> What are your thoughts on implementing reversible patterns?

I'm thinking something like this

    (defprotocol Pattern
      (extract [this input] "=> bindings")
      (replace [this input bindings] "=> new-input"))

This should obey some rule like:

    (submap? (extract p (replace p input bindings)) bindings)

That would make the data-grammar/generic-traversal experiments much nicer and also cleanly delineates patterns from views. It may also help when I try to do things like generative testing.

Not quite sure how that would work for (or ...) and (not ...) patterns, so it still needs more hammock time.