range(a,b,c) not optimized for (at compile time) unknown step value

[robertwb]

We optimized range(a) and range(a,b), why not this one?

[scoder]

We do. You may just have hit one of these cases:

            if not isinstance(step.constant_result, (int, long)):
                # cannot determine step direction
                return node
            step_value = step.constant_result
            if step_value == 0:
                # will lead to an error elsewhere
                return node

Could you provide example code that doesn't work?

[robertwb]

Hmm... it was probably the first case there.

[robertwb]

Actually, we can turn that second version into a c loop, it will just have to have a test that depends on the (runtime) sign of the step.

[mark]

Replying to robertwb:

Actually, we can turn that second version into a c loop, it will just have to have a test that depends on the (runtime) sign of the step.

What prange does is simple arithmetic, i.e. compute the number of steps that will be taken and compute the index variable from that. That way you won't have to use 'if' and duplicate the loop body:

nsteps = (stop - start) / step; for (i = 0; i < nsteps; i++) { index = start + step * i; ... }