[From mail from Alan Bishop:]

Steve Heller has been thinking about changes to Ref tracing that affects
this report, I think.

From a Tue Jul 16 10:25:34 -0700 1996
To: sbb, terryc
CC: dlong, fresko
Subject: Warning about refs


There remains a problem with refs that probably won't get fixed for
1.0.k. It might affect what cleanup you do at state transitions.

If there is a subgraph of objects that is only reachable via a Ref,
and there is a reference from one of the subgraph's objects back to
the "subroot" (i.e. the object pointed to by the Ref), then the ref
will never get cleared.

I.e.


  +------+
  | Ref |
  +------+
    |
    V
  +------+
  | Obj |
  +------+
    |
    V
  +------+
  | Obj |
  +------+

is ok, and the ref will get cleared.

However,

  +------+
  | Ref |
  +------+
    |
    V
  +------+
  | Obj | <-----+
  +------+ |
    | |
    V |
  +------+ |
  | Obj | ------+
  +------+

will not, as the cycle involving the "subroot" will convince the
garbage collector that there's a hard ref and that clearing the soft
one is a bad idea.

If the cycle doesn't involve the subroot, then you're ok, i.e.

  +------+
  | Ref |
  +------+
    |
    V
  +------+
  | Obj |
  +------+
    |
    V
  +------+
  | Obj | <-----+
  +------+ |
    | |
    V |
  +------+ |
  | Obj | ------+
  +------+