On Mon, Jan 20, 2025 at 10:29:53AM +0100, Richard Biener wrote:
> That said, the testcase in the PR is a bit artificial, so we might get away
> with
> ignoring the issue for GCC 15 and go for fixing tree_could_trap_p?
I'm afraid code like that can be seen in the real-world, it doesn't have to
be written by the user to clearly do out of bounds access, it is enough if
there is some inline function testing e.g. the array index for some
particular large value for whatever reason and then jump threading
creating bb with that clear out of bounds access because of that.
So I'd feel unsafe to leave this unfixed for GCC 15.
Jakub
