Predicting the maintainability of XSL transformations

XSLT is a popular language for implementing both presentation templates in Web applications as well as document and message converters in enterprise applications. The widespread adoption and popularity of XSLT raises the challenge of efficiently managing the evolution of significant amounts of XSLT code. This challenge calls for guidelines and tool support for developing maintainable XSLT code. In this setting, this paper addresses the following question: Can the maintainability of XSL transformations, measured in terms of code churn in the next revision of a transformation, be predicted using a combination of simple metrics? This question is studied using a dataset extracted from open-source software project repositories. An outcome of this empirical study is a set of statistical models for predicting the maintainability of XSL transformations with relatively high accuracy. In addition, by analyzing the major influencers of code churn in these models, the paper identifies guidelines for designing XSL transformations with reduced future churn.

Research highlights
► We study models to predict the churn of the next revision of an XSL transformation.
► Models trained using simple count metrics can predict high churn with 65% accuracy.
► Models are suitable for ranking transformations according to their expected churn.
► System-independent models achieve an accuracy comparable to system-specific models.
► High number of expressions without wildcards or function calls indicate high churn.