> It has some chance to fail, but it fails more often on Cog. I think the last
> assert should be removed.

But we want to ensure that it does do a pointer-swap, instead of
removing it, shouldn't it be changed to:

  deny:  a == b