restructure

Rewrite nested Clojure data with a declared shape.

over compiles a selector and rewrite body into code in the style of update/mapv/update-vals, but keeps the shape in one place. It visits only what you select.

API

(over selector body)         ; => rewritten value
(compile-over selector body) ; => fn of one argument
(over-plan selector body)    ; => compiler plan data

Examples

1) Update nested numbers without a full walk

(def data
  {:a [{:aa 1 :bb 2}
       {:cc 3}]
   :b [{:dd 4}]})

(over [{_ [{_ n}]} data]
  {n (cond-> n (even? n) inc)})

;; => {:a [{:aa 1 :bb 3} {:cc 3}]
;;     :b [{:dd 5}]}

Map + vector traversal with one binding.

Plain Clojure:

(update-vals data
             #(mapv (fn [m]
                      (update-vals m (fn [n] (cond-> n (even? n) inc))))
                    %))

2) Filter map entries + normalize a field

(require '[clojure.string :as str])

(def users
  {:alice {:active true  :email "ALICE@EXAMPLE.COM"}
   :bob   {:active false :email "bob@example.com"}
   :cara  {:active true  :email "CARA@EXAMPLE.COM"}})

(over [{_ {:keys [active email] :as u}} users]
  {u?    active
   email (str/lower-case email)})

;; => {:alice {:active true :email "alice@example.com"}
;;     :cara  {:active true :email "cara@example.com"}}

Map-entry traversal with a guard (u?) to drop entries.

Plain Clojure:

(->> users
     (reduce-kv (fn [m id u]
                  (if (:active u)
                    (assoc m id (update u :email str/lower-case))
                    m))
                {}))

3) Traverse a nested vector and drop bad items

(def order
  {:id 42
   :lines [{:sku "A" :qty 2}
           {:sku ""  :qty 1}
           {:sku "B" :qty 0}]})

(over [{:keys [lines]} order
       [{:keys [sku qty] :as line}] lines]
  {line? (seq sku)
   qty   (or qty 0)})

;; => {:id 42
;;     :lines [{:sku "A" :qty 2}
;;             {:sku "B" :qty 0}]}

Sequential traversal; each element can be rewritten or dropped.

Plain Clojure:

(update order :lines
        (fn [lines]
          (->> lines
               (filter (comp seq :sku))
               (mapv #(update % :qty (fnil identity 0))))))

Selector semantics (query, not destructuring)

A selector is a vector of alternating pattern and source:

(over [pattern1 source1
       pattern2 source2
       ...]
  body-map)

• The first source is the input expression.
• Later sources must be previously bound symbols.
• The selector looks like destructuring, but it is a query: it binds and traverses; it does not introduce locals the way Clojure destructuring does.

Supported patterns

• Map-entry traversal: {kpat vpat}
  • Requires a map value; iterates entries.
  • kpat binds the key (no traversal). vpat may traverse.
• Sequential traversal: [pat]
  • Requires a sequential or set value; iterates elements.
• Plain symbol: sym
  • Binds the current value.
• Map destructuring form: {:keys [...], :strs [...], :as sym, :or {...}}
  • No traversal; binds keys and/or :as.
  • Nested destructuring inside :keys/:strs is rejected.
• Records are treated as maps and may come back as plain maps after rewrite.

Unsupported or ambiguous patterns are rejected with ex-info including :phase, :path, and :pattern.

Rewrite/body semantics

The body is a map from target symbols to expressions.

• A key x rewrites the binding x.
• A key x? is a guard. When false, the corresponding structural position is elided.

Elision rules:

• Map-entry traversal: false guard removes the entry.
• Sequential traversal: false guard removes the element.
• Map destructure key: false guard removes that key.
• :as binding: false guard removes the whole element at that level.

Traversal is post-order: children are processed before parents. Guards see post-child values.

Tradeoffs

Plain Clojure is already a win when:

• A couple of update calls are clearer than any DSL.
• update-vals + mapv is easier to scan for simple shapes.

clojure.walk/postwalk is better when:

• You need full-tree traversal.
• You don’t know the shape ahead of time.

over fits when:

• You have a known shape and want to touch a few nested spots.
• You want the traversal shape and the rewrite adjacent.
• You want compiled code that avoids full walks and keeps structural sharing.

Non-goals

• Not a general tree rewriting engine.
• Not a replacement for all data transformation.
• Not arbitrary destructuring: selectors are queries, not local binding forms.
• No :let or computed traversal paths.

Notes

• over-plan returns the compiler plan as data. Useful for debugging.

Running tests

clojure -M:test

Lint and format

clj -M:lint --lint src test
clj -M:fmt

Check (format + test + lint)

clj -M:check

License

MIT