dosync Archive Pages Categories Tags

A Tool For Thought

03 June 2016

As ClojureScript nears its fifth birthday I find myself reflecting on its suitability as a "tool for thought". Certainly reading the Structure and Interpretation of Computer Programs nearly thirteen years ago painted an ideal image of Lisp as a truly interactive and tangible approach to computing. That experience eventually lead me to Clojure and later ClojureScript. Now that I use both for the day job, has that wonderful image of computing that Sussman & Abelson mapped out faded over the ensuing years?

Looking around we see many compelling new languages with a client story in the functional space especially of the typed variety. Some of these are truly new endeavors and others are simply old faces with JavaScript backends. Elm is notably exciting with its principled stance on simplicity and approachability while delivering on the benefits of static typing. Whatever your opinions of Scala, many programmers find it a comfortable train ride across the OOP / FP border and Scala.js means you can bring your experiences to the client. PureScript has a strong following in the Haskell community and of course if you really just want Haskell or OCaml on the client you have a few options as well.

But frankly I don't have any interest in any of these initiatives beyond that of the language geek. For me debugging and profiling, advanced optimization, IDE support, rich tooling across all desirable JavaScript targets are equally important considerations to weigh against static guarantees when choosing the programming language for clients.

Still I'm not a staunch defender of the dynamic nor the static stance and I think people with strong opinions either way are being more than a bit dishonest to themselves about the realities of industrial software development. Personally the thing I find most compelling about statically typed languages is that well designed ones give you a tool for thinking. As Simon Peyton Jones sez in Coders At Work, it's not about correctness it's about crispness.

As a Clojure(Script) programmer I would love to tap into that deep well of crispness somehow. Unfortunately idiomatic Clojure code presents many challenges to typing, just ask my buddy Ambrose Bonnaire-Sergeant.

As Clojurists we live in a sea of lists, symbols, maps, keywords, vectors, and sets and some of these structures represent data and others represent code. I think it's no small part of the attraction of Lisps that the use/mention distinction becomes at least slightly blurred. The problem as I think many of us can attest is when things get so blurry you lose your way. This isn't to say that you can't build big sophisticated programs in this way, for example, ClojureScript itself:

And while I'm sure Kent Dybvig is nodding as well, we shouldn't stop wondering if we can't the move the goal post a little bit closer.

I think clojure.spec moves that goal post a lot closer.


In my humble opinion clojure.spec is the most VPRI-worthy feature Rich Hickey has shipped since delivering fast persistent data structures. Yes, yes you now have Schema like validations, but this is but the tip of an iceberg. clojure.spec takes Matt Might et. al. Parsing with Derivatives and really, really runs with it. By casting validation as fundamentally a parsing problem (computer science!), we get a wonderfully expressive language for crisply describing our Clojure programs without changing how we joyfully write them.

So instead of thinking about typical application domain examples, lets instead consider how we might spec our old friend let. Intuitively we know let must be made of some obvious basic parts:

(require '[cljs.spec :as s])

(s/def ::let
     :name     '#{let}
     :bindings ::bindings
     :forms    (s/* (constantly true))))

Ah wishful thinking! How we adore thee. We know the first part of a let expression has to be the symbol let (Duh!), we're ignoring ::bindings for now, and we know that we'll have zero or more forms after the bindings. After some hammocking we give ::bindings a go —

(s/def ::bindings


Let's try it:

(s/conform ::let '(let [x 1] (+ x y)))
;; {:name let, :bindings [x 1], :forms [(+ x y)]}


... hmm, actually not so fast.

(s/conform ::let '(let [x 1 y] (+ x y)))
;; {:name let, :bindings [x 1 y], :forms [(+ x y)]}

Ugh. That's not right, :bindings must be an even number of forms:

(s/def ::bindings
  (s/and vector? #(-> % count even?)))

Let's try again:

(s/conform ::let '(let [x 1 y] (+ x y)))
;; :cljs.spec/invalid

That's more like it but good tools for thinking should be a little bit more forthcoming:

(s/explain ::let '(let [x 1 y] (+ x y)))
;; In: [1] val: [x 1 y] fails spec: 
;;   :cljs.user/bindings at: [:bindings] predicate:
;;   (-> % count even?)


(s/conform ::let '(let [1 y] (+ x y)))
;; {:name let, :bindings [1 y], :forms [(+ x y)]}

Ah we're not crisp enough yet. Let's make a ::binding spec to control what can appear in the vector:

(s/def ::binding
    :name  symbol?
    :value (constantly true)))

That looks right, now let's fix ::bindings to use it:

(s/def ::bindings
  (s/and vector?
         #(-> % count even?)
         (s/* ::binding)))

Now let's try it:

(s/conform ::let '(let [y 1] (+ x y)))
;; {:name let, :bindings [{:name y, :value 1}], :forms [(+ x y)]}

Yes I am trying to trick you into reimplementing ClojureScript from scratch.

Obviously we could go further but the above should suffice to show how much clojure.spec can help when thinking about the shape of the problem.


While the Clojurist loves to wax poetic about hammock driven development, Clojure(Script) the language did not offer much in the way of helping us really work through a design while sitting at a keyboard. Without belittling the power of pencil and paper, the whole point of a computer is to give us interactive ways of exploring our thoughts in ways that pencil and paper can't! So I think clojure.spec gives us a pretty damn nice hammock away from the hammock and I look forward to hearing more about it from the larger community.

Happy hacking!