Comment 10 for bug 1460152

FYI, the hash approach is slow for the normal case since we always have to perform an sum. Furthermore it doesn't take into account #include'd files that might also change (eg, apparmor is updated and has a different base abstraction). For the workaround, I guess it is ok since the slowdown will only be for a couple of profiles but I would have rather seen us unconditionally invalidating the cache when switching from a to b or vice versa.