Testing Process Spheres
Let's look at each component of the testing process individually.
We'll begin our discussion of the testing process with test planning, the most important activity in the testing process. It involves assessing risk, identifying and prioritizing test requirements, estimating testing resource requirements, developing a testing project plan, and assigning testing responsibilities to test team members. These ingredients can be structured as a formal test plan or they can be developed individually and used at the appropriate times.
The traditional idea of a test plan is the who, what, when, where, how, and how long of the testing process. Since using Rational RequisitePro's capabilities, we have adjusted our thoughts on what a test plan is and how it is used. We can import the software requirements document into tools such as Rational RequisitePro and then develop test scenarios directly from such tools into a test plan that is constructed using the RUP test plan template, which we have modified extensively (see Figure 1.2). From the test plan scenarios, we can create test requirements. They can be constructed directly in the scenario tables in the test plan document, or they can be created in a separate test requirements document using the RUP template. From either of these documents, we can create a test requirements view that we can export to a Comma Separated Values (CSV) file. Then we can open the test requirements during testing in Microsoft (MS) Excel and update it online with the test results and import it back into RequisitePro. This is how we defined our manual testing process. We do not yet have this fully implemented, but have piloted it and it works very well.
The point is that we now see the test plan as a working document derived from software requirements and linked to test requirements and test results. It is a dynamic document used during testing. The old idea of test plan is that it is a planning document. It is a document that forces the tester to think about what he is going to do during testing. From this perspective, it becomes a document that, once the planning stage is finished, sits on the shelf. In our experience, very few people refer back to the test plan during test execution. In fact, we have worked in several companies where test plans were actually created after the fact. With our approach, the test plan is created up front, and it is updated when the software requirements are updated; subsequently, the updates are reflected in the test requirements, which can actually be used for testing. The test plan standard is based on a modified version of the template contained in RUP, which accompanies the Rational Suite TestStudio.
The following is how we defined the manual system testing process at one client organization. We have also piloted this and it works very well. From the test plan scenarios, we create test case requirements. They are constructed directly in scenario tables created in the test plan document, as well as in separate test requirements documents (requirements grids). From either of these sources, we can create test requirements views that can be exported to CSV files. For manual testing, we open the CSV files in MS Excel. We use the information to guide manual test execution activities, and we update it online with the test results. We import the updated files back into our automated tool for results analysis and reporting.
Figure 1.2 contains a modified version of the RUP test plan template document table of contents (TOC). The TOC has been simplified in that the number of testing types has been reduced to include only Function Testing, Business Cycle Testing, Setup and Configuration Testing, and User Interface Testing.
FIGURE 1.2 RUP Test Plan TOC with Modifications
The purpose of the test plan is to assemble the information extracted from the requirements/design documents into test requirements that can be implemented as test scenarios. The test scenarios are the portion of the test plan that will directly feed the development of test conditions, test cases, and test data.
Desktop tools such as MS Office and MS Project can be used to automate test planning and project management. For example, checklists created in MS Excel spreadsheets can be used to assess and analyze risk and to document test requirements; MS Project can be used to produce the project plan; MS Word can be used to create a formal test plan that ties it altogether. These documents are test-planning artifacts. Just as software development artifacts need configuration management, so do test objects.
Test design includes identifying test conditions that are based on the previously specified test requirements, developing all possible functional variants of those conditions, divining the expected behavior(s) of each variant when executed against the application under test (AUT), and executing manual tests during the design process prior to test automation. The manual testing allows the test design engineer to verify that the test data are appropriate and correct for the automated test in which they will be used. It also allows the test designer to become a "human" tester who can identify errors automation might miss. Test design also embraces the layout of the test data that will be input to the AUT. When designing from a data-driven perspective, these same data also control the navigation of the automated test scripts. Last, the test scripts themselves have to be designed.
Designing the tests and the test data is the most time-consuming portion of the testing process. It is also the most important set of activities. If the tests do not test what the requirements have indicated, then the tests are invalid. If the test data do not reflect the intent of the tests, once again the tests are invalid. Test case design is so important that we have included a section in the appendices devoted to test design techniques and their use.
Test designers can use MS Excel to design and build the tests. If you use this approach, it is best to keep everything in a single workbook and to include one test conditions spreadsheet, one test data spreadsheet, and as many detailed test spreadsheets as are needed to adequately describe the environmental, pretest, and posttest activities that are related to each test. For manual testing, a test log should be constructed that will be used during test execution (used online, not as a printed test log). There are integrated tool suites available that support test data design and creation with mechanisms known as data pools. In all cases the data are stored as CSV files that can be read and interpreted by automated test scripts. Those data can be used for either manual or automated testing.
Test implementation can be subdivided into Test Construction, Test Execution, Test Results Capture and Analysis, and Test Result Verification. We'll discuss each activity separately.
Test Construction Test construction is facilitated using the same set of tools that test design employs. The test data are constructed in a spreadsheet in the same workbook as the test conditions. Those data can then be exported to CSV files that can be used at test execution. When the tests are executed via an automated framework, test construction also includes writing the test scripts. Automated test scripts are software programs. As such, they have their own programming languages and/or language extensions that are required to accommodate software testing events. The scripting language is usually embedded in a capture/playback tool that has an accompanying source code editor. The flavors of the languages vary by vendor and, as the associated syntax/semantics vary, so does the difficulty of using a specific product. In addition, some vendors' scripting languages and their recording tools are more robust than others.
The more specialized the commands available as part of the language, the more control the test engineer has over the test environment and the AUT. Specialized tests are built into the languages as commands that, when executed, test specific itemsfor example, graphical user interface (GUI) object properties and data and window existenceand do file comparisons. Some of the built-in test cases are useful and most are very powerful for GUI testing, but they are not all useful for functional testing. We implement many of the tests through executing external test data (data that reflect the test requirements) and verification of the results. Those data control how the test script behaves against the application; the data contain values the test script uses to populate the target application input data fields.
The design and implementation of the test scripts is left to the script-writer. If there are no guidelines for developing test scripts, the scripts that are created will most likely be badly structured and each will be the product of the personality of the individual who coded it. We have seen this happen when several people on the test team were given specific sections of an application and asked to write automated test scripts for their portion. We even gave them basic templates as starting points and, still, no two were alike. We have written and implemented a set of automated test script writing guidelines that is described in Chapter 8.
If possible, test script coding should be completed in parallel with test data development. Using an approach such as Archer Group's CSDDT allows the test scriptwriters to work independently of the test data designers. This is doable because the data drive the test scripts. The scripts are designed and built to specifics that create a general test-processing engine that does not care about test data content.
It is very important to have test script coding conventions in use if the workload is spread across a number of scriptwriters. It is also important that the right personnel are assigned to this set of activities. We have found that test designers do not like coding test scripts and that test implementers do not enjoy designing and building test conditions and test data. In fact, they both do crappy work when their roles are reversed.
Script writing, as it is used here, is the coding of test scripts in a test tool's proprietary test scripting language, in one of the extant application programming languages such as Java or Visual Basic; in one of the standard scripting languages such as Perl, CGI, or VB Script; or in the operating system's command procedure language (for example, coding Unix shell scripts that execute test procedures). As far as test script writing is concerned, you really need people who enjoy programming. Bruce was an engineer who moved into the software industry as a programmer before becoming a test scriptwriter. That type of experience makes a person a natural test script-writer. Test script coding requires either prior programming experience or current training in programming concepts such as logic, syntax, and semantics. It also requires an eye for developing complex logic structures. Because of these requirements, don't expect nontechnical testers to write test scripts and, furthermore, don't expect them to be able to create effective test scripts using the capture/playback facilities most tool suites provide. Test scripts developed by that method are a maintenance nightmare.
Test Execution Test execution can be manual, automated, or automated-manual. It is common wisdom in the testing industry that manual tests and automated tests each find different classes of errors. Thus, the experts say that we should do both. We agree to a great extentthat to do full-blown manual system tests and then follow up with automated regression tests is a neat idea, but most testing efforts have the resources to do only one or the other. What we suggest is a combination of both where manual testing occurs in parallel with test case design and construction.
The test designers should have the application open as they design and construct the test data, and they should execute it against the application fea-ture(s) it is designed to test. This accomplishes two things. First, it performs a validation of the data that will eventually be used in the automated regression tests and, second, it implements manual functional (system-level) tests of each application feature prior to the automated regression tests. Many errors can and will be found during test case design and construction when this process is followed. We have successfully applied this approach.
Test Results Capture and Analysis For manual testing, a test log
should be developed and implemented online. At the very least it should be a spreadsheet in the same workbook where the test conditions and test data are stored. Ideally, a test results database is used as a central repository for permanent storage. This is the case when the tests are executed using capture/playback tools such as Rational Robot or Mercury Interactive's Win-Runner. The test results are trapped as they occur and are written to the repository where later they can be viewed and reports can be printed.
Bruce has developed an automated manual test script (using Rational Robot and the SQABasic language) that does this very thing for manual tests (this tool is fully described in Chapter 9). It displays the manual test scripts much as a teleprompter does, and each step in the script can be selected during test execution. When execution is completed, the test can be logged as pass, fail, or skipped. The manual test script is based on one that was published several years ago by the Phoenix Arizona SQA Users Group (13). It has been completely rewritten and enhanced to add custom comment entries in the test log. This test script is included with the utilities that are on the FTP site that supports this book. It can easily be adapted to scripting languages other than SQABasic.
Test Results Verification Test results verification can be accomplished either manually or through test automation. Manually eyeballing the results and making subjective assessments of correctness is one way of verifying the test results. Test results can further be compared to sets of expected output data values, compared to values that should have been created in database rows based on the transactions tested, and compared against stored files and reports.
Test results verification is a task ripe for automation. Test results should be captured in a test log that is stored in a test repository. The test log should store results from previous test runs that can serve as baseline behaviors with which to compare the current test results. For manual tests, the test log can be an MS Excel workbook where new spreadsheets are created for each new test iteration. It does not matter how the test results are stored. What is important is that comparisons can be made. If a baseline is not established, then the assessment of a pass or fail status is based on a guess at what the results should be. Being able to define and store baseline behaviors is an important advantage and a strong argument for automated testing.
Commercially available automated testing tool suites offer a variety of automated verification methods. For example, Rational Robot uses SQA-Basic, which has special test cases called verification points that can be used to trap AUT characteristics; these can then be used as baselines for future regression tests.
Test Reporting Test reporting is an essential part of the testing process because it documents the test results and their analysis. Two levels of reporting are requireda summary report should be generated for middle- and upper-level technical managers and for customers, and a detailed report should be compiled and presented to the development team members as feedback.
These reports should be presented in standard formats that can be edited and tweaked for each individual test project report. We have employed the reporting template that is in RUP's documentation, but you can create your own. Two versions of this report can be createda summary report and a detailed report. You can also find templates and examples of test reporting on the World Wide Web.
An important reporting item is also defect tracking information. Defect tracking reports can be generated separately using tools such as Rational ClearQuest and/or MS Excel. Defect information should also be summarized and included in both the detailed and summary test evaluation reports. It is imperative to include a list of known defects that have not been addressed and that will be in the software upon its release. The information in the list should be grouped according to severity. Information such as this can be used to make intelligent release decisions, and help desk personnel can use it after the software goes into production.