Sidekicks and superheroes: a look into student reasoning about concurrency with threads versus actors

When computations consist of different processes or threads that may execute at the same time or in an interleaved manner, we say the computation is concurrent or that it exhibits concurrency. University students often struggle to reason about concurrency. Researchers have presented students with natural language descriptions of scenarios involving concurrency and have evaluated students' natural language responses to explore their prior knowledge and to determine characteristics of student reasoning about concurrency. However, natural language responses are necessarily abstract and pertain to the design stage of software development, while studies in which students are asked to implement solutions provide a more concrete view of student reasoning and reveal difficulties that arise when putting design into practice. In this work we build upon prior work and ask: What are common and problematic features in student approaches to solving concurrency-related problems? How do these features differ from the design phase to the implementation phase of the software development life cycle? In this study we asked students enrolled in a jointly offered upper division undergraduate / MS level course on "Programming with Concurrency" to implement a solution to the same problem (a sidekick/superhero version of the party matching problem) using two different approaches: Threads-based in Java, and Actors-based in Scala. We performed qualitative feature analysis of their implementations and their written reflections to better understand student reasoning when programming concurrent solutions. The feature of additional complexity was found in our earlier natural language study in which students described their designs. Additional complexity was also represented in student implementations in this study and adversely impacted student success as their programs became cumbersome and hard to trace. Student reflections provided insight into their resistance to modeling prior to implementation, and indicated that they found formal modeling techniques difficult. These findings suggest that lighter weight interventions that address student difficulties earlier in the development process may be more palatable to students and thus have substantial impact.