@User: It's possible (likely?) that klayout is working just fine. Magic has a fundamentally different way of describing non-Manhattan geometry, using "split tiles", such that all angle corners are by definition on the grid. A "wire" statement in the GDS would just describe the centerline path plus the width of the wire. It would be up to the tool to render the wire from the path and ensure that all points are on-grid. So I think it is likely that it's still Magic's fault; it is reading a wire that describes a path such that the corner points are not on-grid, and it snaps them to the grid in an inconsistent manner, maybe dependent on the quadrant the points are in, or maybe simple round-off error. So if you could post the original GDS for that simple example with metal1 and metal2, I can look into how magic reads the GDS paths and see if there is a round-off issue that needs fixing.