I strongly object to removing gfan. -Marshall
On Oct 19, 9:53 am, William Stein <wst...@gmail.com> wrote: > On Tue, Oct 19, 2010 at 6:35 AM, mhampton <hampto...@gmail.com> wrote: > > I support replacing cddlib with PPL for the default computation of > > exact polyhedra. > > Does gfan depend on cddlib? If we remove cddlib, would we then also have to > remove gfan? From deps: > > $(INST)/$(GFAN): $(BASE) $(INST)/$(MPIR) $(INST)/$(CDDLIB) > $(INSTALL) "$(SAGE_SPKG) $(GFAN) 2>&1" "tee -a > $(SAGE_LOGS)/$(GFAN).log" > > I would personally be fine with removing gfan, unless some people strongly > object. > > > > > > > Eventually I would really like to see lrs as a standard component as > > well, because: > > 1) It is small and compiles on all our supported platforms. > > 2) David Avis has always been very responsive to any emails about lrs. > > 3) It uses a very different algorithm from PPL; sometimes it performs > > much better, and always with relatively little memory use. > > 4) It has been an optional package for a couple of years now, with no > > known problems. > > > PPL can be the default, ideally with lrs as an optional backend. > > > For numerical polytopes we should keep cddlib for now. Qhull might be > > the better long-term choice for numerical polytopes, since it was > > written from the beginning with that in mind. For now I am working on > > Qhull as an optional package (but its not very high on my priority > > list). > > > -Marshall > > > On Oct 18, 7:13 am, Volker Braun <vbraun.n...@gmail.com> wrote: > > > === Introduction === > > > > I would like to have the Parma Polyhedra Library (PPL) included > > > as a standard spkg. My goal is to rewrite as much of the > > > sage.geometry.* modules on top of a Cython PPL wrapper as opposed > > > to piping ASCII to/from cddlib. Some of the reasons are: > > > > * PPL is already faster by itself, and having a Cython wrapper reduces > > > overhead. I'm seeing about 20x speedup on medium-sized problems. > > > > * PPL is the only polyhedral computation toolkit that is from the > > > ground up designed to be used as a shared library.(1) > > > > * PPL is mature, well tested, has an active development including bug > > > tracker and mailing list. It is used in a variety of other projects, > > > including gcc's Graphite loop optimizer. It has been tested on > > > Linux, FreeBSD, OpenBSD, Solaris, IRIX64, Mac OS X, Cygwin, DEC > > > OSF/1. > > > > cddlib was the first really useful software package back in the day > > > and it was hugely influential for many subsequent codes. But nowadays > > > there are alternatives that combine the same algorithms with better > > > implementations and software development practices.(2) > > > > Note that I'm not proposing to remove cddlib entirely. It is still a > > > requirement for gfan and the only program to do polyhedral > > > computations over floating point numbers.(3) > > > > === Work Plan === > > > > I have written a PPL spkg and a Cython (C++) wrapper that should cover > > > all functions necessary for dealing with polyhedra in Sage. You can > > > find it here: > > > >http://trac.sagemath.org/sage_trac/ticket/10039 > > > > Using this wrapper, I patched sage.geometry.cone: > > > >http://trac.sagemath.org/sage_trac/ticket/10140 > > > > I think that the PPL spkg and the Cython wrapper is ready for > > > inclusion in the next Sage release (4.6.1?). In the subsequent release > > > we can then rebase sage.geometry modules on top of the wrapper. > > > > Right now I would like to invite any comments on this plan! > > > > === Notes === > > > > (1) cddlib also comes with a shared library. But its poorly documented > > > and, according to valgrind, leaks memory. Other toolkits like lrs and > > > qhull don't build shared libraries but have minimal instructions for > > > statically linking with parts of their code. > > > > (2) In particular, cddlib has no bug tracker, no new release in the > > > last 5 years, and segfaults on some of its own testsuite. > > > > (3) Though in a numerically unstable way, see Trac #10046, #10037. > > > -- > > To post to this group, send an email to sage-devel@googlegroups.com > > To unsubscribe from this group, send an email to > > sage-devel+unsubscr...@googlegroups.com<sage-devel%2bunsubscr...@googlegroups.com> > > For more options, visit this group at > >http://groups.google.com/group/sage-devel > > URL:http://www.sagemath.org > > -- > William Stein > Professor of Mathematics > University of Washingtonhttp://wstein.org -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
