Affiliation:
1. Picnic Technologies, University of Amsterdam, Amsterdam, Netherlands
2. Picnic Technologies, Amsterdam, Netherlands
3. University of Amsterdam, Amsterdam, Netherlands
Abstract
Software projects use a significant number of external libraries to speed up development and deliver high-quality software in a cost-effective way. Requirements placed on external libraries can change over time and libraries may cease to be maintained or evolve in incompatible ways. Consequently, a library may no longer be fit for purpose and require replacement. Such a
library migration
is generally performed manually and may be complex, error-prone, and time-consuming, especially for large codebases. It follows that an automated approach to library migrations is preferred.
We consider the suitability of Error Prone and Refaster, two Google tools capable of refactoring Java programs, for automated library migration. For validation we consider a migration in a proprietary industrial codebase as well as an actively developed open source codebase. We show that using out-of-the-box Error Prone and Refaster full migration of RxJava 2-referencing files is possible in only 35% and 0% of such files, respectively.
To increase the capabilities and be able to migrate more Rx-Java 2-referencing files completely, we create extensions to Error Prone and Refaster. Therefore we propose and create extensions that improve the accuracy with which Refaster templates are matched, as well as the correctness of their replacement operations. We additionally propose a set of Error Prone extensions to facilitate rewriting method return types. Together they enable a new class of rewrite operations especially relevant to library migrations. Lastly, several generalizations and improvements of existing Error Prone plugins are discussed. Together these improvements enable a new class of rewrite operations, especially relevant to library migrations. Using the proposed extensions, we manage to fully migrate an additional 35% of the RxJava 2-referencing files in the industrial codebase, and an initial 39% of such files in the open source codebase.
Publisher
Association for Computing Machinery (ACM)
Subject
Industrial and Manufacturing Engineering
Cited by
1 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献