https://gcc.gnu.org/bugzilla/show_bug.cgi?id=30475
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #64 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
The X + 100 > X case is optimized with:
/* X + Y < Y is the same as X < 0 when there is no overflow. */
(for op (lt le gt ge)
(simplify
(op:c (plus:c@2 @0 @1) @1)
(if (ANY_INTEGRAL_TYPE_P (TREE_TYPE (@0))
&& TYPE_OVERFLOW_UNDEFINED (TREE_TYPE (@0))
&& !TYPE_OVERFLOW_SANITIZED (TREE_TYPE (@0))
&& (CONSTANT_CLASS_P (@0) || single_use (@2)))
(op @0 { build_zero_cst (TREE_TYPE (@0)); }))))
So, if we wanted to maintain -Wstrict-overflow=*, we'd need to add it to this
spot and to other 40+ or so other spots that check for TYPE_OVERFLOW_UNDEFINED.
In this particular case, it would need to differentiate between whether @0 is a
constant (then it should warn at WARN_STRICT_OVERFLOW_ALL level), or not,
then it should warn at WARN_STRICT_OVERFLOW_COMPARISON level).
Though, -fsanitize=undefined really is the preferred way how to check code,
warnings will necessarily have lots of false positives, while the X + 100 > X
case when written that way is easy to avoid, if it is e.g. after inlining it
might not be that easy and there can be much more complex cases.
