WebSite Testing

Dr. Edward Miller, Software Research, Inc.

Introduction

The nearly instant worldwide audience makes a WebSite's quality and reliability crucial to its success. The nature of the WWW and of WebSite software pose unique software testing challenges. Webmasters, WWW applications developers, and WebSite quality assurance managers need tools and methods that meet very specific needs. Our technical approach, based on extending existing WWW browsers, offers many attractive benefits in meeting these needs.

Background

Within minutes of going live, a WWW application can have many thousands more users than a conventional, non-WWW application. The immediacy of a WebSite creates immediate expectations of quality, but the technical complexities of a WebSite and variances in the available browsers make testing and quality control that much more difficult, than "conventional" client/server or application testing. Automated testing of WebSites thus is both an opportunity and a significant challenge.

Defining Website Quality and Reliability

Like any complex piece of software there is no single quality measure to fully characterize a WebSite.

There are many dimensions of quality, and each measure will pertain to a particular WebSite in varying degrees. Here are some of them:

• Timeliness: How much has a WebSite changed since the last upgrade?

• Structural Quality: Are all links inside and outside the WebSite working? Do all of the images work?

• Content: Does the content of critical pages match what is expected?

• Accuracy and Consistency: Are today's copies of the pages downloaded the same as yesterday's?

• Response Time and Latency: Does the WebSite server respond to a browser request within certain parameters? In an E-commerce context, what is the end to end response time after a SUBMIT?

• Performance: Is the Browser -> Web -> WebSite -> Web -> Browser connection quick enough? How does the performance vary by time of day, by load and usage?

Clearly, “Quality” is in the mind of the WebSite user. A poor-quality WebSite, one with many broken pages and faulty images, with Cgi-Bin error messages, etc. may cost in poor customer relations, lost corporate image, and even in lost sales revenue. Very complex WebSites can sometimes overload the user.

Website Architectural Factors

A WebSite can be quite complex, and that complexity can be a real impediment in assuring WebSite Quality.

What makes a WebSite complex? These are the issues test systems have to contend with:

• Browser. There is a kind of de facto standard: the WebSite must use only those constructs that work with the majority of browsers. But this still leaves room for a lot of creativity, and a range of technical difficulties.

• Display Technologies. What you see in your browser is actually composed from many sources:

  • HTML: Various versions of HTML must be supported.

  • Java, JavaScript, ActiveX: Obviously JavaScript and Java applets are likely parts of a WebSite, and the quality process must support these.

  • Cgi-Bin Scripts: All of the different types of Cgi-Bin Scripts (perl, awk, shell-scripts, etc.) need to be handled; tests will need to check “end to end” operation.

• Navigation. Navigation in a WebSite often is complex and has to be quick and error free.

• Object Mode. The display you see in a browser changes dynamically; the only constants are the "objects" that make up the display. Testing ought to be in terms of these objects.

• Server Response. How fast the WebSite host responds influences whether a user moves, continues, or gives up.

Website Test Automation Requirements

Assuring WebSite quality automatically requires conducting sets of tests, automatically and repeatably, that demonstrate required properties and behaviors. Here are some required elements of tools that aim to do this.

• Browser Independent Tests should be realistic, but not be dependent on a particular browser.

• No Buffering, Caching. Local caching and buffering should be disabled so that timed experiments are a true measures of performance.

• Object Mode. Object mode operation is essential to protect an investment in test suites and to assure that test suites continue operating when WebSite pages experience change.

• Tables and Forms. Even when the layout of a table or form varies in the browser's view, tests of it should continue independent of these factors.

Tests need to operate from the browser level for two reasons: (1) this is where users see a WebSite, so tests based in browser operation are the most realistic; and (2) tests based in browsers can be run locally or across the Web equally well. Local execution is fine for quality control, but not for performance measurement work, where response time including Web-variable delays reflective of real-world usage is essential.

Website Dynamic Validation

Confirming validity of what is tested is the key to assuring WebSite quality the most difficult challenge of all. Here are four key areas where test automation will have a significant impact.

1. Operational Testing. Individual test steps may involve a variety of checks on individual pages in the WebSite:

   • Page Consistency: Is the entire page identical to a prior version? Are key parts of the text the same or different?

   • Table, Form Consistency: Are all of the parts of a table or form present? Correctly laid out? Can you confirm that selected texts are in the "right place"

   • Page Relationships: Are all of the links on a page the same as they were before? Are there new or missing links? Are there any broken links?

   • Performance Consistency, Response Times: Is the response time for a user action the same as it was (within a range)?

2. Test Suites. Typically you may have dozens or hundreds (or thousands?) of tests, and you may wish to run these tests in a variety of modes: unattended, distributed across many machines, background, etc.

3. Content Validation. Apart from how a WebSite responds dynamically, the content should be checkable either exactly or approximately. Here are some ways that content validation could be accomplished:

   • Structural: All of the links and anchors should match with prior baseline data.

   • Checkpoints, Exact Reproduction: One or more text elements in a page should be markable as "required to match".

   • Selected Images/Fragments: The tester should be able to rubber-band sections of an image and require that the selection image match later during a subsequent rendition of it.

4. Load Simulation. Load analysis needs to proceed by having a special purpose browser act like a human user. This assures that the performance checking experiment indicates true performance -- not performance on simulated but unrealistic conditions. There are many "http torture machines" that generate large numbers of http requests, but that is not necessarily the way real-world users generate requests.

Testing System Characteristics

Considering all of these disparate requirements, it seems evident that a single product that supports all of these goals will not be possible. However, there is one common theme: the majority of the work seems to be based on "...what does it [the WebSite] look like from the point of view of the user?" That is, from the point of view of someone using a browser to look at the WebSite.

This observation led our group to conclude that it would be worthwhile trying to build certain test features into a "test enabled web browser", which we called CAPBAK/Web in the expectation that this approach would let us do the majority of the WebSite quality control functions using that engine as a base.

Browser Based Solution - With this as a starting point, we determined that the browser based solution had to meet these additional requirements:

• Commonly Available Technology Base. The browser had to be based on a well known base (there appear to be only two or three choices).

• Some Browser Features Must Be Deletable. At the same time, certain requirements imposed limitations on what was to be built. For example, if we were going to have accurate timing data we had to be able to disable caching because otherwise we are measuring response times within the client machine rather than "across the web."

• Extensibility Assured. To permit meaningful experiments, the product had to be extensible enough to permit timings, static analysis, and other information to be extracted.

Taking these requirements into account, and after investigation of W3C's Amaya Browser and the open-architecture Mozilla/Netscape Browser we chose the IE Browser as our initial base for our implementation of CAPBAK/Web.

User Interface - How the user interacts with the product is very important, in part because in some cases the user will be someone very familiar with WebSite browsing and not necessarily a testing expert. The design we implemented takes this reality into account.

• "C" Scripting. We use interpreted "C" language as the control language because the syntax is well known, the language is fully expressive of most of the needed logic, and because it interfaces well with other products.

• Files Interface. We implemented a set of dialogs to capture critical information and made each of them recordable in a text file. The dialogs are associated with files that are kept in parallel with each browser invocation:

  • Keysave File: This is the file that is being created the file is shown line by line during script recording as the user moves around the candidate WebSite.

  • Timing File: Results of timings are shown and saved in this file.

  • Messages File: Any error messages encountered are delivered to this file. For example, if a file can't be downloaded within the user-specified maximum time an error message is issued and the playback continues. (This helps preserve the utility of tests that are partially unsuccessful.)

  • Event File: This file contains a complete log of recording and playback activities that are useful primarily to debug a test recording session or to better understand what actually went on during playback.

• Operational Features.Based on prior experience, the user interface for CAPBAK/Web had to provide for several kinds of capabilities already known to be critical for a testing system. Many of these are critically important for automated testing because they assure an optimal combination of test script reliability and robustness.

  • Script Capture/Replay: We had to be able to capture a user's actual behavior online, and be able to create scripts by hand.

  • Object Mode: The recording and playback had to support pure-Object Mode operation. This was achieved by using internal information structures in a way that lets the scripts (either recorded or constructed) refer to objects that are meaningful in the browser context.

  • [Adjustable] True-Time Mode: We assured realistic behavior of the product by providing for recording of user-delays and for efficient handling of delays by incorporating a continuously variable "playback delay multiplier" that can be set by the user.

  • Playback Synchronization: For tests to be robust there must be a built-in mode that assures synchronization so that Web-dependent delays don't interfere with proper WebSite checking. CAPBAK/Web does this using a proprietary playback synchronization method that waits for download completion (except if a specified maximum wait time is exceeded).

  • Timer Capability: To make accurate on-line performance checks we built in a 1 millisecond resolution timer that could be read and reset from the playback script.

  • Validate Selected Text Capability: A key need for WebSite content checking, as described above, is the ability to capture an element of text from an image so that it can be compared with a baseline value. This feature was implemented by digging into the browser data structures in a novel way (see below for an illustration). The user highlights a selected passage of the web page and clicks on the "Validate Selected Text" menu item.
    
    

    Figure 1. Illustration of CAPBAK/Web Validate Selected Text Feature.

    What results is a recorded line that includes the ASCII text of what was selected, plus some other information that locates the text fragment in the page. During playback if the same text is not found at the same location an error message is generated.

  • Multiple-playback: We confirmed that multiple playback was possible by running separate copies of the browser in parallel. This solved the problem of how to multiply a single test session into a number of test sessions to simulate multiple users each acting realistically.

• Test Wizards - In most cases manual scripting is too laborious to use and making a recording to achieve a certain result is equally unacceptable. We built in several test wizards that mechanize some of the most common script-writing chores.

  • Link Wizard: This wizard creates a script based on the current Web page that visits every link in the page. Scripts created this way are the basis for "link checking" test suites that confirm the presence (but not necessarily the content) of URLs.

  • FORM Wizard: For E-Commerce testing which involves FORMS we included in the system a FORM Wizard that generates a script that:
    • Initializes the form.
    • Presses each pushbutton by name.
    • Presses each radio button by name.
    • Types a pre-set script fragment into each text field.
    • Presses SUBMIT.

    Here is a sample of the output of this wizard, applied to our standard test page:

    void name() { /* Produced by CAPBAK/Web [IE] Ver. 1.5 Form Wizard */ /* (c) Copyright 1999 by Software Research, Inc. */ WT_InitLink("http://www.testworks.com/Products/Web/CAPBAK/example1/"); WT_SubmitForm(FORM:0:12, "RESET FORM"); WT_SelectOneRadio(FORM:0:0, "now", "TRUE"); WT_SelectOneRadio(FORM:0:1, "next", "TRUE"); WT_SelectOneRadio(FORM:0:2, "look", "TRUE"); WT_SelectOneRadio(FORM:0:3, "no", "TRUE"); WT_SelectCheckBox(FORM:0:4, "concerned", "TRUE"); WT_SelectCheckBox(FORM:0:5, "info", "TRUE"); WT_SelectCheckBox(FORM:0:6, "evaluate", "TRUE"); WT_SelectCheckBox(FORM:0:7, "send", "TRUE"); WT_FormTextInput(FORM:0:8, "TestWorks"); WT_FormTextInput(FORM:0:9, "TestWorks"); WT_FormTextInput(FORM:0:10, "TestWorks"); WT_FormTextInput(FORM:0:11, "TestWorks"); WT_SubmitForm(FORM:0:13, "SUBMIT FORM"); }

    
    Figure 2. Sample of Output of FORM Test Wizard.
    

    The idea is that this script can be processed automatically to produce the result of varying combinations of pushing buttons. As is clear, the wizard will have pushed all buttons, but only the last-applied one in a set of radio buttons will be left in the TRUE state.

  • Text Wizard: For detailed content validation this wizard yields up a script that includes in confirmation of the entire text of the candidate page. This script is used to confirm that the content of a page has not changed (in effect, the entire text content of the subject is recorded in the script).

Example Uses

Early applications of the CAPBAK/Web system have been very effective in producing experiments and collecting data that is very useful for WebSite checking. While we expect CAPBAK/Web to be the main engine for a range of WebSite quality control and testing activities, we've chosen two of the most typical and most important applications to illustrate how CAPBAK/Web can be used.

• Performance Testing Illustration - To illustrate how CAPBAK/Web measures timing we have built a set of Public Portal Performance Profile TestSuites that have these features:

  • Top 20 Web Portals: We selected 20 commonly available WebSites on which to measure response times. These are called the "P4" suites.

  • User Recording: We recorded one user's excursion through these suites and saved that keysave file (playback script).

  • User Recording: We played back the scripts on a 56 kbps modem so that we had a realistic comparison of how long it would take to make this very-full visit to our selected 20 portals.

  • P4 Timings: We measured the elapsed time it took for this script to execute at various times during the day. The results from one typical day's executions showed a playback time range of from 457 secs. to 758 secs (i.e. from -19% of the average to +36% of the average playback time).

  • Second Layer Added: We added to the base script a set of links to each page referenced on the same set of 20 WebSites. This yielded the P4+ suite that visits some 1573 separate pages, or around 78 per WebSite. The testsuite takes around 20,764 secs (~5 Hrs 45 mins) to execute, or an average of 1038 secs per WebSite.

  • Lessons Learned: It is relatively easy to configure a sophisticated test script that visits many links in a realistic way, and provides realistic user-perceived timing data.

• E-Commerce Illustration - This example shows a typical E-Commerce product ordering situation. The script automatically places an order and uses the Validate Selected Text sequence to confirm that the order was processed correctly. In a real-world example this is the equivalent of (i) selecting an item for the shopping basket, (ii) ordering it, and (iii) examining the confirmation page's order code to assure that the transaction was successful. (The final validation step of confirming that the ordered item was actually delivered to a specific address is not part of what CAPBAK/Web can do.)

  • Typical Order Form: We based this script on a typical order form that collects customer information including, for example, a code number (a credit card number).

  • Type-In with Code Number: Starting with the FORM Wizard generated script, we modify it to include only the parts we want, and include the code number 8889999 (See Figure 1).

  • Response File: Once the playback presses the SUBMIT button the WebServer response page shows up displaying the response to the code number. We use the Validate Selected Text (See Figure 1) to capture the response number text.

  • Error Message Generated: If the Cgi-Bin scripts make a mistake this will be caught during playback because the expected exact text 8889999 will not be present, it will be something else.

  • Completed TestScript: Here is the complete testscript for CAPBAK/Web that illustrates this sequence of activities.

    void name() { /* Recording by CAPBAK/Web [IE] Ver. 1.5 (c) Copyright 1999 by Software Research, Inc. */ WT_InitLink("http://www.soft.com/Products/Web/CAPBAK/example1/example1broken.html"); WT_SelectOneRadio(FORM:1:0, "buying-now", "TRUE"); WT_SelectOneRadio(FORM:1:1, "next-month", "FALSE"); WT_SelectOneRadio(FORM:1:2, "just-looking", "FALSE"); WT_SelectOneRadio(FORM:1:3, "no-interest", "FALSE"); WT_SelectOneRadio(FORM:1:4, "Yes", "TRUE"); WT_SelectOneRadio(FORM:1:5, "Yes", "TRUE"); WT_SelectOneRadio(FORM:1:6, "Yes", "TRUE"); WT_SelectOneRadio(FORM:1:7, "Yes", "TRUE"); WT_FormTextInput(FORM:1:8, "Mr. Software"); WT_FormTextInput(FORM:1:9, "415-550-3020"); WT_FormTextInput(FORM:1:10, "[email protected]"); WT_FormTextInput(FORM:1:11, "8889999"); WT_SubmitForm(FORM:1:13, "SUBMIT FORM"); WT_Wait(3425); WT_ValidateText(12, 143, "88899999"); }

    Figure 3. Script for E-Commerce Test Loop.

  • Lessons Learned: This examples illustrates how it is possible to automatically validate a website using CAPBAK/Web by detecting when an artificial order is mis-processed.

Summary

All of these needs and requirements impose constraints on the test automation tools used to confirm the quality and reliability of a WebSite. The CAPBAK/Web approach offers some significant benefits and technical advantages when dealing with complicated WebSites. Better, more reliable WebSites should be the result.

Resources

A more complete version of this paper can be found at WebSite Testing

You can learn more about the CAPBAK/WEB system by taking a Tour of CAPBAK.

There is a detailed description of the P4 Family of CAPBAK/Web Examples.