* this file should go away before asking for a review :) * I think we should wait for indexing-by-arrays-2, since this would clean up the iterator interface * For getitem, we need to reuse parent getitem, but with some hook that recomputes indexes. That hook would be used for any sort of access, be it slices or be it integers, but in general I would like to avoid code duplication, since the indexing is getting slowly fairly complex. * iterating over a transposed array still fails. * next_skip_x is implemented badly. It still iterates over items (calls next_skip n times) and is jit.unroll_safe. This means that the JIT will assume skip is a constant and compile a linear trace unrolling all iterations. This is *BAD*. Instead it would be something like: for arrays: res.offset = self.offset + skip for views something slightly more complex that does appropriate modulo. Note that next() has unroll_safe, because the loop only dependas on the length of shape which is a constant for assembler.