Application programming interfaces (APIs) are a common and industrially-relevant means for third-party software developers to reuse external functionality. Several techniques have been proposed to help migrate client code between library versions with incompatible APIs, but it is not clear how well these perform in an absolute sense. We present a retroactive study into the presence and nature of API incompatibilities between several versions of a set of Java-based software libraries; for each, we perform a detailed, manual analysis to determine what the correct adaptations are to migrate from the older to the newer version. In addition, we investigate whether any of a set of adaptation recommender techniques is capable of identifying the correct adaptations for library migration. We find that a given API incompatibility can typically be addressed by only one or two recommender techniques, but sometimes none serve. Furthermore, those techniques give correct recommendations, on average, in only about 20% of cases.

Automated refactoring is a key feature of modern IDEs. Existing refactorings rely on the transformation of source code declarations, in which references may also be transformed as a side effect. However, there exist situations in which a declaration is not available for refactoring or would be inappropriate to transform, for example, in the presence of dangling references or where a set of references should be retargeted to a different declaration.

Software developers who extend or repair existing software systems spend considerable effort in understanding how their modifications will require follow-on changes in order to work correctly. Tool support for this process is available for single, popular languages, but does not suffice for less popular languages, uncommon language variants, or arbitrary combinations of languages and connection technologies. We have created the DSketch tool so that developers can create a lightweight pattern specification for how dependencies can be heuristically identified in their systems. We performed two case studies involving industrial developers who applied our tool for conducting polylingual dependency analysis in software systems; the developers found it easy to configure the tool for their needs, were able to adapt their patterns to new contexts, and had sufficiently accurate dependency predictions for their work.

Software dependency analysis is an important step in determining the potential impact of changes. Existing tool support for conducting dependency analysis does not sufficiently support systems written in more than one language. Tools based on semantic analyses are expensive to create for combinations of multiple languages, while lexical tools provide poor accuracy and rely heavily on developer skill. This paper reports on an investigation into the application of a series of incrementally-better island grammars to an industrial, open-source polylingual system to determine the cost-to-accuracy relationship involved in developing and applying island grammars for dependency analysis. The results of our study suggest the effort-cost in writing richer island grammars rises faster than the resulting accuracy.