Boost logo

Boost :

Subject: Re: [boost] GGL Review
From: Brandon Kohn (blkohn_at_[hidden])
Date: 2009-11-17 08:35:31


I think this discussion has deviated a bit of course from what Jonathan
and I were discussing. I should have been a bit more explicit when I
said 'Why do we need another buggy library.'. If you follow our
discussion, I started by stating my concerns about the floating point
robustness of GGL with respect to the boolean operations. The point I
was trying to make was that I believe that GGL would be the first
library of its kind in Boost that walks the line of being provably
correct (WRT boolean operations with FP coordinates) and that this
perhaps means that more care needs to be taken in reviewing it. Jonathan
  replied that for his needs it didn't matter if the algorithms were
correct, only that they worked most of the time. My own view is that
something that only works part of the time really doesn't work. From
these two polarized views came my statement which was essentially meant
to say: Why would we really want an algorithm that only works some/most
of the time in Boost? Of course a good argument to that is that we can
use GMP as the coordinate type if we need the added stability. Fine.
This still comes back to my original post/reply that (IMO) these
considerations are treading new ground in a review process and that the
author of the library needs to address these concerns by providing tests
proving any (implied) claims of an algorithm's correctness. If this
cannot be done at the start, perhaps the boolean operations should be
omitted until such a time as they can be properly proven. Luke seems to
have much expertise in the area of boolean operations, and has already
suggested a battery of tests. I think this would be an excellent place
to start.

Regards,

Brandon


Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk