https://bugs.kde.org/show_bug.cgi?id=242137
--- Comment #33 from Julian Seward <jsew...@acm.org> ---
Those checks are enabled now by --expensive-definedness-checks=yes, which
was introduced in 3.11.0.
--expensive-definedness-checks=no|yes
Use extra-precise definedness tracking
[no]
There's still a problem with LLVM optimised code though, and to some extent
also with gcc. Both of these compilers will now sometimes compile
if (a && b) ...
as if it was written
if (b && a)
in the case that they can prove that a is always false whenever b is undefined.
The transformation is correct but it means that there is a branch on
uninitialised
data and so Memcheck complains (wrongly.)
--
You are receiving this mail because:
You are watching all bug changes.
