Section: New Results

Logical Foundations of Programming Languages

Participants : Esaie Bauer, Rémi Douence, Marie Kerjean, Ambroise Lafont, Maxime Lucas, Étienne Miquey, Guillaume Munch-Maccagnoni, Nicolas Tabareau.

Classical Logic

Continuation-and-environment-passing style translations: a focus on call-by-need

The call-by-need evaluation strategy for the $\lambda$-calculus is an evaluation strategy that lazily evaluates arguments only if needed, and if so, shares computations across all places where it is needed. To implement this evaluation strategy, abstract machines require some form of global environment. While abstract machines usually lead to a better understanding of the flow of control during the execution, easing in particular the definition of continuation-passing style translations, the case of machines with global environments turns out to be much more subtle. The main purpose of [21] is to understand how to type a continuation-and-environment-passing style translations, that it to say how to soundly translate a classical calculus with environment into a calculus that does not have these features. To this end, we focus on a sequent calculus presentation of a call-by-need λ-calculus with classical control for which Ariola et. al already defined an untyped translation and which we equipped with a system of simple types in a previous paper. We present here a type system for the target language of their translation, which highlights a variant of Kripke forcing related to the environment-passing part of the translation. Finally, we show that our construction naturally handles the cases of call-by-name and call-by-value calculi with environment, encompassing in particular the Milner Abstract Machine, a machine with global environments for the call-by-name $\lambda$-calculus.

Revisiting the duality of computation: an algebraic analysis of classical realizability models

In an impressive series of papers, Krivine showed at the edge of the last decade how classical realizability provides a surprising technique to build models for classical theories. In particular, he proved that classical realizability subsumes Cohen’s forcing, and even more, gives rise to unexpected models of set theories. Pursuing the algebraic analysis of these models that was first undertaken by Streicher, Miquel recently proposed to lay the algebraic foundation of classical realizability and forcing within new structures which he called implicative algebras. These structures are a generalization of Boolean algebras based on an internal law representing the implication. Notably, implicative algebras allow for the adequate interpretation of both programs (i.e. proofs) and their types (i.e. formulas) in the same structure. The very definition of implicative algebras takes position on a presentation of logic through universal quantification and the implication and, computationally, relies on the call-by-name λ-calculus. In [13], we investigate the relevance of this choice, by introducing two similar structures. On the one hand, we define disjunctive algebras, which rely on internal laws for the negation and the disjunction and which we show to be particular cases of implicative algebras. On the other hand, we introduce conjunctive algebras, which rather put the focus on conjunctions and on the call-by-value evaluation strategy. We finally show how disjunctive and conjunctive algebras algebraically reflect the well-known duality of computation between call-by-name and call-by-value.

Models of programming languages mixing effects and resources

Efficient deconstruction with typed pointer reversal

Building on the connection between resource management in systems programming and ordered logic we established previously, we investigate a pervasive issue in the languages C++ and Rust whereby compiler-generated clean-up functions cause a stack overflow on deep structures. In [17], we show how to generate clean-up algorithms that run in constant time and space for a broad class of ordered algebraic datatypes such as ones that can be found in C++ and Rust or in future extensions of functional programming languages with first-class resources.

Resource safety in OCaml

Building on our investigations for a resource-management model for OCaml, we have proposed several preliminary improvements to the OCaml language. We contributed to the design and implementation of new resource management primitives (PRs #2118, #8962), resource-safe C APIs (PRs #8993, #8997, #9037), and core runtime capabilities (PR #8961). (#2118 has been merged into OCaml 4.08 and #8993 and #9037 have been merged into OCaml 4.10.)

We continued to interact with L. White and S. Dolan (Jane Street), on the design of resource management and exception safety in multicore OCaml.

Syntax and Rewriting Systems

Reduction Monads and Their Signatures

In [1], we study reduction monads, which are essentially the same as monads relative to the free functor from sets into multigraphs. Reduction monads account for two aspects of the lambda calculus: on the one hand, in the monadic viewpoint, the lambda calculus is an object equipped with a well-behaved substitution; on the other hand, in the graphical viewpoint, it is an oriented multigraph whose vertices are terms and whose edges witness the reductions between two terms. We study presentations of reduction monads. To this end, we propose a notion of reduction signature. As usual, such a signature plays the role of a virtual presentation, and specifies arities for generating operations-possibly subject to equations-together with arities for generating reduction rules. For each such signature, we define a category of models; any model is, in particular, a reduction monad. If the initial object of this category of models exists, we call it the reduction monad presented (or specified) by the given reduction signature. Our main result identifies a class of reduction signatures which specify a reduction monad in the above sense. We show in the examples that our approach covers several standard variants of the lambda calculus.

Modules over monads and operational semantics

[22] is a contribution to the search for efficient and high-level mathematical tools to specify and reason about (abstract) programming languages or calculi. Generalising the reduction monads of Ahrens et al., we introduce operational monads, thus covering new applications such as the-calculus, Positive GSOS specifications, and the big-step, simply-typed, call-by-value-calculus. Finally, we design a notion of signature for operational monads that covers all our examples.

Modular specification of monads through higher-order presentations

In their work on second-order equational logic, Fiore and Hur have studied presentations of simply typed languages by generating binding constructions and equations among them. To each pair consisting of a binding signature and a set of equations, they associate a category of `models', and they give a monadicity result which implies that this category has an initial object, which is the language presented by the pair. In [10], we propose, for the untyped setting, a variant of their approach where monads and modules over them are the central notions. More precisely, we study, for monads over sets, presentations by generating (`higher-order') operations and equations among them. We consider a notion of 2-signature which allows to specify a monad with a family of binding operations subject to a family of equations, as is the case for the paradigmatic example of the lambda calculus, specified by its two standard constructions (application and abstraction) subject to $\beta$- and $\eta$-equalities. Such a 2-signature is hence a pair ($\Sigma$,E) of a binding signature $\Sigma$ and a family E of equations for $\Sigma$. This notion of 2-signature has been introduced earlier by Ahrens in a slightly different context. We associate, to each 2-signature ($\Sigma$,E), a category of `models of ($\Sigma$,E); and we say that a 2-signature is `effective' if this category has an initial object; the monad underlying this (essentially unique) object is the `monad specified by the 2-signature'. Not every 2-signature is effective; we identify a class of 2-signatures, which we call `algebraic', that are effective. Importantly, our 2-signatures together with their models enjoy `modularity': when we glue (algebraic) 2-signatures together, their initial models are glued accordingly. We provide a computer formalization for our main results.

The Diamond Lemma for non-terminating rewriting systems

In [16], we study the confluence property for rewriting systems whose underlying set of terms admits a vector space structure. For that, we use deterministic reduction strategies. These strategies are based on the choice of standard reductions applied to basis elements. We provide a sufficient condition of confluence in terms of the kernel of the operator which computes standard normal forms. We present a local criterion which enables us to check the confluence property in this framework. We show how this criterion is related to the Diamond Lemma for terminating rewriting systems

Differential Linear Logic

Higher-order distributions for differential linear logic

Linear Logic was introduced as the computational counterpart of the algebraic notion of linearity. Differential Linear Logic refines Linear Logic with a proof-theoretical interpretation of the geometrical process of differentiation. In [24], we construct a polarized model of Differential Linear Logic satisfying computational constraints such as an interpretation for higher-order functions, as well as constraints inherited from physics such as a continuous interpretation for spaces. This extends what was done previously by Kerjean for first order Differential Linear Logic without promotion. Concretely, we follow the previous idea of interpreting the exponential of Differential Linear Logic as a space of higher-order distributions with compact-support, and is constructed as an inductive limit of spaces of distributions on Euclidean spaces. We prove that this exponential is endowed with a co-monadic like structure, with the notable exception that it is functorial only on isomorphisms. Interestingly, as previously argued by Ehrhard, this still allows one to interpret differential linear logic without promotion.

Chiralities in topological vector spaces

Chiralities are categories introduced by Mellies to account for a game semantics point of view on negation. In [23], [20], we uncover instances of this structure in the theory of topological vector spaces, thus constructing several new polarized models of Multiplicative Linear Logic. These models improve previously known smooth models of Differential Linear Logic, showing the relevance of chiralities to express topological properties of vector spaces. They are the first denotational polarized models of Multiplicative Linear Logic, based on the pre-existing theory of topological vector spaces, in which two distinct sets of formulas, two distinct negations, and two shifts appear naturally.

Distributed Programming

Chemical foundations of distributed aspects.

Distributed applications are challenging to program because they have to deal with a plethora of concerns, including synchronisation, locality, replication, security and fault tolerance. Aspect-oriented programming (AOP) is a paradigm that promotes better modularity by providing means to encapsulate cross-cutting concerns in entities called aspects. Over the last years, a number of distributed aspect-oriented programming languages and systems have been proposed, illustrating the benefits of AOP in a distributed setting. Chemical calculi are particularly well-suited to formally specify the behaviour of concurrent and distributed systems. The join calculus is a functional name-passing calculus, with both distributed and object-oriented extensions. It is used as the basis of concurrency and distribution features in several mainstream languages like C# (Polyphonic C#, now C$\omega$), OCaml (JoCaml), and Scala Joins. Unsurprisingly, practical programming in the join calculus also suffers from modularity issues when dealing with crosscutting concerns. We propose the Aspect Join Calculus [9], an aspect-oriented and distributed variant of the join calculus that addresses crosscutting and provides a formal foundation for distributed AOP. We develop a minimal aspect join calculus that allows aspects to advise chemical reactions. We show how to deal with causal relations in pointcuts and how to support advanced customisable aspect weaving semantics.