Automatic detection and resolution of deadlocks in Go programs

• Autor(a) principal: Coelho, Jorge Gabriel Grave
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/155578
• topic: Deadlock detection; Deadlock resolution; Static analysis; Go programming language; Message-passing; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

The Go programming language is acquiring momentum in the development of concurrent software. Even though Go supports the shared-memory model, the message-passing alternative is the favoured idiomatic approach. Naturally, this practice is not exempt of the usual difficulties: programs may deadlock and the language run-time has only very basic support for deadlock detection. Previous research on deadlock detection mainly focused on shared-memory concurrency models. For mainstream languages, tools and approaches specific to the message-passing paradigm are scarce and incipient. There is however a large body of work on models of concurrency that only recently started to be applied to languages like Go. Since the Go run-time lets many deadlocks pass unnoticed, and the existing solutions provided by third party tools detect many deadlocks but only try to fix a limited set of specific patterns, imposing severe conditions to do so, there is a clear need for more general deadlock resolution strategies, going beyond prevention and avoidance. To gain insight on real-world deadlock bugs, we first built and categorized a collection of bugs sourced from high-profile open-source Go projects. Next, we extended and implemented an algorithm that takes an abstraction of the communication behaviour of the program and, when all the communication operations on channels necessary for progress are present, but a deadlock is possible, presents the problem and offers a possible resolution for the error. The extensions allows our approach to analyse a much wider range of real world programs. We conclude with an evaluation, comparing with two other state-of-the-art solutions.