Specter

Clojure has fantastic facilities for doing immutable programming, with a rich library of persistent data structures and efficient mechanisms for manipulating and traversing them. However, Clojure's story is incomplete. Once you nest data structures – which is extremely common – Clojure becomes cumbersome and complex. Clojure even lacks a facility for a basic task like transforming every value in a generic sequence without changing the type or order of that sequence.

Specter, available for both Clojure and ClojureScript, provides a high performance abstraction called navigators which complete the story around immutable programming and make it easy to transform and query nested data structures. It allows you to concisely specify what you want to change within a data structure, and get a new data structure back with only your changes applied – everything else is reconstructed and the types of data structures throughout don't unexpectedly change.

Consider these examples:

Example 1: Increment every even number nested within map of vector of maps

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

;; Manual Clojure
(defn map-vals [m afn]
  (->> m (map (fn [[k v]] [k (afn v)])) (into {})))

(map-vals data
  (fn [v]
    (mapv
      (fn [m]
        (map-vals
          m
          (fn [v] (if (even? v) (inc v) v))))
      v)))

;; Specter
(transform [MAP-VALS ALL MAP-VALS even?] inc data)

Example 2: Append a sequence of elements to a nested vector

(def data {:a [1 2 3]})

;; Manual Clojure
(update data :a (fn [v] (into (if v v []) [4 5])))

;; Specter
(setval [:a END] [4 5] data)

Example 3: Increment the last odd number in a sequence

(def data [1 2 3 4 5 6 7 8])

;; Manual Clojure
(let [idx (reduce-kv (fn [res i v] (if (odd? v) i res)) nil data)]
  (if idx (update data idx inc) data))

;; Specter
(transform [(filterer odd?) LAST] inc data)

Example 4: Map a function over a sequence without changing the type or order of the sequence

;; Manual Clojure
(map inc data) ;; doesn't work, becomes a lazy sequence
(into (empty data) (map inc data)) ;; doesn't work, reverses the order of lists

;; Specter
(transform ALL inc data) ;; works for all Clojure datatypes with near-optimal efficiency

Specter has performance rivaling hand-optimized code (see this benchmark). Clojure's only comparable built-in operations are get-in and update-in, and the Specter equivalents are 30% and 85% faster respectively (while being just as concise). Under the hood, Specter uses advanced dynamic techniques to strip away the overhead of composition. Additionally, the built-in navigators use the most efficient means possible of accessing data structures. For example, ALL uses mapv on vectors, the IMapIterable interface on small maps, and reduce-kv in conjunction with transients on larger maps.

The most important aspect of Specter is its composability. Specter navigators can be composed with any other navigators, so the supported use cases grow combinatorially. And because Specter is completely extensible, it can be used to navigate any data structure or object you have.

Latest Version

The latest release version of Specter is hosted on Clojars:

Learn Specter

• Introductory blog post: Functional-navigational programming in Clojure(Script) with Specter
• Presentation about Specter: Specter: Powerful and Simple Data Structure Manipulation
• List of navigators with examples: This wiki page provides a more comprehensive overview than the API docs about the behavior of specific navigators and includes many examples.
• Core operations and defining new navigators: This wiki page provides a more comprehensive overview than the API docs of the core select/transform/etc. operations and the operations for defining new navigators.
• API docs
• Performance guide: This post provides an overview of how Specter achieves its performance.

Specter's API is contained in these files:

• macros.clj: This contains the core select/transform/etc. operations as well as macros for defining new navigators.
• specter.cljc: This contains the built-in navigators and functional versions of select/transform/etc.
• transients.cljc: This contains navigators for transient collections.
• zipper.cljc: This integrates zipper-based navigation into Specter.

Questions?

You can ask questions about Specter by opening an issue on Github.

You can also find help in the #specter channel on Clojurians.

Examples

Increment all the values in maps of maps:

user> (use 'com.rpl.specter)
user> (transform [MAP-VALS MAP-VALS]
              inc
              {:a {:aa 1} :b {:ba -1 :bb 2}})
{:a {:aa 2}, :b {:ba 0, :bb 3}}

Increment all the even values for :a keys in a sequence of maps:

user> (transform [ALL :a even?]
              inc
              [{:a 1} {:a 2} {:a 4} {:a 3}])
[{:a 1} {:a 3} {:a 5} {:a 3}]

Retrieve every number divisible by 3 out of a sequence of sequences:

user> (select [ALL ALL #(= 0 (mod % 3))]
              [[1 2 3 4] [] [5 3 2 18] [2 4 6] [12]])
[3 3 18 6 12]

Increment the last odd number in a sequence:

user> (transform [(filterer odd?) LAST]
              inc
              [2 1 3 6 9 4 8])
[2 1 3 6 10 4 8]

Increment all the odd numbers between indices 1 (inclusive) and 4 (exclusive):

user> (transform [(srange 1 4) ALL odd?] inc [0 1 2 3 4 5 6 7])
[0 2 2 4 4 5 6 7]

Replace the subsequence from indices 2 to 4 with [:a :b :c :d :e]:

user> (setval (srange 2 4) [:a :b :c :d :e] [0 1 2 3 4 5 6 7 8 9])
[0 1 :a :b :c :d :e 4 5 6 7 8 9]

Concatenate the sequence [:a :b] to every nested sequence of a sequence:

user> (setval [ALL END] [:a :b] [[1] '(1 2) [:c]])
[[1 :a :b] (1 2 :a :b) [:c :a :b]]

Get all the numbers out of a data structure, no matter how they're nested:

user> (select (walker number?)
              {2 [1 2 [6 7]] :a 4 :c {:a 1 :d [2 nil]}})
[2 1 2 1 2 6 7 4]

Navigate via non-keyword keys:

user> (select [(keypath "a") (keypath "b")]
              {"a" {"b" 10}})
[10]

Reverse the positions of all even numbers between indices 4 and 11:

user> (transform [(srange 4 11) (filterer even?)]
              reverse
              [0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15])
[0 1 2 3 10 5 8 7 6 9 4 11 12 13 14 15]

Append [:c :d] to every subsequence that has at least two even numbers:

user> (setval [ALL
               (selected? (filterer even?) (view count) #(>= % 2))
               END]
              [:c :d]
              [[1 2 3 4 5 6] [7 0 -1] [8 8] []])
[[1 2 3 4 5 6 :c :d] [7 0 -1] [8 8 :c :d] []]

When doing more involved transformations, you often find you lose context when navigating deep within a data structure and need information "up" the data structure to perform the transformation. Specter solves this problem by allowing you to collect values during navigation to use in the transform function. Here's an example which transforms a sequence of maps by adding the value of the :b key to the value of the :a key, but only if the :a key is even:

user> (transform [ALL (collect-one :b) :a even?]
              +
              [{:a 1 :b 3} {:a 2 :b -10} {:a 4 :b 10} {:a 3}])
[{:b 3, :a 1} {:b -10, :a -8} {:b 10, :a 14} {:a 3}]

The transform function receives as arguments all the collected values followed by the navigated to value. So in this case + receives the value of the :b key followed by the value of the :a key, and the transform is performed to :a's value.

The four built-in ways for collecting values are VAL, collect, collect-one, and putval. VAL just adds whatever element it's currently on to the value list, while collect and collect-one take in a selector to navigate to the desired value. collect works just like select by finding a sequence of values, while collect-one expects to only navigate to a single value. Finally, putval adds an external value into the collected values list.

Increment the value for :a key by 10:

user> (transform [:a (putval 10)]
              +
              {:a 1 :b 3})
{:b 3 :a 11}

For every map in a sequence, increment every number in :c's value if :a is even or increment :d if :a is odd:

user> (transform [ALL (if-path [:a even?] [:c ALL] :d)]
              inc
              [{:a 2 :c [1 2] :d 4} {:a 4 :c [0 10 -1]} {:a -1 :c [1 1 1] :d 1}])
[{:c [2 3], :d 4, :a 2} {:c [1 11 0], :a 4} {:c [1 1 1], :d 2, :a -1}]

"Protocol paths" can be used to navigate on polymorphic data. For example, if you have two ways of storing "account" information:

(defrecord Account [funds])
(defrecord User [account])
(defrecord Family [accounts-list])

You can make an "AccountPath" that dynamically chooses its path based on the type of element it is currently navigated to:

(defprotocolpath AccountPath [])
(extend-protocolpath AccountPath
  User :account
  Family [:accounts-list ALL])

Then, here is how to select all the funds out of a list of User and Family:

user> (select [ALL AccountPath :funds]
        [(->User (->Account 50))
         (->User (->Account 51))
         (->Family [(->Account 1)
                    (->Account 2)])
         ])
[50 51 1 2]

The next examples demonstrate recursive navigation. Here's one way to double all the even numbers in a tree:

(defprotocolpath TreeWalker [])

(extend-protocolpath TreeWalker
  Object nil
  clojure.lang.PersistentVector [ALL TreeWalker])

(transform [TreeWalker number? even?] #(* 2 %) [:a 1 [2 [[[3]]] :e] [4 5 [6 7]]])
;; => [:a 1 [4 [[[3]]] :e] [8 5 [12 7]]]

Here's how to reverse the positions of all even numbers in a tree (with order based on a depth first search). This example uses conditional navigation instead of protocol paths to do the walk:

(def TreeValues
  (recursive-path [] p
    (if-path vector?
      [ALL p]
      STAY
      )))


(transform (subselect TreeValues even?)
  reverse
  [1 2 [3 [[4]] 5] [6 [7 8] 9 [[10]]]]
  )
;; => [1 10 [3 [[8]] 5] [6 [7 4] 9 [[2]]]]

Future work

• Integrate Specter with other kinds of data structures, such as graphs. Desired navigations include: reduction in topological order, navigate to outgoing/incoming nodes, to a subgraph (with metadata indicating how to attach external edges on transformation), to node attributes, to node values, to specific nodes.
• Make it possible to parallelize selects/transforms
• Any connection to transducers?

License

Copyright 2015-2016 Red Planet Labs, Inc. Specter is licensed under Apache License v2.0.