Mateusz Loskot wrote:
Jeffrey Hellrung wrote: [...]
FWIW, I have working implementations of Schewchuk's expansion arithmetic algorithms (in free functions and classes utilizing these free functions), both for fixed-sized expansions (with error estimates) and arbitrary-sized expansions, if you (or anyone) is interested. It's not self-contained (depends on some other components of our code base), but could be a good starting point if you're looking in that direction. We currently use it in an adaptive context to "robustify" our geometric predicates. Of course, it isn't 100% robust, as overflow and underflow could happen, but so far it has not failed. It's also not going to be as efficient as the "tricks" Shewchuk uses in his triangulation (I'm guessing), but it still gets you somewhere.
Jeffrey, I'd be interested in taking look at it myself. Perhaps you could push it to the Sandbox?
I don't have a boost svn-sandbox account, but seeing as how the code is not (currently) self-contained and depends on a larger code base, I don't think it belongs in the sandbox. I have zipped up just the expansion arithmetic code and uploaded it to www.math.ucla.edu/~jhellrun/sake-svn/sake/expansion_arithmetic.7z If you are interested in actually using this code, we should discuss. I imagine it could be excised out of its containing code base without too much difficulty. - Jeff