Strategies for Program Transformation
Stratego Tutorial at ETAPS
Full Day Tutorial on April 14, 2002 at
ETAPS 2002 in Grenoble, France
Handouts are now available from the
StrategoDocumentation page
Abstract
Program transformation has applications in many areas of software
engineering including compilation, optimization, refactoring, program
synthesis, software renovation, and reverse engineering. Program
transformation increases programmer productivity by automating
programming tasks, thus enabling programming at a higher-level of
abstraction, and increasing maintainability and re-usability.
This tutorial gives an introduction to principles and practice of
program transformation with rewriting strategies. Programmable
rewriting strategies allow the separation of transformation rules from
the strategies for applying them, thus providing support for concise,
declarative, and reusable specification of program transformation
systems.
The tutorial discusses all aspects of transformation with strategies;
after a taxonomy of program transformation and a discussion of program
representation, it covers specification of program transformations
with rewrite rules, combinators for composing rewriting strategies, in
particular combinators for generic traversal, exchange of
context-dependent information through scoped dynamic rewrite rules,
pragmatics of programming with strategies in Stratego, and
applications of these techniques.
Contents
Introduction
Program transformation has applications in many areas of software
engineering including compilation, optimization, refactoring, program
synthesis, software renovation, and reverse engineering. The aim of
program transformation is to increase programmer productivity by
automating programming tasks, thus enabling programming at a
higher-level of abstraction, and increasing maintainability and
re-usability.
Rewrite rules provide a good formalism for specification of basic
transformation steps. However, since sets of rewrite rules for a
programming language are usually not confluent and terminating,
standard rewriting techniques are not adequate for implementing
program transformation systems.
Stratego is a modular language
for the specification of fully automatic program transformation
systems based on the paradigm of rewriting strategies. Basic
transformation steps are defined using _labeled conditional rewrite
rules_. Rules are combined into complete transformations by means of
programmable
rewriting strategies. An important aspect of these
strategies are combinators for generic traversal.
Scoped dynamic rewrite rules overcome the limitations posed by the context-free
nature of rewrite rules. Together these features support concise and
declarative specification of program transformation systems. The
Stratego distribution contains the Stratego Compiler and the Stratego
Standard Library, and is distributed under the GNU General Public
License.
This full day tutorial gives an introduction to principles and
practice of program transformation with rewriting strategies.
Material
This section outlines the material presented in the tutorial. First an
overview of the tutorial is given in the form of a schedule.
Program
| 9:00--9:30 | Introduction: A taxonomy of program transformation |
| 9:30--10:00 | Program representation |
| 10:15--11:00 | Term rewriting and program transformation |
| 11:15--12:30 | Rewriting strategies |
| 14:00--14:45 | Scoped dynamic rewrite rules |
| 15:00--15:20 | Pragmatics |
| 15:20--16:20 | Lab session: try it out yourself |
| 16:30--17:30 | Applications |
The lab session in the afternoon is optional and depends on the
interest of the audience and the availability of computers. It can be
skipped in favour of a longer session about applications.
A Taxonomy of Program Transformation
Program transformation is used in many areas of software engineering,
including compiler construction, software visualization, documentation
generation, and automatic software renovation. At the basis of all
these different applications lie the main program transformation
scenarios of translation and rephrasing. These main scenarios can be
refined into a number of typical
sub-scenarios.
In a
translating scenario a program is transformed from a source
language into a program in a
different target language. Examples of
translating scenarios are synthesis, migration, compilation, and
analysis. In
ProgramSynthesis an implementation is derived from
a high-level specification such that the implementation satisfies the
specification. A prime example of program synthesis is parser
generation. In
ProgramMigration a program is transformed to
another language. For example, transforming a Fortran77 program to an
equivalent Fortran90 program.
ProgramCompilation is a form
of synthesis in which a program in a high-level language is
transformed to a program in a lower-level language. In
ProgramAnalysis a program is reduced to some property, or
value. Type-checking is an example of program analysis.
In a rephrasing scenario a program is transformed into a different
program in the
same language, i.e., source and target language are
the same. Examples of rephrasing scenarios are normalization,
renovation, refactoring, and optimization. In a
normalization a
program is reduced to a program in a sub-language. In
renovation
some aspect of a program is improved. For example, repairing a Y2K
bug. A
refactoring is a transformation that improves the design of
a program while preserving its functionality. An
optimization is
transformation that improves the run-time and/or space performance of
the program.
The introduction session gives a short overview of the transformation
taxonomy with typical examples of the various transformations to put
the rest of the material into perspective.
Program Representation
Although some transformation systems work directly on text, in general
a textual representation is not adequate for performing complex
transformations. Therefore, a structured representation is used by
most systems. Since programs are written as texts by programmers,
parsers are needed to convert from text to structure and unparsers are
needed to convert structure to text.
A number of issues should be considered when choosing a program
representation: to use parse trees or abstract syntax trees, trees or
graphs, how to represent variables and variable bindings, and how to
exchange programs between transformation components.
In this session these issues are discussed and the ATerm
representation used in Stratego is presented. The ideas are
illustrated with the abstract syntax of several (small) languages that
will be used in the rest of the tutorial.
Term Rewriting and Program Transformation
Rewrite rules provide a good formalism for expressing program
transformations. A rewrite rule defines a local transformation
derived from an algebraic equality of programs.
In this session the basics of term rewriting are reviewed including
rewrite rules, conditional rewrite rules, substitution, standard
(exhaustive) rewriting strategies, and the notions of confluence and
termination. The application of rewriting to program transformation
is illustrated using several examples.
Rewriting Strategies
Exhaustive application of all rules to the entire abstract syntax tree
of a program is not adequate for most transformation problems. The
system of rewrite rules expressing basic transformations is often
non-confluent and/or non-terminating.
An ad hoc solution that is often used is to encode control over the
application of rules into the rules themselves by introducing
additional function symbols. This intertwining of rules and strategy
obscures the underlying program equalities, incurs a programming
penalty in the form of rules that define a traversal through the
abstract syntax tree, and disables the reuse of rules in different
transformations.
The paradigm of programmable rewriting strategies solves the problem
of control over the application of rules while maintaining the
separation of rules and strategies. A strategy is a little program
that makes a selection from the available rules and defines the order
and position in the tree for applying the rules. Thus rules remain
pure, are not intertwined with the strategy, and can be reused in
multiple transformations.
Support for strategies is provided by a number of transformation
systems in various forms. In
TAMPR a transformation is
organized as a sequence of canonical forms. For each canonical form a
tree is normalized with respect to a subset of the rules in the
specification.
ELAN provides non-deterministic sequential
strategies. Stratego provides generic primitive traversal operators
that can be used to compose generic tree traversal schemas. For more
information about related formalisms see
a survey of rewriting strategies in program transformation systems.
In this central session of the tutorial the notion of a strategy is
defined and the operators for combining strategies are presented. In
particular, the specification of (generic) traversal strategies will
get much attention.
Dynamic Rewrite Rules
Another problem of rewriting is the context-free nature of rewrite
rules. A rule has only knowledge of the construct it is
transforming. However, transformation problems are often
context-sensitive. For example, when inlining a function at a call
site, the call is replaced by the body of the function in which the
actual parameters have been substituted for the formal parameters.
This requires that the formal parameters and the body of the function
are known at the call site, but these are only available higher-up in
the syntax tree.
There are many similar problems in program transformation; examples
are bound variable renaming, typechecking, constant and copy
propagation, and dead code elimination. Although the basic
transformations in all these applications can be expressed by means of
rewrite rules, they need contextual information.
The usual solution to this problem is to extend the traversal over the
tree (be it hand-written or generic) such that it distributes the data
needed by transformation rules. The disadvantage of these solutions
is that the traversal strategy becomes data heavy instead of just
handling control flow. That is, all traversal functions become
infected with additional parameters carrying context information.
The concept of
scoped dynamic rewrite rules is an extension of
rewriting strategies that overcomes the context-free nature of rewrite
rules. A dynamic rule is a normal rewrite rule that is generated at
run-time and that can access information from its generation context.
For example, to define an inliner, a rule that inlines function calls
for a specific function can be generated at the point where the
function is declared, and used at call sites of the function.
Dynamic rules are introduced using several examples: bound variable
renaming, function inlining, and dead code elimination.
Pragmatics
In this short session the pragmatics of working with Stratego are
discussed and demonstrated. First it is shown how to write, compile
and run Stratego programs. Then a short overview of the architecture
of transformation systems is given, including the specification of
parsers and pretty-printers using tools from
XT a bundle of program transformation tools.
Lab Session
The lab session is an optional component of the tutorial. Depending
on feasibility (see below) and interest from the participants this
session can occupy the rest of the tutorial or be skipped in favour of
a longer session on applications.
In the lab session participants can try out the techniques presented
in the earlier sessions by writing several small Stratego programs
solving small transformation problems.
The lab session depends on the availability of computers. This
requires a room with computers running Linux or Unix with an
installation of XT-0.9 and the tutorial package. If it is not
feasible to provide a computer room for just one hour, an alternative
solution is to have participants bring their own laptops. It is
conceivable that two participants can work together on one laptop,
such that not all participants need computers. This requires that
participants install the software prior to the conference. A page on
the stratego-language.org site will provide download and installation
instructions. Alternatively, the lab session can be done in the form
of a demonstration.
Applications
The earlier sessions present the principles of transformation with
strategies using many examples. To consolidate the understanding of
the participants and to show more possibilities this session presents
some further applications from the following areas (depending on the
interest of the audience):
- Compilation by transformation
- Optimization
- Refactoring
- Document transformation (XML)
- Program visualization
Audience
Relevance to ETAPS 2002
The material presented in this tutorial is closely related to the
themes of the CC and ESOP conferences and the LDTA workshop.
Intended Audience
The tutorial is directed at researchers in the field of program
transformation and potential users, both academic and industrial.
Researchers
Researchers working on tools or formal methods for program
transformation can use the tutorial to get a complete overview of the
ideas underlying the strategic programming idiom. Either as a direct
source of study (the sources of Stratego and XT are available for
experimentation) or as inspiration for their own setting.
Potential Users
Academic and industrial developers of language processing systems such
as compilers, optimizers, program generators, and program
visualization tools can use the tutorial to learn to use a powerful
set of tools for compact implementation of their systems. The tools
are readily available and distributed under GNU GPL.
Required Background
Participants are expected to have a basic understanding of compiler
construction and programming language syntax and semantics.
Learning Objectives
After the tutorial participants will understand the principles of
programming transformation systems with rewriting strategies. In
particular, they will understand the ideas and advantages of generic
term traversal and dynamic rules. With this understanding the
participants should be able to start writing Stratego specifications.
History of the Tutorial
The tutorial as proposed above has not been given before. The material
for the tutorial has evolved over the last couple of years for
numerous seminar presentations, several university lectures, and
tutorials at Stratego Users Days.
The
First Stratego Users Day was held in March 2000 at CWI, Amsterdam,
The Netherlands and included a two hour tutorial on the basics of
program transformation with rewriting strategies.
Prior to the
Second Stratego Users Day that was held in February 2001
at Utrecht University a full day tutorial was given about Stratego.
The tutorial was given in demonstration mode involving the
participants in solving several transformation problems.
The material for this tutorial draws on the material recently
developed for the course on
Software Generation taught at
Utrecht University in the academic years 2000--2001 and 2001-2002.
Instructor
Eelco Visser is
- One of developers of the XT bundle of transformation tools
- Designer of the Syntax Definition Formalism SDF2 used in XT
- Author of numerous papers on program transformation, including papers on the design, implementation, and application of Stratego, and a recent survey exploring the role of rewriting strategies in program transformation
- Teacher of program transformation at Utrecht University
- Founder of program-transformation.org, a collaborative web-site based on wiki technology dedicated to surveying the field of program tranformation.
- Program committee member of the LDTA workshops in 2001 and 2003 and program chair of the RULE workshop on rule-based programming in 2002.