Theses

During the development of a software system, large amounts of new information, such as source code, work items and documentation, are produced continuously. As a developer works, one of his major activities is to consult portions of this information pertinent to his work to answer the questions he has about the system and its development. Current development environments are centered around models of the artifacts used in development, rather than of the people who perform the work, making it difficult and sometimes infeasible for the developer to satisfy his information needs. We introduce two developer-centric models, the degree-of-knowledge (DOK) model and the information fragments model, which support developers in accessing the small portions of information needed to answer the questions they have. The degree-of-knowledge model computes automatically, for each source code element in the development environment, a real value that represents a developer's knowledge of that element based on a developer's authorship and interaction data. We present evidence that shows that both authorship and interaction information are important in characterizing a developer's knowledge of code. We report on the usage of our model in case studies on expert finding, knowledge transfer and identifying changes of interest. We show that our model improves upon an existing expertise finding approach and can accurately identify changes for which a developer should likely be aware. Finally, we discuss the robustness of the model across multiple development sites and teams. The information fragment model automates the composition of different kinds of information and allows developers to easily choose how to display the composed information. We show that the model supports answering 78 questions that involve the integration of information siloed by existing programming environments. We identified these questions from interviews with developers. We also describe how 18 professional developers were able to use a prototype tool based on our model to successfully and quickly answer 94% of eight of the 78 questions posed in a case study. The separation of composition and presentation supported by the model, allowed the developers to answer the questions according to their personal preferences.

Previous research in computer science shows that software developers are typically deluged by an enormous volume of information daily. Improving the effectiveness of developers to filter this information may yield significant productivity improvements. To combat this overload, we introduce an indicator, called degree-of-knowledge (DOK), which is a real value indicating how much knowledge a developer has with a source code element. A developer's DOK values for a source code base can be computed automatically from authorship data mined from the source revision systems and from interaction data collected as the developer works. This indicator may help reduce information overload by, for instance, filtering the source code to only show the elements for which a developer has high knowledge. We describe our implementation of an efficient framework for computing DOK values in a development environment.

Software development environments have evolved to make it easy for third parties to integrate a variety of tools into the environment. Previous research has shown that developers often do not use or know all of the tools available in development environments that they regularly use. The most common solution to alleviate this problem is to provide a means to search through passive help documents. However, this approach requires a developer to be able to express her desires in a form understood by a search engine. To overcome this limitation, we propose using an active help system to automatically recommend to a developer tools that are appropriate to the developer’s current goal. We introduce an active help system to recommend navigation tools because previous research has shown a substantial amount of effort expended by a developer on a software evolution task is related to navigation. Our approach infers which structurally related pieces of information are of interest to a developer and determines when the navigation performed by the developer between the related information is suboptimal. In suboptimal navigation situations, our approach recommends a tool that would enable more efficient navigation. Our approach was implemented as a prototypical active help system, called Spy glass, in IBM’s Rational Team Concert development environment. To investigate whether and under what conditions Spyglass helps or hinders a developer in navigation, two empirical studies were conducted: a longitudinal user study in an undergraduate software engineering course and a laboratory user study comparing Spyglass to a passive tutorial. The longitudinal study provided feedback to improve the user interface of Spyglass and provided evidence that the developers who were novice to the development environment liked the idea of having an active help system in the environment. The laboratory study confirmed that Spyglass can help developers navigate between code elements more efficiently. The developers in the laboratory study also confirmed their desire for a system like Spyglass. The laboratory study also revealed that Spyglass requires improvements to its accuracy.