Please note: This master’s thesis presentation will be given online.
Yangtian
Zi,
Master’s candidate
David
R.
Cheriton
School
of
Computer
Science
Supervisor: Professor Gregor Richards
Gradual type systems with monotonic dynamic semantics, such as HiggsCheck implementing SafeTypeScript, are able to achieve decent performance, making them a viable option for real-world applications. However, the type restrictions for objects in monotonic dynamic semantics are, as the name suggests, monotonic. Once a typed reference is defined or assigned to refer to a object, the contract carrying the type obligation of the reference is as part of the object for the remainder of execution. In some cases, such contracts can be "zombies": the reference that justifies a contract is out of scope, yet the object still retains the type obligation.
In this thesis, we propose a novel idea and an implementation of contract liveness: contracts must be justified by live stack references defined with associated type obligations. Our implementation, taking inspiration from how garbage collectors approximate object liveness by reachability of objects, approximates contract liveness by reachability of contracts. Then, to achieve a much closer approximation to contract liveness, we introduce a poisoning process: we nullify the stack references justifying the violated contract, and associate the location which triggered the contract violation with a poisoned reference for blame.
The implementation is compared with the original implementation of HiggsCheck. The comparison showed that our system is fully compatible with code which raised no errors, and only incurred an 8.14% slowdown on average. We also discuss the performance of the contract removal process, and possible worst cases for the liveness-based system. Also, the semantics of HiggsCheck SafeTypeScript is modified to formalize the liveness-based type system. Our work proves that relaxations of contractual obligations in a gradually typed system with monotonic semantics are viable and realistic.
To join this master’s thesis presentation on Zoom, please go to https://uwaterloo.zoom.us/j/93166120476?pwd=aVZ6LzNIZXppclFWTFZibmx2SlFSUT09.