On 04/24/2017 05:35 PM, Martin Sebor wrote:
Bug 80497 brings to light that my fix for PR 80364 where I corrected
the handling for int128_t was incomplete. I handled the non-constant
case but missed the INTEGER_CST case just a few lines above. The
attached patch also corrects that problem plus one more elsewhere
in the pass.
Both of the changes in this patch seem safe enough to make even now
in GCC 7 but since they are ice-on-invalid-code perhaps it's better
to wait for 7.1?
Martin
gcc-80497.diff
PR tree-optimization/80497 - ICE at -O1 and above on valid code on
x86_64-linux-gnu in tree_to_uhwi
gcc/ChangeLog:
PR tree-optimization/80497
* gimple-ssa-sprintf.c (get_int_range): Avoid assuming all integer
constants are representable in HOST_WIDE_INT.
(parse_directive): Ditto.
gcc/testsuite/ChangeLog:
PR tree-optimization/80497
* gcc.dg/tree-ssa/builtin-sprintf-warn-17.c: New test.
OK for the trunk. Jakub's call on when it's OK for the branch -- one
can easily argue this is a regression.
jeff
