23.7 Cognitive Aids

23.7 Cognitive Aids

Quality activities often require examining and understanding complex artifacts, from requirements statements to program code to test execution logs. Information clutter and nonlocality increase the cognitive burden of these tasks, decreasing effectiveness and efficiency. Even inherently manual tasks that depend on human judgment and creativity can be made more effective by cognitive aids that reduce cognitive burden by gathering and presenting relevant information in a task-appropriate manner, with a minimum of irrelevant and distracting details.

Information that requires a shift of attention (e.g., following a reference in one file or page to a definition on another) is said to be nonlocal. Nonlocality creates opportunities for human error, which lead to software faults, such as inconsistent uses of data values in a program, or inconsistent use of technical terms in a specification document. Not surprisingly, then, quality tasks often involve gathering and analyzing nonlocal information. Human analysis capability is amplified by bringing relevant information together. For example, where a human may be required to make a judgment about consistent use of technical terms, tools can support that judgment by gathering uses of terms together. Often tools synthesize a global view from scattered local information, as, for example, displaying a call graph extracted from many source code files.

Information required for a quality task is often obscured by a mass of distracting irrelevant detail. Tool support for focus and abstraction, delivering and drawing attention to relevant information while suppressing irrelevant detail, improve human effectiveness by reducing clutter and distraction. For example, an inspection tool that displays just the checklist items relevant to a particular inspection task and location in the artifact under inspection increases the efficiency and thoroughness of the human inspector. Similarly, an effective summary report of automated test executions quickly focuses attention on deviations from expected test behavior.

Cognitive aids for browsing and visualization are sometimes available as separate tools, but more often their features are embedded in other tools and customized to support particular tasks. Pretty-printing and program slicing,^[1] for example, improve code readability and make it easier to identify elements of interest. Diagrammatic representations condense presentation of code properties, providing a summary view of nonlocal information.

Diagrammatic and graphical representations are often used to present the results of program analysis, such as data and control flow relations, structural test coverage, distribution of faults and corrections in a program, and source code metrics. Figure 23.1 shows a sample screen shot that visualizes some characteristics of a program. Nodes represent classes and edges inheritance between classes. Node size and background summarize various metrics of the corresponding class. In the diagram of Figure 23.1, width indicates the number of attributes of the class, height indicates the number of methods, and color indicates lines of code, where white represents the smallest classes, black represents the largest, and intermediate sizes are represented by shades of gray. The graphic provides no more information than a table of values, but it facilitates a quicker and fuller grasp of how those values are distributed.

Figure 23.1: Visualization tools can summarize non-local information to facilitate understanding and navigation. The CodeCrawler tool, shown here, uses color, width, and height to represent three static measures of size (number of attributes, number of methods, and lines of code) with connections representing inheritance relations.

^[1]Program slicing is an application of static or dynamic dependence analysis (see Chapter 6) to identify portions of a program relevant to the current focus.