RefCounts are dead. Long live lazy copy-on-write.

[Eric Dean Tribble <tribble>]

Your suggestion strikes me as an interesting variant of a
      copy-on-write implementation.  It covers making ImmuTables from either
      MuTables or ImmuTables.  I don't think it handles making a MuTable
      from an ImmuTable (or MuTable).  

   Why not?

Ah. I understand.  Supporting this case requires trailing backward
deltas for all changes to a MuTable.  I imgined a scheme that only
makes deltas when one collection gets made from another.  In my
scheme, B could be made from A, but if B gets changed, the backwards
delta stored in A is no longer sufficient.

This wasn't too coherent, but I hope you get the gist.

dean