Garbo/Garbo.cdr/pc/source/tile.lzh/tile.0

Text File | 1990-07-26 | 14.0 KB | 378 lines

THREADED INTERPRETIVE LANGUAGE ENVIRONMENT (TILE) [RELEASE 2.0]

June 29, 1990

Mikael R.K. Patel
Computer Aided Design Laboratory (CADLAB)
Department of Computer and Information Science
Linkoping University
S-581 83 LINKOPING
SWEDEN
Email: [e-mail]


1.    INTRODUCTION

TILE Forth is a 32-bit implementation of the Forth-83 Standard
written in C. Thus allowing it to be easily moved between different
computers compared to traditional Forth implementations in assembly.

Most Forth implementations are done in assembly to be able to
utilize the underlying architecture as optimal as possible. TILE
Forth goes another direction. The main idea behind TILE Forth is to
achieve a portable forth implementation for workstations and medium
size computer systems so that new groups of programmers may be exposed
to the flavor of an extensible language such as Forth.

The implementation of TILE Forth is selected so that, in principle,
any C-level procedure may become available on the interactive and
incremental forth level. Other models of implementation of a threaded
interpreter in C are possible but are not as flexible.

TILE Forth is organized as a set of modules to allow the kernel to be
used as a general threading engine for C. Environment dependencies such
as memory allocation, error handling and input/output have been separated
out of the kernel to increase flexibility. The forth application is "just"
an example of how to use the kernel.

Comparing forth implementation using the traditional benchmark such as
the classical sieves calculation is difficult because of difference in
speed between workstations and personal computers. The Byte sieves
benchmark is reported to typically run in 16 seconds on a direct threaded
forth implementation. This benchmark will run in 27 seconds in TILE forth
on a SUN-3/60 and less than 13 seconds on a SUN SPARCstation 1. These times
are the total time for loading TILE forth, compiling and executing the
benchmark. Comparing to, for instance, other interpretive languages such
as Lisp, where one of the classical benchmarks is calculation of the
Fibonacci function, the performance increase is over one magnitude.

The kernel supports the Standard Forth-83 word set except for the
blocks file word set which are not used. The kernel is extended with
many of the concepts from modern programming languages. Here is a list
of some of the extensions; argument binding and local variables, queue
management, low level compiler words, string functions, floating point
numbers, exceptions and multi-tasking. The TILE Forth environment also
contains a set of reusable source files for high level multi-tasking,
data modeling and structuring modules, and a number of programming tools.

To allow interaction and incremental program development TILE Forth
includes a programming environment as a mode in GNU Emacs. This environ-
ment helps with program structuring, documentation search, and program
development. Each vocabulary in the kernel and the source library file is
described by a manual, documentation and test file. This style of
programming is emphasized throughout the environment to increase
understanding and reusability of the library modules. During compilation
TILE Forth's io-package keeps track for which modules have been loaded
so that they are only loaded once even if included by several modules.

Writing a Forth in C gives some possibilities that normally are
not available when performing the same task in assembly. TILE Forth
has been profiled using the available tools under Unix. This information
has been used to optimize the compiler so that it achieves a compilation
speed of over 200.000 lines per minute on my machine (a disk-less SUN
SPARCstation 1). Currently code is only saved in source form and
applications are typically "compile-and-go".

So far TILE Forth has been ported and tested at over forty locations
without any major problems except where C compilers do not allow sub-
routine pointers in data structures.


2.    EXTENSIONS

What is new in the TILE forth? First of all the overall organization
of words. To increase portability and understanding of forth code modules
vocabularies are used as the primary packaging mechanism. New data types
such as rational and floating point numbers are implemented in separate
vocabularies. The vocabularies act as both a program module and an
abstract data type.

2.1    Extendable interpreter

To allow extension of the literal symbol set (normally only integer
numbers) each vocabulary is allowed to have a literal recognition
function. This function is executed by the interpreter when the symbol
search has failed. The literal recognizer for the forth vocabulary is
"?number". This simple mechanism allows modules such as for rational and
floating point numbers, and integer ranges to extend with their own
literal function.

2.2    Data description

As the Forth-83 Standard lack tools for description of data structures
TILE Forth contains a fairly large library of tools for this purpose.
These are described more in detail in the next section.

2.3    Argument binding and local variables

When writing a forth function with many arguments stack shuffling
becomes a real pain. Argument binding and local variables is a nice
way out of these situations. Also for the new-comer to Forth this
gives some support to this at first very cryptic language. Even
the stack function may be rewritten using this mechanism:

    : 2drop { a b } ;
    : 2swap { a b c d } c d a b  ;
    : fac { n } n 0> if n 1- recurse n * else 1 then ;

The argument frame is created on top of the parameter stack and is
disposed when functions is exited. This implementations style of
reduces the cost of binding as most functions have more arguments
then return values. A minimum number of data elements have to be
move to create and manage the argument frame.

2.4     Exception handling

Another extension in TILE Forth is exception handling with multiple
exception handling code block. The syntactical structure is very
close to that of Ada, i.e., any colon definition may contain an error
handling section. Should an error occur during the execution of the
function the stack status is restore to the situation at the call
of the function and the lastest exception block is executed with the
signal or exception as a parameter;

    exception zero-divide ( -- exception)

    : div ( x y -- z)
          /
    exception> ( x y signal -- )
      drop zero-divide raise
        ;

Error situations may be indicated using an exception raise function.
Low level errors, such as zero division, are transformed to exceptions
in TILE Forth.

2.5    Entry visibility and forward declaration

Last, some of the less significant extension are forward declaration
of entries, hidden or private entries, and extra entry modes. Forward
declaration of entries are automatically bound when the entry is later
given a definition. Should a binding not exist at run-time an error
message is given and the computation is aborted.

    forward eval ( ... )

    : apply ( ... ) ... eval ... ;
    : eval ( ... ) ... apply ... ;

Three new entry modes have been added to the classical forth model
(immediate). These allow hiding of entries in different situations.
The first two marks the last defined word's visibility according to
an interpreter state. These two modifiers are called "compilation"
and "execution" and are used as "immediate". A word like "if" is
"compilation immediate" meaning it is visible when compiling and
then always executed.

    compiler forth definitions

    : if ( -- ) compile (?branch) >mark ; compilation immediate

The "private" modifier is somewhat different. It concerns the
visibility across vocabularies (modules and types). If a word is
marked as "private" the word is only visible when the vocabulary in
which it is defined in is "current". This is very close to the concept
of hidden in modules and packages in Modula-2 and Ada.

    4 field +name ( entry -- ptr) private

The above definition will only be visible in the vocabulary it was
defined. The "private" modifier is useful to help isolate implementation
dependencies and reduce the name space which also increases compilation
speed.


3.     SOURCE LIBRARY

The TILE Forth programming environment contains a number of tools to
make programming in Forth a bit easier. If you have GNU Emacs, TILE
Forth may run in a specialized forth-mode. This mode supports automatic
program indentation (pretty printing), documentation search, and
interactive and incremental program development, or "edit-compile-test"
style of program development.

To aid program development there is also a source code library with
manual pages, documentation (glossary), and test and example code.
Most of the source code are data modeling tools. In principle, from
bit field definition to object oriented structures are available. The
source code library also contains debugging tools for tracing, break-
point'ing and profiling of programs.

The first level of data modeling tools are modules for describing;
1) bit fields, 2) structures (records), 3) aggregates of data
(vectors, stacks, buffers, etc), and 4) high level data objects
(lists, sets, etc).

The next level of tools are some tools for high level syntactic sugar
for multi-tasking concepts (semaphores, channels, etc), finite state
machines (FSM), anonymous code block (blocks), a general top down parser
with backtrack and semantic binding, and a simulation package. The source
library will be extended during the coming releases.


4.     PROGRAMMING STYLE

A source code module has, in general, the following structure; the
first section includes any modules needed (these are only loaded once).
Second follows global definitions for the module. Normally this is
a vocabulary for the module. Third comes the search chain to be used
throughout the module. It is important not to change the search order
as 1) it becomes difficult for a reader to understand the code, 2)
any change in the search chain flushes the internal lookup cache
in TILE Forth and reduces compilation speed.

    .( Loading the Library...) cr

    #include someLibrary.f83
    ...

    ( Global data and definitions)

    : someGlobalDefinitions ( -- ) ... ;

    vocabulary theLibrary

    someLibrary ... theLibrary definitions

    ( Local data and definitions)

    : somePrivateDefinitions ( -- ) ... ; private
    ...
    : someDefinitions ( -- ) ... ;

    forth only

To create lexical levels within the same vocabulary the word "restore"
may be used. It stores the vocabulary pointer to the given entry and
thus hides the words defined after this entry. The word "restore" has
much the same action as "forget" but without putting back the dictionary
pointer.


5.    SOURCE FILES

The TILE Forth source is broken down into the following files:

README
   This short documentation of TILE.

COPYING
   The GNU General Public License.

INSTALL
   Some help on how to install TILE Forth.

PORTING
   Some help on how to port TILE Forth and typical problems

Makefile
   Allows a number of compilation styles for debugging, profiling,
   sharing etc. New machines and conditional compilation symbols are
   added here.

src
  The C source library with the kernel code and GNU Emacs forth-mode.

lib
   The Forth-83 source library for data description and management,
   high level tasking, etc.

tst
   Test file for each Forth-83 source code file and a set of benchmarks.

man
   Manual pages for the TILE Forth C kernel and Forth-83 source code
   library.

doc
   Documentation and glossaries for each source code file and kernel
   vocabularies.

bin
   Utility commands and the TILE forth compiler/interpreter.



6.    CONFIGURATION

TILE forth is targeted for 32-bit machines and no special aid is
available to allow it to be compiled for other bit-widths. The
configuration is maintained by a "make" files.

This configuration file allows a number of different modes to support
typical program development phases (on C level) such as debugging,
profiling, optimization and packaging. Please see the information in
these files.


7.    COPYING

This software is offered as shareware. You may use it freely, but
if you do use it and find it useful, you are encouraged to send the
author a contribution (>= $50) to the following address:

    TILE Technology HB
    Stragatan 19
    S-582 67 Linkoping
    SWEDEN

If you send me a contribution, I will send you manual pages and
documentation files and will answer questions by mail. Also your
name will be put on a distribution list for future releases.

For further information about copying see the file COPYING and
the headers in the source code files. Commercial usage of the
kernel is not allowed without a license from the company above.


8.    NOTE

Due to the 32-bit implementation in C a number of Forth-83 definitions
are not directly confirmed. Below is a short list of words that might
give problems when porting Forth code to this environment:

* The Block Word Set is not supported. Source code is saved as text
  files.

* All stacks and words size are 32-bit. Thus special care must be
  taken with memory allocation and access.

* Lowercase and uppercase are distinguished, and all forth words are
  lowercase.

* A word in TILE is allowed arbitrary length as the name is stored as
  as a null terminated string.

* Input such as "key" performs a read operation to the operating system
  thus will echo the characters.

* Variables should not allocate extra memory. "create" should be used.

* Double number arithmetic functions are not available.

Some major changes have been made to the kernel in this second release.
To allow implementation of floating point numbers and increase porting
the kernel is now written in its own extendable typing system. Some
extension have been removed such as the casting operator in the
interpreter.


ACKNOWLEDGMENTS

First of all I wish to express my gratitude to Goran Rydqvist for helped
me out with the first version of the kernel and who implemented the
forth-mode for GNU Emacs.

Second a special thanks to the beta test group who gave me valuable
feedback. Especially Mitch Bradley, Bob Giovannucci Jr., Moises Lejter,
and Brooks David Smith.

Last, I wish to thank the may users that have been in touch after the
first release and given me many comments and encouragements.

Thank you all.