Fix #77: tagged literal issue

[borkdude]

No description provided.

[yuhan0]

Other functions are still broken which rely on a seq of MapEntry's :

(keys #ordered/map ([:a 1])) 
;; => Error printing return value (ClassCastException) at clojure.lang.APersistentMap$KeySeq/first (APersistentMap.java:171).
;;    class clojure.lang.PersistentVector cannot be cast to class java.util.Map$Entry (clojure.lang.PersistentVector is in unnamed module of loader 'app'; java.util.Map$Entry is in module java.base of loader 'bootstrap')

I guess we could further patch the seq implementation

  (seq [this]
    (->> order 
         (keep identity)
         (map #(MapEntry. (nth % 0) (nth % 1)))))

but that seems inefficient and hides the fact that the data structure invariants are being violated, which may break other parts of the code.

[yuhan0]

Namely given m : OrderedMap,

• (.-order m) should be a sequence of MapEntry
• (.-backing-map m) should be a map from Object to MapEntry

These are violated in the case of an OrderedMap constructed from a data literal, which have PersistentVector in place of MapEntry in both those fields

(def yes (ordered-map [[:a 1]]))
(let [m #ordered/map ([:a 1])]
  (def no m))

(-> yes .-order first type) ;; => clojure.lang.MapEntry
(-> no .-order first type) ;; => clojure.lang.PersistentVector

(-> yes .-backing-map (get :a) type) ;; => clojure.lang.MapEntry
(-> no .-backing-map (get :a) type) ;; => clojure.lang.PersistentVector

[borkdude]

I pushed additional "fixes"

[yuhan0]

Yeah, I guess that would work.. feels slightly icky though, and the (no longer accurate) ^MapEntry tags should probably be removed.

[borkdude]

There's actually no reason that MapEntries should have to be used internally, as long as they are produced when called. But not sure if it's worth all this trouble.