Casperjs/PhantomJs vs Selenium

They are attacking different problems. Since PhantomJS runs perfectly on the command-line, it is suitable as the first layer of smoke testing, whether as part of development workflow and/or in a continuous integration server. Selenium targets multiple browsers and hence it is very useful to ensure cross-browser consistency and carry out extensive testings across different operating systems.

If your web application needs to run on a variety of web browsers, running the UI testing only with PhantomJS will not yield the most test coverage. However, it is perfectly fine to launch PhantomJS and exercise some basic sanity checks before doing the in-depth tests. Imagine the madness of testing a finance application where the login screen is unintentionally broken and non-functional!

Note that the line between the two gets slightly blurred with the recent WebDriver support in the latest PhantomJS. It is now possible to quickly run the tests first using PhantomJS and then (assuming there is no serious error encountered) continue to execute the same tests thoroughly in a Selenium setup.

With the recent WebDriver integration (as Ariya has noted), you can now use Selenium to drive PhantomJS.

This is immensely powerful.

You can run a set of fully automated Selenium tests (using PhantomJS as the WebDriver implementation) via your CI on a headless Unix server on every check-in. Then if you want to test browser compatibility you can run your tests locally by changing the underlying WebDriver implementation to Chrome, Firefox etc.

I am currently writing a web extraction framework. I have 524 tests that get data from 250 websites using XPath. Initially the framework used a HTML parser, HTMLCleaner, but I am currently investigating using Selenium because I want Javascript support. I have run the tests against the HtmlUnit, Chrome, Firefox and PhantomJS drivers. Here is a comparison of the time taken and the number of failures for each approach:

                    Failures    Time (secs) HtmlCleaner         0           82  HtmlUnit            169         102 Google Chrome       38          562 Firefox             46          1159    PhantomJS           40          575

Some comments:

• In some cases the "failures" may not be failures at all, it may bethat the extractors are failing because Javascript is re-writing theDOM. I am in the process of analyzing the failures to find the cause.

• That said, HtmlUnit is the fastest Selenium driver but it is also unreliable. This unreliability does not just concern Javascript, there are problems processing "messy, dirty, real-world" HTML because something seems to be broken in the tag balancing algorithm. A couple of issues have been raised about this but they have not been fixed - see HTML-UNIT 1423 and HTML-UNIT 1046.

• Firefox is the slowest Selenium driver, even though I am disabling image loading and stylesheets. This is because it is the slowest to load and initialize, making it considerably slower than Chrome, and every time an extraction fails I need to reload the driver (in the tests I create a pool of 5 drivers to mitigate the URL retrieval delays for all the Selenium web drivers).

• PhantomJS achieves a better accuracy than Firefox, slightly lower than Chrome, but in around half the time of Firefox. What is more, I can run it on my dev box, it does not "take over my machine" by launching multiple browsers so I can get on with work.

I would highly recommend PhantomJS.