Scalable data-race analysis for Go

Develop a dedicated data-race analysis for Go that scales to large programs, thereby enabling the data-race-freedom assumption used to establish soundness of the most-general taint-flow summary to be discharged.

Background

The soundness of the most-general taint-flow summary for Go depends on the analyzed program being data-race-free. In the presence of a data race, an instruction may access memory through an access path outside the model’s input set, potentially making the summary unsound.

The paper notes that this assumption could in principle be discharged by a dedicated data-race analysis, but reports that, to the authors’ knowledge, no such analysis exists for Go at a scale suitable for large programs. Developing one would remove an unvalidated assumption underlying the soundness argument.