Re: GMP code committed -- watch for bugs.

[Kevin Ryde]

Rob Browning <address@hidden> writes:
>
> There may well be bugs

In >, < and =, mpz_cmp_d cannot be called with a NaN, that'll need to
be tested explicitly.

The doco doesn't say much about nans, but you can imagine with a
positive/zero/negative return there's no scope to indicate
"unordered".


Also, unfortunately mpz_cmp_d currently doesn't recognise infinities,
so those will need to be checked before making a call too.

The current code ends up treating infinities as 2^1023 or some such
big value, so the problem will only be seen with bignums larger than
that.

I'd been meaning to add infinities to mpz_cmp_d and friends, I'll see
if that can be in gmp 4.2.  In which case perhaps a bit of a macro
like below (untested) to hide the difference between the versions,


/* mpz_cmp_d only recognises infinities in gmp 4.2 and up.
   For prior versions use an explicit check here.  */
#if __GNU_MP_VERSION < 4                                        \
  || (__GNU_MP_VERSION == 4 && __GNU_MP_VERSION_MINOR < 2)
#define guile_mpz_cmp_d(z, d)                           \
  (xisinf (d) ? (d < 0.0 ? 1 : -1) : mpz_cmp_d (z, d))
#else
#define guile_mpz_cmp_d(z, d)  mpz_cmp_d (z, d)
#endif