Set-Theoretic Types for Polymorphic Variants

Polymorphic variants are a useful feature of the OCaml language whose current definition and implementation rely on kinding constraints to simulate a subtyping relation via unification. This yields an awkward formalization and results in a type system whose behaviour is in some cases unintuitive and/or unduly restrictive. In this work, we present an alternative formalization of poly-morphic variants, based on set-theoretic types and subtyping, that yields a cleaner and more streamlined system. Our formalization is more expressive than the current one (it types more programs while preserving type safety), it can internalize some meta-theoretic properties, and it removes some pathological cases of the current implementation resulting in a more intuitive and, thus, predictable type system. More generally, this work shows how to add full-fledged union types to functional languages of the ML family that usually rely on the Hindley-Milner type system. As an aside, our system also improves the theory of semantic subtyping, notably by proving completeness for the type reconstruction algorithm.

💡 Research Summary

This paper presents a fundamental redesign of the type system for polymorphic variants in OCaml, replacing its current ad-hoc mechanism with a principled foundation based on set-theoretic types and semantic subtyping.

The core issue addressed is the current implementation’s reliance on “kinding constraints” to simulate subtyping within the Hindley-Milner framework. This leads to an awkward formalization with several practical drawbacks: loss of polymorphism where intuitively none should occur (e.g., a constrained identity function id2 no longer being interchangeable with its argument), imprecise typing of pattern-matching branches due to the inability to leverage information from preceding patterns or the pattern itself, and coarse type approximations for matched expressions when no exact variant type exists.

The proposed solution is a new type system, dubbed “S”, built on semantic subtyping. In this system, types are interpreted as sets of values, and the subtyping relation is defined as set inclusion. The type language is enriched with first-class, unrestricted union (|), intersection (&), and negation (~) connectives. This allows the type system to directly express precise set-theoretic operations during type checking.

A key advantage is the elegant and precise typing of pattern matching. The type for a given branch is computed as (type_of_matched_expression & type_matched_by_pattern) \ (union_of_types_matched_by_prior_patterns). This formula directly captures the exact set of values for which the branch will execute, enabling exhaustiveness and redundancy checks to be performed intrinsically as subtyping checks within the type system itself. The paper formally defines the syntax, dynamic semantics, and the deductive type system “S” for a core ML-like language with polymorphic variants and pattern matching.

The authors demonstrate that “S” is a conservative extension of a formalization “K” of OCaml’s current approach: every program typable in “K” is also typable in “S” (often with a more precise type). Furthermore, “S” is strictly more expressive, typing safe programs that “K” rejects. A significant technical contribution is the definition of a type reconstruction algorithm for “S” and the proof of its completeness (in addition to soundness), a property not established for prior related work.

The paper also discusses important extensions and refinements, such as introducing overloaded function types, refining the typing of pattern variables for more precise branch analysis (solving the map function example), and addressing a runtime compatibility issue with OCaml’s lack of type tags. In conclusion, this work provides a cleaner, more intuitive, and more expressive foundation for polymorphic variants and demonstrates a viable path for integrating full-fledged union and intersection types into implicitly-typed ML-family languages.