In this work I will present a new feature proposed for the OCaml programming language, “constructor unboxing”, first suggested by Jeremy Yallop in March 2020. It enables a more efficient representation of certain sum types, but requires a static analysis to forbid certain unboxing requests that would be unsound.

To define this static analysis, one has to solve a problem of normalization of first-order definitions in presence of recursion. In the talk I hope to explain my current understanding of this halting problem, and present an algorithm to compute normal forms and reject (in finite time) non-normalizable definitions.

Is λ-calculus a reasonable machine according to Van Emde Boas' Invariance Thesis? By defining the useful evaluation strategy, B. Accattoli and U. Dal Lago were able to answer positively to this question. However, their characterization doesn't facilitate the use of inductive arguments to reason about it. And, up to this date, there are no quantitative models to capture its time complexity.

In this talk I'll discuss the first inductive characterization of the useful evaluation for open call-by-value. We achieve this by parameterizing the evaluation relation with respect to the information provided by the context in which the computation takes place. First we prove some properties of our notion of useful calculus. Then, we give a quantitative model for the useful strategy, using a non-idempotent intersection type system with tight rules. Therefore, typing a term t gives exact information about how many steps does the computation requires to compute the normal form of t.

Closure conversion is a program transformation at work in compilers for functional languages to turn inner functions into global ones, by building 'closures' pairing the transformed functions with the 'environment' of their free variables. Abstract machines rely on similar and yet different concepts of 'closures' and 'environments'.

In this talk, we analyze the relationship between the two approaches. We adopt a very simple lambda-calculus with tuples as source language and study abstract machines for both the source language and the target of closure conversion.

We overview three contributions. Firstly, a new simple proof technique for the correctness of closure conversion, inspired by abstract machines. Secondly, we show how the closure invariants of the target language allow us to design a new way of handling environments in abstract machines, not suffering the shortcomings of other styles. Thirdly, we analyze the machines from the point of view of sharing and time complexity, dissecting how different aspects of closure conversion impact on the cost of the execution.

Cut-reduction procedures for classical sequent calculus are notoriously non-deterministic and non-confluent, both in the original formulation by Gentzen and in later reformulations. It is natural to ask whether those instances of non-confluence are superficial in nature, i.e. whether syntactically distinct normal forms of the same derivation are in fact correlated in a non-trivial way, as is the case in the intuitionistic and linear versions of sequent calculus. A famous counter-example by Lafont purports to show that the answer is negative, that is, every interpretation of derivations in LK that is invariant under classical cut-elimination must be a trivial one that identifies at least all proofs of the same sequent. A long-standing open question has then been whether it could be possible to work around Lafont's example by natural and non-trivial adjustments of the calculus and/or of cut-reduction steps, without resorting to symmetry-breaking techniques like polarization or embeddings into intuitionistic or linear logic.

Working within the propositional fragment of the context-sharing formulation of LK — where parallel logical rules permute freely — we show that the graph constructed by tracing the history of atomic formula occurrences through axiom and cut rules is invariant under arbitrary rule permutations in cut-free proofs, thus providing a canonical representation of normal-form proofs.

We then introduce a refinement of the notion of axiom-induced graph that allows extending the invariance result to proofs with cuts, although at the cost of a strong assumption on the shape of derivations. Because cut-reduction in this formulation of LK can be implemented entirely by logical rule permutations plus a pair of local rewriting steps that preserve the refined axiom graphs, the result yields a non-trivial invariant of cut-reduction.

Our work introduces a quantitative version of the simple type assignment system, starting from a suitable restriction of non-idempotent intersection types. The key idea is to restrict multiset formation to uniform types, two types being uniform if they differ only in the cardinality of the multisets occurring in it. The resulting system has the same typability power as the simple type assignment system; thus, assigning types to terms supplies the very same qualitative information given by simple types, but at the same time provides some interesting quantitative information. It is well known that typability for simple types is equivalent to unification; we prove a similar result for the newly introduced system. More precisely, we show that typability is equivalent to a unification problem which is a non-trivial extension of the classical one: in addition to unification rules, our typing algorithm makes use of an operation that increases the cardinality of multisets whenever needed.