RFC: fix PR 15364

[Andreas Arnez]

Tom Tromey <tromey@redhat.com> writes:

> My earlier fix for PR 9197 introduced a regression.
> You can see it clearly here:
>
> (gdb) p s2
> $1 = (struct s *) 0x0
> (gdb) p s2 && s2->x == 0
> Cannot access memory at address 0x0
>
> The exception in this case is thrown by binop_promote, but I think that
> a lot of code expects EVAL_AVOID_SIDE_EFFECTS to return a not_lval
> value, so I fixed the problem by changing the various spots I touched
> for 9197 to do this.
>
> Built and regtested on x86-64 Fedora 18.
> New test case included.

Since there hasn't been any objection, I assume that you're going to
commit the patch, right?