July 2, 2014

5 mistakes I made writing my first Clojure application

Learn from my shame.

It’s been about two years since I undertook my first sizable application in Clojure, and really my first experience architecting a system in a functional language. It was a local dating application, and it’s still up and running, despite the sins below. But, reading the code now, a few dire instances of truely bad code emerge. So, I crawled through for some examples, which I now impart to you:

Deeply Nested ifs/lets

Any time you put a let in a let or an if in an if, that’s a strong code smell. It means you’re still writing code in a way that requires you to short-circuit your functions with returns.

You probably need to change your code to do fewer things. Here’s an example, taken straight from my old dumb self:

(defn register [{{:keys [username
                         password
                         email
                         email_repeat
                         gender
                         dob_year
                         dob_month
                         dob_day]} :params :as request}]
(cond

   (or (empty? username) (empty? password) (empty? email) (empty? email_repeat) (empty? gender))
       (fail-registration "Please fill in all fields")

   (not (= email email_repeat))
       (fail-registration "Whoops! You entered the wrong email address.")

   :else
   (let [dob [(Integer/parseInt dob_year) (Integer/parseInt dob_month) (Integer/parseInt dob_day)]
         [usr err] (user/create-user! {
                                      :username username
                                      :password password
                                      :email email})]
     (if err
       (fail-registration err)
       (do
         (auth/login! username password)

         ; Create an initial profile
         (let [prof (merge (user/fetch-user-profile usr) {:name (:username usr)})
               dob [(utils/parse-int dob_year)
                    (utils/parse-int dob_month)
                    (utils/parse-int dob_day)]
               remote-addr (:remote-addr (req/ring-request))
               [prof err] (profile/update-profile! prof {:dob dob :name username :gender gender :last_login_ip remote-addr})]
           (if err
            (session/flash-put! err)
            (if (= (:gender prof) "M")
              (mail/signup-confirmation-m usr)
              (mail/signup-confirmation-f usr)))

         ; Send to profile edit screen
         (resp/redirect "/edit/profile")))))

So what is to be done about that? First, consider what the function is responsible for. It:

  • Checks form validation
  • Constructs a user object
  • Logs the user in
  • Constructs a profile object
  • Sends a confirmation email

My favorite way of refactoring is to first write code as if the functions I need exist already; top-down, you might call it.

(defn register [request]
  (attempt-all
    [user-data (validate-and-clean-user-data (:params request))
     user (create-user! user-data)
     _ (login! (:username user) (:password user))
     _ (create-user-profile! user (:params request)]

    (do
      (send-confirmation-email user)
      (resp/redirect "/edit/profile"))
    #(fail-registration (get-message %))))

Here, attempt-all is a monadic error-checking construct that will short-circuit to the fail-registration condition if any of the bindings fail. If everything makes it to the end, you can go ahead and send that confirmation mail and redirect to the profile edit page.

Now, all that remains is to write validate-and-clean-user-data, create-user!, login!, and create-user-profile!, each a monadic function returning an error state if failed.

No dedicated validation layer

Instead of separating things like validate-user and put-user!, I embedded validation in functions that interacted with the database. In theory, this means I couldn’t accidentally forget to validate something; in practice it made database code more bloated and complicated than it needed to be, and often had to be worked around, creating unnecessary duplication.

Here’s a sample offender, but there are many more.

(defn create-user! [data]
  ; Overwrite the create user function
  (cond
    (not (nil? (fetch-one :users :where {:username {:$regex (str "^" (:username data) "$") :$options "-i"}})))
        [nil "That username is already in use."]
    (not (nil? (fetch-one :users :where {:email {:$regex (:email data) :$options "-i"}})))
        [nil "That email address is already in use."]
    :else
        (let [
              salt (auth/random-hash 64)
              data (merge data {
                          :subs_level nil
                          :auth-token nil
                          :auth-token-expires nil
                          :passwordhash (auth/pbkdf2 (:password data) salt)
                          :salt salt
                          :created (utils/now)})]
          (models/create-record! :users user-schema data))))

At very worst, this function should look like this:

(defn create-user! [userdata]
  (attempt-all
    [user (validate-and-clean-user-data userdata)]
    (models/create-record! :users user-schema data)))

This is a very good reason to make validate-and-clean-user-data idempotent; it would be nice to be able to call it on the same data as many times as we want without issue.

Rolled my own schema checking

I created a schema checking tool to facilitate safer database access, but I really could have just used Prismatic/Schema instead and been happier for it.

(def user-schema
  [
   [:username :str]
   [:passwordhash :str]
   [:subs_level :str?]
   [:experiment_tag :str?]
   [:salt :str]
   [:email :str]
   [:created :datetime?]
   [:auth-token :str?]
   [:auth-token-expires :datetime?]
   [:allow_system_mail :bool?]
   [:allow_mktng_mail :bool?]
   [:allow_retention_mail :bool?]
   [:customer_tok :str?]
   ])

(models/model "user" :users user-schema)

Magic function generation

In the previous code sample, the model macro generates some functions: fetch-user, create-user!, update-user!, and delete-user!. It seemed clever at the time, but it’s pretty obvious upon re-reading just how bad an idea this was. In the best case, the generated functions had to be reloaded anyhow to define more specific logic; in the worst, the file suddenly grew 4 undocumented functions that go on to be used in the API everywhere.

This stings especially badly because of just how thin a wrapper around some existing functions the generated ones were, and how much effort I put in to make it happen. Cripes.

(defmacro model [name coll schema]
  `(let [fetch-fn# (symbol (str "fetch-" ~name))
         create-fn# (symbol (str "create-" ~name "!"))
         update-fn# (symbol (str "update-" ~name "!"))
         delete-fn# (symbol (str "delete-" ~name "!"))
         ]
     ; Fetch
     (intern *ns* fetch-fn#
             (fn [id# & args#] (fetch-record ~coll id#)))

     ; Create
     (intern *ns* create-fn#
             (fn [data#] (create-record! ~coll ~schema data#)))

     ; Update
     (intern *ns* update-fn#
             (fn [obj# data#] (update-record! ~coll ~schema obj# data#)))

     ; Delete
     (intern *ns* delete-fn#
             (fn [obj#] (destroy-record! ~coll obj#)))
     ))

Remember kids: explicit is better than implicit, no matter how many measly characters you save.

Forgetting about reduce.

In most languages, this is how you sum up a list of numbers:

def sum_list(nums):
    total = 0
    for num in nums:
        total += num
    return total

Since this requires mutating total, this is not a good functional way of going about this problem, and Clojure won’t let you do this without a lot of monkeying around. loop and recur might seem to be the idiomatic way around this:

(defn sum-list [nums]
  (loop [total 0 remaining nums]
    (if (nil? remaining)
      total
      (recur (+ total (first nums)) (rest nums)))))

However, to do so would be to forget about the power of reduce:

(defn sum-list [nums] (reduce + 0 nums))

or even:

(def sum-list (partial reduce + 0))

Another good example is the classic mapreduce word count:


;; Loop + recur
(defn count-words [words]
  (loop [counts {} words words]
    (if (nil? words)
        counts
        (recur (update-in counts [(first words)] (fnil inc 0)) (rest words)))))

;; Reduce
(defn count-words [words]
  (reduce
    #(update-in %1 [%2] (fnil inc 0))
    {} words))

;; Bonus actually correct answer
(def count-words frequencies)

There are many more crimes against simplicity in this codebase, but there are some of the most egregious (and fun to read!). I hope you enjoyed it.