Robert Dodier wrote:
> On Wed, Jan 21, 2009 at 9:32 AM, Aleksej Saushev <asau at inbox.ru> wrote:
>
>> Maybe it is better to default to /tmp or ${TMPDIR}?
>
> Well, /tmp is problematic. There is potential for collisions between
> different users, and what's worse, stuff may get erased after a while.
>
> I guess we could make the user name part of the path, to avoid collisions,
> but I don't know a way to work around the file volatility feature.
>
> Robert
Speaking about affine, i think that the best place to compile it is
in "binary-gcl" or "binary-cmucl" subdirectory of affine like it was
previously. This way the installation is shared by all users. The only
thing to be done is to mention in the documentation of affine that
it must be compiled as root or administrator once. For me this solution is
ways simpler and more appropriate than putting binaries in some strange
place, ready for collisions.
While on that i have compiled affine on 5.17.1 by using the indication by
R. Toy somewhere on Google, that is, replace last line by:
(mk:oos "affine" :compile)
and it works no problem. The Grobner basis computation in affine is *much
much* faster than the one in the Grobner package. Little problem, such
basis are not unique, and it is difficult to check results between two
implementations, i have tried comparing with Maple. Unfortunately affine is
very little documented, in particular in the way order of variables is
defined, and their weight. But the result entirely depends on that.
Anyways, speedwise, Maple and affine are in the same league, while the
Grobner package leads to extremely slow computations, for example for
/* A tough example learned from Cox */
poly_grobner([x^5+y^4+z^3-1,x^3+y^3+z^2-1], [x,y,z]);
takes 1mn30s
where affine and Maple answer after around a second.
--
Michel Talon