Ownership You Can Count On

While many forms of memory management have been proposed, the only ones to achieve widespread adoption have been explicit deallocation and garbage collection. This leaves programmers requiring explicit control of memory behavior unable to obtain the software engineering and security benefits of programming in a safe language. We present a new approach to memory management called alias counting which combines type-based object ownership and run-time reference counting of references by non-owning (aliasing) pointers. When an owning pointer is destroyed, if the target object's reference count is zero, the object is destroyed; otherwise a run-time error occurs. We have implemented alias counting in a safe C#-like language called Gel. The Gel compiler includes an effective reference count elimination optimization that takes advantage of ownership semantics. Our anecdotal experience in writing the Gel compiler and some smaller benchmarks in Gel is that the annotation requirements and extra programming effort are small, and the type system catches most ownership errors at compile time. Quantitatively, we compare microbenchmarks written in Gel, Java, C#, and C++, and versions of the Gel compiler written in both Gel and C#. In our benchmarks, Gel programs require at most 40% more memory than the corresponding C++ programs, while C# and Java generally require at least twice as much memory and often considerably more. Run-time performance varies considerably across all languages, with no clear winner, but Gel outperforms either C++ or Java/C# in all but one benchmark. Our reference count optimization typically eliminates 90% of all reference counting operations, speeding Gel programs by 20-40%.