abrooks / topical-tabs

An example Chrome extension using Chromex library (ClojureScript)

Home Page:https://github.com/binaryage/chromex

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

chromex-sample GitHub license

An example extension using Chromex library

This project acts as a code example for chromex library but also as a skeleton with project configuration following best practices. We recommend to use it as a starting point when starting development of your own extension.

chromex-sample has a minimalist background page, popup button and content script:

  • background page listens for connections from popup buttons and content scripts (there can be multiple of them)
  • popup button connects to the background page and sends a simple "HELLO" message after connection
  • content script connects to the background page and sends a simple "HELLO" message after connection
  • content script does a simple page analysis upon launch (it counts number of script tags) and sends an info message to the background page
  • background page listens to tab creation events and notifies all connected clients about new tabs being created

chromex-sample project has following configuration:

  • uses leiningen + lein-cljsbuild
  • integrates cljs-devtools
  • integrates figwheel (for background page and popup buttons)
  • under :unpacked profile (development)
    • background page and popup button
      • compiles with optimizations :none
      • namespaces are included as individual files and source maps work as expected
      • figwheel works
    • content script
      • due to security restrictions, content script has to be provided as a single file
      • compiles with :optimizations :whitespace and :pretty-print true
      • figwheel cannot be used in this context (eval is not allowed)
  • under :release profile
    • background page, popup button and content script compile with optimizations :advanced
    • elides asserts
    • no figwheel support
    • no cljs-devtools support
    • lein package task is provided for building an extension package for release

Local setup

Extension development

We assume you are familiar with ClojureScript tooling and you have your machine in a good shape running recent versions of java, maven, leiningen, etc.

  • clone this repo somewhere:

    git clone https://github.com/binaryage/chromex-sample.git
    cd chromex-sample
  • chromex sample is gets built into resources/unpacked/compiled folder.

    In one terminal session run (will build background and popup pages using figwheel):

    lein fig

    In a second terminal session run (will auto-build content-script):

    lein content
  • use latest Chrome Canary with Custom Formatters enabled

  • In Chrome Canary, open chrome://extensions and add resources/unpacked via "Load unpacked extension..."

Debugging

Chrome extension development is more complex than regular ClojureScript front-end work. You are writing (and debugging) multiple parallel communicating processes: your background page, your popup, and all the browser pages running your content script.

Amazingly, the ClojureScript tooling and Figwheel live coding remain very usable in this environment. But, you need to be aware of a few things, particularly in regard to compiler warnings:

Most warnings do not appear in the repls. Figwheel intercepts them for display in the browser. Warning will appear in the Chrome console and, when possible, as an overlay in the browser window. But, the exact behavior depends upon which part of your code has the error:

Content script: Warnings and errors will appear in the repl running lein content.

popup: Chrome normally closes the popup anytime focus leaves Chrome. So, if you are working in your editor, the popup is closed and you will not see any error messages anywhere. This can be very frustrating but is easy to fix. When you first open the popup, right click on its icon and select Inspect popup. This opens the Chrome inspector/console and keeps the popup open while the inspector remains open. Any errrors will appear in both the console and as the Figwheel overlay in your popup window. Also, of course, this gives you niceties of Figwheel live coding. Your changes will appear immediately, with no need to close and reopen the popup.

background: The background code is running under Figwheel, so no messages will appear in the repl. It also has no visibile window, so no Figwheel overlay can appear. You will only see warnings in the Chrome console. You can open the inspector/console from chrome://extensions. Under your extension, click on the Inspect Views line.

In summary, effective live debugging requires up to five open windows on your screen:

  • Your editor;
  • The shell running lein content, if you are making changes to content script code;
  • The web browser, with open popup and/or content page;
  • A Chrome inspector console, watching the background page; and
  • A Chrome inspector console, watching the popup page.

Extension packaging

Leiningen project has defined "release" profile for compilation in advanced mode. Run:

lein release

This will build an optimized build into resources/release. You can add this folder via "Load unpacked extension..." to test it.

When satisfied, you can run:

lein package

This will create a folder releases/chromex-sample-0.1.0 where 0.1.0 will be current version from project.clj. This folder will contain only files meant to be packaged.

Finally you can use Chrome's "Pack extension" tool to prepare the final package (.crx and .pem files).

Code discussion

Before reading the code below you should get familiar with Chrome Extension System architecture.

Popup page

Let's start with popup button code:

; -- a message loop -------------------------------------------------------------------------------

(defn process-message! [message]
  (log "POPUP: got message:" message))

(defn run-message-loop! [message-channel]
  (log "POPUP: starting message loop...")
  (go-loop []
    (when-some [message (<! message-channel)]
      (process-message! message)
      (recur))
    (log "POPUP: leaving message loop")))

(defn connect-to-background-page! []
  (let [background-port (runtime/connect)]
    (post-message! background-port "hello from POPUP!")
    (run-message-loop! background-port)))

; -- main entry point -----------------------------------------------------------------------------

(defn init! []
  (log "POPUP: init")
  (connect-to-background-page!))

When a popup button is clicked, Chrome creates a new javascript context and runs our code by calling init!. At this point we call runtime/connect to connect to our background page. We get a background-port back which is a wrapper of runtime.Port. background-port implements chromex protocol IChromePort which we can use to post-message! to our background page. background-port also implements core-async/ReadPort so we can treat it as a core.async channel for reading incoming messages sent by our background page. You can see that implemented in run-message-loop! which takes messages off the channel and simply prints them into console (in process-message!).

Marshalling

At this point you might ask. How is it possible that we called API method runtime/connect and got back background-port implementing IChromePort? That is not documented behaviour described in Chrome's extension APIs docs. We would expect a native javascript object of type runtime.Port.

This transformation was done by marshalling subsystem implemented in Chromex library. Marshalling is responsible for converting parameter values when crossing API boundary. Parameter values can be automatically converted to ClojureScript values when returned from native Javascript API calls and in the other direction parameters can be converted to native Javascript values when passed into API calls. This is a way how to ease extension development and promote idiomatic ClojureScript patterns.

Chromex library does not try to do heavy marshalling. You should review marshalling logic in marshalling.clj and marshalling.cljs files to understand which parameter types get converted and how. You can also later use this subsystem to marshall additional parameter types of your own interest. For example automatic calling of js->clj and clj->js would come handy at many places.

Message loop

It is worth noting that core.async channel returns nil when closed. That is why we leave the message loop after receiving a nil message. If you wanted to terminate the message channel from popup side, you could call core.async's close! on the message-channel (it implements core-async/Channel and will properly disconnect runtime.Port).

As a consequence you cannot send a nil message through our channel.

Background page

Let's take a look at background page which is also pretty simple. It just has to handle multiple clients and their individual message loops. Also it maintains one main event loop for receiving events from Chrome. With core.async channels the code reads quite well:

(def clients (atom []))

; -- clients manipulation -------------------------------------------------------------------------

(defn add-client! [client]
  (log "BACKGROUND: client connected" (get-sender client))
  (swap! clients conj client))

(defn remove-client! [client]
  (log "BACKGROUND: client disconnected" (get-sender client))
  (let [remove-item (fn [coll item] (remove #(identical? item %) coll))]
    (swap! clients remove-item client)))

; -- client event loop ----------------------------------------------------------------------------

(defn run-client-message-loop! [client]
  (log "BACKGROUND: starting event loop for client:" (get-sender client))
  (go-loop []
    (when-some [message (<! client)]
      (log "BACKGROUND: got client message:" message "from" (get-sender client))
      (recur))
    (log "BACKGROUND: leaving event loop for client:" (get-sender client))
    (remove-client! client)))

; -- event handlers -------------------------------------------------------------------------------

(defn handle-client-connection! [client]
  (add-client! client)
  (post-message! client "hello from BACKGROUND PAGE!")
  (run-client-message-loop! client))

(defn tell-clients-about-new-tab! []
  (doseq [client @clients]
    (post-message! client "a new tab was created")))

; -- main event loop ------------------------------------------------------------------------------

(defn process-chrome-event [event-num event]
  (log (gstring/format "BACKGROUND: got chrome event (%05d)" event-num) event)
  (let [[event-id event-args] event]
    (case event-id
      ::runtime/on-connect (apply handle-client-connection! event-args)
      ::tabs/on-created (tell-clients-about-new-tab!)
      nil)))

(defn run-chrome-event-loop! [chrome-event-channel]
  (log "BACKGROUND: starting main event loop...")
  (go-loop [event-num 1]
    (when-some [event (<! chrome-event-channel)]
      (process-chrome-event event-num event)
      (recur (inc event-num)))
    (log "BACKGROUND: leaving main event loop")))

(defn boot-chrome-event-loop! []
  (let [chrome-event-channel (make-chrome-event-channel (chan))]
    (tabs/tap-all-events chrome-event-channel)
    (runtime/tap-all-events chrome-event-channel)
    (run-chrome-event-loop! chrome-event-channel)))

; -- main entry point -----------------------------------------------------------------------------

(defn init! []
  (log "BACKGROUND: init")
  (boot-chrome-event-loop!))

Again, main entry point for background page is our init! function. We start by running main event loop by subscribing to some Chrome events. tabs/tap-all-events is a convenience method which subscribes to all events defined in tabs namespace to be delivered into provided channel. Similarly runtime/tap-all-events subscribes all runtime events. We could as well subscribe individual events for example by calling tabs/tap-on-created-events, but subscribing in bulk is more convenient in this case. As you can see we create our own ordinary core.async channel and wrap it in make-chrome-event-channel call. This is an optional step, but convenient. make-chrome-event-channel returns a channel which is aware of Chrome event subscriptions and is able to unsubscribe them when the channel is about to be closed (for whatever reason). This way we don't have to do any book keeping for future cleanup.

Events delivered into the channel are in a form [event-id event-args] where event-args is a vector of parameters which were passed into event's callback function (after marshalling). So you can read Chrome documentation to figure out what to expect there. For example our :chromex.ext.runtime/on-connect event-id is documented under runtime/on-connect event and claims that the callback has a single parameter port of type runtime.Port. Se we get IChromePort wrapper, because marshalling converted native runtime.Port into ClojureScript-friendly IChromePort on the way out.

Ok, when anything connects to our background page, we receive an event with ::runtime/on-connect id. We call handle-client-connection! with event-args. Here we have to do some client-specific work. First, add this new client into a collection of active clients. Second, send a hello message to the client and finally run client-specific event loop for receiving messages from this client. We don't do anything with received messages, we just print them into console with a bit of information about the sender. When our client message channel gets terminated (for whatever reason), we remove client from active clients and forget about it.

Notifying clients about interesting events

We provide an additional separate functionality from maintaining client message loops. When Chrome notifies us about a new tab being created. We simply send a message to all our connected clients by calling tell-clients-about-new-tab!.

More on cleanup

You might be asking why there is no explicit cleanup code here? There should be some .removeListener calls when we are leaving message loops, no?

This cleanup is done under the hood because we are using Chromex wrappers here. Wrappers act as core.async channels but know how to gracefully disconnect when channel is closed (or close channel when client disconnected). In case of client connections you get a runtime.Port wrapper automatically thanks to marshalling. In case of main event loop you created a wrapper explicitly by calling make-chrome-event-channel.

Please keep in mind that you can always access underlying objects and talk to them directly if needed.

Content Script

Our content script is almost copy&paste of popup page code:

; -- a message loop -------------------------------------------------------------------------------

(defn process-message! [message]
  (log "CONTENT SCRIPT: got message:" message))

(defn run-message-loop! [message-channel]
  (log "CONTENT SCRIPT: starting message loop...")
  (go-loop []
    (when-some [message (<! message-channel)]
      (process-message! message)
      (recur))
    (log "CONTENT SCRIPT: leaving message loop")))

; -- a simple page analysis  ----------------------------------------------------------------------

(defn do-page-analysis! [background-port]
  (let [script-elements (.getElementsByTagName js/document "script")
        script-count (.-length script-elements)
        title (.-title js/document)
        msg (str "CONTENT SCRIPT: document '" title "' contains " script-count " script tags.")]
    (log msg)
    (post-message! background-port msg)))

(defn connect-to-background-page! []
  (let [background-port (runtime/connect)]
    (post-message! background-port "hello from CONTENT SCRIPT!")
    (run-message-loop! background-port)
    (do-page-analysis! background-port)))

; -- main entry point -----------------------------------------------------------------------------

(defn init! []
  (log "CONTENT SCRIPT: init")
  (connect-to-background-page!))

Upon launch we connect to the background page, send hello message and start a message loop with background page.

Additionally we call do-page-analysis! which does some simple DOM access, counts number of script tags and sends a reporting message to the background page.

Receiving messages from background page

As you can see, we don't have any interesting logic here for processing messages from background page. In process-message! we simply print the received message into console. It works! You can test it by creating new tabs. Background page should be sending notifications about new tabs being created.

About

An example Chrome extension using Chromex library (ClojureScript)

https://github.com/binaryage/chromex

License:Other


Languages

Language:Clojure 79.2%Language:Shell 16.1%Language:HTML 3.3%Language:JavaScript 1.4%