You love Firebase? I love Firebase.

WARNING: Fun with transducers ahead. Note that the use of transducers could have been totally avoided, but what’s the point if we’re not having a little fun, right? right?

Firebase has made client side web-apps development trivial for me. I can forget about managing my data, pushing it, sanitizing it (may be a little) and retrieving it. These things may seem simple but having them taken care for you puts you in a different State of Mind which takes that data management burden out of your thought process.

I’ve been working on a Firebase Clojure library called pani (hindi word for water). Things are flowing (no pun intended) nicely, although Clojure support is lagging behind ClojureScript (since I tend to use the latter more).

In this post I am going to touch on one aspect of pani. What I intend to do is write a function named listen which listens for events on Firebase refs and provides a nice way of dealing with them. We’d be using core.async to deliver these events.

I originally intended to deconstruct these delivered events using core.match, but it seems like core.async and core.match don’t play nicely together yet.

First things first

The listen function will bind to Firebase events and wait for notifications. The function will return a core.async chan. Received Events will be posted to this channel as Clojure vectors. I will refrain from using certain functions from pani itself to make things clearer (although it could have resulted in more concise code).

I am not going to show you the requires and stuff since this function will be a part of the pani library and requires have already been taken care of.

Basic Structure

I think the function should look something like:

(defn listen
  "Given a Firebase root and a key (or a seq of keys) return a
   channel which will deliver events"
  [root korks]
  (let [c (chan)]

The function takes a Firebase root (can be created with pani/root function and either a single key or a seq of keys). We just declare a chan for now and return it.

Let’s build on it.

Listening for Firebase Events

Although pani already has functions to listen for Firebase events, let’s just re-write them here using some transducers for extra giggles.

Mostly we’re just interested in three Firebase events here: child_added, child_removed and child_changed (let’s collectively decide to not worry about child_moved). For most of my use cases I’ve found that I never need the previous snapshot or node name (let me know if that’s not the case though, or I’ll sooner or later hit one).

First, let’s write a little function which takes a Firebase ref and returns to us a chan which will have the received value posted to it (pani/bind does something similar) after its passed through a provided transducer.

(defn- fb->chan
  "Given a firebase ref, an event and a transducer, binds and posts to returned channel"
  [fbref event td]
  (let [c (chan 1 td)]
    (.on fbref (clojure.core/name event)
         #(put! c [event %]))

This function takes a Firebase ref, an event in the form of a keyword e.g. :child_added and a transducer. The function pushes values as a vector into the channel. The values look like [event-type firebase-snapshot], e.g. [:child_added #js {:firebase "stuff"}] (that #js is called a reader, basically saying what follows needs to be interpreted as a javascript object, makes more sense when you’re reading in code, hence the name).

The transducer then accepts this vector and turns it into a flattened out vector, something like [:child_added "key name" "value"]. Like I said before, we can totally not use a transducer here, but we’re just having a little bit of fun.

Testing this is not a trivial thing to do, so you’d have to take my word for it that it’s working fine. If you really want to test it, you can create a new ClojureScript application, refer to pani and play around with it. Hint: a good starting point for ClojureScript apps is David Nolen’s mies.

listen Machinery

Once we have the fb->chan method, we can starting defining our listen function. It looks something like this for me:

(defn listen
  "Listens for events on the given firebase ref"
  [root korks]
  (let [root    (walk-root root korks)
        events  [:child_added :child_removed :child_changed]
        td      (map (fn [[evt snap]]
                       [evt (.name snap) (.val snap)]))
        chans   (map (fn [event]
                       (fb->chan root event td)) events)]
    (merge chans)))

The function begins by taking care of any child refs we may need to get to using walk-root. This function works by walking up the refs, e.g.

"/" -> (walk-root [:hello :world]) -> "/hello/world"

Next we list the events we are interested in, followed by our transducer which pulls the Firebase snapshot’s name and value out of the snapshot.

[:child_added #js {:name "1" :value "sup"}] ->
    [:child_added "1" "sup"]

A simple transformation.

Finally we map over our events of interest and generate chans for each one of them using the fb->chan method.

We finally return a merged chan.

Using our new function

To begin with, let’s just print what we receive.

(def r (p/root ""))

(let [c (p/listen r [:items])]
  (go-loop [msg (<! c)]
           (println msg)
           (recur (<! c))))

We define r as the root of our Firebase app. We call listen on it and pass it [:items], since here we’re interested in /items ref. Any items added, removed or changed under this ref needs to be told us about. When I run this and simulate adding and removing values, I get output like this:

[:child_added "4" "clojure"]
[:child_added "5" "rocks!"]
[:child_changed "5" "rocks hard!"]
[:child_removed "5" "rocks hard!"]

Ok, so far so good.

Let’s say I want to keep track of values under my Firebase ref. I would start with an empty map and as I receive notifications about items being added, removed or modified, I would update my map accordingly. A nice and pretty way of doing this would be to use clojure.match, but we’re going to stay away from it for now. Instead we’ll play a little bit with multi-methods.

A whirlwind tour of Multi-methods

Multi-methods you say?

A mutli-method in its gist can be thought of as a dispatch table. Based on a certain value one of the defined methods of a multi-method will be called. This certain value is again computed by calling a function called a dispatch function. All arguments passed to the multi-method are passed to this dispatch function. The return value of the dispatch function then determines which of the functions from the dispatch table needs to be called.

Let’s work on a simple example:

(defmulti greet :message)

(defmethod greet :bye [_]

(defmethod greet :hello [_]

Here we declare a multi-method named greet, the second parameter to defmulti is a function. Recall that a keyword can be called as a function on a map:

(:message {:message "what"})
=> "what"

So its totally fine to use a keyword as a dispatch function for a multi-method as long as you’re expecting a map as the only parameter.

Next, we define two cases for calling the multi-method. This is done using the defmethod form. We’re essentially telling Clojure which code to call for which return value of the dispatch function. When the dispatch function returns :bye return "Goodbye" and when it returns :hello, return "Hello". Since, here, we don’t really care about the arguments passed in we just use _ in parameter list. Read some more examples here. Let’s test it out:

(greet {:message :bye})
=> "Goodbye"

(greent {:message :hello})
=> "Hello"

Mutli-methods are much broader in scope though, so if you find them interesting, read more about them. I’ve found that The Joy of Clojure has a pretty enlightening text on these.

Using multi-methods in our case

Before we write our mutli-method, let’s try and see how we’re going to use it.

(let [c (p/listen r [:items])]
  (go-loop [msg (<! c)
            my-data {}]
           (let [new-data (handle-value msg my-data)]
             (println "My data is now " new-data)
             (recur (<! c) new-data))))

We’re running a go-loop here listening for messages on channel c. We then pass the received message and the current value of my-data (our list of items) to our multi-method handle-value. handle-value is then supposed to return us the new state of our data.

We can define our multi-method like so:

(defmethod handle-value #(first %1))

(defmethod handle-value :child_added [[_ k v] data]
  (assoc data k v))

(defmethod handle-value :child_removed [[_ k _] data]
  (dissoc data k))

(defmethod handle-value :child_changed [[_ k v] data]
  (assoc data k v))

We know that handle-value gets passed two arguments. The event type is the first element in the first argument. We pass a #(first %1) function as the dispatch function when defining the multi-method, which pulls the first element out of the first argument passed to the multi-method.

The defmethod form then defines the three situations we want to handle. These methods don’t do much other than associng and disassocing values into passed in data parameter.

Final output

When I hookup all of this machinery and run the code. I get output like this:

My data is now  {2 do}         ;; [:child_added "2" "do"]

My data is now  {3 even, 2 do} ;; [:child_added "3" "even"]

My data is now  {3 even,       ;; [:child_added "4" "eat"]
                 4 eat, 2 do}

My data is now  {3 even,       ;; [:child_added "5" "Hello"]
                 4 eat,
                 5 Hello, 2 do}

My data is now  {3 event,      ;; [:child_changed "3" "event"]
                 4 eat,
                 5 Hello, 2 do}

My data is now  {3 event,      ;; [:child_removed "4" "eat"]
                 5 Hello, 2 do}
My data is now  {3 event,      ;; [:child_removed "2" "do"]
                 5 Hello}

Things look good.


Here are some of the things we discussed:

Until next time!