matrix<double> const&, Matrix<double>&, Matrix<double>&,
> unsigned int)':
> > EVD.cpp:(.text+0x25a): undefined reference to
> `Matrix<double>::eye(unsigned int, double)'
>
> How is EVD.o compiled?
>
> > Now, the only line of code in eigSymmetric() that calls eye() is:
> >
> > int N = a_X.numRows();
> > ...
> > a_V = Matrix<double>::eye(N);
> >
> > so I don't understand why the linker is looking for
> `Matrix<double>::eye(unsigned int, double)' rather than for
> `Matrix<double>::eye(int)'.
>
> At a guess, maybe because the function is declared as something like
> Matrix<T>::eye(unsigned, double = 0.0) i.e. just because you pass the function
> an int doesn't mean that's what the function is declared to take.
>
>
> > There are also many other errors:
> >
> > EVD.cpp:(.text+0x1c6e): undefined reference to
> `Matrix<std::complex<double> >::Matrix()'
> > EVD.cpp:(.text+0x1c7d): undefined reference to `Matrix<double>::Matrix()'
> > EVD.cpp:(.text+0x1f9e): undefined reference to
> `Matrix<std::complex<double> >::Matrix(unsigned int, unsigned int,
> std::vector<std::complex<double>, std::allocator<std::complex<double> > >
> const&, MatrixConst::EOrdering)'
> > EVD.cpp:(.text+0x1fe5): undefined reference to
> `Matrix<std::complex<double> > operator*<std::complex<double>
> >(Matrix<std::complex<double> > const&, Matrix<std::complex<double> >
> const&)'
> > EVD.cpp:(.text+0x2006): undefined reference to
> `Matrix<std::complex<double> > operator-<std::complex<double>
> >(Matrix<std::complex<double> > const&, Matrix<std::complex<double> >
> const&)'
> > EVD.cpp:(.text+0x2080): undefined reference to
> `Matrix<std::complex<double> >::frob2() const'
> > <snip>
>
> It looks like every member of Matrix is missing.
>
> Either the file EVD.o is compiled with -fno-implicit-templates or those functions
> are not defined in EVD.cpp
>
> Function templates must either be defined in every file that uses them or
> explicitly instantiated.
>
> But you haven't provided nearly enough information to know what the specific
> problem is in your case.
>
>
> Click
> https://www.mailcontrol.com/sr/b3FtO7ru!0fGX2PQPOmvUsrLibhXE7+ScsWHL3
> j8xK29vWTEKH!k8KB2zpygqLpOD80j2WpbjaLX0F9OzBqmnw== to report this
> email as spam.
