I spent most of today working with Selenium, throwing out a bunch of emails to its devel list. I'm pretty excited about it.

There's a lot of complicated stuff about Selenium out there, but it's actually really simple. Grig added to his previous post with a shorter description, and Jason Huggans post also had a quick start for Windows/IE. Here's my quick start:

• Check out svn://... somewhere where your web server can see it. Lets say it's in /selenium. This must be the same web server that your application is on.
• Go view /selenium/javascript/TestRunner.html

Voila! Was that so hard? All that server instrumentation stuff, Twisted stuff, etc... really, just don't worry about that. Maybe later, but it's no place to start.

Now, go to selenium/javascript/tests/ and edit TestSuite.html to add more tests (or get rid of the ones that are there), and copy one of the files like TestOpen.html to start your own test (both files are extremely simple -- one's a list of links, the other is a table of commands).

There's more of course, but there's always more. One thing that caught me is that you must refer to links and submit buttons using a limited number of Element Locators -- by id (but I don't give my links ids!), with a Javascript DOM expression (icky domNamesEverywhere!), or with XPath. I chose XPath, even though I've never used XPath, and here's my mini-tutorial:

To follow a link by its text:

//a[text()='link text']

I haven't tried this, but maybe it would work more fuzzily:

//a[contains(text(), 'partial text')]

And for a submit button:

//input[@type='submit']

Or:

//input[@value='Button Description']

Anyway, hopefully we'll get this information into the documentation, and a quick start with Selenium will become more approachable.

I also started working on some infrastructure. The first is expanding and improving a script that takes tcpwatch logs and turns them into test scripts (tcpwatch_scriptgen.py). I previously generated PBP scripts (and in the future probably Twill scripts); now you can also make Selenium scripts.

The second is a CGI-based test suite. You can use a URL like /selenium/javascript/TestRunner.html?test=/path/to/TestSuite.html to use something other than selenium/javascript/tests/TestSuite.html -- which is nice, because you don't want to mix your application tests with the Selenium code. Anyway, you point it to the selenium_suite.py CGI script and it'll index all the suites you put in some directory.

Oh... and that CGI program can also turn comma-separated-value files into HTML tables, so you can write or edit your tests in a spreadsheet (which, I think, is an improvement over an HTML editor). And now tcpwatch_scriptgen.py can create both CSV files and HTML tables.

All of this, temporarily, is located in http://svn.colorstudy.com/home/ianb/webtest_experiement/ -- but it'll move someplace more permanent later. Maybe there will be a place for it in the Selenium repository, otherwise I'll put it in the Webware repository.