Changing your transducer during the runtime of your core.async channel (could use some optimizaiton caching the instantiation)

dynamic-xf-async.clj

(def xf-a
     (atom (map inc)))

(defn xf [& args-out]
  (fn [& args]
    ;; TODO: Cache the inner (apply...)
    (apply (apply @xf-a args-out) args)))

(def ch (chan 1 xf))

(>!! ch 1)
(pr (<!! ch));; => 2
(reset! xf-a (map dec))
(>!! ch 1)
(pr (<!! ch));; => 0

Note: Doesn't properly drain a transducer when the channel closes. Doesnt' work really with stateful transducers. Use with simple map/filter and not much more.