unknown bound: array error.

[Mark J. Turick]

Thats curious that your test application compiled with gcc3.1.  I
grabbed the
latest cvs snapshot and tried again without any better luck. I think
I traced
my problem to gcc/gcc/cp/decl.c: lines 11957-11974.

           /* [dcl.fct]/6, parameter types cannot contain pointers
                 (references) to arrays of unknown bound.  */
              tree t = TREE_TYPE (type);
              int ptr = TYPE_PTR_P (type);

              while (1)
                {
                  if (TYPE_PTR_P (t))
                    ptr = 1;
                  else if (TREE_CODE (t) != ARRAY_TYPE)
                    break;
                  else if (!TYPE_DOMAIN (t))
                    break;
                  t = TREE_TYPE (t);
                }
              if (TREE_CODE (t) == ARRAY_TYPE)
                cp_error ("parameter `%D' includes %s to array of
unknown
bound `%T'",
                          decl, ptr ? "pointer" : "reference", t);

Maybe I'm too dense, but there has to be a lot of code around that
breaks
this rule. After all, its hard to see the difference between a
'**var' and
'*var[]', just look at all the code still around with "main( int
argc, char
*argv[])" and/or "main( int argc, char **argv)" as an obvious
example.

I guess we're stuck using gcc 2.95....


Ingo Krabbe wrote:

> On Mon, 9 Jul 2001, Mark J. Turick wrote:
>
> >
> > I have a large amount of legacy code for a project that we support
> > on
> > various unix platforms. Some of the internal structures are defined
> > as:
> >
> > typedef struct {
> >  int xpos;
> >  int ypos;
> > } XYCoordStruct, (*XYCoordPtr)[];
> >
>         Ok it seems to be another standart problem. While Solaris 8 and
> GCC Compilers try to behave as standart compilers many others try to
> compile code that programmers thought of nice looking code. Its a kind of
> philosophy. I'm not involved in the devlopment of GCC, but I think it goes
> in the right direction, thinking about the future.
>         By the way. I don't get this error with gcc 3.1 (which comes from
> a cvs checkout).
>         typedef struct
>         {
>                 int xpos;
>                 int ypos;
>         } XYCoordStruct, *(XYCoordPtr)[];
>
>         int libdReturnVCoords (
>                 int id_index, int time, int npts,
>                 XYCoordPtr rpts, int *nearest );
>
>         int main( argc, argv )
>                 int argc;
>                 char** argv;
>         {
>                 int x;
>                 XYCoordStruct* a;
>                 libdReturnVCoords( 10, 5, x, &a, &x );
>                 return 0;
>         }
>
> compiles fine, though complaining about missing libdReturnVCoords at the
> linker stage of course. But I cannot declare a Variable to be of type
> XYCoordPtr of course!
>         XYCoordPtr b;
> gives a "array size missing in `b'" error.