HOOPL:
  - get Simon to review Lerner, Ullman & Kam, Kildall
  - implementation section: why?
       . really exists
       . not insane
       . dramatically simpler than original & other implementations
       . can see some of the idea
  - massive failure of parallel structure between sections 3 and 6
    (three bullets)
  - Hypothesis:

      if {P} s --> s' by rewrite, and {P} s {Q} and {P} s' {Q'}, then
      Q \sqsubseteq Q'

      if s {Q} --> s' by rewrite, and {P} s {Q} and {P'} s' {Q}, then
      P' \sqsubseteq P

    Why do we care?  Is this a theorem?  A requirement?  Do we want
    implication (in the logic) and not a lattic op?  How does
    implication translate to lattice ops, anyway?  (disjunction ~ join)



DFOPT: if any extra room, restore Davidson reference!
       then: 'to reduce register pressure'
       then: 2nd dataflow paper
