Title: KF 2.10 minor cosmetic bug Post by: Pauldelbrot on June 28, 2016, 05:30:21 PM During reference orbit and series calculations, KF multiplies the iterations by 100, using 32-bit signed integer arithmetic, before dividing by max-iters. As a result, if there are parts of the image with over 21,474,836 iterations, the R: number will be seen to jump to negative numbers after that iteration, then climb back up towards (and possibly past, if over 40 million iterations are needed) zero. The A: number subsequently does the same thing. However, both numbers then jump to be correct and the image itself appears to be rendered correctly, so this seems to be a cosmetic issue. Using doubles for the intermediate calculations would, of course, fix the bug, as would using 64-bit longs (or at least, that would push the point where the bug occurs up to such ludicrously high numbers of iterations that nobody will likely be going there in the foreseeable future). |