Dynamic feature traces: finding features in unfamiliar code

Abstract

This dissertation shows that existing functional tests of software features can be used by a developer unfamiliar with the software system to identify source code relevant to those features as well as understand why the source code is relevant.There have been prior techniques to locate features in source code which can be roughly broken down into techniques that use static analysis and those that use dynamic analysis.Features, being behaviors of a system, are dynamic in nature. Therefore, this dissertation focuses on dynamic techniques, rather than the use static techniques.The dynamic techniques all require significant knowledge about the system before the technique can be useful. Furthermore, they all suffer in one or both of these respects: they make binary judgments about which source code artifacts relate to a feature without determining to what extent they relate (meaning that the feature must be precisely characterized, or else the technique will return inaccurate results), or they do not provide an idea of why a piece of code is relevant to a feature.Our technique of creating Dynamic Feature Traces improves upon previous work 1) by taking advantage of an existing test suite thereby reducing the amount of system knowledge necessary to use the technique, 2) by ranking the source code artifacts by how strongly they are related to the feature, and 3) by retaining some part of the execution trace of the test suite so that developers can understand why a piece of code is part of a feature. We show that our technique provides benefit over other techniques by applying it to existing software systems and comparing its results to an existing technique.ii

On the effectiveness of accuracy of automated feature location technique

“…Higher recall might be achieved by a longer list of methods, as evaluated in [11], [22]. In this experiment, developers validated 50 methods during five sessions.…”

Topology analysis of software dependencies

“…This feature supports the iterative application of our approach (in which the fuzzy suggestion set can become the set of interest), but also to use the approach with other techniques that produce fuzzy sets of interest (e.g., Dynamic Feature Traces [Eisenberg and De Volder 2005]). We detail how we take into account the fuzziness of the input set in Section 4.4.…”

“…This step helps achieve a more precise and complete description of the code that implements a feature, but requires additional effort on the part of the developers. A number of other techniques have also been proposed that address the same problem while providing additional functionalities, such as a probabilistic ranking of the results or tools to analyze the interaction of features [Antoniol and Guéhéneuc 2005;Eisenberg and De Volder 2005;Salah and Mancoridis 2004].…”

A Survey of Feature Location Techniques

“…Eisenberg et al [15], similarly to Eisenbarth et al [14], present an attempt to deal with the complexity of scenario definition. The approach assumes that the user is unfamiliar with the system and thus should use pre-existing test suites, such as those typically available for systems developed with a Test-Driven Development (TDD) strategy.…”