Thursday, May 24, 2012

Javascript TDD/BDD/Unit Testing

I've been doing some pretty cool stuff at work lately and I think it's time for me to share some of the cool things that we're doing.  Web development isn't anything new, I've been writing HTML since I was 12 years old, but it has changed quite a bit since I wrote my first, very simple, web pages.

Another thing that isn't new for me is Javascript, but I never really got that into it or relied on it to do anything powerful.  Since starting at Best Buy, I've really gotten good at Javascript and have realized that it is quite powerful and can do a lot of good for a website.  I've also realized that it can also do a lot of harm to a website.  One thing that a lot of people omit from Javascript development is testing.  A lot of people consider testing their scripts in the browser to be adequate for their purposes.  I disagree.  Coming from back end systems, I know that thorough testing is very important and I think that this is a component of front end systems that has been lacking across the industry.  That being said, there are ways to automatically test your Javascript without using a browser.

One of the goals of the team I'm on is to introduce an automated testing tool to test Javascript automatically when the project is built.  Our build is written using ant, which I hadn't used since college, but since it's well documented, I didn't have any trouble picking it up again.  Also, I'm using Eclipse as my IDE, so it supports ant syntax highlighting and parsing, which is very helpful.  Ant is very powerful, but since it's written like XML, it can be hard to write complicated constructs like looping and conditions, but we've made it work for us, so far.

I was put in charge of incorporating Code Quality tools into our build.  At first, this was a little overwhelming.  I was just starting to get good at Javascript development and didn't have any experience with testing tools for it, but I started doing research on the internet and quickly found some testing frameworks that I didn't think would be hard to incorporate into our build.  We settled on using Jasmine and I worked on finding a way so that we could call our spec scripts from our build.

This started the frustrating part of our journey.  The people who built Jasmine don't support the different ways to launch the tool (like from an ant build), but only the framework itself.  Okay, I can overcome this, there are some open source tools that I can use listed right on their page.  I started with Jasmine-Node since we were using Node.js for our local web server to test our pages.  I quickly realized that this wouldn't work for us since Jasmine-Node doesn't create a DOM and a lot of our Javascript relies on a DOM in one way or another.

My next attempt was to try Node-Jasmine-DOM.  It has DOM support and Jasmine integration, so I thought I was getting close.  I incorporated it in the build and got it to run in the browser, but there's an aspect to Node development that I didn't realize until it was too late.  Node makes it really easy to use other Node tools as dependencies in the tool you are building.  This is great for reuse, but dependencies are called that for a reason.  One of the dependencies that the Node-Jasmine-DOM tool used had a defect in it and that was blocking our scripts from being tested in the build.  I waited patiently for a couple of weeks after submitting the bug to the owners of the dependent tool, and there was initially some good back and forth about it for a while and someone submitted a workaround for the bug.  I was excited and tried it out on my machine, but their workaround didn't work for mem (in Mac OSX), but worked in Windows.  I went back to the forum and they had closed my defect, so I commented that the workaround didn't work for me and I needed another way to make it work for me.  Another week of waiting without a response and I decided to pursue another way to run our tests.

I started my new search by trying to find a tool that would run Javascript straight from a command-line.  I would do myself what Node-Jasmine-DOM was trying to do (create a DOM and insert Jasmine).  Enter Rhino and Env.js.  I found a blog that had done this before and I was all set.  It took me less than a day to incorporate these new tools into the build and I was immediately running our specs from the build.  At the same time that I was ecstatic to be finally done with this, I was very frustrated that I had wasted so much time trying some of these other tools.

For those curious about what exactly I did, read on.  For those who don't want the technical details, thanks for reading this far!  This next part may not be for you as it will be pretty technical.

Our test target in our ant build does a few steps to run the spec scripts.  The first step is to identify the pages that need testing and has some logic built into it for that.  The next step is to convert these pages into spec runners (example), this means that I needed to insert references to our spec.js files and the Jasmine Javascript files.  Since I wanted to test the Javascript on the page, I just copied the page, changed the extension to *.runner.html and inserted the appropriate spec and Jasmine scripts.

The next step that the build does is to get the list of runner files and run them.  I created an ant macro for this step.  The Build-Doctor guy provided the envjs.bootstrap.js file that I started with, but later customized.  His was very simple and was basically just a for loop that told Envjs to load a page.  I found that that wouldn't work for more than one runner, so I changed it a bit to reload Envjs at the beginning of the loop and then load the runner file.  I also added some console.log entries so that the build is more vocal about what it's doing.

Here's where I got a little bold and changed the Jasmine.js and Env.js files to eliminate some extra console.log calls that I didn't feel were required and I added some in a couple of places where I wanted visibility.  In the end, it looks great in the build and it creates JUnit style reports that our Jenkins CI server can interpret.

I was planning on covering the other major aspect of our build, Freemarker, but this post is getting pretty long, so I'll save that for another day.  Below are some examples of what I did in our build.

This is the macro I wrote in our ant build.
<macrodef name="runRunnerList">
 <attribute name="runnerList" />
  <echo message="Running files: @{runnerList}" />
  <exec executable="java">
   <arg line="-jar ${dir.lib}/java/js.jar -opt -1 lib/javascript/envjs.bootstrap.js @{runnerList}" />
This is the results of running a spec in our build as seen in the terminal.
     [exec] Launching runner: /trunk/bin/debug/_slider/_sliderTest.runner.html
     [exec] Runner Started.
     [exec] _slider : should have 100 hundred items after the method addItem is called 100 times ... 
     [exec] Passed.
     [exec] _slider : should have 99 items after the method removeItem is called ... 
     [exec] Passed.
     [exec] _slider : should have zero items after the method removeAllItems is called ... 
     [exec] Passed.
     [exec] _slider : should show slider when content extends past the width of the slider ... 
     [exec] ***FAILED***
     [exec] _slider : should hide slider when content does not extend past the width of the slider ... 
     [exec] Passed.
     [exec] _slider : should remove padding from the left side of the viewport ... 
     [exec] Passed.
     [exec] _slider : should add the class bby-slider-item-first to the first item in the slider ... 
     [exec] Passed.
     [exec] _slider : should recieve the second item out of five and the HTML should match ... 
     [exec] Passed.
     [exec] _slider: 7 of 8 passed.
     [exec] Runner Finished.
     [exec] Done with runner: /trunk/bin/debug/_slider/_sliderTest.runner.html
This is one of our runner files. The highlighted parts are what the build inserts into the file.
<!DOCTYPE HTML><html lang="en"><head>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
<link href="../css/__common/__common.css" rel="stylesheet" type="text/css" media="all">
<link href="../css/_slider/_slider.css" rel="stylesheet" type="text/css" media="all">
<script type="text/javascript" src="../js/__common/libraries.js"></script>
<script type="text/javascript" src="../js/_slider/_slider.js"></script>
<link rel="stylesheet" type="text/css" href="../js/lib/jasmine-1.2.0/jasmine.css" />
<script src="../js/lib/jasmine-1.2.0/jasmine.js"></script>
<script src="../js/lib/jasmine-1.2.0/jasmine-html.js"></script>
<script type="text/javascript" src="../js/specs/_slider.spec.js"></script>
<script src="../js/lib/jasmine-dom/jasmine-dom-matchers.js"></script>
<script src="../js/lib/jasmine-dom/jasmine-dom-fixtures.js"></script>
<script src="../js/lib/jasmine.console_reporter.js"></script>
<script src="../js/lib/jasmine.junit_reporter.js"></script>
<script type="text/javascript">
$(function() {
    jasmine.getEnv().addReporter(new jasmine.JUnitXmlReporter());
    jasmine.getEnv() = "__common";
I will note here that most of our runner files have actual content in the body. The specs for this runner insert some test content into the body before the spec is run. The scripts in the body are inserted just before the </body> tag.

No comments:

Post a Comment