Comment by nikic

That's a great point. I initially thought we could assume no exposure for loads with non-pointer-compatible TBAA, but you are right that this is not correct if the memory has been laundered through memcpy.