https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111800
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |rsandifo at gcc dot gnu.org
--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Ah, I see what's going on.
tree-ssa-ccp.cc uses widest_int extensively (I believe it is a mistake and
should be using wide_int instead) and in this case has a widest_int with value
-30 (get_len () == 1, get_precision () before my changes 576, with those
changes 32640 bits).
Now, print_hex (but also print_decu/print_decs/print_dec as they always print
larger precision (and when get_len () > 1) using print_hex) prints all numbers
according to the precision, so my checks to ensure the buffer is big enough:
char buf[WIDE_INT_PRINT_BUFFER_SIZE], *p = buf;
unsigned len = wi.get_len ();
if (UNLIKELY (len > WIDE_INT_MAX_INL_ELTS))
p = XALLOCAVEC (char, len * HOST_BITS_PER_WIDE_INT / 4 + 4);
are not correct, they'd need to check precision for wi::neg_p (wi) values.
