Correctness of data mined from CVS

Abstract

Source code repositories managed by the popular CVS are frequently mined by researchers to validate various hypotheses about how the development of a software product progressed. This paper presents a study where the development process of 17 student teams was followed. It will show that in an environment where the teams were permitted to manage their own CVS repositories that several errors emerged that could lead to the misinterpretation of a hypothesis. These were classified into three types: type one errors which relate to the non-use of the system; type two errors that emerged from the direct manipulation of the repository; and type three errors from the limitation of CVS not to record file name changes.

On mining sensor network software repositories

“…Each transaction contains a number of commits, an author, a log message, and a date. Note that we mine CVS repositories with the obvious issues, e. g., file naming, commit transaction, described in [7,17]. Since CVS does not keep track of transactions, we run a commit algorithm on the CVS history as described in [7].…”

Maintaining Fine-Grained Code Metadata Regardless of Moving, Copying and Merging

“…Subversion does provide the original name of a file when the file was renamed with Subversion's rename facility, but only then. CVS does not record file name changes at all, which leads to misinterpretations [5]. For example, Subversion's blame command is useful when origins of code are needed.…”