Browser-based testing tool which is not broken?

Question

Sorry for the controversial title here!

I've done a fair bit of browser-based testing (of JS-heavy web pages), using Selenium and Selenium-based libraries like Capybara. It has always been a big pain in the neck to get the test suites to work consistently. Yes, I know that rather than inserting calls to sleep() here and there, you are supposed to wait for DOM elements to become visible, or meet other conditions, yadda yadda. That's all good, and it does help a bit. I'd think that I wasn't doing it quite right, if it was not for...

If it was not for the time, before I had ever written Selenium code myself, when the company I was consulting for hired a firm which specializes exclusively in writing Selenium test suites. They sent a guy who was supposedly an expert on Selenium, and he wrote a full test suite for our web app. Need I tell you what happened when we actually used the test suite?

I've concluded that the whole approach which tools like Selenium use is subject to a big race condition. You have a test process, and a separate browser process which talk over a socket. The test process sends queries to the browser to find out the state of the DOM, and as the browser continues running, computes what to do. Then it sends a separate message to the browser to simulate a click, or keystroke, etc.

If you have ever written multi-threaded programs, you know that just doesn't work. It's a "check-then-act" race. If the test process could freeze JS execution in the browser, query its state, issue simulated clicks/keystrokes, and then unfreeze the browser, then it would be OK. (Like using a lock to protect the critical section of a multi-threaded program.)

The only way I can think of getting around the race condition would be for each test case to be written in JS. That precludes test cases which span navigation across multiple pages, of course. You also need a way of resetting the page to a "clean" state at the beginning of each test case.

Does anyone know a tool/approach for testing in-browser JS "apps", which allows mortals like me to write test suites which run 100% consistently, without tearing out all my hair?

(PS. Another conclusion from experience with Selenium: A lot of free jQuery plugins have bugs which don't appear when humans are using the page, simply because they can't click as fast as Selenium can!)

Answer 1

Have you seen casperjs, which can run on top of phantomjs (a headless webkit browser) or slimerjs (a headless gecko browser)?

The unit testing syntax is clean and friendly and it executes your tests in the context of the browser sequentially, using a method called evaluate().