4 Search Results for "Xie, Ningning"

This artifact contains the mechanical formalization of the calculi associated with the paper Union Types with Disjoint Switches. All of the metatheory has been formalized in Coq theorem prover. We provide a docker image as well the code archive. The paper studies a union calculus ({λ_{u}}). Primary idea of {λ_{u}} calculus is a type based disjoint switch construct for the elimination of union types. We also study several extensions of the {λ_{u}} calculus including intersection types, distributive subtyping, nominal types, parametric polymorphism and an extension for the empty types.

Union types are nowadays a common feature in many modern programming languages. This paper investigates a formulation of union types with an elimination construct that enables case analysis (or switches) on types. The interesting aspect of this construct is that each clause must operate on disjoint types. By using disjoint switches, it is possible to ensure exhaustiveness (i.e. all possible cases are handled), and that none of the cases overlap. In turn, this means that the order of the cases does not matter and that reordering the cases has no impact on the semantics, helping with program understanding and refactoring. While implemented in the Ceylon language, disjoint switches have not been formally studied in the research literature, although a related notion of disjointness has been studied in the context of disjoint intersection types. We study union types with disjoint switches formally and in a language independent way. We first present a simplified calculus, called the union calculus (λ_u), which includes disjoint switches and prove several results, including type soundness and determinism. The notion of disjointness in λ_u is in essence the dual notion of disjointness for intersection types. We then present a more feature-rich formulation of λ_u, which includes intersection types, distributive subtyping and a simple form of nominal types. This extension reveals new challenges. Those challenges require us to depart from the dual notion of disjointness for intersection types, and use a more general formulation of disjointness instead. Among other applications, we show that disjoint switches provide an alternative to certain forms of overloading, and that they enable a simple approach to nullable (or optional) types. All the results about λ_u and its extensions have been formalized in the Coq theorem prover.

Polymorphism and subtyping are important features in mainstream OO languages. The most common way to integrate the two is via 𝖥_{< :} style bounded quantification. A closely related mechanism is row polymorphism, which provides an alternative to subtyping, while still enabling many of the same applications. Yet another approach is to have type systems with intersection types and polymorphism. A recent addition to this design space are calculi with disjoint intersection types and disjoint polymorphism. With all these alternatives it is natural to wonder how they are related. This paper provides an answer to this question. We show that disjoint polymorphism can recover forms of both row polymorphism and bounded polymorphism, while retaining key desirable properties, such as type-safety and decidability. Furthermore, we identify the extra power of disjoint polymorphism which enables additional features that cannot be easily encoded in calculi with row polymorphism or bounded quantification alone. Ultimately we expect that our work is useful to inform language designers about the expressive power of those common features, and to simplify implementations and metatheory of feature-rich languages with polymorphism and subtyping.

Gradual typing is an effective approach to integrate static and dynamic typing, which supports the smooth transition between both extremes via the (programmer-controlled) precision of type annotations [Jeremy Siek and Walid Taha, 2006; Siek et al., 2015]. Imprecision is normally introduced via the unknown type ?, e.g. function type Int → Bool is more precise than ? → ?, and both more precise than ?. Gradual typing relates types of different precision using consistent type relations, such as type consistency (resp. consistent subtyping), the gradual counterpart of type equality (resp. subtyping). For instance, ? → Int is consistent with Bool → ?. This approach has been applied in a number of settings, such as objects [Jeremy Siek and Walid Taha, 2007], subtyping [Jeremy Siek and Walid Taha, 2007; Ronald Garcia et al., 2016], effects [Bañados Schwerter et al., 2014; Bañados Schwerter et al., 2016], ownership [Ilya Sergey and Dave Clarke, 2012], typestates [Roger Wolff et al., 2011; Ronald Garcia et al., 2014], information-flow typing [Tim Disney and Cormac Flanagan, 2011; Luminous Fennell and Peter Thiemann, 2013; Matías Toro et al., 2018], session types [Igarashi et al., 2017], refinements [Nico Lehmann and {É}ric Tanter, 2017], set-theoretic types [Castagna and Lanvin, 2017], Hoare logic [Johannes Bader et al., 2018], parametric polymorphism [Amal Ahmed et al., 2011; Ahmed et al., 2017; Ina and Igarashi, 2011; Igarashi et al., 2017; Ningning Xie et al., 2018; Matías Toro et al., 2019], and references [Jeremy Siek and Walid Taha, 2006; Herman et al., 2010; Siek et al., 2015]. In particular, gradual typing for mutable references has seen the elaboration of various possible semantics: invariant references [Jeremy Siek and Walid Taha, 2006], guarded references [Herman et al., 2010], monotonic references [Siek et al., 2015], and permissive references [Siek et al., 2015]. Invariant references are a form of references where reference types are invariant with respect to type consistency. Guarded references admit variance thanks to systematic runtime checks on reference reads and writes; the runtime type of an allocated cell never changes during execution. Guarded references have been formulated in a space-efficient coercion calculus, which ensures that gradual programs do not accumulate unbounded pending checks during execution. Hereafter, we refer to this language as HCC. Monotonic references favor efficiency over flexibility by only allowing reference cells to vary towards more precise types. This allows reference operations in statically-typed regions to safely proceed without any runtime checks. Permissive references are the most flexible approach, in which reference cells can be initialized and updated to any value of any type at any time. These four developments reflect different design decisions with respect to gradual references: is the reference type constructor variant under consistency? Can the programmer specify a precise bound on the static type of a reference, and hence on the corresponding heap cell type? Can the heap cell type evolve its precision at runtime, and if yes, how? There is obviously no absolute answer to these questions, as they reflect different tradeoffs such as in efficiency and precision. This work explores the semantics that results from the application of a systematic methodology to gradualize static type systems. Currently we can find in the literature two methodologies to gradualize statically-typed languages: Abstracting Gradual Typing (AGT) [Ronald Garcia et al., 2016], and the Gradualizer [Matteo Cimini and Jeremy Siek, 2016]. In this work, we consider the AGT methodology as it naturally scales to auxiliary structures such as a mutable heap. The AGT methodology helps to systematically construct gradually-typed languages by using abstract interpretation [Cousot and Cousot, 1977] at the type level. In brief, AGT interprets gradual types as an abstraction of sets of possible static types, formally captured through a Galois connection. The static semantics of a gradual language are then derived by lifting the semantics of a statically-typed language through this connection, and the dynamic semantics follow by Curry-Howard from proof normalization of the type safety argument. The AGT methodology has been shown to be effective in many contexts: records and subtyping [Ronald Garcia et al., 2016], type-and-effects [Bañados Schwerter et al., 2014; Bañados Schwerter et al., 2016], refinement types [Nico Lehmann and Éric Tanter, 2017; Niki Vazou et al., 2018], set-theoretic and union types [Castagna and Lanvin, 2017; Matías Toro and Éric Tanter, 2017], information-flow typing [Matías Toro et al., 2018], and parametric polymorphism [Matías Toro et al., 2019]. However, this methodology has never been applied to mutable references in isolation. Although Toro et al. [Matías Toro et al., 2018] apply AGT to a language with references, they only gradualize security levels of types (e.g. Ref Int_?), not whole types (e.g. Ref ? is not supported). In this article we answer the following open questions: Which semantics for gradually-type references follows by systematically applying AGT? Does AGT justify one of the existing approaches, or does it suggest yet another design? Can we recover other semantics for gradual references, if yes, how? This article first reviews the different existing gradual approaches to mutable references through examples. It then presents the semantics for gradual references that is obtained by applying AGT, and how to accommodate the other semantics. More specifically, this work makes the following contributions: - We present λ_REF~, a gradual language with support for mutable references. We derive λ_REF~ by applying the AGT methodology to a fully-static simple language with mutable references called λ_REF. This is the first application of AGT that focuses on gradually-typed mutable references. - We prove that λ_REF~ satisfies the gradual guarantee of Siek et al. [Siek et al., 2015]. We also present the first formal statement and proof of the conservative extension of the dynamic semantics of the static language [Siek et al., 2015], for a gradual language derived using AGT. - We prove that the derived language, λ_REF~, corresponds to the semantics of guarded references from HCC. Formally, given a λ_REF~ term and its compilation to HCC^+ (an adapted version of HCC extended with conditionals and binary operations) we prove that both terms are bisimilar, and that consequently they either both terminate, both fail, or both diverge. - We observe that λ_REF~ and HCC^+ differ in the order of combination of runtime checks. As a result, HCC is space efficient whereas λ_REF~ is not: we can write programs in λ_REF~ that may accumulate an unbounded number of checks. We formalize the changes needed in the dynamic semantics of λ_REF~ to achieve space efficiency. This technique to recover space efficiency is in fact independent from mutable references, and is therefore applicable to other gradual languages derived with AGT. - We formally describe how to support other gradual reference semantics in λ_REF~ by presenting λ_REF~^𝗉𝗆, an extension that additionally supports both permissive and monotonic references. Finally, we prove for the first time that monotonic references satisfy the dynamic gradual guarantee, a non-trivial result that requires careful consideration of updates to the store. Additionally, we implemented λ_REF~ as an interactive prototype that displays both typing derivations and reduction traces. All the examples mentioned in this paper are readily available in the online prototype available at https://pleiad.cl/grefs. As a result, this paper sheds further light on the design space of gradual languages with mutable references and contributes to deepening the understanding of the AGT methodology.