api: avoid premature weight index substitution

devito/finite_differences/differentiable.py

@@ -144,13 +144,6 @@ def coefficients(self):
         key = lambda x: coeff_priority.get(x, -1)
         return sorted(coefficients, key=key, reverse=True)[0]
 
-    @cached_property
-    def _coeff_symbol(self, *args, **kwargs):
-        if self._uses_symbolic_coefficients:
-            return W
-        else:
-            raise ValueError("Couldn't find any symbolic coefficients")
-
     def _eval_at(self, func):
         if not func.is_Staggered:
             # Cartesian grid, do no waste time
@@ -435,10 +428,6 @@ def highest_priority(DiffOp):
     return sorted(DiffOp._args_diff, key=prio, reverse=True)[0]
 
 
-# Abstract symbol representing a symbolic coefficient
-W = sympy.Function('W')
-
-
 class DifferentiableOp(Differentiable):
 
     __sympy_class__ = None

devito/finite_differences/finite_difference.py

@@ -158,8 +158,6 @@ def make_derivative(expr, dim, fd_order, deriv_order, side, matvec, x0, coeffici
     # `coefficients` method (`taylor` or `symbolic`)
     if weights is None:
         weights = fd_weights_registry[coefficients](expr, deriv_order, indices, x0)
-    elif wdim is not None:
-        weights = [weights._subs(wdim, i) for i in range(len(indices))]
 
     # Enforce fixed precision FD coefficients to avoid variations in results
     weights = [sympify(w).evalf(_PRECISION) for w in weights]
@@ -193,6 +191,8 @@ def make_derivative(expr, dim, fd_order, deriv_order, side, matvec, x0, coeffici
         deriv = DiffDerivative(expr*weights, {dim: indices.free_dim})
     else:
         terms = []
+        if wdim is not None:
+            weights = [weights._subs(wdim, i) for i in range(len(indices))]
         for i, c in zip(indices, weights):
             # The FD term
             term = expr._subs(dim, i) * c