update polymake interface to 2.14-rc1

[tkluck]

I've written a new interface for polymake using its shared library interface, while trying to stay backward compatible with what there was before.

I've classified this as a defect, since the current interface doesn't work anymore.

http://www.polymake.org/lib/exe/fetch.php/download/polymake-2.14r1-minimal.tar.bz2

Depends on #13768
Depends on #13767

CC: @williamstein @mkoeppe

Component: geometry

Work Issues: source tar link, fix doctests, info for how to install, add doctests

Author: Timo Kluck

Branch/Commit: public/ticket/14416 @ 50f6da9

Reviewer: Karl-Dieter Crisman, Frédéric Chapoton, Matthias Köppe

Issue created by migration from https://trac.sagemath.org/ticket/14116

[tkluck]

comment:1

Attachment: trac_14116_polymake_upgrade.patch.gz

A related effort is https://bitbucket.org/burcin/pypolymake

[kcrisman]

comment:2

Needs to be rebased to #13432.

[kcrisman]

comment:3

Uploading rebased patch next. But:

Compiling sage/geometry/polymake/polymake.pyx because it changed.
Cythonizing sage/geometry/polymake/polymake.pyx

Error compiling Cython file:
------------------------------------------------------------
...
    def save(self, filename):
        self.thisptr.save(<string><char*>filename)

    def __getattr__(self, propertyname):
        propertyname = propertyname.upper()
        return <str>get_property_wrapper(self.thisptr[0], <string><char*>propertyname).c_str()
                                                                                           ^
------------------------------------------------------------

sage/geometry/polymake/polymake.pyx:247:92: default encoding required for conversion from 'char const *' to 'str object'

This is presumably related to your comment about getting rid of std::string cpp_string = <string><char*>python_string stuff; the Cython in Sage is now 5.19.

Error compiling Cython file:
------------------------------------------------------------
...
                    p = i.numerator() if callable(i.numerator) else i.numerator
                    q = i.denominator() if callable(i.denominator) else i.denominator
                    entries.push_back(Rational(p,q))
                else:
                    entries.push_back(<Rational><int>i)
            matrix = new Matrix[Rational](rows, cols, entries.begin())
                                                                  ^
------------------------------------------------------------

sage/geometry/polymake/polymake.pyx:268:67: Cannot assign type 'iterator' to 'vector[Rational]'

No idea, though the message seems like it would be clear to someone who knew Cython.

[kcrisman]

Attachment: trac_14116-polymake-upgrade-rebased.patch.gz

[tkluck]

comment:4

kcrisman: I'm considering replacing my own work by what Burcin has done with pypolymake. He works around some cython issues in a much nicer way. Have you looked at that, too?

[kcrisman]

comment:5

Have you looked at that, too?

No. It's not exactly clear to me what to do (though this seems likely), and the forums suggest that it doesn't have as much functionality as the interface here, though I may be misinterpreting that. It would be nice to have it in an spkg form, though I understand that is different from the lmonade philosophy.

Also note this polymake user's attempt, though apparently they later just used pypolymake. I assume that user is not yourself?

[kcrisman]

comment:6

In particular, the following things seems like hacks:

ln -s /usr/local/sage-4.7.1/devel/sage/c_lib/include /usr/local/sage-4.7.1/local/include/sage

and this apparently required patch

diff --git a/sage-env b/sage-env
--- a/sage-env
+++ b/sage-env
@@ -219,6 +219,9 @@
     mkdir -p "$MPLCONFIGDIR"
 fi
 
+LD_PRELOAD="$SAGE_LOCAL/lib/libpolymake.so"
+export LD_PRELOAD
+
 LD_LIBRARY_PATH="$SAGE_ROOT/local/lib/:$LD_LIBRARY_PATH" && export LD_LIBRARY_PATH
 # The following is needed for openmpi:
 LD_LIBRARY_PATH="$SAGE_ROOT/local/lib/openmpi:$LD_LIBRARY_PATH" && export LD_LIBRARY_PATH

which I'll point out would no longer apply directly, as the LD_LIBRARY_PATH code has changed, though one could just add the lines, of course. And of course sage-env doesn't live in local/bin any more, but in spkg/bin.

[tkluck]

comment:7

I just pushed some initial work to

https://github.com/tkluck/sage/tree/polymake-upgrade

It integrates pypolymake with Sage, while keeping backwards compatibility with the old interface. Additionally, it adds an interface allowing you to specify polytopes by Sage expressions.

I'm having some trouble getting the doctests to work, since polymake has a habit of "spamming" stderr. (Actually, it outputs citation information, which is worthwhile. We do need to deal with it, though.) Also, trying to be backwards compatible both with Sage and with pypolymake leads to ugliness such as having both n_points and num_points methods.

I'm doing development in the new integrated git repository, which is supposed to replace the mercurial repositories sometime this Summer. Here are instructions for reviewing my branch. They include cloning the repository and building Sage from source, so this takes a lot of time. I will also make mercurial patches available for review at some point.

$ # clone the repository
$ git clone git://github.com/tkluck/sage.git
$ cd sage
$ git checkout -t polymake-upgrade
$ # get the upstream package for polymake
$ cd upstream
$ wget http://polymake.org/lib/exe/fetch.php/download/polymake-2.12-rc3.tar.bz2
$ mv polymake-2.12{-rc3,}.tar.bz2
$ cd ..
$ # build sage
$ MAKE="make -j4" make
$ # install the polymake package
$ ./sage -i polymake
$ # build the polymake interface, now that the package is available
$ ./sage -b