TMap - A good regression test is invaluable.

Contact Home Sitemap

Home / Archief / A good regression test is invaluable.

A GOOD REGRESSION TEST IS INVALUABLE.

In practice, regression tests are often poorly designed. A good method to build, use and control regression testing has a number of principles which makes it possible to:

Specify and implement test cases on the basis of priority (which is an important test case or test case which covers major functionality/risk);
quickly and adequately report on the progress of test specification and/or test performance;
accurately plan and budget test routes;
fast and variable determining regression tests;
easy handling of changes in the test object in the tests.

A good regression test is invaluable. Especially in the management situation, testing provides confidence that the new version of the system or package as a whole is working. Structured test methods offer many tools and handles to create good regression test sets. Thereof tools support the (mostly) automated execution.

The principle is that from each test case a collection of additional data is determined into the test cases for the regression test are ‘classified’. Using these classifications all cross sections along the subsets of test cases can form the total test that are selected.

Examples of classifications are:

Application.
Part object.
Function.
Risk class.
Process (line).
Release.
Requirement.
Transaction.
Severity category.

A good selection of the correct classifications and the classification of the test cases will determine the usefulness of this concept. A key question is the classification by severity category. This classification gives the 'weight' of the test case in the test set and makes it possible to use a variable in test depth when to draft a risk-based regression test. The application of these severity categories in the selection of regression tests is as follows (classification in three categories):

Select per object part only the test cases in category 1, a small regression test is created. This subset is used for some object that no adjustments are made (or for an pre test for a new or substantially changed object parts).
The test cases of category 2 (including category = 1) provide a standard regression test, for example, some object part to which adjustments were made.
The test cases in category 3 (= including category 1 and 2) cover the entire object part and applies to new or substantially modified object parts.

The degree of detail used in the specification of the test cases has no demands. When it is anticipated that test cases will be executed by testers without matter knowledge, the test cases need to be specified in detail.

Only on one point the concept has a specific requirement set for the test cases. They must be independent of each other, as described in the specification of the physical test cases. This is the so-called independence principle of the concept. The test cases also need to be executed parallel with each other. Test cases that exclusively use the test environment for a certain period of time hinder the performance of other test cases. This makes planning the duration of the requirement test.

Application of this concept makes the size of the (regression) testing and related activities in the test path measurable.

As with the test ware in general, there should also be considered when, how and by whom the tests may be updated.
For a fuller explanation see the white-paper Test Cube