On 1/17/23 13:48, Jakub Jelinek wrote:
On Tue, Jan 17, 2023 at 01:43:35PM +0100, Christophe Lyon wrote:
As a follow-up to this, I ran the full testsuite with -fstack-protector-all
and this results in lots of failures (~65000 in gcc.sum alone).
I guess that is way too much.
Since you also mentioned -fstack-protector-strong, I ran the full testsuite
with it, which results in more failures too but the difference is much
smaller than with -fstack-protector=all (from 126 FAIL to 309)
But this could be doable by adding explicit -fno-stack-protector options
to test that can't handle those.
For instance, I see many failures with -fstack-protector-strong in:
gcc.target/aarch64/sve/pcs/
It looks like you have them too, according to the logs I downloaded from
your link above.
So is it worth adding -fno-stack-protector to my few new testcases?
(I can, no problem, but just wondering why you appear to notice the problem
with my new tests, and not with the ones in gcc.target/aarch64/sve/pcs/)
Because I mainly look for regressions (compare the test_summary
dumps against older gcc build); if something fails for years, it doesn't
show up in the regression diffs.
OK that's what I thought, thanks for confirming.
I'll add -fno-stack-protector to my tests.
Thanks,
Christophe
Jakub
