Taking Remote Screenshots With Selenium And The Robot Framework

.Screenshot are incredibely helpful when doing user interface tests, to get fast feedback why a test has failed. Selenium provides several great opportunities to do exactly that. But, there are many problems related to that at the moment, from which some will discussed, even solved in the following post.

Update (2010-02-15): This post refers to SeleniumLibrary version 2.2. In the meantime, the proposals made to be included for remote screenshots have been included in version 2.3. (Release Notes ). Thanks to the Robot Framework Team!

Our preferred way to integrate Selenium into our automated acceptance tests is the Robot Framework. This enables us to drive tests through all layers of the application, from the database up to the UI. Since it’s automated, this happens daily (or better nightly) or even on every commit. And this is where the first problem occurs. Our CI-Server (Hudson) runs on Linux … but without an XServer. Thus, the selenium server cannot run on the same machine, from with the acceptance tests are triggered. Another reason to run the selenium server on a different host is, that it’s quite difficult to run the Internet Explorer on Linux. A browser that still in wide use, especially in corporate networks, despite the numerous security warnings and rendering issues.

Many screenshot methods that are offered by the selenium server, take the file to save the screenshot to as an argument. This makes it difficult to access that screenshots (scp, fileshares, …). You’ll waste precious time just to access the file, while the analysis and debugging should be as easy as possible. But there’s also two methods in Selenium , that return the Screenshot as String base64 encoded PNG). This looks like a promising path 🙂

However, the existing SeleniumLibrary for the Robot Framework does not expose those methods as keyword. So we have to help ourselves, until it will be officially supported (Issue 89 ). So please download the latest Source-Distribution of the Robot SeleniumLibrary 2.2.2 , and edit the file __init__.py. Please excuse my first, amateurish python coding trials:

The functions _absnorm and _write_to_file I borrowed from the OperatingSystem Robot Library (and also slightly modified them). To install the modified Selenium Library, just execute “setup.py install”.

How can the new keyword be used in a Robot Testcase? I find it most helpful to get a browser screenshot after a failing test, to debug the problem quickly. You can use the testcase teardown for that, which invokes the keyword when the test failed:

Testing

Setting	Value
Library	SeleniumLibrary	10	localhost	4444
Library	OperatingSystem

Variable	Value
${BROWSER}	ff

Test Case	Action	Arguments
TestSelenium PASS	Open Browser	http://www.google.de	${BROWSER}
	[Teardown]	Selenium Teardown
TestSelenium FAIL	Open Browser	http://www.google.de	${BROWSER}
	Page Should Contain	AAAAAAAAAAAAA-IMPOSSIBLETEXT
	[Teardown]	Selenium Teardown

Keyword	Action	Arguments
SeleniumTeardown	Run Keyword If Test Failed	Take Screenshot
	Close Browser
Take Screenshot	Run Keyword If	‘${BROWSER}’ != ‘*iexplore’ and ‘${BROWSER}’ != *’ie’ and ‘${BROWSER}’ != ‘*internetexplorer’ and ‘${BROWSER}’ != ‘*iehta’	Capture Remote Screenshot

Now, in every teardown it is checked if the test failed. If this is the case, the keyword “Take Screenshot” will be called. This checks for the correct browser, and only if its not the Internet Explorer, it calls the new Keyword “Capture Remote Screenshot”, which embeds the screenshot in the robot logfile:

In order to get screenshots working with internet explorer, a few hacks were necessary. At first I had to realize, that I have to call the correct selenium method, otherwise the screenshot is completely black (not only for IE, but for all Browsers). If you use the method selenium.capture_screenshot_to_string(), you screenshot will be black. It uses Java to take a screenshot from the entire screen (not only the browser content). This works with all browsers, unless the selenium server runs in the background. We are running the selenium server on a remote desktop, and in order to get non-black screenshots you have to say logged in. An alternative setup was using VNC, but this seemed also unpractical to us (via stackoverflow ):

What we do is launch everything from under the context of a VNC session. On Windows, configure VNC to launch a session upon startup. Then make sure the user auto-logs in. Then place a .bat file in Program Files->Startup that launches Selenium RC. It’s kind of a pain, but it’s the most reliable way I’ve found for ensuring that Selenium RC starts in an environment that supports screenshots, launching IE, interacting with native events, etc.

The last alternative is, to not capture the entire screen, but only the browser content. This has the advantage, that you will see the complete website, even if it is larger than the screen. This method uses javascript to get hold of the rendered browser content. Problem: Works perfectly with Firefox, but not at all with Internet Explorer. Not yet. This can be fixed — so I thought. The following turned out to be a dead end. If somebody knows how to fix it, I’d be very interested to hear about it.

WIN: Screenshots with IE

This is the only working method, that I could figure out to take Screenshots with the internet explorer. Run the selenium server in “singleWindow” mode. In order to do that, execute:

java -jar selenium-server.jar -singleWindow

Additionally, selenium has to use the browser profile “*iexploreproxy”, that’s why the testcase above filters for all other IE-profiles.

Note: Eventually, you still have to install the latest SnapsIE library (see below). I have not tested, if the above works when I deinstall it again.

Achtung (2010-02-15): SnapsIE 0.2 indeed has to be installed to make screenshots work with IE.

FAIL: Screenshots with IE

This did not work at all. As I said, if you know why, let me know:

1. Download and install SnapsIE 0.2
2. Unpack selenium-server-1.0.1.jar and
   • Replace core\lib\snapsie.js with the version from SnapsIE 0.2
   • Edit the file core\scripts\selenium-api.js according to the proposal by Elf in the Selenium-Forum. I extended the changes a little, so here’s the complete new method::