gcc.dg/noncompile/pr16876 should actually compile with a warning, rather
than fail with an error.

Some work on related issues has been done before in:
http://gcc.gnu.org/ml/gcc-patches/2004-10/msg00772.html

I suppose in this code:
    case ic_argpass:
    case ic_argpass_nonproto:
      /* ??? This should not be an error when inlining calls to
         unprototyped functions.  */
      error ("incompatible type for argument %d of %qE", parmnum, rname);
      break;
we should emit a warning rather than an error for the ic_argpass_nonproto case.
However, since we are then not set up to fail the compilation, we should not
return an error_mark_node, as this could mask subsequent actual
constraint violations;
Instead, we should return a value of the required type; e.g. we could
make NULL pointer of type pointer to type, and generate and INDIRECT_REF.