Large repositories of source code available over the Internet, or within large organizations, create new challenges and opportunities for data mining and statistical machine learning. Here we first develop Sourcerer, an infrastructure for the automated crawling, parsing, fingerprinting, and database storage of open source software on an Internet-scale. In one experiment, we gather 4,632 Java projects from SourceForge and Apache totaling over 38 million lines of code from 9,250 developers. Simple statistical analyses of the data first reveal robust power-law behavior for package, method call, and lexical containment distributions. We then develop and apply unsupervised, probabilistic, topic and author-topic (AT) models to automatically 123 Mining and searching internet-scale software repositories 301 discover the topics embedded in the code and extract topic-word, document-topic, and AT distributions. In addition to serving as a convenient summary for program function and developer activities, these and other related distributions provide a statistical and information-theoretic basis for quantifying and analyzing source file similarity, developer similarity and competence, topic scattering, and document tangling, with direct applications to software engineering an software development staffing. Finally, by combining software textual content with structural information captured by our CodeRank approach, we are able to significantly improve software retrieval performance, increasing the area under the curve (AUC) retrieval metric to 0.92-roughly 10-30% better than previous approaches based on text alone. A prototype of the system is available at: http://sourcerer.ics.uci.edu.

Developers often learn to use APIs (Application Programming Interfaces) by looking at existing examples of API usage. Code repositories contain many instances of such usage of APIs. However, conventional information retrieval techniques fail to perform well in retrieving API usage examples from code repositories. This paper presents Structural Semantic Indexing (SSI), a technique to associate words to source code entities based on similarities of API usage. The heuristic behind this technique is that entities (classes, methods, etc.) that show similar uses of APIs are semantically related because they do similar things. We evaluate the effectiveness of SSI in code retrieval by comparing three SSI based retrieval schemes with two conventional baseline schemes. We evaluate the performance of the retrieval schemes by running a set of 20 candidate queries against a repository containing 222,397 source code entities from 346 jars belonging to the Eclipse framework. The results of the evaluation show that SSI is effective in improving the retrieval of examples in code repositories.

We develop and apply statistical topic models to software as a means of extracting concepts from source code. The effectiveness of the technique is demonstrated on 1,555 projects from SourceForge and Apache consisting of 113,000 files and 19 million lines of code. In addition to providing an automated, unsupervised, solution to the problem of summarizing program functionality, the approach provides a probabilistic framework with which to analyze and visualize source file similarity. Finally, we introduce an information-theoretic approach for computing tangling and scattering of extracted concepts, and present preliminary results.

No abstract

No abstract

This paper presents an analysis of a year long usage log of Koders, the first commercially available Internet-Scale code search engine (http://www.koders.com). The usage log comprises about ten million activities from more than three million users. Analysis of the usage data shows that despite of attracting a large number of visitors, Koders has a very sparse usage and that it lacks regular usage from many of its users. When compared to Web search, search behavior in Koders showed many similar patterns. A topic modeling analysis of the usage data shows what topics users of Koders are looking for. Observations on the prevalence of these topics among the users, and observations on how search and download activities vary across topics, lead to the conclusion that users who find code search engines usable are those who already know to a high level of specificity what to look for. This paper also presents a general categorization of these topics that provides insights on the different ways code search engine users express their queries. It identifies various forms of queries in Koders's log and the kinds of results addressed by the queries. It also provides several suggestions for improvements in code search engines based on the analysis of usage, topics, and query forms. The work presented in this paper is the first of its kind that reveals several insights on the usage of an Internet-Scale code search engine.

We present an analysis of modularity in aspect oriented design using the theory of modular design developed by Baldwin and Clark [10]. We use the three major elements of that theory, namely: i) Design Structure Matrix (DSM), an analysis and modeling tool; ii) Modular Operators, units of variations for design evolution; and iii) Net Options Value (NOV), a quantitative approach to evaluate design. We study the design evolution of a Web Services application where we observe the effects of applying aspect oriented modularization.Based on our analysis we get to the following three main conclusions. First, on the structural part, it is possible to apply the DSM to aspect oriented modularizations in a straightforward manner, i.e. without modifications to DSMs basic model. This shows that aspects can, in fact, be treated as modules of design. Second, the evolution of a design into including aspect modules uses the modular operators proposed by Baldwin and Clark, with a variant of the Inversion operator. This variant captures taking redundant, scattered information hidden in modules and moving it down or keeping it at the same level in the design hierarchy. Third, when calculating and comparing NOVs of the different designs of our application, we obtained higher NOV for the design with aspects than for the design without aspects. This shows that, under this theory of modularity, certain aspect oriented modularizations can add value to the design.

No abstract