Publications by Dr Olaf Chitil

Observing intermediate values helps to understand what is going on when your program runs. Gill presented an observation method for lazy functional languages that preserves the program's semantics. However, users need to define for each type how its values are observed: a laborious task and strictness of the program can easily be affected. Here we define how any value can be observed based on the structure of its type by applying generic programming. Furthermore we present an extension to specify per observation point how much to observe of a value.

We present two implementations of Oppen's pretty-printing algorithm in Haskell that meet the efficiency of Oppen's imperative solution but have a simpler, clear structure. We start with an implementation that uses lazy evaluation to simulate two co-operating processes. Then we present an implementation that uses higher-order functions for delimited continuations to simulate co-routines with explicit scheduling.

The tracer Hat records in a detailed trace the computation of a program written in the lazy functional language Haskell. The trace can then be viewed in various ways to support program comprehension and debugging. The trace was named the augmented redex trail. Its structure was inspired by standard graph rewriting implementations of functional languages. Here we describe a model of the trace that captures its essential properties and allows formal reasoning. The trace is a graph constructed by graph rewriting but goes beyond simple term graphs. Although the trace is a graph whose structure is independent of any rewriting strategy, we define the trace inductively, thus giving us a powerful method for proving its properties.

There are several purely functional libraries for converting tree structured data into indented text, but they all make use of some backtracking. Over twenty years ago, Oppen published a more efficient imperative implementation of a pretty printer. This article shows that the same efficiency is also obtainable without destructive updates by developing a similar but purely functional Haskell implementation with the same complexity bounds. At its heart lie two lazy double ended queues.

Deforestation optimises a functional program by transforming it into another one that does not create certain intermediate data structures. Short cut deforestation is a deforestation method which is based on a single, local transformation rule. In return, short cut deforestation expects both producer and consumer of the intermediate structure in a certain form. Warm fusion was proposed to automatically transform functions into this form. Unfortunately, it is costly and hard to implement. Starting from the fact that short cut deforestation is based on a parametricity theorem of the second-order typed lambda-calculus, we show how the required form of a list producer can be derived through the use of type inference. Typability for the second-order typed lambda-calculus is undecidable. However, we present a linear-time algorithm that solves a partial type inference problem and that, together with controlled inlining and polymorphic type instantiation, suffices for deforestation. The resulting new short cut deforestation algorithm is efficient and removes more intermediate lists than the original.

A comprehensive semantics for functional programs is presented, which generalizes the well-known call-by-value and call-by-name semantics. By permitting a separate choice between call-by value and call-by-name for every argument position of every function and parameterizing the semantics by this choice we abstract from the parameter-passing mechanism. Thus common and distinguishing features of all instances of the sigma-semantics, especially call-by-value and call-by-name semantics, are highlighted. Furthermore, a property can be validated for all instances of the sigma-semantics by a single proof. This is employed for proving the equivalence of the given denotational (fixed-point based) and two operational (reduction based) definitions of the sigma-semantics. We present and apply means for very simple proofs of equivalence with the denotational sigma-semantics for a large class of reduction-based sigma-semantics. Our basis are simple first-order constructor-based functional programs with patterns.

Type error messages of compilers of statically typed functional languages are often inaccurate, making type error debugging hard. Many solutions to the problem have been proposed, but most have been evaluated only with short programs, that is, of fewer than 30 lines. In this paper we note that our own tool for delta debugging type errors scales poorly for large programs. In response we present a new tool that applies a new algorithm for segmenting a large program before the delta debugging algorithm is applied. We propose a framework for quantifying the quality of type error debuggers and apply it to our new tool demonstrating substantial improvement.

Existing methods for generating a detailed trace of a computation of a lazy functional program are complex. These complications limit the use of tracing in practice. However, such a detailed trace is desirable for understanding and debugging a lazy functional program. Here we present a lightweight method that instruments a program to generate such a trace, namely the augmented redex trail introduced by the Haskell tracer Hat. The new method is a major step towards an omniscient debugger for real-world Haskell programs.

A computation tree of a program execution describes computations of functions and their dependencies. A computation tree describes how a program works and is at the heart of algorithmic debugging. To generate a computation tree, existing algorithmic debuggers either use a complex implementation or yield a less informative approximation. We present a method for lazy functional languages that requires only a simple tracing library to generate a detailed computation tree. With our algorithmic debugger a programmer can debug any Haskell program by only importing our library and annotating suspected functions.

Existing algorithmic debuggers for Haskell require a transformation of all modules in a program, even libraries that the user does not want to debug and which may use language features not supported by the debugger. This is a pity, because a promising ap- proach to debugging is therefore not applicable to many real-world programs. We use the cost centre stack from the Glasgow Haskell Compiler profiling environment together with runtime value observations as provided by the Haskell Object Observation Debugger (HOOD) to collect enough information for algorithmic debugging. Program annotations are in suspected modules only. With this technique algorithmic debugging is applicable to a much larger set of Haskell programs. This demonstrates that for functional languages in general a simple stack trace extension is useful to support tasks such as profiling and debugging.

Until now there has been no support for specifying and enforcing contracts within a lazy functional program. That is a shame, because contracts consist of pre- and post-conditions for functions that go beyond the standard static types. This paper presents the design and implementation of a small, easy-to-use, purely functional contract library for Haskell, which, when a contract is violated, also provides more useful information than the classical blaming of one contract partner. From now on lazy functional languages can profit from the assurances in the development of correct programs that contracts provide.

Lazy functional programming languages need lazy assertions to ensure that assertions preserve the meaning of programs. Examples in this paper demonstrate that previously proposed lazy assertions nonetheless break basic semantic equivalences, because they include a non-deterministic disjunction combinator. The objective of this paper is to determine ''correct'' definitions for lazy assertions. The starting point is our formalisation of basic properties such as laziness, taking them as axioms of our design space. We develop the first denotational semantics for lazy assertions; assertions denote subdomains. We define a weak disjunction combinator and together with a conjunction combinator assertions form a bounded distributive lattice. From the established laws we derive an efficient prototype implementation of lazy assertions for Haskell as a library.

Functional programming is a programming paradigm like object-oriented programming and logic programming. Functional programming comprises both a specific programming style and a class of programming languages that encourage and support this programming style. Functional programming enables the programmer to describe an algorithm on a high-level, in terms of the problem domain, without having to deal with machine-related details. A program is constructed from functions that only map inputs to outputs, without any other effect on the program state. Thus a function will always return the same output, regardless of when and in which context the function is used. These functions provide clear interfaces, separate concerns and are easy to reuse. A small and simple set of highly orthogonal language constructs assists in writing modular programs.

Algorithmic debuggers for higher-order functional languages have to display functional values. Originally functional values had been represented as partial applications of function and constructor symbols, but a recent approach represents functional values as finite maps. The two representations require the computation tree that is central to algorithmic debugging to be structured rather differently. In this paper we present a unifying framework that formally defines algorithmic debugging for both representations in an implementation-independent way. On this basis we prove the soundness of algorithmic debugging with finite maps. Our framework shows how a single implementation can support both forms of algorithmic debugging. The proof exposed that algorithmic debugging with finite maps does not handle arbitrary functional programs, but in current practice the problematic ones are excluded by Haskell's type system. Both framework and proof suggest variations of algorithmic debugging with finite maps and thus are tools for further improvement of this form of debugging.

Assertions test expected properties of run-time values without disrupting the normal computation of a program. We present a library for enriching Haskell programs with assertions. Expected properties can be specified in a parser-combinator like language. The assertions are lazy: they do not force evaluation but only examine what is evaluated by the program. They are also prompt: assertion failure is reported as early as possible. The implementation is based on lazy observations and continuation-based coroutines.

This paper formally presents a model of tracing for functional programs based on a small-step operational semantics. The model records the computation of a functional program in a graph which can be utilised for various purposes such as algorithmic debugging. The main contribution of this paper is to prove the correctness of algorithmic debugging for functional programs based on the model. Although algorithmic debugging for functional programs is implemented in several tracers such as Hat, the correctness has not been formally proved before. The difficulty of the proof is to find a suitable induction principle and a more general induction hypothesis.

Assertions test expected properties of run-time values without disrupting the normal computation of a program. Here we present a library for enriching programs in the lazy language Haskell with assertions. Expected properties are written in an expressive pattern logic that combines pattern matching with logical operations and predicates. The presented assertions are lazy: they do not force evaluation but only examine what is evaluated by other parts of the program. They are also prompt: assertion failure is reported as early as possible, before a faulty value is used by the main computation.

Currently, program slicing and algorithmic debugging are two of the most relevant debugging techniques for declarative languages. They help programmers to find bugs in a semiautomatic manner. On the one hand, program slicing is a technique to extract those program fragments that (potentially) affect the values computed at some point of interest. On the other hand, algorithmic debugging is able to locate a bug by automatically generating a series of questions and processing the programmer's answers. In this work, we show for functional languages how the combination of both techniques produces a more powerful debugging schema that reduces the number of questions that programmers must answer to locate a bug.

Tracing a computation is a key method for program comprehension and debugging. Hat is a tracing system for Haskell 98 programs. During a computation a trace is recorded in a file; then the user studies the trace with a collection of viewing tools. Different views are complementary and can productively be used together. Experience shows that users of the viewing tools find it hard to keep orientation and navigate to a point of interest in the trace. Hence this paper describes a new viewing tool where navigation through the trace is based on the program source. The tool combines ideas from algorithmic debugging, traditional stepping debuggers and dynamic program slicing.

Assertions test expected properties of run-time values without disrupting the normal working of a program. So in a lazy functional language assertions should be lazy - not forcing evaluation, but only examining what is evaluated by other parts of the program. We explore the subtle semantics of lazy assertions and describe sequential and concurrent variants of a method for checking lazy assertions. All variants are implemented in Haskell.

A lightweight approach to debugging functional logic programs by observations is presented, implemented for the language Curry. The Curry Object Observation System (COOSy) comprises a portable library plus a viewing tool. A programmer can observe data structures and functions by annotating expressions in his program. The possibly partial values of observed expressions that are computed during program execution are recorded in a trace file, including information on non-deterministic choices and logical variables. A separate viewing tool displays the trace content. COOSy covers all aspects of modern functional logic multiparadigm languages such as lazy evaluation, higher order functions, non-deterministic search, logical variables, concurrency and constraints. Both use and implementation of COOSy are described.

It is a very undesirable situation that today's software often contains errors. One motivation for using a functional programming language is that it is more difficult (or even impossible) to make low-level mistakes, and it is easier to reason about programs. But even the most advanced functional programmers are not infallible; they misunderstand the properties of their own programs, or those of others, and so commit errors.

The type systems of most typed functional programming languages are based on the Hindley-Milner type system. A practical problem with these type systems is that it is often hard to understand why a program is not type correct or a function does not have the intended type. We suggest that at the core of this problem is the difficulty of explaining why a given expression has a certain type. The type system is not defined compositionally. We propose to explain types using a variant of the Hindley-Milner type system that defines a compositional type explanation graph of principal typings. We describe how the programmer understands types by interactive navigation through the explanation graph. Furthermore, the explanation graph can be the foundation for algorithmic debugging of type errors, that is, semi-automatic localisation of the source of a type error without even having to understand the type inference steps. We implemented a prototype of a tool to explore the usefulness of the proposed methods.

In this paper we compare three systems for tracing and debugging Haskell programs: Freja, Hat and Hood. We evaluate their usefulness in practice by applying them to a number of moderately complex programs in which errors had deliberately been introduced. We identify the strengths and weaknesses of each system and then form ideas on how the systems can be improved further.

In a non-strict functional programming language functions that yield the same result for all total arguments can still differ for partial arguments, that is, they differ in their strictness. Here a tool for Haskell is presented that enables the programmer to easily check whether a given function is least-strict; if it is not least-strict, then the tool suggests how to make it less strict.

We have proved the correctness of algorithmic debugging if the traces are acyclic. For cyclic traces, however, does algorithmic debugging still work? There does not exist a common understanding of how to debug cyclic traces in functional programming communities for a long time. In this paper we give two small examples to demonstrate that it is extremely difficult to find a generic algorithmic debugging scheme for cyclic traces. We conjecturre that it is impossible to have a generic scheme for cyclic traces, because the examples are very small and the choices of reasonable debugging trees are very limited. We also present acyclic traces in which constants are shared unless shared constants result in a cycle. The normal algorithmic debugging scheme works fine for acyclic traces and the proof is very similar to our previous paper.

The purpose of the document is to prove the correctness of Algorithmic Debugging where the traces for local functions are generated in a new way. The processes of generating computation graphs follow exactly what we might do by hand. Therefore, we can be confident that the graphs are correct.We do not need to justify the graphs by comparing ?-lifted programs.

As the name states, trusted functions do not have bugs. It is up to user to specify which function is trusted. Commonly used functions in standard library are normally trusted. In the process of algorithmic debugging, we search for faulty nodes to locate bugs. Since a trusted function cannot be a faulty node, there is no point to keep trusted func- tions in Evaluation Dependency Trees (EDT) for algorithmic debugging. In this report, we create smaller tree structures by removing trusted func- tions. There are two different ways to achieve this: generating a smaller tree structure directly from the original trace; or creating a smaller trace first and then from which generating a smaller tree structure.

In non-strict functional programming languages such as Haskell, it happens often that some parts of a program are not evaluated because their values are not demanded. In practice, those unevaluated parts are often replaced by a placeholder (e.g. ₎ in order to keep the trace size smaller. In the process of algorithmic debugging, one needs to answer several questions in order to locate a program fault. Replacing unevaluated parts makes these questions shorter and semantically clearer. In this paper, we present a formal model of tracing in which unevaluated parts are replaced by the symbol _. The most important property, the correctness of algorithmic debugging, is proved.

Pretty printing is the task of nicely formatting tree structured data within a given line width limit. In 1980 Oppen published a pretty printing algorithm that takes time linear in the size of the input, independent of the line width, and uses only limited look-ahead. This work inspired the development of a number of purely functional pretty printing libraries in Haskell. Here I present a new functional pretty printing algorithm that has all the nice properties of Oppen's and is surprisingly simple. A double-ended queue of delimited continuations is the key to addressing all aspects of the problem explicitly.

The Glasgow Haskell compiler (GHC) with its over 40.000 lines of code is quite daunting for a newcomer. Here we give a short practical introduction based on our experiences in how to add an optimisation pass to GHC. Thus we hope to encourage other developers of optimisations to implement them in GHC. These notes are meant to be extended and updated from time to time. Hence observe the date shown below. Unpublished, November 4, 1997, 16 pages.

A comprehensive semantics for functional programs is presented, which generalizes the well-known call-by-value and call-by-name semantics. By permitting a separate choice between call-by value and call-by-name for every argument position of every function and parameterizing the semantics by this choice we abstract from the parameter-passing mechanism. Thus common and distinguishing features of all instances of the sigma-semantics, especially call-by-value and call-by-name semantics, are highlighted. Furthermore, a property can be validated for all instances of the sigma-semantics by a single proof. This is employed for proving the equivalence of the given denotational (fixed-point based) and two operational (reduction based) definitions of the sigma-semantics. We present and apply means for very simple proofs of equivalence with the denotational sigma-semantics for a large class of reduction-based sigma-semantics. Our basis are simple first-order constructor-based functional programs with patterns.

The cause of a type error can be very difficult to find for the Hindley-Milner type system. Consequently many solutions have been proposed, but they are hardly used in practice. Here we propose a new solution that provides counter-factual type error messages; these messages state what types specific subexpressions in a program should have (in contrast to the types they actually have) to remove a type error. Such messages are easy-to-understand, because programers are already familiar with them. Furthermore, our solution is easy-to-implement, because it reuses an existing type checker as a subroutine. We transform an ill-typed program into a well-typed program with additional λ-bound variables. The types of these λ-bound variables yield actual and counter-factual type information. That type information plus intended types added as type annotations direct the search of the type debugger.

Debugging type errors is a necessary process that programmers, both novices and experts alike, face when using statically typed functional programming languages. All compilers often report the location of a type error inaccurately. This problem has been a subject of research for over thirty years. We present a new method for locating type errors: We apply the Isolating Delta Debugging algorithm coupled with a blackbox compiler. We evaluate our implementation for Haskell by comparing it with the output of the Glasgow Haskell Compiler; overall we obtain positive results in favour of our method of type error debugging.

Many different approaches to type error debugging were developed independently. In this paper, we describe a new common framework for several type error debugging approaches. For this purpose, we introduce expected types from the outer context and propose a method for obtaining them. Using expected types, we develop three type error debugging approaches: enumeration of type error messages, type error slicing and (improved) interactive type error debugging. Based on our idea we implemented prototypes and confirm that the framework works well for type debugging.

The Hindley-Milner type system is a foundation for most statically typed functional programming languages, such as ML, OCaml and Haskell. This type system has many advantageous, but it does make type debugging hard: If a program is not well-typed, it can be difficult for the programmer to locate the cause of the type error, that is, to determine where to change the program how. Many solutions to the problem have been proposed. Here we propose a new solution with two distinctive advantages: It is easy to use for the functional programmer, because it appears to be only a minor extension of the type error messages they are already familiar with. It is easy to implement, because it does not require the implementation of a new type checker, but instead reuses any existing one as a subroutine (like [2]).

The type error message of a standard type checker for a functional language gives only a single location as potential cause of the type error. If that location is not the cause, which often is the case, then the type error message hardly helps in locating the real cause. Here we present a method that uses a standard type checker to enumerate locations that potentially cause the type error, each with an actual and a counter-factual type for the given location. Adding our method to existing compilers requires only limited effort but improves type error debugging substantially.

Observing intermediate values helps to understand what is going on when your program runs. Gill presented an observation method for lazy functional languages that preserves the program's semantics. However, users need to define for each type how its values are observed: a laborious task and strictness of the program can easily be affected. Here we define how any value can be observed based on the structure of its type by applying generic programming frameworks. Furthermore we present an extension to specify per observation point how much to observe of a value. We discuss especially functional values and behaviour based on class membership in generic programming frameworks.

Using a separate editor and interpreter provides many distracting obstacles for inexperienced students learning a programming language. Professional interactive development environments, however, confuse these students with their excessive features. Hence this paper presents Heat, an interactive development environment specially designed for novice students learning the functional programming language Haskell. Based on teaching experience, Heat provides a small number of features and is easy to use. Heat proves that a small portable interactive development environment can be implemented on top of but independent of a particular Haskell interpreter. Heat with Hugs has been used in teaching functional programming at the University of Kent for the past three years.

Assertions test expected properties of run-time values without disrupting the normal computation of a program. Here we present a library for enriching programs in the lazy language Haskell with assertions. Expected properties are written in an expressive pattern logic that combines pattern matching with logical operations and predicates. The presented assertions are lazy: they do not force evaluation but only examine what is evaluated by other parts of the program. They are also prompt: assertion failure is reported as early as possible, before a faulty value is used by the main computation.

Functional values are used naturally in higher order functional programs, as they are commonly passed around or returned by other function. As such any debugger for these languages must be capable of conveying information about functional values to the user. Current debuggers are limited to displaying the name of the function, or a partial application. In this paper we study two alternative methods of displaying functional values, and their effect on a popular debugging method (Algorithmic Debugging).

In a non-strict functional programming language functions that yield the same result for all total arguments can still differ for partial arguments, that is, they differ in their strictness. Here a Haskell library is presented that enables the programmer to easily check whether a given function is least-strict; if it is not least-strict, then the tool suggests how to make it less strict.

In functional programming languages such as Haskell, it happens often that some parts of a program are not evaluated because their values are not demanded. In practice, those unevaluated parts are often replaced by a placeholder (e.g. ₎ in order to keep the trace size smaller. For algorithmic debugging, this also makes the questions shorter and clearer. In this paper, we present a formal model of tracing in which unevaluated parts are replaced by the symbol _. Some properties such as the correctness of algorithmic debugging will also be proved.

Algorithmic debugging is a semi-automatic method for locating bugs in programs. An algorithmic debugger asks a user a series of questions about the intended behaviour of the program. Here we present two new methods that reduces the number of questions a user must answer to locate a bug. First, we describe a heuristic based on comparing computations of the same program with different inputs. Besides a computation that exhibits some erroneous behaviour, we use information from computations that produce correct results. The heuristic uses program slices to identify areas of code that are likely to be correct. Secondly, we describe a method of compressing the search tree that guides the questions of an algorithmic debugger. This compression is particularly successful when used in combination with our heuristic. Both heuristic and tree-compression are applicable to algorithmic debugging in general. We have implemented it for locating bugs in Haskell programs.

The tracer Hat records in a detailed trace the computation of a program written in the lazy functional language Haskell. The trace can then be viewed in various ways to support program comprehension and debugging. Here we describe a model of the trace that captures its essential properties and allows formal reasoning. The trace structure was inspired by standard graph rewriting implementations of functional languages. The trace is a graph which is constructed by graph rewriting but goes beyond simple term graphs.

This paper formally presents a model of tracing for functional programs based on a small-step operational semantics. The model records the computation of a functional program in a graph which can be utilised for various purposes such as algorithmic debugging. The main contribution of this paper is to prove the correctness of algorithmic debugging for functional programs based on the model. Although algorithmic debugging for functional programs is implemented in several tracers such as Hat, the correctness has not been formally proved before. The difficulty of the proof is to find a suitable induction principle and a more general induction hypothesis.

We outline two methods for locating bugs in a program. This is done by comparing computations of the same program with different input. At least one of these computations must produce a correct result, while exactly one must exhibit some erroneous behaviour. Firstly reductions that are thought highly likely to be correct are eliminated from the search for the bug. Secondly, a program slicing technique is used to identify areas of code that are likely to be correct. Both methods have been implemented. In combination with algorithmic debugging they provide a system that quickly and accurately identifies bugs.

Experience shows that users of the Hat viewing tools find it hard to keep orientation and navigate to a point of interest in the trace. Hence this paper describes a new viewing tool where navigation through the trace is based on the program source. The tool combines ideas from algorithmic debugging, traditional stepping debuggers and dynamic program slicing.

Hat is a programmer's tool for generating a trace of a computation of a Haskell 98 program and viewing such a trace in various different ways. Applications include program comprehension and debugging. The trace viewing tools show expressions and equations of a computation, but they hardly refer to the source program. This disregard of the program is odd, because the computation follows from the program and the usually familiar source program can help orientation in a complex computation. Hence I started the development of new trace viewing tools that are based on showing the source program with various changing markings.

Hat is a programmer's tool for generating a trace of a computation of a Haskell 98 program and viewing such a trace in various different ways. Applications include program comprehension and debugging. A new version of Hat uses a stand-alone program transformation to produce self-tracing Haskell programs. The transformation is small and works with any Haskell 98 compiler that implements the standard foreign function interface. We present general techniques for building compiler independent tools similar to Hat based on program transformation. We also point out which features of Haskell 98 caused us particular grief.

Hat is a programmer's tool for generating a trace of a computation of a Haskell 98 program and viewing such a trace in various different ways. Applications include program comprehension and debugging. A new version of Hat uses a stand-alone program transformation to produce self-tracing Haskell programs. The transformation is small and works with any Haskell 98 compiler that implements the standard foreign function interface. We present general techniques for building compiler independent tools similar to Hat based on program transformation. We also point out which features of Haskell 98 caused us particular grief.

We define a small step operational semantics for a core of Haskell. We modify this semantics to generate traces, specifically Augmented Redex Trails. This small and direct definition of Augmented Redex Trails shall improve our understanding of them and shall help to extend them systematically.

Different tracing systems for Haskell give different views of a program at work. In practice, several views are complementary and can productively be used together. Until now each system has generated its own trace, containing only the information needed for its particular view. Here we present the design of a trace that can serve several views. The trace is generated and written to file as the computation proceeds. We have implemented both the generation of the trace and several different viewers.

In modern statically typed functional languages, type inference is used to determine the type of each function automatically. Whenever this fails, the compiler emits an error message that is often very complex. Sometimes the expression mentioned in the type error message is not the one that is wrong. We therefore implement an interactive tool that allows programmers to browse through the source code of their program and query the types of each expression. If a variable cannot be typed, we would like to present a set of possible types from which the user can decide which is wrong. This should help finding the origin of type errors without detailed knowledge of type inference on the user side.

In this paper we compare three systems for tracing and debugging Haskell programs: Freja, the Redex Trail System and Hood. We identify the similarities and differences of these systems and we evaluate their usefulness in practice by applying them to a number of small to medium programs in which errors had deliberately been introduced.

Deforestation optimises a functional program by transforming it into another one that does not create certain intermediate data structures. Short cut deforestation is a deforestation method which is based on a single, local transformation rule. In return, short cut deforestation expects both producer and consumer of the intermediate structure in a certain form. Starting from the fact that short cut deforestation is based on a parametricity theorem of the second-order typed lambda-calculus, we show how the required form of a list producer can be derived through the use of type inference. Type inference can also indicates which function definitions need to be inlined. Because only limited inlining across module boundaries is practically feasible, we develop a scheme for splitting a function definition into a worker definition and a wrapper definition. For deforestation we only need to inline the small wrapper definition.

Deforestation optimises a functional program by transforming it into another one that does not create certain intermediate data structures. Our type-inference based deforestation algorithm performs extensive inlining, but only limited inlining across module boundaries is practically feasible. Therefore we here present a type-inference based algorithm that splits a function denition into a worker denition and a wrapper denition. For deforestation we only need to inline the small wrappers which transfer the required information. We show that we even can deforest denitions of functions that consume their own result with the worker/wrapper scheme, in contrast to the original algorithm with inlining.

Deforestation optimises a functional program by transforming it into another one that does not create certain intermediate data structures. In [ICFP'99] we presented a type-inference based deforestation algorithm which performs extensive inlining. However, across module boundaries only limited inlining is practically feasible. Furthermore, inlining is a non-trivial transformation which is therefore best implemented as a separate optimisation pass. To perform short cut deforestation (nearly) without inlining, Gill suggested to split definitions into workers and wrappers and inline only the small wrappers, which transfer the information needed for deforestation. We show that Gill's use of a function build limits deforestation and note that his reasons for using build do not apply to our approach. Hence we develop a more general worker/wrapper scheme without build. We give a type-inference based algorithm which splits definitions into workers and wrappers. Finally, we show that we can deforest more expressions with the worker/wrapper scheme than the algorithm with inlining.

Common subexpression elimination is a well-known compiler optimisation that saves time by avoiding the repetition of the same computation. In lazy functional languages, referential transparency renders the identification of common subexpressions very simple. More common subexpressions can be recognised because they can be of arbitrary type whereas standard common subexpression elimination only shares primitive values. However, because lazy functional languages decouple program structure from data space allocation and control flow, analysing its effects and deciding under which conditions the elimination of a common subexpression is beneficial proves to be quite difficult. We developed and implemented the transformation for the language Haskell by extending the Glasgow Haskell compiler. On real-world programs the transformation showed nearly no effect. The reason is that common subexpressions whose elimination could speed up programs are uncommon in lazy functional languages.

Common subexpression elimination is a well-known compiler optimisation that saves time by avoiding the repetition of the same computation. To our knowledge it has not yet been applied to lazy functional programming languages, although there are several advantages. First, the referential transparency of these languages makes the identification of common subexpressions very simple. Second, more common subexpressions can be recognised because they can be of arbitrary type whereas standard common subexpression elimination only shares primitive values. However, because lazy functional languages decouple program structure from data space allocation and control flow, analysing its effects and deciding under which conditions the elimination of a common subexpression is beneficial proves to be quite difficult. We developed and implemented the transformation for the language Haskell by extending the Glasgow Haskell compiler and measured its effectiveness on real-world programs.

In lazy functional programming modularity is often achieved by using intermediate data structures to combine separate parts of a program. Each intermediate data structure is produced by one part and consumed by another one. Deforestation optimises a functional program by transformation into a program which does not produce such intermediate data structures. In this thesis we present a new method for deforestation, which combines a known method,short cut deforestation, with a new analysis that is based on type inference. Short cut deforestation eliminates an intermediate list by a single, local transformation. In return, short cut deforestation expects both producer and consumer of the intermediate list in a certain form. Whereas the required form of the consumer is generally considered desirable in a well-structured program anyway, the required form of the producer is only a crutch to enable deforestation. Hence only the list-producing functions of the standard libraries were defined in the desired form and short cut deforestation has been confined to compositions of these functions. Here, we present an algorithm which transforms an arbitrary producer into the required form. Starting from the observation that short cut deforestation is based on a parametricity theorem of the second-order typed lambda-calculus, we show how the construction of the required form can be reduced to a type inference problem. Typability for the second-order typed lambda-calculus is undecidable, but we only need to solve a partial type inference problem. For this problem we develop an algorithm based on the well-known Hindley-Milner type inference algorithm. The transformation of a producer often requires inlining of some function definitions. Type inference even indicates which function definitions need to be inlined. However, only limited inlining across module boundaries is practically feasible. Therefore, we extend the previously developed algorithm to split a function definition into a worker definition and a wrapper definition. We only need to inline the small wrapper definition, which transfers all information required for deforestation. The flexibility of type inference allows us to remove intermediate lists which original short cut deforestation cannot remove, even with hand-crafted producers. In contrast to most previous work on deforestation, we give a detailed proof of completeness and semantic correctness of our transformation.

This book constitutes the thoroughly refereed post-proceedings of the 19th International Workshop on Implementation and Applications of Functional Languages, IFL 2007, held in Freiburg, Germany in September 2007. The 15 revised full papers presented went through two rounds of reviewing and improvement and were selected from 33 submissions. The papers address all current theoretical and methodological issues on functional and function-based languages such as type checking, contract checking, compilation, parallelism, development and debugging, data structures, parsing as well as various performance related concepts.

Book review