The projection-based strictness analysis of Wadler and Hughes is elegant and theoretically satisfying except in one respect: the need for lifting. The domains and functions over which the analysis is performed need to be transformed, leading to a less direct correspondence between analysis and program than might be hoped for. In this paper we shall see that the projection analysis may be reformulated in terms of partial projections, so removing this infelicity. There are additional benefits of the formulation: the two forms of information captured by the projection are distinguished, and the operational significance of the range of the projection fits exactly with the theory of unboxed types.

A number of difficulties in the formalism of Pure Type Systems (PTS) is discussed and an alternative classification system for typed calculi is proposed. In the new approach the main novelty is that one first explicitly specifies the dependencies that may occur. This is especially useful to describe constants, but it also facilitates the description of other type theoretic features like dependent sums.

A generic functional program is one which is parameterised by datatype. By installing specific choices, for example lists or trees, different programs are obtained that are, nevertheless, abstractly the same. The purpose of this paper is to explore the possibility of deriving generic programs. Part of the theory of lists that deals with segments is recast as a theory about ‘segments’ in a wide class of datatypes, and then used to pose and solve a generic version of a well-known problem.

We develop an operational model for a language based on linear logic. Our semantics is ‘low-level’ enough to express sharing and copying while still being ‘high-level’ enough to abstract away from details of memory layout, and thus can be used to test potential applications of linear logic for analysis of programs. In particular, we demonstrate a precise relationship between type correctness for the linear-logic-based language and the correctness of a reference-counting interpretation of the primitives, and formulate and prove a result describing the possible run-time reference counts of values of linear type.

This paper presents a purely syntactic account of type generativity and sharing – two key mechanisms in the SML module system – and shows its equivalence with the traditional stamp-based description of these mechanisms. This syntactic description recasts the SML module system in a more abstract, type-theoretic framework.

The Bologna Optimal Higher-order Machine (BOHM) is a prototype implementation of the core of a functional language based on (a variant of) Lamping's optimal graph reduction technique (Lamping, 1990; Gonthier et al., 1992a; Asperti, 1994). The source language is a sugared λ-calculus enriched with booleans, integers, lists and basic operations on these data types (following the guidelines of Interaction Systems – Asperti and Laneve (1993b, 1994), Laneve (1993)). In this paper, we shall describe BOHM's general architecture (comprising the garbage collector), and give a large set of benchmarks and experimental results.

Explicit substitutions were proposed by Abadi, Cardelli, Curien, Hardin and Lévy to internalise substitutions into λ-calculus and to propose a mechanism for computing on substitutions. λν is another view of the same concept which aims to explain the process of substitution and to decompose it in small steps. It favours simplicity and preservation of strong normalisation. This way, another important property is missed, namely confluence on open terms. In spirit, λν is closely related to another calculus of explicit substitutions proposed by de Bruijn and called CλξΦ. In this paper, we introduce λν, we present CλξΦ in the same framework as λν and we compare both calculi. Moreover, we prove properties of λν; namely λν correctly implements β reduction, λν is confluent on closed terms, i.e. on terms of classical λ-calculus and on all terms that are derived from those terms, and finally λν preserves strong normalisation in the following sense: strongly β normalising terms are strongly λν normalising.

We study the operational semantics of an extension of Girard's System Fω with two control operators: an abort operation that abandons the current control context, and a callcc operation that captures the current control context. Two classes of operational semantics are considered, each with a call-by-value and a call-by-name variant, differing in their treatment of polymorphic abstraction and instantiation. Under the standard semantics, polymorphic abstractions are values and polymorphic instantiation is a significant computation step; under the ML-like semantics evaluation proceeds beneath polymorphic abstractions and polymorphic instantiation is computationally insignificant. Compositional, type-preserving continuation-passing style (cps) transformation algorithms are given for the standard semantics, resulting in terms on which all four evaluation strategies coincide. This has as a corollary the soundness and termination of well-typed programs under the standard evaluation strategies. In contrast, such results are obtained for the call-by-value ML-like strategy only for a restricted sub-language in which constructor abstractions are limited to values. The ML-like call-by-name semantics is indistinguishable from the standard call-by-name semantics when attention is limited to complete programs.

We introduce a positive supercompiler, a version of Turchin's supercompiler maintaining only positive information during transformation, and using folding without generalization. The positive supercompiler can also be regarded as a variant of Wadler's deforestation maintaining an increased amount of information. We compare our algorithm to deforestation and, in less detail, to partial evaluation, Turchin's supercompiler, Generalized Partial Computation (GPC), and partial deduction by classifying these transformers by the amount of information they maintain during transformation. This factor is significant, as a differentiating example reveals: positive supercompilation, Turchin's supercompiler, GPC and partial deduction can specialize a general pattern matcher with respect to a fixed pattern to obtain an efficient matcher very similar to the Knuth–Morris–Pratt algorithm. Deforestation and traditional partial evaluation achieve this effect only after a non-trivial hand rewriting of the general matcher.

First-generation heap profilers for lazy functional languages have proved to be effective tools for locating some kinds of space faults, but in other cases they cannot provide sufficient information to solve the problem. This paper describes the design, implementation and use of a new profiler that goes beyond the two-dimensional ‘who produces what’ view of heap cells to provide information about their more dynamic and structural attributes. Specifically, the new profiler can distinguish between cells according to their eventual lifetime, or on the basis of the closure retainers by virtue of which they remain part of the live heap. A bootstrapping Haskell compiler (nhc) hosts the implementation: among examples of the profiler's use we include self-application to nhc. Another example is the original heap-profiling case study clausify, which now consumes even less memory and is much faster.

In this article, we extend the Barendregt Cube with ∏-conversion (which is the analogue of β-conversion, on product type level) and study its properties. We use this extension to separate the problem of whether a term is typable from the problem of what is the type of a term.

The resource constrained shortest path problem is an NP-hard problem for which many ingenious algorithms have been developed. These algorithms are usually implemented in Fortran or another imperative programming language. We have implemented some of the simpler algorithms in a lazy functional language. Benefits accrue in the software engineering of the implementations. Our implementations have been applied to a standard benchmark of data files, which is available from the Operational Research Library of Imperial College, London. The performance of the lazy functional implementations, even with the comparatively simple algorithms that we have used, is competitive with a reference Fortran implementation.

The paper explores the application of a lazy functional language, Haskell, to a series of grid-based scientific problems—solution of the Poisson equation, and Monte Carlo simulation of two theoretical models from statistical and particle physics. The implementations introduce certain abstractions of grid topology, making extensive use of the polymorphic features of Haskell. Updating is expressed naturally through use of infinite lists, exploiting the laziness of the language. Evolution of systems is represented by arrays of interacting streams.

Over 25 implementations of different functional languages are benchmarked using the same program, a floating-point intensive application taken from molecular biology. The principal aspects studied are compile time and execution time for the various implementations that were benchmarked. An important consideration is how the program can be modified and tuned to obtain maximal performance on each language implementation. With few exceptions, the compilers take a significant amount of time to compile this program, though most compilers were faster than the then current GNU C compiler (GCC version 2.5.8). Compilers that generate C or Lisp are often slower than those that generate native code directly: the cost of compiling the intermediate form is normally a large fraction of the total compilation time. There is no clear distinction between the runtime performance of eager and lazy implementations when appropriate annotations are used: lazy implementations have clearly come of age when it comes to implementing largely strict applications, such as the Pseudoknot program. The speed of C can be approached by some implementations, but to achieve this performance, special measures such as strictness annotations are required by non-strict implementations. The benchmark results have to be interpreted with care. Firstly, a benchmark based on a single program cannot cover a wide spectrum of ‘typical’ applications. Secondly, the compilers vary in the kind and level of optimisations offered, so the effort required to obtain an optimal version of the program is similarly varied.

We present a comprehensive analysis of all the components of creation, access and disposal of heap-allocated and stack-allocated activation records. Among our results are:

•Although stack frames are known to have a better cache read-miss rate than heap frames, our simple analytical model (backed up by simulation results) shows that the difference is too trivial to matter.

•The cache write-miss rate of heap frames is very high; we show that a variety of miss-handling strategies (exemplified by specific modern machines) can give good performance, but not all can.

•Stacks restrict the flexibility of closure representations (for higher-order functions) in important (and costly) ways.

•The extra load placed on the garbage collector by heap-allocated frames is small.

•The demands of modern programming languages make stacks complicated to implement efficiently and correctly.

Overall, the execution cost of stack-allocated and heap-allocated frames is similar; but heap frames are simpler to implement and allow very efficient first-class continuations.

Brodal recently introduced the first implementation of imperative priority queues to support findMin, insert and meld in O(1) worst-case time, and deleteMin in O(log n) worst-case time. These bounds are asymptotically optimal among all comparison-based priority queues. In this paper, we adapt Brodal's data structure to a purely functional setting. In doing so, we both simplify the data structure and clarify its relationship to the binomial queues of Vuillemin, which support all four operations in O(log n) time. Specifically, we derive our implementation from binomial queues in three steps: first, we reduce the running time of insert to O(1) by eliminating the possibility of cascading links; second, we reduce the running time of findMin to O(1) by adding a global root to hold the minimum element; and finally, we reduce the running time of meld to O(1) by allowing priority queues to contain other priority queues. Each of these steps is expressed using ML-style functors. The last transformation, known as data-structural bootstrapping, is an interesting application of higher-order functors and recursive structures.

This paper describes a compiling graph reduction system which realizes the reduction semantics of a fully-fledged applied λ-calculus. High-level functional programs are conceptually executed as sequences of program transformations governed by full β-reductions. They may be carried out step-by-step, and intermediate programs may be displayed in high-level notation, rendering the system suitable for interactive program design, high-level debugging, and also for teaching basic programming language concepts and language interpretation. Run-time efficiency for production runs is achieved by means of an abstract stack machine ASM which serves as an intermediate level of code generation. It employs multiple stacks for reasonably fast function calls, optimized tail-end recursions, and earliest possible releases of subgraphs that are no longer needed. The ASM involves an interpreter if and only if potential naming conflicts need to be resolved when reducing partial function applications.

Compiler generation based on Mosses' action semantics has been studied by Brown, Moura, and Watt, and also by the second author. The core of each of their systems is a handwritten action compiler, producing either C or machine code. We have obtained an action compiler in a much simpler way: by partial evaluation of an action interpreter. Even though our compiler produces Scheme code, the code runs as fast as that produced by the previous action compilers.

The Third Homomorphism Theorem is a folk theorem of the constructive algorithmics community. It states that a function on lists that can be computed both from left to right and from right to left is necessarily a list homomorphism – it can be computed according to any parenthesization of the list. We formalize and prove the theorem, and use it to improve an O(n2) sorting algorithm to O(n log n).

This paper describes a scheme for constructing parsers based on the top-down combinator approach. In particular, it describes a set of combinators for parsing expressions described by ambiguous grammars with precedence and associativity rules. The new combinators embody the mechanical grammar manipulations typically employed to remove left-recursion and hence help to avoid the possibility of a non-terminating parser. A number of approaches to the problem are described—the most elegant and efficient method is based on continuation passing. As a practical demonstration, a parser for the expression part of the C programming language is presented. The expression combinators are general, and may be constructed from any suitable set of top-down combinators. A comparison with parser generators shows that the combinator approach is most applicable for rapid development.