The simplifier produces regexps with some simple invariants:

- DSM's are only top-level, never appearing in the body of a DSM,
  repeat, sequence, choice, or submatch.

- A repeat's body is not a repeat, trivial match, or empty match.

- A choice's body contains more than one element; no element is
  - a choice,
  - a DSM, or
  - an empty-match.
  
- A choice contains 0 or 1 char-set, bos, and eos elements.

- A sequence's body contains more than one element; no element is
  - a sequence,
  - a DSM, 
  - a trivial match, or
  - an empty-match
  
- There are no empty matches in the regexp unless the entire regexp
  is either an empty match, or a dsm whose body is an empty match.
  (This is good, because there is no way to write an empty match
  in Posix notation in a char-set independent way -- you have to
  use the six-char "[^\000-\177]" for ASCII.)

To see these invariants:

- We can always bubble up empty matches:
  - If a sequence has one, the whole sequence is reduced to an empty match.
  - They can be deleted from a choice; if the choice reduces to 0 elements,
    the choice can be reduced to an empty match.
  - A repeat of an empty match is either an empty match or a trivial match,
    depending upon whether FROM is >0 or 0, respectively.
  - DSM of an empty match: the DSM itself can be bubbled upwards (see below).

- We can always bubble up DSM regexps:
  - If an elt of a choice or sequence is a DSM, it can be "absorbed"
    into the element's relocation offset.
  - Repeat commutes with DSM.
  - A DSM body can be "absorbed" into a submatch record by increasing the
    submatch's DSM0 count.
  - Nested DSM's can be collapsed together.