<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"> <html> <head> <meta content="text/html; charset=ISO-8859-1" http-equiv="content-type"> <title>Developing Drivers</title> </head> <body> <h2>Overview</h2> This page details important information for people developing drivers for Selenium.<br> <h2>Same Origin Policy</h2> This is a security issue that affects all modern browsers. It is well described <a href="http://www.mozilla.org/projects/security/components/same-origin.html">here</a>, but for our purposes, it constrains the way that a JavaScript in a browser may interoperate with other open frames and windows. In short, the Selenium JavaScript app must come from (or appear to) the same origin as the AUT.<br> <h2>The Driver</h2> The driver has a number of duties. These are typically.. <br> <ul> <li>Instantiating the web server with a dynamic app (servlets for Java) </li> <li>Launching a browser with a with a URL that makes sense for connecting to that server </li> <li>providing an API to such that commands can be routed though the dynamic app to the browser</li> </ul> <h2>Local, Remote and URLs<br> </h2> An application may be testable in a remote location by humans, but for scripted testing and the need for the driver to dynamically drive the browser, it is optimal for the driver to be on the same machine as that serving the AUT and the browser testing it. <br> <br> Thus, an application as deployed ...<br> <br> <img alt="http://localhost:8080/" src="images/stockmeister.png" style="width: 361px; height: 68px;"><br> <br> ... would appear like so for the purposes of testing ...<br> <br> <img alt="http://localhost:8080/" src="images/localhostAut.png" style="width: 346px; height: 43px;"><br> <br> As with the standalone version of Selenium, there are a number of files (HTML and JavaScript) that comprise the bulk of the testing framework and sit in the browser as testing occurs. It makes most sense to put these in a virtual directory on the same server ...<br> <br> <img alt="http://localhost:8080/selenium/" src="images/localhostSelenium.png" style="width: 326px; height: 57px;"><br> <br> The dynamic webapp needs to be similarly mounted ...<br> <br> <img alt="http://localhost:8080/selenium/driver?..." src="images/localhostDriver.png" style="width: 318px; height: 52px;"><br> <br> As the dynamic is the link between what is happening in the browser and the driving process, it we need to somehow have an instance reference to it. This is easier in some languages and web servers than others. Also full programatic start/stop control over the web server is not always possible for some larger web servers implementations.<br> <h2>Reply/Request Architecture</h2> Because a browser cannot open a socket and listen on it, we must effectively initiate communication thru the driver on the browser side. Ignoring the possibilities granted by keep-alive, we simply poll from the browser to the server and pick up commands to process inside the browser. Results of commands are also passed back to the dynamic hander over the same mechanism. These are in fact done in the same HTTP request. The results from the <span style="font-style: italic;">previous</span> command go back as query string parameters, and the <span style="font-style: italic;">next</span> command is communicated in a text/plain document<br> <br> The previous/next business introduces some complexity on the driver side of the design. Namely hidden behind the API, the driver must manage queues for the outgoing commands and the (some time later) incoming responses. Java, until 1.5, did not have a blocking queue to make this easy. Most other languages did.<br> <h2><span style="font-weight: bold;">Selenese</span></h2> <table style="width: 100%;" border="1" cellpadding="2" cellspacing="2"> <tbody> <tr> <td>Selenese is the contrived (and mostly hidden) wire language that the driver uses to speak to the browser-bot through the dynamic handler. It uses HTTP for its transport, and is quite simple. <br> <br> Responses come from the browser to the driver in a query string like <br> <br> <span style="font-family: monospace;">commandResult=OK</span>, <br> <br> Commands go from the driver to the browser-bot in a text/plain document:<br> <br> <span style="font-family: monospace;">| open | /foo/bar.html | |</span><br style="font-family: monospace;"> <br> This two way communication is of course invisible to the observer.</td> <td style="width: 30%; background-color: rgb(255, 153, 102);">The BrowserBot, by understanding Selenese, allows a process other than the browser itsself to direct events on the Application Under Test. <br> <br> The Selenese language is simple enough to be commandable by any language that has an API that can handle HTTP requests. <br> <br> Thus, Selenese allows many different open, free or closed license drivers to interoperate with the BrowserBot.</td> </tr> </tbody> </table> <h2>Choregraphy</h2> The driver cleary has some fairly heavy things to do. It is important for some robustness to be built into the design. For example it may take a long time for the browser to be instantiated on a particular platform. It is appropriate for wait timeouts to be generous here. Similarly whilst issuing individual commands it is important for the driver to wait a sufficient amount of time for a command to execute (and its result to come back). For the most part on a localhost setup, it will be because the AUT is slow, but it could be because some break in the app means that there will be no subsequent response. Some timeout regime should be able to recover from, mark a test as failed (for reasons of timeout), and start again with the next test.<br> <br> <br> </body> </html>