Re-added the assert on the exponent, but replacing MPFR_EMIN_MIN by

MP_EXP_T_MIN (this makes more sense): an assertion failed would mean
that the exponent is not representable (an undefined behavior in the
ISO C standard). If need be, we could choose to return MP_EXP_T_MIN
in such a case, or perhaps MP_EXP_T_MAX to signal an error. The
mantissa would still be meaningful.


git-svn-id: svn://scm.gforge.inria.fr/svn/mpfr/trunk@2459 280ebfd0-de03-0410-8827-d642c229c3f4

1 files changed, 5 insertions, 4 deletions

diff --git a/get_z_exp.c b/get_z_exp.c
index 5eaa3c62f..763c5a9b0 100644
--- a/get_z_exp.c
+++ b/get_z_exp.c
@@ -65,10 +65,11 @@ mpfr_get_z_exp (mpz_ptr z, mpfr_srcptr f)
 
   SIZ(z) = MPFR_SIGN(f) < 0 ? -fn : fn;
 
-  /* This always fails for very small "f", ie. when MPFR_EXP(f) is equal to
-     or only just above MPFR_EMIN_MIN.
-  MPFR_ASSERTN((mp_exp_unsigned_t) MPFR_GET_EXP (f) - MPFR_EMIN_MIN
+  /* Test if the result is representable. Later, we could choose
+     to return MP_EXP_T_MIN if it isn't, or perhaps MP_EXP_T_MAX
+     to signal an error. The mantissa would still be meaningful. */
+  MPFR_ASSERTN((mp_exp_unsigned_t) MPFR_GET_EXP (f) - MP_EXP_T_MIN
                >= (mp_exp_unsigned_t) MPFR_PREC(f));
-  */
+
   return MPFR_GET_EXP (f) - MPFR_PREC (f);
 }