gmp3

[Henry Cejtin]

I'll  definitely  check  it out quickly.  The only worry is that they changed
the allocation behavior of some of the mpn stuff.  If they did then it  would
still call malloc, which would be bad.

The GMP stuff is very nice, but the non-allocating layer isn't optimial.  For
one thing it only deals with natural numbers (which has nothing  to  do  with
allocation).   More  seriously,  some  of  the  mpn  routines  (like  gcd) DO
allcation.

I'm curious about your mentioning of  gmp2  bugs.   My  quick  scan  of  gmp3
revealed some algorithmic improvement, but I haven't seen any bug stuff.