Compiler intermediate representations have to strike a balance between being high-level enough to allow for easy translation of surface languages into them and being low-level enough to make code generation easy. An intermediate representation based on a logical system typically has the former property and additionally satisfies several meta-theoretical properties which are valuable when optimizing code. Recently, classical sequent calculus, which is widely recognized and impactful within the fields of logic and proof theory, has been proposed as a natural candidate in this regard, due to its symmetric treatment of data and control flow. For such a language to be useful, however, it must eventually be compiled to machine code. In this paper, we introduce an intermediate representation that is based on classical sequent calculus and demonstrate that this language can be directly translated to conventional hardware. We provide both a formal description and an implementation. Preliminary performance evaluations indicate that our approach is viable.

The sequent calculus is a proof system which was designed as a more symmetric alternative to natural deduction. The λμμ˜-calculus is a term assignment system for the sequent calculus and a great foundation for compiler intermediate languages due to its first-class representation of evaluation contexts. Unfortunately, only experts of the sequent calculus can appreciate its beauty. To remedy this, we present the first introduction to the λμμ˜-calculus which is not directed at type theorists or logicians but at compiler hackers and programming-language enthusiasts. We do this by writing a compiler from a small but interesting surface language to the λμμ˜-calculus as a compiler intermediate language.

Translating programs into continuation-passing style is a well-studied tool to explicitly deal with the control structure of programs. This is useful, for example, for compilation. In a typed setting, there also is a logical interpretation of such a translation as an embedding of classical logic into intuitionistic logic. A naturally arising question is whether there is an inverse translation back to direct style. The answer to this question depends on how the continuation-passing translation is defined and on the domain of the inverse translation. In general, translating programs from continuation-passing style back to direct style requires the use of control operators to account for the use of continuations in non-trivial ways. We present two languages, one in direct style and one in continuation-passing style. Both languages are typed and equipped with an abstract machine semantics. Moreover, both languages allow for non-trivial control flow. We further present a translation to continuation-passing style and a translation back to direct style. We show that both translations are type-preserving and also preserve semantics in a very precise way giving an operational correspondence between the two languages. Moreover, we show that the compositions of the translations are well-behaved. In particular, they are syntactic one-sided inverses on the full language and full syntactic inverses when restricted to trivial control flow.

Effect handlers are a high-level abstraction that enables programmers to use effects in a structured way. They have gained a lot of popularity within academia and subsequently also in industry. However, the abstraction often comes with a significant runtime cost and there has been intensive research recently on how to reduce this price. A promising approach in this regard is to implement effect handlers using a CPS translation and to provide sufficient information about the nesting of handlers. With this information the CPS translation can decide how effects have to be lifted through handlers, i.e., which handlers need to be skipped, in order to handle the effect at the correct place. A structured way to make this information available is to use a calculus with a region system and explicit subregion evidence. Such calculi, however, are quite verbose, which makes them impractical to use as a source-level language. We present a method to infer the lifting information for a calculus underlying a source-level language. This calculus uses second-class capabilities for the safe use of effects. To do so, we define a typed translation to a calculus with regions and evidence and we show that this lift-inference translation is typability- and semantics-preserving. On the one hand, this exposes the precise relation between the second-class property and the structure given by regions. On the other hand, it closes a gap in a compiler pipeline enabling efficient compilation of the source-level language. We have implemented lift inference in this compiler pipeline and conducted benchmarks which indicate that the approach is indeed working.

Effect handlers are a language feature which enjoys popularity in academia and is also gaining traction in industry. Programs use abstract effect operations and handlers provide meaning to them in a delimited scope. Each effect operation is handled by the dynamically closest handler. Using an effect operation outside of a matching handler is meaningless and results in an error. A type-and-effect system prevents such errors from happening. Lexical effect handlers are a recent variant of effect handlers with a number of attractive properties. Just as with traditional effect handlers, programs use effect operations and handlers give meaning to them. But unlike with traditional effect handlers, the connection between effect operations and their handler is lexical. Consequently, they typically have different type-and-effect systems. The semantics of lexical effect handlers as well as their implementations use multi-prompt delimited control. They rely on the generation of fresh labels at runtime, which associate effect operations with their handlers. This use of labels and multi-prompt delimited control is theoretically and practically unsatisfactory. Our main result is that typed lexical effect handlers do not need the full power of multi-prompt delimited control. We present the first CPS translation for lexical effect handlers to pure System F. It preserves well-typedness and simulates the traditional operational semantics. Importantly, it does so without requiring runtime labels. The CPS translation can be used to study the semantics of lexical effect handlers as well as as an implementation technique.