Maxima 5.16.3 release --- enourmous slowdown of gf-package

On 8/26/08, Oliver Kullmann <O.Kullmann at swansea.ac.uk> wrote:

>  field : [2,8,x^8+x^4+x^3+x+1];
>  apply(gf_set,field);
>  Evaluation took 0.1280 seconds (0.1272 elapsed) using 840.867 KB.
>  (%o3) true
>
>  and now in 5.16.3:
>
>  apply(gf_set,field);
>  Evaluation took 4.7883 seconds (4.9496 elapsed) using 42.714 MB.
>  (%o58) true

Oliver, I believe the difference in behavior can be attributed
to a change in the behavior of the ifactors function:
in Maxima 5.15, ifactors(1) => [] (an empty list), while in 5.16,
ifactors(1) => [[1, 1]]. (This new return value makes sense to me
so I don't see any need to change it back.)

The [[1, 1]] from ifactors makes gf_findprim act differently: there
is a loop it goes through many times.  GF_VERBOSE:true shows that.

I find the following change to gf.mac seems to speed up gf greatly
without changing the expected behavior as indicated by gf_test.

I'll commit this patch to cvs trunk and 5.16 release branch,
although I'm not promising to make another release on the branch.

Please try this modification and let me know how it turns out for you.

best

Robert Dodier

PS. Here is the patch.

--- share/contrib/gf/gf.mac 16 Jul 2008 05:36:45 -0000  1.3
+++ share/contrib/gf/gf.mac 28 Aug 2008 03:32:29 -0000
@@ -411,7 +411,7 @@
   block([f,fastf,slowf,found:false,fastCase,
          lf,lfastf,lslowf,decomp,main_powers,q,elt,leng,i,test,fast_test,j,blist,sig,genn2l],
   sig:?even(gf_exp),
-  fastf : ifactors(gf_char-1), /* "fast" factors of gf_char-1 */
+  fastf : if gf_char=2 then [] else ifactors(gf_char-1), /* "fast"
factors of gf_char-1 */
   lfastf : length(fastf),
   fastf : makelist(fastf[i][1],i,1, lfastf),
   fastfex : makelist([fastf[i],true,(gf_char-1)/fastf[i]],i,1,lfastf),