linking error for static member variables

[aaragon]

Hi everyone,

I have a linking error when using gcc4.2 and static member variables.
The class template definition is something around the following:

template<>
class Element<L2_t> : public Element_common<L2, Side<2,2> > {

public:

static const ubyte_t dim_ = 2;
static const ubyte_t num_nodes_ = 2;
static const ubyte_t gauss_pts_= 3;

// member functions
};

Undefined symbols:
"fea::Element<(fea::Element_type)1>::gauss_pts_", referenced from:
__ZN3fea7ElementILNS_12Element_typeE1EE10gauss_pts_E
$non_lazy_ptr in element.o
"fea::Element<(fea::Element_type)2>::gauss_pts_", referenced from:
__ZN3fea7ElementILNS_12Element_typeE2EE10gauss_pts_E
$non_lazy_ptr in element.o
"fea::Element<(fea::Element_type)3>::gauss_pts_", referenced from:
__ZN3fea7ElementILNS_12Element_typeE3EE10gauss_pts_E
$non_lazy_ptr in element.o
ld: symbol(s) not found

The strange thing is that I have many other static variables with the
same kind of definition (same type and const as you can see) but it
only complaints about the gauss_pts_ member variable. I solved the
problem by moving the definition to the cpp file as follows:

// in cpp file
const ubyte_t Element<L2_t>::gauss_pts_ = 3;

But I want to understand really what is happening here. When I compile
the code with gcc4.3, the problem disappears. So, is this a bug in the
compiler?

Thanks for the help,

aa

 

[aaragon]

WTH is 'ubyte_t'?

just a type definition to cope with the way I name things.

Not much.  Any static member variable/constant has to be defined if used
outside of the class.  Since you didn't show us how those constants are
used, the assumption is that they *are* used outside and *must* be
defined or you're in violation of the ODR.

It is defined, when inside a class you do something like

static const int test = 2;

you're defining inline the value of the constant. That is exacly what
I was doing before, but the linker couldn't find it. Weird thing is
that only happened with one varible, the other static constants were
defined in the same way, but they didn't give me problems. So there
must be a problem with the copmiler.

 

[aaragon]

Well, the language prohibits initialisations of static const data
members *unless* they are of an *integral type*.  If your 'ubyte_t' is
not integral, your code is ill-formed.

u_byte is just a type definition for an unsigned short, which is of an
integral type. The rest of the static constant member variables were
also of the same type but only this one game the problem. Weird.

Nope.  It's declared and initialised.  It's not defined.  It's only
defined when you can say that the memory for it is allocated and *where*
it is allocated.  Unless you do

   const int <classname>::test;

in the *namespace* scope somewhere in a translation unit, the data
member is *undefined*.

 > That is exacly what


Sure.  What else could it be?..  Stupid compilers!

Indeed, stupid compilers!

V

Thanks for answering my post.

aa