Supporting Developers in Porting Software via Combined Textual and Structural Analysis of Software Artifacts

In the engineering and scientific domains software commonly has a long lifespan, lasting decades instead of years. Due to this lifespan, software often outlives the current generation of hardware, and in turn needs to be modified to execute on newer classes of hardware architectures [1]. Supporting developers in this difficult software maintenance activity is very important in order to improve their productivity, reduce bugs, and prevent architectural erosion [3]. Software engineering researchers have found that developers commonly begin maintenance tasks by first identifying and then comprehending the relevant program units for that task [5]. Program comprehension at the outset of maintenance tasks has been well studied and supported by a number of tools, utilizing static and dynamic information extracted from the code base. Research in the last decade in applying textual analysis approaches to source code, adapted from the fields of natural language processing and information retrieval, has produced notable results and resulted in the proliferation of program comprehension tools based on this type of information, often combined with static or dynamic sources [2]. Textual approaches have now matured to be applicable to the problem of enabling long-lasting high performance scientific and engineering software, a domain where natural language information is usually more challenging to extract than in general purpose software. Based on these opportunities, our position is that the use of textual and structural analyses coupled with platform usage patterns extracted by mining similar open source projects can improve the quality and efficiency of scientific and engineering software maintenance. Our position is based on two key insights: 1) combining structural and textual analyses can reveal information to support scientific and engineering software maintenance; 2) mining software repositories of projects that exist within the same platform can reveal patterns that could aid developers in identifying and comprehending relevant program elements for porting. We envision that research in these directions could yield software comprehension and recommendation tools that could be applied to a code base in order to improve porting to new platforms during maintenance. In the following we discuss each of the two key ideas in turn. Combining structural and textual analyses. Natural language processing of identifiers and comments embedded in source code has become recognized as a necessary tool in aiding developers accomplish certain software engineering tasks, such as feature location [2], software modulariza-tion [4], and automated code summarization [6]. Numerous text analysis algorithms and approaches, including topic modeling, …