| CONTENTS:
INTRODUCTION, THE DEFINITION, TEXTS, COURSES, MAILING LIST, LIBRARY,
IMPLEMENTATIONS.
INTRODUCTION
Standard ML is a statically scoped interactive functional language with a
polymorphic static type system, polymorphic references, polymorphic exceptions,
a sophisticated modules system and a formal semantics. The ML project won
the British Computer Society's Technical Award for 1987. It is a general
purpose programming language which is currently used for formal verification,
VLSI work, microprocessor design, graphical interfaces, and compilers.
THE DEFINITION.
Robin Milner, Mads Tofte and Robert Harper
The Definition of Standard ML
MIT, 1990.
Robin Milner and Mads Tofte are writing a commentary on the Definition.
TEXTS.
The first book is quite slow-paced and is aimed at people learning to program.
It doesn't cover the modules system.
Ake Wikstrom
Functional Programming Using Standard ML
Prentice Hall 1987
ISBN: 0-13-331661-0
The next book goes at a faster pace, and includes an introduction to the
modules system. It also includes sections on denotational semantics,
lambda calculus and implementation techniques.
Chris Reade
Elements of Functional Programming
Addison-Wesley 1989
ISBN: 0-201-12915-9
The following report is available from the LFCS (Dorothy McKie,
[email protected]) and costs 5 pounds or 10 US dollars. It covers all of
Standard ML.
Robert Harper
Introduction to Standard ML
LFCS Report Series ECS-LFCS-86-14
Laboratory for Foundations of Computer Science
Department of Computer Science
University of Edinburgh
Nov. 1986 (revised Jan. 1989 by Nick Rothwell and Kevin Mitchell)
The following report is available from the LFCS (Dorothy McKie,
[email protected]) and is free. It includes an introduction to Standard ML
and three lectures on the modules system.
Mads Tofte
Four Lectures on Standard ML
LFCS Report Series ECS-LFCS-89-73
Laboratory for Foundations of Computer Science
Department of Computer Science
University of Edinburgh
March 1989
The following report is available from the LFCS (Dorothy McKie,
[email protected]) and is free. It introduces Extended ML, a language
for writing (non-executable) specifications of Standard ML programs and
for formally developing Standard ML programs from such specifications.
Don Sannella
Formal program development in Extended ML for the working programmer.
LFCS Report Series ECS-LFCS-89-102
Laboratory for Foundations of Computer Science
Department of Computer Science
University of Edinburgh
December 1989
Larry Paulson and Stefan Sokolowski both have books on Standard ML
in preparation.
COURSES
The LFCS runs week-long courses on Standard ML twice a year. The first
two days are optional and cover the core language; the other three days
cover the modules system, with an introduction to specification and
refinement in the Extended ML algebraic specification language.
Contact [email protected] for more information.
MAILING LIST
The worldwide sml mailing list covers all aspects of Standard ML.
Subscription requests from Europe to [email protected].
Subscription requests from elsewhere to [email protected]
Contributions to [email protected].
LIBRARY
The Edinburgh Standard ML Program Library is currently available on
beta-release from the LFCS and AT&T. It provides a consistent
set of functions on the basic types of the language and on vectors (both
constant and variable) and a few extras. It includes a consistent set of
parsing and unparsing functions.
The library consists of a set of signatures with sample portable
implementations, two (fairly basic) documentation files, and
implementations for Poly/ML and SML-NJ 0.56 that use some of the
non-standard primitives available in those systems.
IMPLEMENTATIONS
There are 6 implementations of Standard ML available now, which are described
below. There are another 4 being written, which are also described below.
They are mostly interactive incremental compilers. They all implement most
of the standard faithfully; at present they differ slightly in some obscure
corners. I've tried to give some indication of their relative performance
on reasonably sized programs, but performance depends on many factors
(such as the size of real memory on your machine).
Poly/ML.
Poly/ML produces native code for Sun3 UNIX systems. A SPARC version
is being developed. A VAX version exists, but isn't currently supported.
Poly/ML is about the speed of a good compiled Common Lisp. For my uses it
needs 3-6M of heap space.
Poly/ML uses a persistent store, supports arbitrary precision integer
arithmetic. It comes with a make system and a function that lets you define
functor bodies interactively. Abstract Hardware Ltd. are developing an X11
interface, and are improving the speed of the persistent storage system.
Poly/ML is distributed by Abstract Hardware Ltd. ([email protected]).
It costs 500 pounds for an academic site licence and 2,000 pounds per
machine for industrial users with multiple and site licences by negotiation.
Standard ML of New Jersey.
Standard ML of New Jersey is a complete and robust implementation
developed jointly at AT&T Bell Laboratories and Princeton University. It is
an open system (source code is freely available) implemented in Standard ML
that emphasizes quality code generation. Version 0.65 (final beta-release)
of Standard ML of New Jersey generates native code for Vax, 68020, SPARC,
and MIPS (big and little endian) architectures under various versions of the
Unix operating system (BSD, Ultrix, SunOS, Mach), and soon the Macintosh
OS. It needs about twice as much heap space as Poly/ML, though this
varies with the application. It produces code that runs about three
as fast as Poly/ML, and takes slightly longer to compile it.
SML of NJ comes with ML implementations of LEX and YACC. It provides
an interpretive mode with fast compilation, it has profiling and separate
compilation facilities, and it can produce stand-alone executable applications.
Typed first-class continuations are supported, as well as Unix signal handling
in ML. An ML debugger and an X-windows interface are in preparation.
SML of NJ is copyrighted by AT&T but the system, including source
code, is freely distributable. It is available by anonymous ftp from
research.att.com and princeton.edu. Login as "anonymous" with your user
name as password. Put ftp in binary mode and copy the (compressed tar)
files you need from the directory dist/ml (pub/ml on princeton.edu). You
only need the mo.*.tar.Z files for your machines. Alternatively mail
[email protected]. In the UK, it is available from the LFCS (send e-mail
to [email protected]).
Poplog ML.
Standard ML is supported as part of the Poplog system, which also
provides incremental compilers for Pop-11, Common Lisp and Prolog in a
common environment with shared data-structures, so that mixed language
programming is possible. The integrated editor and HELP mechanism
support online teaching aids. The Poplog X Windows interface will be
provided from 2nd Quarter 1990 (May or June we hope). Performance
comparable to Poly/ML (when last tested) but depends on program.
Process size starts from about 1.5Mbytesd (including Editor) on Sun3.
Poplog is available for VAX+VMS, VAX+Ultrix, VAX+Bsd 4.2/3,
Sun-2,3,4, Sun386i, SPARCstation, Solbourne, Sequent Symmetry (with Dynix),
HP M680?0+Unix workstations and Apollo+Unix. Versions for MAC-II with A/UX,
DECstation 3100 and MIPS will be available shortly.
UK educational users should contact the School of Cognitive and
Computing Sciences, University of Sussex ([email protected]).
People in the USA or Canada should contact Computable Functions Inc.
([email protected]. All others should contact Integral Solutions Ltd.
([email protected]). Commercial prices start around 7,500 pounds.
Educational discount 85% . Special prices for UK academics.
Edinburgh ML 4.0.
Edinburgh ML 4.0 is an implementation of the core language (without
the module system). It uses a bytecode interpreter, which is written in C
and runs on any machine with 32 bit words, a continuous address space and
a correct C compiler. The bytecode interpreter can be compiled with
switches to avoid the buggy parts of the C compilers that we've used it with
(as far as I know none of them worked correctly). Ed. ML 4.0 typically
uses 1.5-4M of heap space. It is about a quarter of the speed of Poly/ML.
Ed. ML 4.0 is available from us. People in the UK can get by e-mail;
we also distribute it on Sun/Apollo cartridge or 1/4" tape (tar or VMS format)
for a small fee (50 pounds for academic sites and LFCS affiliates, 100 pounds
for other commercial organisations). Mail [email protected] for more information.
ANU ML
ANU ML is descended from Cardelli's ML Pose 3. It implements the
core language of the standard and an old version of modules. It incrementally
compiles to native code on Sun-3, Vax/Ultrix, Pyramid and MacII/AUX. (It
is intended to standardize modules and do the port to Sun-4 in the near
future.)
ANU ML has a program development system with strong support for
debugging (tracing, automatic retesting etc.) and has been extended with
a built-in type complex. The designers claim that it is as economical with
memory as Edinburgh ML but closer in speed to NJ ML.
ANU ML is still considered to be in beta release since exceptions
have been standardized quite recently. It is available from Malcolm Newey,
CS Dept., Australian National University ([email protected]) by arrangement;
soon to be available by ftp.
Rutherford ML.
Rutherford ML is an implementation of the core language that runs
on Franz Lisp. It is no longer maintained, and is only distributed with
the Cambridge LCF system (which it supports).
Rutherford also supply an SML-YACC parser generator, which runs on
Edinburgh ML and Poly/ML, and will soon run on New Jersey ML.
Both SML-YACC and Cambridge LCF are available from Brian Matthews
([email protected]).
The Kit Compiler.
The LFCS are developing an implementation to serve as a base for
experiments in language design. The code reflects the semantics as directly
as possible.
At present the Kit Compiler can only be run on another implementation
of Standard ML. Run like this it is slow and needs at least 16M of heap
space. Eventually we hope to have a version about the size and speed of
Edinburgh ML.
The Kit Compiler is not available yet.
Ten15 ML.
Harlequin limited is developing a Standard ML compiler for the Royal
Signals and Radar Establishment, Malvern. The target for the compiler is an
algebra specifying an abstract machine. The algebra is called Ten15 and was
developed by the RSRE. The ML compiler produces an encoding in Ten15 which
can then be translated into machine code for a variety of machines.
Harlequin MLWorks.
Harlequin limited are currently developing MLWorks to support
programming in Standard ML. This product will provide an integrated
design and development environment supporting programming in the small
and in the large. The environment will also support the specification
of ML programs using Extended ML. Harlequin's product is being
developed with both the industrial and academic user in mind. MLWorks
will be available for use on a large number of hardware platforms.
Integrated within MLWorks will be an industrial-strength compiler
for Standard ML. Harlequin plans to launch the compiler separately during
the first quarter of 1991.
LCS (a Language for Communicating Systems)
LCS implements the core language (without modules), extended with
High Order CCS agents. Agents are "first-class" values and are defined as
Standard ML values of specific types that may be turned into processes.
Agent constructs include all those of CCS, with some extensions. LCS is
implemented as a byte-code interpreter, written in C; it runs on virtually
all Unix machines and on Apple Macintoshes (Finder ok). Core images are
portable across machines.
Two interactive user interfaces are provided: an executive and a
simulator. In addition to SML, the executive allows users to start and manage
processes, either foreground or background (the top-level runs itself as
an LCS process). The simulator implements a set of commands for interactively
expanding LCS agents, under full control of the user.
LCS runs between 6 times slower than SML-NJ on toy examples and 2
times slower on large examples (including garbage collection time, measured
on a locally used 7000 lines ML application). Message passing between
processes is about 5 times slower than functional parameter passing.
Availability is scheduled for the last quarter of 1990. Earlier
versions of the system have been running locally for about two years.
Contact Bernard Berthomieu ([email protected]) for more information.
OBSOLETE VERSIONS
Edinburgh ML 3.5.
Edinburgh ML 3.5 is a predecessor of Edinburgh ML 4.0. It uses the
old style exceptions, and includes an obsolete version of the modules system.
It is mentioned here because it is available on the Macintosh.
Edinburgh ML 3.5 is available in the UK by e-mail from the LFCS. We
will also distribute it on a tape, as for Edinburgh ML 4.0. It is also
available from Meta Software (Peter Hendersen, [email protected]).
|