Software engineers working on projects involving systems integration, reuse, and maintenance spend a large fraction of their time trying to understand unfamiliar code. A particularly hard problem is understanding code that contains delocalized plans, that is, the design for some program task is spread across several non-adjacent subroutines or modules.

Modern testing tools collect program traces from test runs to evaluate test coverage thoroughness. This same trace data may provide insight for program comprehension at no extra cost. This report describes a very simple method that uses execution count data from different test runs to partition the code into equivalence classes that may represent delocalized plans. A small case study of one program was performed in which the classes identified by the method were evaluated by the original programmer to see if they represented meaningful plans. Results were mixed; the method identified some interesting design structures, but also often reported code segments that had no clear relationship.

A tool based on this method could be useful to a software engineer since it is easy to implement and would require minimal effort to use. A mouse click on a confusing code segment could reveal other segments in the same equivalence class. The software engineer could quickly determine if the class provides insight or not.

The method of analyzing trace data described here is very simple. The investigation of other program comprehension techniques for using this readily available information could be a fruitful area for further research.

This report may be cited as SERC-TR-81-F, Software Engineering Research Center, University of Florida, CIS Department, Gainesville, FL 32611, July, 1996.

Download a compressed Postscript file of this report.

(This page last modified December 11, 1998.)