Since I really need the #1518273 to be fixed ASAP,
I prepared the fix myself.
Attached patch does two things:
1. Casting non-pointer to generic pointer now works -
the far pointer is implied.
2. adjustArithmeticResult() should not assume the
left-arg is always a pointer if the result is. It
seems like CSE breaks that assumption, so it should
not be made.
Can this be applied?