[Haskell-cafe] Re: bizarre memory usage with data.binary

[Anatoly Yakovenko]

> ...and indeed it can't be done, except by the naive brute-force method
> of comparing every subtree, possibly optimized by cryptographically
> hashing a representation of every subtree, since sharing isn't an
> observable property.

i was thinking that instead of having a reference to a node, each node
just holds an index from an array of nodes.  Traversal would take an
extra step, but it should fix the problem with encode/decode.