On Wed, 2014-12-17 at 11:38 +0000, Barak A. Pearlmutter wrote: > > I think this must be a bug in the Loongson FPU (my guess since the > > assertion is related to floating point math). It builds successfully on > > all the non-loongson boards I've tried. > > Should I allow the build to proceed even when the test fails? > I'm reluctant to do that because people might prefer to avoid doing > geometric calculations on a known-buggy platform.
I think it's best to leave it failing at the moment since the assertion will probably fail when clippoly is run anyway, and like I said I don't think this is really a bug in clippoly. > > On a side note, it might be better to replace the entire function with > > the constant time method of finding the intersection of two lines: > > https://en.wikipedia.org/wiki/Line%E2%80%93line_intersection#Intersection_of_two_lines > > Patches welcome! > Qhull does go to some trouble to deal correctly with floating point > issues, so it might be handling tricky cases better than the method > you're pointing to. Possibly. It just struck me as strange to use a recursive method when a perfectly good constant time one exists (I haven't done any testing / coding with it though). Thanks, James > Should be picked up by the test suite though... > > Cheers, > > --Barak.
