FishMarket 1.0/FishMarket v1.0.iso/fishies/501-525/disk_525/siod/siod.doc

Text File | 1992-05-06 | 18KB | 522 lines

 *                        COPYRIGHT (c) 1989 BY                             *
 *        PARADIGM ASSOCIATES INCORPORATED, CAMBRIDGE, MASSACHUSETTS.       *
 *        See the source file SLIB.C for more information.                  *

Documentation for Release 2.4 27-APR-90, George Carrette

[Release Notes:]

1.4 This release is functionally the same as release 1.3 but has been
remodularized in response to people who have been encorporating SIOD
as an interpreted extension language in other systems.

1.5 Added the -g flag to enable mark-and-sweep garbage collection.
    The default is stop-and-copy.

2.0 Set_Repl_Hooks, catch & throw.

2.1 Additions to SIOD.SCM: Backquote, cond.

2.2 User Type extension. Read-Macros. (From C-programmer level).

2.3 save-forms. load with argument t, comment character, faster intern.
    -o flag gives obarray size. default 100.

2.4 speed up arithmetic and the evaluator. fixes to siod.scm. no_interrupt
    around calls to C I/O. gen_readr.

[e-mail]
George Carrette


Paradigm Associates Inc          Phone: [phone]
29 Putnam Ave, Suite 6
Cambridge, MA 02138

[Files:]

 siod.h    Declarations
 slib.c    scheme library.
 siod.c    a main program.
 siod.scm  Some scheme code
 pratt.scm A pratt-parser in scheme.


[Motivation:]

The most obvious thing one should notice is that this lisp implementation
is extremely small. For example, the resulting binary executable file
on a VAX/VMS system with /notraceback/nodebug is 17 kilo-bytes.

Small enough to understand, the source file slib.c is 30 kilo-bytes.

Small enough to include in the smallest applications which require
command interpreters or extension languages.

We also want to be able to run code from the book "Structure and
Interpretation of Computer Programs."

Techniques used will be familiar to most lisp implementors.  Having
objects be all the same size, and having only two statically allocated
spaces simplifies and speeds up both consing and gc considerably.  the
MSUBR hack allows for a modular implementation of tail recursion,
an extension of the FSUBR that is, as far as I know, original.
The optional mark and sweep garbage collector may be selected at runtime.

Error handling is rather crude. A topic taken with machine fault,
exception handling, tracing, debugging, and state recovery which we
could cover in detail, but is clearly beyond the scope of this
implementation. Suffice it to say that if you have a good symbolic
debugger you can set a break point at "err" and observe in detail all
the arguments and local variables of the procedures in question, since
there is no casting of data types. For example, if X is an offending
or interesting object then examining X->type will give you the type,
and X->storage_as.cons will show the car and the cdr.

[Invocation:]

siod [-hXXXXX] [-iXXXXX] [-gX] [-oXXXXX]
 -h where XXXXX is an integer, to specify the heap size, in obj cells,
 -i where XXXXX is a filename to load before going into the repl loop.
 -g where X = 1 for stop-and-copy GC, X = 0 for mark-and-sweep.
 -o where XXXXX is the size of the symbol hash table to use, default 100.

  Example:
   siod -isiod.scm -h100000

[Garbage Collection:]

There are two storage management techniques which may be chosen at runtime
by specifying the -g argument flag.

 -g1 (the default) is stop-and-copy. This is the simplest and most
     portable implementation. GC is only done at toplevel.
 -g0 is mark-and-sweep. GC is done at any time, required or requested.
     However, the implementation is not as portable.

Discussion of stop-and-copy follows:

As one can see from the source, garbage collection is really quite an easy
thing. The procedure gc_relocate is about 25 lines of code, and
scan_newspace is about 15.

The real tricks in handling garbage collection are (in a copying gc):
 (1) keeping track of locations containing objects
 (2) parsing the heap (in the space scanning)

The procedure gc_protect is called once (e.g. at startup) on each
"global" location which will contain a lisp object.

That leaves the stack. Now, if we had chosen not to use the argument
and return-value passing mechanism provided by the C-language
implementation, (also known as the "machine stack" and "machine
procedure calling mechanism) this lisp would be larger, slower, and
rather more difficult to read and understand. Furthermore it would be
considerably more painful to *add* functionality in the way of SUBR's
to the implementation.

Aside from writing a very machine and compiler specific assembling language
routine for each C-language implementation, embodying assumptions about
the placement choices for arguments and local values, etc, we
are left with the following limitation: "YOU CAN GC ONLY AT TOP-LEVEL"

However, this fits in perfectly with the programming style imposed in
many user interface implementations including the MIT X-Windows Toolkit.
In the X Toolkit, a callback or work procedure is not supposed to spend
much time implementing the action. Therefore it cannot have allocated
much storage, and the callback trampoline mechanism can post a work
procedure to call the garbage collector when needed.

Our simple object format makes parsing the heap rather trivial.
In more complex situations one ends up requiring object headers or markers
of some kind to keep track of the actual storage lengths of objects
and what components of objects are lisp pointers.


Discussion of mark-and-sweep follows:

In a mark-and-sweep GC the objects are not relocated. Instead
one only has to LOOK at objects which are referenced by the argument
frames and local variables of the underlying (in this case C-coded)
implementation procedures. If a pointer "LOOKS" like it is a valid
lisp object (see the procedure mark_locations_array) then it may be marked,
and the objects it points to may be marked, as being in-use storage which
is not linked into the freelist in the gc_sweep phase.

Another advantage of the mark_and_sweep storage management technique is
that only one heap is required.

This main disadvantages are:
(1) start-up cost to initially link freelist.
    (can be avoided by more general but slower NEWCELL code).
(2) does not COMPACT or LOCALIZE the use of storage. This is POOR engineering
    practice in a virtual memory environment.
(2) the entire heap must be looked at, not just the parts with useful storage.

In general, mark-and-sweep is slower in that it has to look at more
memory locations for a given heap size, however the heap size can
be smaller for a given problem being solved. More complex analysis
is required when READ-ONLY, STATIC, storage spaces are considered.
Additionally the most sophisticated stop-and-copy storage management
techniques take into account considerations of object usage temporality.

[Compilation:]

The code has been compiled and run by the author on Sun III and IV,
Encore Multimax, 4.3BSD VAX, VAX/VMS, AMIGA 500 using the Lattice C
compiler, MacIntosh using THINK C version 4.0

On all unix machines use (with floating-point flags as needed)

  %cc -O -c siod.c
  %cc -O -c slib.c
  %cc -o siod siod.o slib.o

on VAX/VMS:

  $ cc siod
  $ cc slib
  $ link siod,slib,sys$input:/opt
  sys$library:vaxcrtl/share
  $ siod == "$" + F$ENV("DEFAULT") + "SIOD"

on AMIGA 500, ignore warning messages about return value mismatches,
  %lc siod.c
  %lc slib.c
  %blink lib:c.o,siod.o,slib.o to siod lib lib:lcm.lib,lib:lc.lib,lib:amiga.lib

in THINK C.
  The siod project must include siod.c,slib.c,siod_m.c, Mactraps, ansi, unix.


[System:]

The interrupts called SIGINT and SIGFPE by the C runtime system are
handled by invoking the lisp error procedure. SIGINT is usually caused
by the CONTROL-C character and SIGFPE by floating point overflow or underflow.

[Syntax:]

The only special characters are the parenthesis and single quote.
Everything else, besides whitespace of course, will make up a regular token.
These tokens are either symbols or numbers depending on what they look like.
Dotted-list notation is not supported on input, only on output.

[Special forms:]

The CAR of a list is evaluated first, if the value is a SUBR of type 9 or 10
then it is a special form.

(define symbol value) is presently like (set! symbol value).

(define (f . arglist) . body) ==> (define f (lambda arglist . body))

(lambda arglist . body) Returns a closure.

(if pred val1 val2) If pred evaluates to () then val2 is evaluated else val1.

(begin . body) Each form in body is evaluated with the result of the last
returned.

(set! symbol value) Evaluates value and sets the local or global value of
the symbol.

(or x1 x2 x3 ...) Returns the first Xn such that Xn evaluated non-().

(and x1 x2 x3 ...) Keeps evaluating Xj until one returns (), or Xn.

(quote form). Input syntax 'form, returns form without evaluation.

(let pairlist . body) Each element in pairlist is (variable value).
Evaluates each value then sets of new bindings for each of the variables,
then evaluates the body like the body of a progn. This is actually
implemented as a macro turning into a let-internal form.

(the-environment) Returns the current lexical environment.

[Macro Special forms:]

If the CAR of a list evaluates to a symbol then the value of that symbol
is called on a single argument, the original form. The result of this
application is a new form which is recursively evaluated.

[Built-In functions:]

These are all SUBR's of type 4,5,6,7, taking from 0 to 3 arguments
with extra arguments ignored, (not even evaluated!) and arguments not
given defaulting to (). SUBR's of type 8 are lexprs, receiving a list
of arguments. Order of evaluation of arguments will depend on the
implementation choice of your system C compiler.

consp cons car cdr set-car! set-cdr!

number? + - * / < > eqv?
The arithmetic functions all take two arguments.

eq?, pointer objective identity. (Use eqv? for numbers.)

symbolconc, takes symbols as arguments and appends them.

symbol?

symbol-bound? takes an optional environment structure.
symbol-value also takes optional env.
set-symbol-value also takes optional env.

env-lookup takes a symbol and an environment structure. If it returns
non-nil the CAR will be the value of the symbol.

assq

read,print

eval, takes a second argument, an environment.

copy-list. Copies the top level conses in a list.

oblist, returns a copy of the list of the symbols that have been interned.

gc-status, prints out the status of garbage collection services, the
number of cells allocated and the number of cells free. If given
a () argument turns gc services off, if non-() turns gc services on.
In mark-and-sweep storage management mode the argument only turns on
and off verbosity of GC messages.

gc, does a mark-and-sweep garbage collection. If called with argument nil
does not print gc messages during the gc.

load, given a filename (which must be a symbol, there are no strings)
will read/eval all the forms in that file. An optional second argument,
if T causes returning of the forms in the file instead of evaluating them.

save-forms, given a filename and a list of forms, prints the forms to the
file. 3rd argument is optional, 'a to open the file in append mode.

quit, will exit back to the operating system.

error, takes a symbol as its first argument, prints the pname of this
as an error message. The second argument (optional) is an offensive
object. The global variable errobj gets set to this object for later
observation.

null?, not. are the same thing.

*catch tag exp, Sets up a dynamic catch frame using tag. Then evaluates exp.

*throw tag value, finds the nearest *catch with an EQ tag, and cause it to
return value.

[Procedures in main program siod.c]

edit is a VMS specific function that takes a single filename argument
and calls the EDT editor (as a procedure) to edit the file.

cfib is the same as standard-fib. You can time it and compare it with
standard-fib to get an idea of the overhead of interpretation.

[Utility procedures in siod.scm:]

Shows how to define macros.

cadr,caddr,cdddr,replace,list.

(defvar variable default-value)

And for us old maclisp hackers, setq and defun, and progn, etc.

call-with-current-continuation

Implemented in terms of *catch and *throw. So upward continuations
are not allowed.

A simple backquote (quasi-quote) implementation.

cond, a macro.

append

nconc

[A streams implementation:]

The first thing we must do is decide how to represent a stream.
There is only one reasonable data structure available to us, the list.
So we might use (<stream-car> <cache-flag> <cdr-cache> <cdr-procedure>)

the-empty-stream is just ().

empty-stream?

head

tail

cons-stream is a special form. Wraps a lambda around the second argument.

*cons-stream is the low-level constructor used by cons-stream.

[Benchmarks:]

A standard-fib procedure is included in siod.scm so that everyone will
use the same definition in any reports of speed. Make sure the return
result is correct. use command line argument of
 %siod -h100000 -isiod.scm

(standard-fib 10) => 55 ; 795 cons work.
(standard-fib 15) => 610 ; 8877 cons work.
(standard-fib 20) => 6765 ; 98508 cons work.

[Porting:]

The only code under #ifdef is the definition of myruntime, which
should be defined to return a double float, the number of cpu seconds
used by the process so far. It uses the the tms_utime slot, and assumes
a clock cycle of 1/60'th of a second.

The stack and register marking code used in the mark-and-sweep GC
is unlikely to work on machines that do not keep the procedure call
stack in main memory at all times. It is assumed that setjmp saves
all registers into the jmp_buff data structure.

There is a bit of type casting in close_open_files and vload. The
pname of an un-interned symbol is used as a pointer to FILE. This
saves the code (a conser, a print case, and two gc cases) of defining
a new data type for keeping track of binary data.

If the stack is not always aligned (in LISP-PTR sense) then the
gc_mark_and_sweep procedure will not work properly.

Example, assuming a byte addressed 32-bit pointer machine:

stack_start_ptr: [LISP-PTR(4)]
                 [LISP-PTR(4)]
                 [RANDOM(4)]
                 [RANDOM(2)]
                 [LISP-PTR(4)]
                 [LISP-PTR(4)]
                 [RANDOM(2)]
                 [LISP-PTR(4)]
                 [LISP-PTR(4)]
stack_end:       [LISP-PTR(4)]

As mark_locations goes from start to end it will get off proper alignment
somewhere in the middle, and therefore the stack marking operation will
not properly identify some valid lisp pointers.

Fortunately there is an easy fix to this. A more aggressive use of
our mark_locations procedure will suffice.

For example, say that there might be 2-byte random objects inserted into
the stack. Then use two calls to mark_locations:

 mark_locations(((char *)stack_start_ptr) + 0,((char *)&stack_end) + 0);
 mark_locations(((char *)stack_start_ptr) + 2,((char *)&stack_end) + 2);

If we think there might be 1-byte random objects, then 4 calls are required:

 mark_locations(((char *)stack_start_ptr) + 0,((char *)&stack_end) + 0);
 mark_locations(((char *)stack_start_ptr) + 1,((char *)&stack_end) + 1);
 mark_locations(((char *)stack_start_ptr) + 2,((char *)&stack_end) + 2);
 mark_locations(((char *)stack_start_ptr) + 3,((char *)&stack_end) + 3);


[Interface to other programs:]

If your main program does not want to actually have a read/eval/print
loop, and instead wants to do something else entirely, then use
the routine set_repl_hooks to set up for procedures for:

 * putting the prompt "> " and other info strings to standard output.

 * reading (getting) an expression

 * evaluating an expression

 * printing an expression.

The routine get_eof_val may be called inside your reading procedure
to return a value that will cause exit from the read/eval/print loop.

In order to call a single C function in the context of the repl loop,
you can do the following:

int flag = 0;

void my_puts(st)
 char *st;
{}

LISP my_reader()
{if (flag == 1)
  return(get_eof_val());
 flag == 1;
 return(NIL);}

LISP my_eval(x)
 LISP x;
{call_my_c_function();
 return(NIL);}

LISP my_print(x)
 LISP x;
{}

do_my_c_function()
{set_repl_hooks(my_puts,my_read,my_eval,my_print);
 repl_driver(1, /* or 0 if we do not want lisp's SIGINT handler */
             0);}


If you need a completely different read-eval-print-loop, for example
one based in X-Window procedures such as XtAddInput, then you may want to
have your own input-scanner and utilize a read-from-string kind of
function.


[User Type Extension:]

There are 5 user types currently available. tc_user_1 through tc_user_5.
If you use them then you must at least tell the garbage collector about
them. To do this you must have 4 functions,
 * a user_relocate, takes a object and returns a new copy.
 * a user_scan, takes an object and calls relocate on its subparts.
 * a user_mark, takes an object and calls gc_mark on its subparts or
                it may return one of these to avoid stack growth.
 * a user_free, takes an object to hack before it gets onto the freelist.

set_gc_hooks(user_relocate_fcn,
             user_scan_fcn,
             user_mark_fcn,
             user_free_fcn,
             &kind_of_gc);

kind_of_gc should be a long. It will receive 0 for mark-and-sweep, 1 for
stop-and-copy. Therefore set_gc_hooks should be called AFTER process_cla.
You must specify a relocate function with stop-and-copy. The scan
function may be NULL if your user types will not have lisp objects in them.
Under mark-and-sweep the mark function is required but the free function
may be NULL.

See SIOD.C for a very simple string-append implementation example.

You might also want to extend the printer. This is optional.

set_print_hooks(fcn);

The fcn receives the object which should be printed to its second
argument, a FILE*.

The evaluator may also be extended, with the "application" of user defined
types following in the manner of an MSUBR.

Lastly there is a simple read macro facility.

set_read_hooks("All-Readmacros",
               "Token-Ending-Readmacros",
                read_macro_handling_fcn);

The read_macro_handling_fcn will receive the character used to trigger
it and the struct gen_readio * being read from. It should return a lisp object.