perm filename MCMAN3.S78[206,LSP]1 blob
sn#345121 filedate 1978-03-31 generic text, type T, neo UTF8
**DRAFT** The System **DRAFT**
Part 3 - The System
Table of Contents
1. The System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-1
1.1 The Top Level Function . . . . . . . . . . . . . . . . . . . . . . . 3-1
1.2 Breakpoints . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-5
1.3 Control Characters . . . . . . . . . . . . . . . . . . . . . . . . . 3-9
1.4 Exceptional Condition Handling . . . . . . . . . . . . . . . . . . .3-15
1.4.1 The LISP Error System . . . . . . . . . . . . . . . . . . . . . . .3-15
1.4.2 User Interrupts . . . . . . . . . . . . . . . . . . . . . . . . . .3-16
1.4.3 Catalogue of User Interrupt Channels . . . . . . . . . . . . . . . .3-19
1.4.4 Autoload . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-26
1.5 Debugging . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-28
1.5.1 Binding, Pdl Pointers, and the Evaluator . . . . . . . . . . . . . 3-28
1.5.2 Functions for Debugging . . . . . . . . . . . . . . . . . . . . . .3-28
1.5.3 The Trace Package . . . . . . . . . . . . . . . . . . . . . . . . .3-35
1.5.4 The Stepper . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-40
1.5.5 The MAR Break Feature . . . . . . . . . . . . . . . . . . . . . . .3-56
1.6 Storage Management . . . . . . . . . . . . . . . . . . . . . . . . .3-60
1.6.1 Garbage Collection . . . . . . . . . . . . . . . . . . . . . . . . .3-60
1.6.2 Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-61
1.6.3 Storage Control Functions . . . . . . . . . . . . . . . . . . . . .3-64
1.6.4 Dynamic Space and Pdl Expansion . . . . . . . . . . . . . . . . . .3-65
1.6.5 Initial Allocation . . . . . . . . . . . . . . . . . . . . . . . . .3-66
1.7 Implementing Subsystems with Maclisp . . . . . . . . . . . . . . . .3-69
1.7.1 Entering LISP . . . . . . . . . . . . . . . . . . . . . . . . . . .3-69
1.7.2 Saving an Environment . . . . . . . . . . . . . . . . . . . . . . .3-70
1.7.3 Gaining and Keeping Control . . . . . . . . . . . . . . . . . . . .3-73
1.7.4 Purity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3-74
1.8 Miscellaneous Functions . . . . . . . . . . . . . . . . . . . . . .3-79
1.8.1 The Status Functions . . . . . . . . . . . . . . . . . . . . . . . .3-79
1.8.2 Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-100
1.8.3 Escaping from LISP . . . . . . . . . . . . . . . . . . . . . . . . 3-101
1.8.4 Additional Functions . . . . . . . . . . . . . . . . . . . . . . . 3-103
December 6, 1977 Page 3-1
**DRAFT** The System **DRAFT**
1. The System
1.1 The Top Level Function
The following function is an approximation to what Maclisp does when it is at
its "top level."
December 6, 1977 ∪3-1. Page 3-1
**DRAFT** Maclisp Reference Manual **DRAFT**
(defun standard-top-level nil
(prog (↑q ↑w ↑r evalhook base ibase ... )
errors ;errors, uncaught throws, etc. come here
↑g ;ctrl/G quits come here
(reset-bound-vars-and-restore-pdls)
(setq ↑q nil)
(setq ↑w nil)
(setq evalhook nil)
(nointerrupt nil)
(do-delayed-tty-and-alarmclock-interrupts)
;Recall that errors do (setq // errlist) so lambda-binding
; errlist will work properly. See errlist.
(mapc (function eval) //)
(or (status linmode)(terpri))
(do ((eof (list nil)) ;internal variables
(prt '* *))
(nil) ;do forever (until ↑g or error)
(setq * (cond ((status toplevel)
(eval (status toplevel)))
(t (terpri)
(cond (prin1 (funcall prin1 prt))
(t (prin1 prt)))
(typ 40)
(setq (do ((form))(nil)
(setq form
(cond (read
(funcall read
eof))
(t (read eof))))
(or (eq form eof)
(return form))
(terpri)))
(and (null read)
(atom -)
(is-a-space (tyipeek))(tyi))
((lambda (+) (eval -))
(prog2 nil + (setq + -)))))))))
which causes a "read-eval-print loop," i.e. each S-expression that is typed in
gets evaluated and the value is printed, then the next S-expression is read.
Errors and ↑g quit to top level. That is they reinitialize and then re-enter
Page 3-2 ∪3-1.1 December 6, 1977
**DRAFT** The System **DRAFT**
this loop, printing a * (but not destroying the value of the variable *).
Notice that there is a place in the middle where the user can insert his own
special form to be evaluated, using (sstatus toplevel). It is also possible to
change just the reader or just the printer by setq'ing read or prin1. See the
sstatus function (page 3-79).
Variables used by the top-level read-eval-print loop:
* VARIABLE
Contains the last S-expression printed out by the read-eval-print loop,
that is, the value of the last form typed in. This is true even after an
error return to top level, allowing one to refer to the value printed out
before the aborted computation.
+ VARIABLE
Contains the last S-expression typed in. This can be used to edit it or to
do it over again. (Notice how + is bound in the read-eval-print loop.
This causes + to receive the correct value even if the evaluation aborts,
since an error or ↑g quit will undo the binding.)
- VARIABLE
Contains the current S-expression typed in. This can be used by user-
written error handlers. It can't be usefully accessed by expressions typed
in, since it is set before the expression is evaluated.
By special arrangement the values of +, *, and - are preserved across a
break. When the break is first entered these have the values for the last top-
level operation, during the break they behave the same as at top level, and
after the break returns they are restored to the values for the top level loop.
(See break, page 3-5).
/ VARIABLE
/ is used to temporarily hold the value of errlist when an error returns to
top level. This is so that lambda-binding errlist will have an effect
(assuming no one lambda-binds /). Note that / must be typed in as // since
the slash character is special to the LISP reader.
December 6, 1977 ∪3-1.1 Page 3-3
**DRAFT** Maclisp Reference Manual **DRAFT**
errlist VARIABLE
The value of errlist is a list of forms which are evaluated when control
returns to top level either because of an error or when an environment is
initially started. It doesn't apply if the environment started up was
saved using (suspend). This feature is used to provide special error
handling for subsystems written in LISP.
The symbol errlist is evaluated to get the list of forms in the binding
context in which the error occurred, but the forms themselves are evaluated
in the top-level binding context.
Example:
((lambda (errlist)
(putprop 'foo 'bar 'baz)
(hack)
(remprop 'foo 'baz))
(cons '(remprop 'foo 'baz)
errlist))
The property list of foo will be properly restored even if the computation
(hack) is aborted.
Page 3-4 ∪3-1.1 December 6, 1977
**DRAFT** The System **DRAFT**
1.2 Breakpoints
Breakpoints are a mechanism to allow the user to gain control at any point in
a program. Use of the function break causes a read-eval-print loop, similar to
the one at top level, to be entered. (This is also called a break loop.) The
user may evaluate any S-expressions, inspect the bindings of variables, and exit
from the break in several ways. Normal execution then proceeds. (See , , and
page 3-31)
This mechanism can be used to permit human intervention when an unexpected
condition occurs. It is used in this way by the Maclisp error system. See the
section on Exceptional Condition Handling, page 3-15. A break loop makes the
full power of the LISP interpreter available for debugging.
break FSUBR
(break tag pred) evaluates pred, but not tag. If the value of pred is not
nil, the state of the I/O system is saved, ";bkpt tag" is typed out, and
control returns to the terminal. We say that a "break loop" has been
entered. tag may be any object. It is used only as a message typed out
(using princ) to identify the break. It is not evaluated. If pred is
omitted, t is assumed. Thus (break tag) is equivalent to (break tag t).
(break tag nil) returns nil, and produces no action whatsoever.
A break loop is a read-eval-print loop similar to top level. break does an
errset so that errors cannot cause an abnormal return from the break. A ↑x
quit, which causes an ordinary error, will thus return to the break loop if
used to interrupt a computation started in the break loop. A ↑g quit,
however, returns back to LISP top level, resetting the environment using
the errlist, as described above.
Two forms, $P and (return x), may be typed in a break loop. If $P is typed
in, break returns nil and execution continues. This "$P" is <dollar> P in
the Multics implementation, but <altmode> P in the PDP-10 implementations,
followed of course in either case by a <space> or <newline> as appropriate
(see (status linmode)). (An atom other than $P can be used to perform this
function by changing the value of $P to another (non-nil) atom. The
initial value of $P is always $P).
If (return x) is typed in, break evaluates x and returns that value. If as
December 6, 1977 ∪3-1.2 Page 3-5
**DRAFT** Maclisp Reference Manual **DRAFT**
a result of the evaluation of a typed-in form, (throw x break) is
evaluated, break returns x as its value. (Notice the distinction -
executing a form (return x) does not return from break unless it was typed
directly at the break loop.) See return, .
When break returns, the state of the I/O system is restored.
An approximate LISP definition of what break does follows. Note that the
user program can modify this by using (sstatus breaklevel).
Page 3-6 ∪3-1.2 December 6, 1977
**DRAFT** The System **DRAFT**
(defun break fexpr (x)
(*break (eval (cadr x)) (car x)) ;note argument reversal
(declare (special ↑q ↑w evalhook * + -))
(defun *break (breakp breakid)
(and breakp
(do ((↑q nil) (↑w nil) (evalhook nil) (terpri t) (* *) (+ +) (- -))
() ;bind key variables
(terpri msgfiles) ;msgfiles arguments
(princ '|;bkpt | msgfiles) ; used for
(princ breakid msgfiles) ; Newio only
(terpri msgfiles)
(setq + -) ;last form typed
(return
(prog2 nil
(catch
(do () (nil) ;do forever (until throw)
(errset
(do ((eof (list nil)) (form))
(nil)
(cond ((status breaklevel)
(eval (status breaklevel)))
(t (setq form (cond (read (funcall read eof))
(t (read eof))))
(and (null read)
(atom form)
(is-a-space (tyipeek))
(tyi))
(cond ((eq form eof) (terpri))
((and $P (eq form '$P)) (throw nil break))
((eq (car form) 'return)
(throw (eval (cadr form)) break))
(t (setq - form)
(print
(setq * ((lambda (+) (eval form))
(prog2 nil + (setq + -)))))
(terpri))))))))
break)
(terpri)
))))
December 6, 1977 ∪3-1.2 Page 3-7
**DRAFT** Maclisp Reference Manual **DRAFT**
The arguments to break are a breakpoint identification and (optionally) a
break switch. If the break switch evaluates to nil, then nil is returned.
Otherwise, the variables ↑q, ↑w, evalhook and terpri are bound to nil, the
variables *, +, and - are bound to their current values, and the message ";bkpt
<breakid>" is printed. A read-eval-print loop similar to the top level loop is
then entered. This break loop is surrounded by an errset. Errors or typing ↑x
merely cause the break loop to be re-entered. The value of (status breaklevel)
serves a function similar to that of (status toplevel) in the top level loop.
As each form is read in the default break loop, there are four cases:
1. End of file. For console input this merely indicates rubout beyond the
number of input characters. Whether input is from console or elsewhere,
the (terpri) is done and the reader is entered.
2. The form is the atom $P or eq to the non-nil value of $P. nil is
returned from the break.
3. The form is (return value). The form value is evaluated and returned
from the break.
4. Otherwise the form is evaluated and the result printed out in a manner
analogous to the top level read-eval-print loop. The variables +, -,
and * are updated appropriately. (Recall, however, that they were bound
on entry to *break, and so will be restored eventually.)
The way to return from a break is to do a throw with a tag of break; this
will return from the catch which surrounds the break loop. This is how cases 2
and 3 return their values; case 4 may also cause a return from the break.
Page 3-8 ∪3-1.2 December 6, 1977
**DRAFT** The System **DRAFT**
1.3 Control Characters
LISP can be directed to take certain actions by entering "control characters"
from the terminal. The difference between control characters and normal input
is that control characters take effect as soon as they are entered, while normal
input only takes effect when LISP asks for it, by use of functions such as read,
or by being in the top level read-eval-print loop or in a break loop.
Control characters can be typed in from the terminal according to some
procedure that depends on the implementation. A program can mimic the effects
of the various control characters by directly calling the function associated
with the particular control key (see below).
Although control characters are usually processed as soon as they are typed,
they may be delayed if there is a garbage collection in progress or LISP is in
(nointerrupt tty) mode - see the nointerrupt function (page 3-19).
Entering Control Characters in ITS LISP
In the ITS implementation of Maclisp, control characters are entered by means
of the "CTRL" key on the terminal. For example, CTRL/g is entered by holding
down "CTRL" and striking the "G" key. Control characters normally echo as an
uparrow or circumflex followed by the character.
In Newio, any character at all may be made an interrupt character. See
(sstatus tty) and (sstatus ttyint).
Interrupt characters are also read as part of the terminal input stream.
Normally they are marked in the readtable as "worthless" characters.
Entering Control Characters in TOPS-10 LISP
Most control characters may be entered in the same way as in ITS LISP if LISP
is currently read'ing from the terminal. If a LISP program is actively running,
it is necessary to first gain its attention by typing the CTRL/c character one
or two times, thereby returning to the monitor. The monitor command REENTER may
then be used to re-enter the LISP. LISP will print "?↑" and read a character,
which may be a control character or a character whose "control" meaning is to be
used. Thus typing either ↑g or "G" will cause the ↑g interrupt to occur. If
LISP is not ready to take an interrupt, there may be a delay before the "?↑" is
printed.
December 6, 1977 ∪3-1.3 Page 3-9
**DRAFT** Maclisp Reference Manual **DRAFT**
Entering Control Characters in Multics LISP
In the Multics implementation of Maclisp, one signals one's desire to enter a
"control" character by hitting the "attention" key on the terminal. (This is
called "break," "interrupt," "attn," "quit," etc. on different terminals. If
Multics is being accessed through the ARPA network, an "interrupt process" (@S S
or @S I P from a TIP) signal should be transmitted.) LISP responds by typing out
"CTRL/". Now you may type one letter from the list later in this section, which
will be interpreted to have its "control" meaning. This control character must
be followed by a newline.
It is also possible to enter "control" characters from an input character
stream, which may have its source at the terminal or in an exec←com, without the
use of the "attention" key. The desired control character is prefixed by a \036
character. If two of these prefix characters occur together, one \036 character
is read and no "control" action is performed. Otherwise, the character
following the \036 is processed as a control character, then reading continues.
Control characters will be accepted in upper or lower case. All characters
other than those with defined meanings are rejected with an error message. Only
one control character may be entered at a time.
Example for Multics LISP:
(lines containing user input are preceded by ">>>")
>>> (defun loop (x) (loop (add1 x)))
loop
>>> (loop 0)
function runs for a long time,
>>> <ATTN> then user hits attention button.
>>> CTRL/B LISP types "CTRL/", user types "B"
>>> ;bkpt ↑b system enters break loop
>>> x user looks at value of x
4067
>>> <ATTN> user hits attention button again
>>> CTRL/G and returns to top level
Quit
*
When a "user interrupt" is caused, if the interrupt is not enabled nothing
happens. If the interrupt is enabled, then a user-specified function is called.
The interrupt may be enabled by binding the appropriate symbol to the function
to handle it, or by using the (sstatus ttyint) function (page 3-80).
Page 3-10 ∪3-1.3 December 6, 1977
**DRAFT** The System **DRAFT**
In the following descriptions of control characters, those which can always
be processed immediately, even during a garbage collection or in (nointerrupt
'tty) mode, will be indicated by an "!" for the ITS and TOPS-10 implementations,
and an "*" for the Multics Implemention. When appropriate, equivalent LISP code
is given for producing the same result from a user program.
Control Characters that have initially defined meanings in all implementations:
B !* runs a break loop with breakid "↑b" (see break, page 3-5). (In the
PDP-10 Oldio implementation ↑h is used instead of ↑b.) (break ↑b).
C !* sets the value of the atom ↑d to nil, turning off garbage collector
messages. Because ↑c is trapped by DEC-10 monitors, this control
character can only be typed using the REENTER mechanism and typing
"C". (setq ↑d nil).
D !* sets the value of the atom ↑d to t, turning on garbage collector
messages. (setq ↑d t)
G quits back to the top level of LISP, rebinding all variables to their
global values, resetting various system variables, and evaluating the
errlist forms. This is used to stop a running program when there is
no intention of restarting it again. (Prints out a *; see the "top
level" function and the ↑g function.) H is used instead of ↑b in some
implementations (see above).
Q ! sets the value of the atom ↑q to t, enabling input from the source
selected by the value of infile, or selected by use of the function
uread. In the PDP-10 Newio implementation, this is not an interrupt
character; instead it is a macro character, and takes effect only if
processed by read. (setq ↑q t).
R ! sets the value of the atom ↑r to t, enabling output to the
destinations selected by the value of outfiles, or selected by use of
the uwrite function. (setq ↑r t).
S ! turns off typeout until input is read. This is used to suppress the
rest of the typeout from the current request, without affecting
typeout from the next request that is typed in. It is implemented by
setting ↑w to t, then putting a macro character in the input stream
which sets ↑w to nil and does a (terpri) when it is read.
December 6, 1977 ∪3-1.3 Page 3-11
**DRAFT** Maclisp Reference Manual **DRAFT**
T ! sets the value of the atom ↑r to nil, disabling output to the
destinations that CTRL/r enables. (setq ↑r nil).
U causes the current call to read to be restarted from the beginning.
(Not available in PDP-10 implementations).
V ! sets the value of the atom ↑w to nil, enabling output to the terminal.
(setq ↑w nil).
W ! sets the value of the atom ↑w to t, disabling output to the terminal.
(setq ↑w t) (and possible also (clear-output tyo)).
X causes an error which can be caught by errset. This is a less drastic
"quit" than CTRL/g. If it is typed within a break loop, it will
return no further than the break loop, since break uses errset.
(error 'quit).
Z !* On ITS returns to ITS command level, i.e. DDT. On Multics returns to
Multics command level. (start re-enters LISP.) On TOPS-10 goes to DDT
if a DDT has been loaded with LISP.
The following control characters only exist in the Multics implementation.
. !* does nothing, and is used merely to speed up a slow process by causing
an interaction.
? !* asks the LISP subsystem what it is doing: running, waiting for input,
collecting garbage, or running with tty-interrupts masked off.
The following control characters only exist in PDP-10 implementations with
the "moby I/O" capability. (For now, this means only at the MIT AI Laboratory.)
(see (sstatus ttyscan), page 3-82).
F Cause graphics display slave to seize a display.
N Turn on display for character output.
O Turn off display for character output.
Y Cause display slave to release display.
The following control characters only work in the PDP-10 implementation.
Page 3-12 ∪3-1.3 December 6, 1977
**DRAFT** The System **DRAFT**
They are not really interrupts, but occur only when processed by the terminal
input prescanner (see (sstatus ttyscan), page 3-82).
K redisplay the current input. This allows you to get a clean copy of
your input after rubouts have been used. (On a display console, LISP
will attempt to make rubbed-out characters actually disappear from the
screen. ↑k is still useful if a timing error disrupts this process.)
L erases the screen if the terminal is a display, then does a CTRL/k.
Control-Character Functions
↑g SUBR 0 args
Produces a quit to top level just as if a CTRL/g had been typed.
These functions exist only in the PDP-10 Oldio and the Multics
implementations. They are being phased out, so using them in new programs is
not recommended.
ioc FSUBR
The argument to ioc is processed as if it were a "control character" that
had been typed in. Numbers are taken as a whole, atomic symbols' pnames
are processed character by character, except that nil is ignored.
Examples:
(ioc 1) causes user interrupt 1.
(ioc vt) switches output to the terminal.
(ioc q) switches input to a file.
(ioc g) quits back to the top level of LISP.
If ioc returns, its value is t.
December 6, 1977 ∪3-1.3 Page 3-13
**DRAFT** Maclisp Reference Manual **DRAFT**
iog FSUBR
iog first saves the values of the I/O switches ↑q, ↑r, and ↑w. Then it
processes its first argument the same as ioc. Next the remaining arguments
to iog are evaluated, from left to right. The values of the variables ↑q,
↑r, and ↑w are restored, and the value of the last argument is returned.
Example:
(iog vt (princ "A Message."))
gets a message to the console no matter what the I/O system is doing. It
evaluates to "A Message."
(iog a x1 x2 ... xn)
can also be written
((lambda (↑q ↑r ↑w)
(ioc a)
x1
x2
...
xn)
nil nil nil)
Page 3-14 ∪3-1.3 December 6, 1977
**DRAFT** The System **DRAFT**
1.4 Exceptional Condition Handling
1.4.1 The LISP Error System
The errors detected by the LISP system are divided into two types:
correctable and uncorrectable. The uncorrectable errors will be explained first
since they are simpler.
An uncorrectable error is an error that causes the evaluation in which it
occurs to be aborted. When an uncorrectable error occurs, the first thing that
happens is the printing of an error message. In Oldio, the error message goes
to the terminal and nowhere else (unless suppressed, see errset, page 3-21), no
matter how the I/O switches and variables are set. In Newio, the variable
msgfiles is a list of the files to which error messages should be routed; this
initially is just the terminal. The error message consists of some explanatory
text and (usually) the object or form that caused the error.
After the error message has been printed, control is returned to the most
recent error-catcher. There is an error-catcher at top level, and error-
catchers are set up by the function errset (and break, which uses errset). All
variable bindings between the error-catcher and the point where the error
occurred are restored. Thus all variables are restored to the values they had
at top level or at the time the errset was done, unless they were setq'ed free
(without being bound).
What happens next depends on how the error-catcher was set up. At top level,
the forms on the errlist are evaluated and the top level loop (or a user
specified top level form) is re-entered. (The symbol errlist is evaluated prior
to the above restoration of bindings, and saved in the variable /. In this way
the errlist used is the one current at the time of the error, despite the
restoration of bindings.) If an error returns to break, it simply re-enters its
read-eval-print loop. In the Multics implementation the fact that break has
caught an error is signalled by ringing the bell on the terminal. If an error
returns to errset, errset returns nil and evaluation proceeds. If an error
returns to top level, the state of the world is reset and * is typed.
The above description is slightly simplified. The user can request an
interrupt to occur between the typing of the message, and the unwinding of
bindings and return of control to an error-catcher. If the error is going to
December 6, 1977 ∪3-1.4 Page 3-15
**DRAFT** Maclisp Reference Manual **DRAFT**
return to top level, the *rset-trap user interrupt is signalled. This user
interrupt is initially a system-supplied break loop which allows the user to
examine the values of variables before the bindings are restored, in hope of
finding the cause of the error. In (*rset t) mode a break loop is entered, but
in (*rset nil) mode the user interrupt is ignored. If the error is going to
return to a break or an errset, and *rset is non-nil, the errset user interrupt
is signalled. The initial environment contains a null handler for this
interrupt, but the user may supply a break loop or other handler.
Correctable errors are errors which may be corrected by user intervention.
If such an error is properly corrected, evaluation will proceed as if no error
had occurred. If the option to correct the error is not exercised, this type of
error will be handled as if it were an uncorrectable error.
When a correctable error occurs, a user interrupt is signalled. See page 3-
19 for user interrupt channel assignments for these errors. The initial
environment contains handlers for these errors which print an error message
similar to the message printed for an uncorrectable error and then enter a break
loop.
The argument passed to the user interrupt handler is an S-expression
describing the error. See section 1.4.2 for details. If the user interrupt
handler is nil, or if it returns a non-list, the error is treated as an
uncorrectable error. But if the handler returns a list, the first element of
that list is used to correct the error in a way which depends on the particular
error which occurred.
If the most recent error-catcher is not top-level, correctable errors will be
treated as uncorrectable errors unless there is a non-null handler for the
errset interrupt. This is to prevent confusing "multiply nested" error breaks
unless the user indicates that he is sophisticated by setting up a handler for
the errset interrupt. (The errset handler itself is only invoked if *rset is
non-nil, however.)
See the functions error, err, and errset.
1.4.2 User Interrupts
LISP provides a number of "user interrupts," which are a mechanism by which a
user procedure may temporarily gain control when an exceptional condition
Page 3-16 ∪3-1.4.1 December 6, 1977
**DRAFT** The System **DRAFT**
happens. The exceptional conditions that use the user interrupt system include
certain control characters, the alarmclock timers, asynchronous I/O conditions,
the garbage collector, and many of the errors that are detected by the
interpreter or by the system functions. Errors detected by user functions can
use this mechanism also.
The user interrupts are divided up into several channels. Each channel has
associated with it a service function. If the service function is nil,
interrupts on that channel will be ignored. If the service function is not nil,
it is a function which is called with one argument when the user-interrupt
occurs. (A few interrupt handlers take more than one argument. See the
specific descriptions.) The nature of the argument depends on which channel the
interrupt is on; usually it is an S-expression which can be used to localize the
cause of the interrupt. Some user interrupts use the value returned by the
service function to decide what to do about the cause of the interrupt.
Interrupts can be either synchronous (e.g. errors and garbage collector
interrupts) or asynchronous (control characters, alarmclock, etc.). To prevent
timing errors, asynchronous interrupts are always run in (nointerrupt t) mode.
A handler for an asynchronous interrupt must explicitly do (nointerrupt nil) to
permit other asynchronous errors to interrupt it. (For example, the system-
supplied ↑b handler does this so that control character interrupts can be used
within the ↑b break loop.)
The service functions for most user interrupts are kept as the values of
symbols with mnemonic names. A list of these symbols begins on page 3-19.
There are also user interrupts for control characters. The service functions
for these are declared using (sstatus ttyint). See page 3-80.
The initial values for the service functions of the various interrupts are
provided by the system as break loops for some interrupts and nil for others
(except for some control characters).
There are some special considerations for user interrupts signalled by
correctable error conditions. The argument to the service function is a
description of the error whose exact form is described in the catalogue at the
end of this section. If the service function returns nil (or any other atom),
the normal error procedure occurs -- control returns to the most recent errset,
or to top level if there was no errset. If the service function returns a list,
the first element of the list is used to attempt recovery from the error. The
exact way that it is used is described in the catalogue. If recovery is
successful execution proceeds from the point where the error occurred. If
recovery is unsuccessful another error is signalled.
December 6, 1977 ∪3-1.4.2 Page 3-17
**DRAFT** Maclisp Reference Manual **DRAFT**
Here is an example of a user interrupt service function. This is the one
supplied by the system for unbound variable errors when the user does not
specify one. Note that the system-supplied error service functions consistently
bind args to the argument supplied. The user can check the value of this
variable to see what is wrong. Note too that the system-supplied error handlers
restore readtable and obarray before breaking.
(defun +internal-ubv-break (args)
(declare (special args))
(errprint nil msgfiles) ;print error message
((lambda (readtable obarray)
(nointerrupt nil)
(break unbnd-vrbl))
(get 'readtable 'array)
(get 'obarray 'array)))
(setq unbnd-vrbl '+internal-ubv-break)
alarmclock SUBR 2 args
alarmclock is a function for controlling timers. It can start and stop
two separate timers; one is a real-time timer (which counts seconds of
elapsed time) and the other is a cpu-time timer (which counts microseconds
of machine run time). The first argument to alarmclock indicates which
timer is being referred to: it may be the atom time to indicate the real-
time timer or the atom runtime to indicate the cpu-time timer.
The second argument to alarmclock controls what is done to the selected
timer. If it is a non-negative number (fixnum or flonum) the timer is
started. Thus if n is a positive fixnum or flonum, evaluating (alarmclock
'time n) sets the real-time timer to go off in n seconds, and (alarmclock
'runtime n) sets the cpu-time timer to go off in n microseconds. If the
timer was already running the old setting is lost. Thus at any given time
each timer can only be running for one alarm, but the two timers can run
simultaneously.
If the second argument to alarmclock is not a positive number, the timer
is shut off, so (alarmclock x nil) or (alarmclock x -1) shuts off the x
timer.
alarmclock returns t if it starts a timer, nil if it shuts it off.
Page 3-18 ∪3-1.4.2 December 6, 1977
**DRAFT** The System **DRAFT**
When a timer goes off, the alarmclock user interrupt occurs. The
service function is run in (nointerrupt t) mode so that it will not be
interrupted while it is performing its service. If it wants to allow
interrupts, other timers, etc. it can evaluate (nointerrupt nil). In any
case the status of the nointerrupt flag will be restored when the service
function returns. The argument passed to the user interrupt service
function is the atom time or the atom runtime, depending on which timer
went off. See also the function nointerrupt.
nointerrupt SUBR 1 arg
(nointerrupt t) shuts off LISP interrupts. This prevents alarmclock
timers from going off and prevents the use of control characters such as
CTRL/g and CTRL/b. Any of these interrupts that occur are simply saved.
(nointerrupt t) mode is used to protect critical code in large subsystems
written in LISP. A similar deferral technique is used by the LISP system
itself to protect against interrupts in the garbage collector.
(nointerrupt 'tty) prevents control characters (typed on the terminal,
or "tty") from causing interrupts; however, alarmclock interrupts (and
other asynchronous interrupts) are still allowed. Any non-tty asynchronous
interrupts which were saved will now go off.
(nointerrupt nil) turns interrupts back on. Any interrupts which were
saved will now get processed. This is the normal, initial state.
The result returned from nointerrupt is the previous interrupt status -
nil, t, or tty.
Example:
((lambda (oldstatus)
<protected code>
(nointerrupt oldstatus))
(nointerrupt t))
1.4.3 Catalogue of User Interrupt Channels
December 6, 1977 ∪3-1.4.2 Page 3-19
**DRAFT** Maclisp Reference Manual **DRAFT**
Each user interrupt channel (except some associated with files) has a
variable whose value is a functional form, the service function for that
channel. The name of the interrupt channel is the same as the name of the
variable. The following lists the user interrupt channels in alphabetical
order. The argument to which the service function is applied and the value
which it should return are described. By convention, almost all service
functions receive one argument. Some user interrupts are initially set to a
system-supplied handler which binds the variable args to this argument and
enters a break loop. The name of the interrupt is used as the break identifer.
Some user interrupts ignore the value returned by the service function, while
others distinguish two cases. If the value is atomic, the service function was
not able to recover from the condition that caused the interrupt. LISP will
take its default action, such as returning control to the most recent errset.
If the value is a list, the car of that list is used to recover from the
condition that caused the interrupt. It is usually a new piece of data to be
used in place of the one that was being complained about, or a new form to be
evaluated in place of the form that erred.
If the value of the service-function variable is nil instead of a functional
form, the user interrupt is considered to be turned off. The system behaves as
if the function had run and returned nil.
Some user interrupts are asynchronous in nature, and are executed in
(nointerrupt t) mode to prevent timing errors. The interrupt handler may choose
to run in (nointerrupt nil) mode, however, as the initial ↑b handler does. The
nointerrupt mode is restored after the handler is run. Such interrupts are
themselves deferred by (nointerrupt t) mode.
alarmclock VARIABLE
The value of alarmclock is the service function for the user interrupt signalled
when a timer set up by the alarmclock function goes off. The argument is the
name of the timer which went off, time or runtime. The returned value is
ignored. The service function is executed in (nointerrupt t) mode. This
interrupt is initially turned off.
autoload VARIABLE
The value of autoload is the service function for the user interrupt which
provides automatic loading of program packages into the environment. The
Page 3-20 ∪3-1.4.3 December 6, 1977
**DRAFT** The System **DRAFT**
argument is (function-name . autoload-property). The returned value is ignored.
See page 3-26 for details. This interrupt is initially set to a function which
simply loads a file.
cli-message VARIABLE
The value of cli-message service handler for the user interrupt signalled when
another job has interrupted the LISP job via the CLI device. The argument is
nil and the returned value is ignored. A user handler is expected to open a
file on the CLA device and read the message from the other job. The service
function is run in (nointerrupt t) mode. This interrupt is initially turned
off. Currently, it exists only in the ITS Newio implementation.
errset VARIABLE
The value of errset is the service function for the user interrupt which is
signalled when an error is caught by an errset and *rset is non-nil. The
argument is nil and the returned value is ignored. This user interrupt is
initially off. Turning it on affects the behavior of the error system (see page
3-16).
fail-act VARIABLE
The value of fail-act is the service function for the user interrupt which is
signalled when any of a large variety of miscellaneous error conditions occurs.
The argument is a list whose first element is generally a symbol which describes
the type of error condition. The rest of the list contains various objects
related to the error. The returned value depends on the error. These are not
standardized and will not be described here. This interrupt is initially set to
a break loop.
gc-daemon VARIABLE
The value of gc-daemon is the service function for the user interrupt which is
signalled after each garbage collection. The argument is a list of items; in
the PDP-10 implementation each item is of the form (space-name free-before
free-after size-before size-after) and in the Multics implementation, each item
is of the form (space-name free-before . free-after). The returned value is
ignored. This interrupt is initially turned off.
December 6, 1977 ∪3-1.4.3 Page 3-21
**DRAFT** Maclisp Reference Manual **DRAFT**
gc-lossage VARIABLE
The value of gc-lossage is the service function for the user interrupt which is
signalled when there is no more available address space or when the Time Sharing
Monitor rejects a request for more memory. In the Multics implementation, there
is always enough memory, so this user interrupt never occurs. In the PDP-10
implementation the argument is the name of the space that lost, and the returned
value is ignored. This interrupt is initially set to a break loop.
gc-overflow VARIABLE
The value of gc-overflow is the service function for the user interrupt which is
signalled when a space overflows its gcmax. (see alloc and (status gcmax).) The
argument is the name of the space. The returned value is ignored. This
interrupt is initially set to a break loop.
io-lossage VARIABLE
The value of io-lossage is the service function for the user interrupt which is
signalled when the I/O system encounters an error (for example, a file which was
being opened was not found). The argument is a list of the name of the function
which erred and its arguments, which may have been standardized or otherwise
partially digested. The returned value is a list of a new form to be evaluated
in place of the call to the function which erred. This interrupt is initially
set to a break loop.
machine-error VARIABLE
The value of machine-error is the service handler for the user interrupt
signalled when some difficulty is experienced by the host machine. The service
function receives four arguments instead of one. The first is an atomic symbol
indicating the type of error:
eval illegal machine operation
examine attempt to reference non-existent memory
deposit attempt to write into read-only memory
oddp parity error
The other three arguments are fixnums which are addresses of memory locations.
The second is the location of the error; the third is the program counter when
Page 3-22 ∪3-1.4.3 December 6, 1977
**DRAFT** The System **DRAFT**
the error occurred; and the fourth is the JPC (the program counter as of the
last jump instruction before the error occurred). The machine-error handler may
signal a different kind of error or a ↑g quit (see the ↑g function) if desired,
or enter a break loop. The subr function (see page 3-103) may be useful in
decoding the three fixnum arguments. If the handler returns, the value is
ignored and the erroneous operation is retried. If the user provides no
machine-error handler (the interrupt is initially turned off), the error is
handled in the default manner for the host machine. On ITS, this puts the user
in DDT. This currently exists only in the PDP-10 Newio implementation.
Example:
(defun machine-error-handler (type loc pc jpc)
((lambda (args terpri)
(declare (special args terpri))
(terpri msgfiles)
(cond ((eq type 'examine)
(princ '|;REFERENCE TO NON-EXISTENT
MEMORY| msgfiles))
((eq type 'deposit)
(princ '|;WRITE INTO READ-ONLY MEMORY|
msgfiles))
((eq type 'eval)
(princ '|;ILLEGAL MACHINE OPERATION|
msgfiles))
((eq type 'oddp)
(princ '|;MEMORY PARITY ERROR|
msgfiles))
(t (princ '|;UNKNOWN MACHINE ERROR|
msgfiles)))
(princ '| FROM LOCATION | msgfiles)
(princ pc)
(princ '| IN FUNCTION | msgfiles)
(prin1 (subr pc))
(break machine-error))
(list type loc pc jpc)
t))
mar-break VARIABLE
The value of mar-break is the service handler for the user interrupt signalled
when the memory location specified by (sstatus mar) (see page 3-97) has been
December 6, 1977 ∪3-1.4.3 Page 3-23
**DRAFT** Maclisp Reference Manual **DRAFT**
accessed in the specified manner. The argument is nil and the returned value is
ignored. The service function is run in (nointerrupt t) mode. Also, LISP
implicitly performs (sstatus mar 0 nil) before running the user interrupt; this
helps to prevent infinite loops. This interrupt is initially turned off. It
currently exists only in the ITS Newio implementation. See page 3-56 for more
information on using this interrupt.
pdl-overflow VARIABLE
The value of pdl-overflow is the service function for the user interrupt which
is signalled when a pushdown list exceeds its pdlmax. (see alloc and (status
pdlmax).) The argument is the spacename of the pushdown list. The returned
value is ignored. This interrupt is initially set to a break loop.
sys-death VARIABLE
The value of sys-death is the service handler for the user interrupt signalled
when the time-sharing system is about to go down, has been revived from that
state, or is being debugged. The argument is nil and the returned value is
ignored. A user handler may wish to examine the result of (status its) to
determine the state of the system. The service function is run in (nointerrupt
t) mode. This interrupt is initially turned off. This currently exists only in
the ITS Newio implementation.
tty-return VARIABLE
The value of tty-return is the service handler for the user interrupt signalled
when control of the terminal is returned to the LISP job by its superior. This
allows LISP to determine that the display screen may have been changed by other
jobs. The argument is nil and the returned value is ignored. The service
function is run in (nointerrupt t) mode. This interrupt is initially turned
off. This currently exists only in the ITS Newio implementation.
unbnd-vrbl VARIABLE
The value of unbnd-vrbl is the service function for the user interrupt which is
signalled when an attempt is made to evaluate an atomic symbol which does not
have a value (an unbound variable.) The argument is a list of the symbol which
Page 3-24 ∪3-1.4.3 December 6, 1977
**DRAFT** The System **DRAFT**
could not be evaluated. The returned value is a list of a new symbol to be
evaluated in its place. This interrupt is initially set to a break loop.
undf-fnctn VARIABLE
The value of undf-fnctn is the service function for the user interrupt which is
signalled when an attempt is made to apply an undefined function. The argument
is a list of the functional form which could not be applied. The returned value
is a list of a new functional form to take its place. This interrupt is
initially set to a break loop.
unseen-go-tag VARIABLE
The value of unseen-go-tag is the service function for the user interrupt which
is signalled when go or throw is used with a tag which does not exist in the
current prog body or in any catch, respectively. The argument is a list of the
erroneous tag. The returned value is a list of a new tag to replace it. This
interrupt is initially set to a break loop.
wrng-no-args VARIABLE
The value of wrng-no-args is the service function for the user interrupt which
is signalled when a function is called with the wrong number of arguments. The
argument is a list of two items: First, a list of the function and the arguments
that were passed. Second, the lambda-list if the function was interpreted, or
the same dotted pair as args returns if the function was compiled, or the atom ?
if this information could not be determined. The returned value is a list of a
new form to be evaluated in place of the losing one. This interrupt is
initially set to a break loop.
wrng-type-arg VARIABLE
The value of wrng-type-arg is the service function for the user interrupt which
is signalled when an argument is passed to a system function which is not
acceptable to that function. The argument is a list of the argument which was
not accepted. The returned value is a list of a new argument to replace it.
That is, directly an argument, not a form to be evaluated to get an argument.
This interrupt is initially set to a break loop.
December 6, 1977 ∪3-1.4.3 Page 3-25
**DRAFT** Maclisp Reference Manual **DRAFT**
*rset-trap VARIABLE
The value of *rset-trap is the service function for the user interrupt which is
signalled when an error returns control to top level, just before the bindings
are restored. By convention, the handler for this interrupt should not do
anything unless the variable *rset is non-nil. This is so that the user will
not be bothered unless he has put LISP in debugging mode. The argument is nil
and the returned value is ignored. This interrupt is initially set to a
function which enters a break loop if *rset is non-nil.
There are other interrupt handlers which are associated with I/O files or
inferior jobs. See eoffn, endpagefn, (sstatus ttyint), and create-job.
1.4.4 Autoload
The autoload feature provides the ability for a function not present in the
environment to be automatically loaded in from a file the first time it is
called. When eval, apply, funcall, or the version of apply used by compiled
LISP searches the property list of an atom looking for a functional property,
and fails to find one, it looks for a property under the indicator autoload, and
it it finds one, automatic loading will occur.
Automatic loading is performed by means of the autoload user interrupt; thus
the user may assert any desired degree of control over it. When the autoload
property is encountered, the user interrupt handler is called with one argument,
which is a dotted pair whose car is the atomic symbol which is the function
being autoload'ed, and whose cdr is the value of the autoload property. The
system-supplied handler for this user interrupt could have been defined by:
(setq autoload
(function (lambda (x) (load (cdr x)) )))
From this one can see that the value of the autoload property should be the name
of the file which contains the definition of the function. Note: in the TOPS-
10 implementations the system autoload handler presently uses fasload rather
than load because the load function requires the Newio feature. This affects
the form of an autoload property.
When the interrupt handler returns, it should have put a functional property
on the property list of the function being autoloaded. If not, an undf-fnctn
error will occur with a message such as "function undefined after autoload."
Page 3-26 ∪3-1.4.3 December 6, 1977
**DRAFT** The System **DRAFT**
Examples of setting up functions to be autoloaded:
In the Multics implementation:
(putprop 'foo ">udd>AutoProg>Library>foo-function" 'autoload)
In the PDP-10 Oldio implementation:
(putprop 'foo '(foo fasl dsk me) 'autoload)
In the PDP-10 Newio implementation:
(putprop 'foo '((dsk me) foo) 'autoload)
or (putprop 'foo '|dsk:me;foo fasl| 'autoload) or the Oldio version also
works.
December 6, 1977 ∪3-1.4.4 Page 3-27
**DRAFT** Maclisp Reference Manual **DRAFT**
1.5 Debugging
1.5.1 Binding, Pdl Pointers, and the Evaluator
The Maclisp evaluator is based on a push down list (pdl), or stack, which
holds bindings, evaluation frames, and sundry internal data. Bindings are
values of atomic symbols which are saved when the symbols are used as lambda
variables, prog variables, or do variables. Evaluation frames are constructed
when a non-atomic form is evaluated or when apply is used. They correspond to
function calls.
As the evaluator recursively evaluates a form, information is pushed onto the
pdl and later popped off. When the *rset and nouuo flags are t this information
is sufficiently detailed to be of use in debugging. (See the variables *rset
and nouuo in the next section.)
A position within the pdl may be named by means of a "pdl pointer", which is
a negative fixnum whose value has meaning to the evaluator. nil is also
accepted as a pdl pointer; it means the top of the stack, i.e. the most recent
evaluation. Note that this is different from nil as a binding context pointer,
which means the bottom of the stack or the outermost evaluation. 0 is also
accepted as a pdl pointer; it designates the frame at the bottom of the stack.
Pdl pointers may be used as arguments to several debugging functions described
in the next section. Since the fixnum value of a pdl pointer has only internal
meaning, generally a pdl pointer cannot be obtained from user input, except by
the user typing in a pdl pointer chosen from a list of pdl pointers typed out at
him. The "frame" functions described in the next section may be used to obtain
pdl pointers.
An important thing to note about pdl pointers is their limited scope of
validity. If the information on the pdl which is named by a pdl pointer has
been popped off since the pdl pointer was created, the pdl pointer no longer has
valid meaning.
1.5.2 Functions for Debugging
Page 3-28 ∪3-1.5 December 6, 1977
**DRAFT** The System **DRAFT**
*rset SUBR 1 arg
(*rset x) sets the *rset switch to nil if x is nil, or to t if x is non-
nil, and returns the value it set it to. (See below). This function
exists primarily for user typing convenience.
*rset SWITCH
If the *rset switch is non-nil, extra information is kept by the
interpreter to allow the debugging functions, such as baktrace and
evalframe, to work. In addition, the interpreter will make extra checks
such as checking the number of arguments passed to a subr or lsubr and
checking that array subscripts lie within the declared bounds. Generally,
the *rset switch being on means "I am debugging"; this is known as "*rset
mode". The initial state of the switch is nil.
nouuo SUBR 1 arg
(nouuo t) sets the nouuo switch.
(nouuo nil) turns off the nouuo switch. (This is the initial state.)
nouuo returns t or nil according to whether it turned the nouuo switch
on or off. (See below.) This function exists primarily for user typing
convenience.
nouuo SWITCH
If the nouuo switch is on, function calls made by compiled functions to
compiled functions or system functions are forced to go through the
interpreter each time. This aids in debugging. If the nouuo switch is
off, which is the normal case, compiled calls can be made to go directly,
which is much faster.
The nouuo switch may be turned off at any time. Each compiled function
call will only go through the interpreter once more, at which time it will
be linked directly. If the compiled code has been reloaded into the system
with the PURE option (see page 3-74) then this direct link may be unsnapped
and the Interpreter route re-established by (sstatus uuolinks). Because
December 6, 1977 ∪3-1.5.2 Page 3-29
**DRAFT** Maclisp Reference Manual **DRAFT**
the PURE option requires an amount of extra space and time, it is not
normally on; thus links snapped in code loaded as non-PURE cannot be
unlinked.
The trace package turns this switch on when a function is traced, in
order to ensure that tracing will work even for compiled functions.
Compiled function calls which have been "snapped" to go directly do not
push debugging information in *rset mode and cannot be traced. See also
(status uuolinks).
baktrace LSUBR 0 to 2 args
baktrace displays the stack of pending function calls. It gives
detailed information only in (*rset t) mode. The first argument is a pdl
pointer, as with evalframe. If it is omitted, nil is assumed, which means
start from the top of the pdl. The second argument is the maximum number
of lines to be typed; if it is omitted the entire stack is displayed. (The
second argument is currently permitted only in the Multics implementation.)
The information printed by baktrace is not the same as that obtained with
evalframe; both should be used to get the maximum amount of debugging
information.
baklist LSUBR 0 to 1 arg
baklist returns a list containing the information which baktrace would
print. (This is available only on the PDP-10 implementations.)
errframe SUBR 1 arg
errframe returns a list describing an error which has been stacked up
because of a user interrupt. The list has the form (err pdlptr message
bcp), where pdlptr is a number which describes the location in the pdl of
the error, message is a list of from one to three things which, given to
the error function, could have caused this same error, and bcp (binding
context pointer) is a number which can be used as a second argument to eval
or a third argument to apply to cause evaluation using the bindings in
effect just before the error occurred.
The argument to errframe can be nil, which means to find the error at
Page 3-30 ∪3-1.5.2 December 6, 1977
**DRAFT** The System **DRAFT**
the top of the stack; i.e. the most recent error. It can also be a pdl
pointer, in which case the stack is searched downward from the indicated
position. Thus if there are recursive calls to the error handler, the
second error back down the stack may be found by:
(errframe (cadr (errframe nil)))
The argument to errframe may also be a positive number, which is the
negative of a pdl pointer. This means start from the position in the stack
marked by the pdl pointer and search upwards.
If no error is found, errframe returns nil.
errprint LSUBR 1 to 2 args
errprint treats its argument the same as errframe. The second argument,
in Newio implementation only, is the file(s) into which to print the
information (see print). The message portion of the error frame is
princ'ed. errprint returns t if a message was typed out and nil if no
error frame was found.
evalframe SUBR 1 arg
The argument to evalframe is a pdl pointer, as with errframe. The pdl
is searched for an evaluation of a function call, using the same rules
about starting point and direction as errframe uses. evalframe always
skips over any calls to itself that it finds in the pdl.
The value is a list (type pdlptr form bcp), where type is eval or apply,
pdlptr is a pdl pointer to the evaluation in the stack, suitable for use as
an argument to evalframe or errframe or baktrace, form is the form being
evaluated or a list of the name of the function being applied and the
arguments it was applied to, and bcp is a binding context pointer which can
be used with eval to evaluate something in the binding context just before
the evaluation found by evalframe.
evalframe returns nil if no evaluation can be found.
evalframe only works in (*rset t) mode, since no extra frame information
is saved otherwise.
December 6, 1977 ∪3-1.5.2 Page 3-31
**DRAFT** Maclisp Reference Manual **DRAFT**
freturn SUBR 2 args
(freturn p x) returns control to the evaluation designated by the pdl
pointer p, and forces it to return x. This "non-local-goto" function can
be used to do fancy recovery from errors.
evalhook VARIABLE
If the value of evalhook is non-null, then special things happen in the
evaluator. When a form (even an atom) is to be evaluated, evalhook is
bound to nil and the functional form which was its value is applied to one
argument - the form that was trying to be evaluated. The value it returns
is then returned from the evaluator. This feature is used by the Stepper
package described later in this section.
evalhook is bound to nil by break, and setq'ed to nil by errors that go
back to top level and print *. This provides the ability to escape from
this mode if something bad happens.
In order not to impair the efficiency of the LISP interpreter, several
restrictions are imposed on evalhook. It only applies to evaluation -
whether in a read-eval-print loop, internally in evaluating arguments in
forms, or by explicit use of the function eval. It does not have any
effect on compiled function references, on use of the function apply, or on
the "mapping" functions. Also, as a special case, the array reference
which is the first argument to store is never seen by the evalhook
function; however, the subexpressions of the array reference (the indices)
will be seen. (This special treatment avoids a problem with the way store
works.) Normally the evaluator does not check the value of evalhook, in
order to save time. To make it check, you must both be in (*rset t) -
debugging - mode, and have done (sstatus evalhook t). Not all
implementations need both of those, but you should always do both to be
sure. If you use the Stepper package, you need not worry; it does this
automatically.
evalhook LSUBR 2 or 3 args
(evalhook form hook) is a function which helps exploit the evalhook
feature. The form is evaluated with evalhook lambda-bound to the
functional form hook. The checking of evalhook is bypassed in the
Page 3-32 ∪3-1.5.2 December 6, 1977
**DRAFT** The System **DRAFT**
evaluation of form itself, but not in any subsidiary evaluations, for
instance of arguments in the form. This is like a "one-instruction
proceed" in a machine-language debugger. If all three arguments are
present, the second is a binding context pointer and is used as the second
argument to eval, and the third argument is the hook.
Example:
(defun hook fexpr (x) ;called as (hook <form>)
((lambda (*rset)
(prog2 (sstatus evalhook t)
;magic sstatus
(evalhook (car x) 'hook-function)
;evaluate form
(sstatus evalhook nil)))
;more magic
t))
(defun hook-function (f)
((lambda (terpri)
(declare (special terpri))
(princ '|form: | msgfiles)
(prin1 f msgfiles)
((lambda (v)
(terpri msgfiles)
(princ '|value: | msgfiles)
(prin1 v msgfiles)
v)
(evalhook f 'hook-function)))
t))
;this is how to eval the
; form so as to hook
; sub-forms
The following output might be seen from (hook (cons (car '(a . b)) 'c):
form: (cons (car (quote (a . b))) (quote c))
form: (car (quote (a . b)))
form: (quote (a . b))
value: (a . b)
value: a
December 6, 1977 ∪3-1.5.2 Page 3-33
**DRAFT** Maclisp Reference Manual **DRAFT**
form: (quote c)
value: c
value: (a . c)
(a . c)
The following functions only exist in the Multics implementation.
baktrace1 LSUBR 0 to 2 args
baktrace1 is the same as baktrace except that binding context pointers
suitable for use with eval and apply are displayed along with the function
names.
baktrace2 LSUBR 0 to 2 args
baktrace2 is the same as baktrace1 except that pdl pointers, suitable
for use with baktrace and evalframe, are displayed along with the function
names and binding context pointers.
Page 3-34 ∪3-1.5.2 December 6, 1977
**DRAFT** The System **DRAFT**
1.5.3 The Trace Package
The LISP trace package provides the ability to perform various actions at the
time a function is called and at the time it returns. This can be used for
traditional tracing or for more sophisticated debugging actions.
The trace package is not part of the initial environment; however, it is
automatically loaded in on the first reference to the function trace. (See
autoload.)
The LISP trace package consists of three main functions, trace, untrace, and
remtrace, all of which are fexprs.
A call to trace has the following form:
(trace trace←specs)
A trace←spec in turn is either an atom (the name of the function to be traced)
or a list:
(function-name option1 option2 ...)
where the options are as follows:
break pred causes a break after printing the entry trace (if any) but
before applying the traced function to its arguments, if and
only if pred evaluates to non-nil.
cond pred causes trace information to be printed for function entry and/or
exit if and only if pred evaluates to non-nil.
wherein fn causes the function to be traced only when called from the
specified function fn. The user can give several trace specs to
trace, all specifying the same function but with different
wherein options, so that the function is traced in different
ways when called from different functions. Note that if fn is
already being traced itself, the wherein option probably will
not work as desired. (Then again, it might.) Note that fn may
not be a compiled function.
argpdl pdl specifies an atom pdl whose value trace initially sets to nil.
A list of the current recursion level for the function, the
function's name, and a list of the arguments is cons'ed onto the
December 6, 1977 ∪3-1.5.3 Page 3-35
**DRAFT** Maclisp Reference Manual **DRAFT**
pdl when the function is entered, and cdr'ed back off when the
function is exited. The pdl can be inspected from a breakpoint,
for example, and used to determine the very recent history of
the function. This option can be used with or without printed
trace output. Each function can be given its own pdl, or one
pdl may serve several functions.
entry list specifies a list of arbitrary S-expressions whose values are to
be printed along with the usual entry-trace. The list of
resultant values, when printed, is preceded by a \\ to separate
it from the other information.
exit list similar to entry, but specifies expressions whose values are
printed with the exit-trace. Again, the list of values printed
is preceded by \\.
arg that the function's arguments, resultant value, both, or
specify
value are to be traced. If not specified, the default is
neither
both Any "options" following one of these four are assumed to
both.
nil arbitrary S-expressions whose values are to be printed on
be
both entry and exit to the function. However, if arg is
specified, the values are printed only on entry, and if value,
only on exit. Note that since arg, value, both, and nil swallow
all following expressions for this purpose, whichever one is
used should be the last option specified. Any such values
printed will be preceded by a // and will follow any values
specified by entry or exit options.
If the variable arglist is used in any of the expressions given for the cond,
break, entry, or exit options, or after the arg, value, both, or nil option,
when those expressions are evaluated the value of arglist will effectively be a
list of the arguments given to the traced function. Thus
(trace (foo break (null (car arglist))))
would cause a break in foo if and only if the first argument to foo is nil.
Similarly, the variable fnvalue will effectively be the resulting value of
the traced function. For obvious reasons, this should only be used with the
exit option.
The trace specifications may be "factored." For example,
Page 3-36 ∪3-1.5.3 December 6, 1977
**DRAFT** The System **DRAFT**
(trace ((foo bar) wherein baz value))
is equivalent to
(trace (foo wherein baz value) (bar wherein baz value))
All output printed by trace can be ground into an indented, readable format,
by simply setting the variable sprinter to t. Setting sprinter to nil changes
the output back to use the ordinary print function, which is faster and uses
less storage but is less readable for large list structures.
Examples of the use of trace:
(1) To trace function foo, printing both arguments on entry and result on
exit:
(trace foo)
or (trace (foo)) or (trace (foo both)).
(2) To trace function foo only when called from function bar, and then
only if (cdr x) is nil:
(trace (foo wherein bar cond (null (cdr x))))
or (trace (foo cond (null (cdr x)) wherein bar))
As this example shows, the order of the options makes no difference, except
for arg, value, both, or nil, which must be last.
(3) To trace function quux, printing the resultant value on exiting but no
arguments on entry, printing the value of (car x) on entry, of foo1, foo2,
and (foo3 bar) on exit, and of zxcvbnm and (qwerty shrdlu) on both entry and
exit:
(trace (quux entry ((car x)) exit (foo1 foo2 (foo3 bar))
both zxcvbnm (qwerty shrdlu)))
(4) To trace function foo only when called by functions bar and baz,
printing args on entry and result on exit, printing the value of (quux barf
barph) on exit from foo when called by baz only, and conditionally breaking
when called by bar if a equals b:
December 6, 1977 ∪3-1.5.3 Page 3-37
**DRAFT** Maclisp Reference Manual **DRAFT**
(trace (foo wherein bar break (equal a b))
(foo wherein baz exit ((quux barf barph))))
(5) To trace functions phoo and fu, never printing anything for either,
but saving all arguments for both on a common pdl called foopdl, and breaking
inside phoo if x is nil:
(trace (phoo argpdl foopdl break (null x) cond nil nil)
(fu argpdl foopdl cond nil nil))
The "cond nil" prevents anything at all from being printed. The second
nil in each trace spec specifies that no args or value are to be printed;
although the cond nil would prevent the printout anyway, specifying this too
prevents trace from even setting up the mechanisms to do this.
trace returns as its value a list of names of all functions traced; for
any functions traced with the wherein option, say (trace (foo wherein bar)),
instead of returning just foo it returns a 3-list (foo wherein bar). If
trace finds a trace spec it doesn't like, instead of the function's name it
returns a list whose car is ? and whose cdr is an error message. The error
messages are:
(? wherein foo) trace couldn't find an expr, fexpr, or macro property for the
function specified by the wherein option.
(? argpdl foo) The item following the argpdl option was not a non-nil atomic
symbol.
(? foo not function) Indicates that the function specified to be traced was
non-atomic, or had no functional property. (Valid functional
properties are expr, fexpr, subr, fsubr, lsubr, and macro.)
(? foo) foo is not a valid option.
Thus a use of trace such as
(trace (foo wherein (nil)) (bar argpdl nil))
would return, without setting up any traces,
((? wherein (nil)) (? argpdl nil))
Page 3-38 ∪3-1.5.3 December 6, 1977
**DRAFT** The System **DRAFT**
If you attempt to specify to trace a function already being traced, trace
calls untrace before setting up the new trace. If an error occurs, causing (?
something) to be returned, the function for which the error occurred may or may
not have been untraced. Beware!
It is possible to call trace with no arguments. (trace) evaluates to a list
of all the functions currently being traced.
untrace is used to undo the effects of trace and restore functions to their
normal, untraced state. The argument to untrace for a given function should be
what trace returned for it; i.e. if trace returned foo, use (untrace foo); if
trace returned (foo wherein bar) use (untrace (foo wherein bar)). untrace will
take multiple specifications, e.g. (untrace foo quux (bar wherein baz) fuphoo).
Calling untrace with no arguments will untrace all functions currently being
traced.
remtrace, oddly enough, expunges the entire trace package. It takes no
arguments.
December 6, 1977 ∪3-1.5.3 Page 3-39
**DRAFT** Maclisp Reference Manual **DRAFT**
1.5.4 The Stepper
The LISP "stepping" package is intended to give the LISP programmer a
facility analogous to the Instruction Step mode of running a machine language
program.
The Rich Stepper
The Rich stepper package provides a simple, small debugging capability. It
is available as a library program in the ITS implementation.
↓_How to Use the STEP Facility_↓
The package contains two compiled functions which are loaded by
(fasload step fasl dsk liblsp)
The user interface is through the function (fexpr) step, which sets switches to
put the LISP interpreter in and out of "stepping" mode. The basic commands are:
(step t) ;Turn on stepping mode.
(step nil) ;Turn off stepping mode.
These commands are usually typed at top level, and will take effect immediately
(i.e. the next S-expression typed in will be evaluated in stepping mode). Also
↑g, in addition to returning to top level, turns off stepping mode.
In stepping mode, the LISP evaluator will print out each S-expression to be
evaluated before evaluation, and the returned value after evaluation, calling
itself recursively to display the stepped evaulation of each argument, if the S-
expression is a function call. After displaying each S-expression, the
evaluator will wait before evaluation for a command character from the console:
<space> Continue stepping recursively.
<rubout> Show returned value from this level only, and continue
stepping upward.
Page 3-40 ∪3-1.5.4 December 6, 1977
**DRAFT** The System **DRAFT**
<cr> or <tab> Turn off stepping mode. (but continue evaluation
without stepping).
P Redisplay current form in full (i.e. without prinlevel
or prinlength)
B Get breakpoint; proceed with $P
M See advanced features under Stepping Macro Expansions.
↓_More Advanced Features_↓
Selectively turning on step:
(step foo1 foo2 ...)
If this command is typed at top level, stepping will not commence immediately,
but rather when the evaluator first encounters a S-expression whose car is one
of foo1, foo2, etc. This form will then display at the console, and the
evaluator will be in stepping mode waiting for a command character.
Stepping Macro Expansions:
If the stepper is proceeded with a <space>, it will not step the execution of
macro expansions, but will rather just show the result of the macro of expansion
and wait for another command.
To see the execution of the macro expansion itself, proceed the stepper with
an M instead of a <space>.
Using step with breakpoints:
The above description applies to turning stepping on and off globally at top
level. More detail is necessary to use step flexibly in and out of breakpoints
(e.g. together with trace).
If stepping is turned on by (step t) at top level, the evaluator will NOT be
in stepping mode within a breakpoint loop. If you wish to use stepped
evaluation within a break loop you must turn it on locally by (step t).
Conversely, if stepping was not turned on at top level and it is turned on by
(step t) in a break loop, it will NOT be on when return is made from the break
loop by $P.
December 6, 1977 ∪3-1.5.4 Page 3-41
**DRAFT** Maclisp Reference Manual **DRAFT**
However, executing (step nil) inside a break loop will turn off stepping
globally, i.e. within the break loop, and after return has be made by $P.
The most useful feature is the following, however:
(step) Command at top level has no immediate effect.
After (step) has been executed at top level, a subsequent (step t) inside of
a break loop will have the effect of turning on stepping mode both inside the
break loop and globally, i.e. the evaluator will start to step as soon as the
return is made from the break loop by $P. Thus, for instance, one could set
trace to break at some special place, and then use the break to turn on
stepping.
prinlevel and prinlength:
In the present version, for convenience, prinlevel and prinlength are lambda-
bound inside the hooking function to 3 and 5 respectively. These could be
changed by editing the expr code and recompiling.
When the P command is used, prinlevel and prinlength are temporarily bound to
nil, and the toplevel printer (the value of atom prin1) is used to redisplay the
current form.
Overhead of Stepping:
If stepping mode has been turned off by ↑g, the execution overhead of having
the stepping packing in your LISP is exactly nil.
If stepping mode has been turned off by (step nil, every call to eval incurs
a small overhead--several machine instructions, corresponding to the compiled
code for a simple cond and one function pushdown.
From an overhead point of view, running with (step) entered at top level is
the same as running with (step nil).
Stopping stepping by responding <tab> incurs the same continued overhead as
(step nil).
Running with (step foo1 foo2 ...) can be more expensive, since a member of
the car of the current form into the list (foo1 foo2 ...) is required at each
call to eval.
Page 3-42 ∪3-1.5.4 December 6, 1977
**DRAFT** The System **DRAFT**
In terms of memory requirements, the total compiled stepping package occupies
about 423 words of binary program storage.
Interaction with debug and trace:
No special interactions of the step package with debug, trace, or any other
system packages are known.
The Morgenstern Stepper
The Morgenstern Stepper package provides debugging capabilities for
interpreted LISP programs that are comparable to the capabilities provided by
DDT for assembler code. These capabilities include:
1) Single stepping through the evaluation of a function and over or into other
interpreted functions, when called, on a selective basis as determined by
the user. Each such form and its resulting value may be displayed.
2) Dynamic breakpointing on one or more of the following conditions: the form
or atom about to be evaluated matches a pattern you provide; the form being
evaluated involves a specified function; a given atomic symbol evaluates to
a given value; a given atomic symbol is to be bound in a prog, either type
of do, or an eval'd lambda-expression; or upon an arbitrary condition
specified by a predicate written as LISP code.
3) Returning a different value for a given S-expression. This allows for
changing the action that would be selected by conditionals in the program
and/or by go's in a prog or do. You can also go to any tag inside the
current prog.
4) These capabilities may be requested when the program is initially started by
a top-level form, or they may be initiated at any other point in the course
of execution - either from the terminal while in a breakpoint, or directly
by the program.
The stepper may be invoked initially by using the function mev as one would
use eval of one argument; e.g. (mev '(fcn arg1 arg2)). From a breakpoint or in
a program, the stepper may be turned on by invoking (hkstart) with no arguments.
It may be turned off by the q command described below, or of course by a CTRL/g
December 6, 1977 ∪3-1.5.4 Page 3-43
**DRAFT** Maclisp Reference Manual **DRAFT**
break. After mev evaluates its argument, it returns the value and turns off the
stepper. Note that in the above example the form given as an argument to mev
was quoted. If, say, the value of f was the S-expression (fcn arg1 arg2), then
one could use (mev f) instead.
At any point during the stepping, one may inspect the values of other
variables, and even reapply mev to any form. This may be done in either of
three ways. Each command will be prompted for by //, usually following the last
form printed out. Any S-expression that is typed which is not recognized as a
command will be eval'd (within an errset to catch errors). Alternatively, the e
command to eval any expression, or the h command to get a nice type of CTRL/h
break. (This is really a CTRL/b break, but it used to be CTRL/h so the command
happens to be called h.)
in the its implementation each command must be followed by a space (unless
the command is a list). in the multics implementation each command must be
followed by a newline. actually, this depends not on the implementation but on
(status linmode). each form and result which is printed out will be followed by
#number indicating the relative level of evaluation (i.e. stack depth since
invocation).
the primary commands are:
d (mnemonic for down) go down to the next deeper level of evaluation and
display the first form there before evaluating it. e.g. if the form is a
function call, this will display the first argument of the function if it has
arguments in the call; otherwise it will display the first s-expression of
the body of the function. it then prompts for the next command.
e (eval) can be used to evaluate an arbitrary expression. it starts a new
line, waits for you to type the expression, then eval's it within an errset,
and prints the result. this is comparable to just typing the expression or
atom after the //, but cannot be confused with a command, and the format is
nicer.
h (control-h) enters a break loop, and when $p'ed displays the current form.
within the break, one can inspect the values of variables, etc., and even
reapply mev to any form.
n (next) Display the next form at this level, without showing or inspecting the
evaluation of the lower levels of the current form. The value of the current
form is displayed first. If you wish a condition to be tested for at lower
levels, use nn instead.
Page 3-44 ∪3-1.5.4 December 6, 1977
**DRAFT** The System **DRAFT**
nn Like n but slower since it inspects the lower levels. Use instead of n when
testing for a condition.
u (up) Go up to the next higher level of evaluation and show the next form at
that level. The form(s) at the current and lower levels are evaluated
without display. As an example of its use, after you have seen the
evaluation of the arguments to a function, the next form to be evaluated, if
the function is being interpreted, will be the first S-expression of the
function; to avoid seeing how the function is evaluated internally, you can
type u. Note that the lower levels are not inspected - thus if a condition
is to be tested for at these levels, use uu.
(u num) If num is positive (or zero), forms are not inspected nor displayed
until level num is returned to. If negative, it goes up (abs num) levels
relative to the current level. Thus (u -1) is equivalent to u.
uu Like u but slower. Use if testing for a condition.
(uu num) Like (u num) but slower. Use if testing for a condition.
q (quit) Exit from the stepper.
s (show or display mode) For datapoints and other display terminals, this
gives a nice easily read output of selected levels that constitute the
context of the current evaluation. Specifically, it selects the current
level for sprint'ing (pretty printing) as a "header", and as you go deeper,
the local context is abbreviate-printed under this header, and the current
output will be sprint'ed. s may be used as often as you like. Headers will
automatically be popped when you return. The command (s num) selects a
particular level as a header. It and the command sn and several user
settable parameters are described in the more detailed section below.
(= s-exp) The S-expression is substituted for the current form and another
command is prompted for (i.e. you can step into or over the new form if you
want to). When the resulting value is returned it will be as if the original
form had yielded that value. For example, you can change the apparent truth
or falsity of predicates or bypass a (go label), as well as just returning
different values for an S-expression.
(cond ...) Tests for conditions prior to evaluation of each future form, and
when satisfied will print a message, display the form, and wait for another
command (which may of course be h for a break). The argument to this cond is
December 6, 1977 ∪3-1.5.4 Page 3-45
**DRAFT** Maclisp Reference Manual **DRAFT**
an arbitrary S-expression or symbol which is evaluated like a predicate.
This is similar to the cond feature of the trace package.
In specifying the predicate, the form about to be evaluated may be obtained
as the value of the variable %%form. The expression (hooklevel) returns the
relative level of evaluation. More than one predicate may be given, in which
case they are or'ed together, except when two arguments form a special test
as described in the more detailed section below. The condition will remain
active at all levels that are inspected by the stepper until explicitly
turned off by (cond nil).
(matchf ...) is a function which will pattern match against the current form.
It may be used in the predicate of the cond. (Also see its related use as a
command.) The argument to matchf is compared to %%form element by element
from left to right, and succeeds when each element of the pattern succeeds.
Of importance, the pattern need not include the entire form. * matches
anything. The procedure is applied recursively to sublists, unless the
sublist is of the form (# ...) in which case # is bound to the current
element of %%form and the cdr (not cadr) of the #-list is evaluated as the
test on that element. Except in this case, atoms and lists should be given
as in the original code since they are not evaluated. Some simple examples
are:
(matchf xyz) succeeds if the atom xyz is about to be evaluated.
(matchf (setq alpha)) succeeds if the atom alpha is about to be setq'd.
(matchf (putprop name * 'source)) succeeds if the property source is
about to be putprop'd on the atom pointed to by (i.e. the value of)
name.
(matchf (setq (# member # '(alpha beta s3)))) succeeds if either alpha,
beta, or s3 is about to be setq'd.
(matchf (rplacd * '(* 9))) matches (rplacd (last urlist) '(2 9 4)).
(matchf ((# member # '(foo bar)))) succeeds if a function call to either
foo or bar is about to be evaluated (more precisely if the car of the
form about to be evaluated is either foo or bar).
nil (cond nil) turns the condition off and saves the current non-nil condition.
Page 3-46 ∪3-1.5.4 December 6, 1977
**DRAFT** The System **DRAFT**
(cond) When no argument is given, the last non-nil condition (which is the old
property of %%cond) is established as the current condition (which is the
value of %%cond). (If the previous condition was not nil then it is saved as
the old property, thus allowing for alternation of two conditions.)
(matchf ...) is equivalent to (cond (matchf ...)), see above.
↓_The following functions are useful in connection with the stepper_↓.
(hkstart) will initiate stepping when encountered in a program or typed from
a breakpoint. (hkstop) will act like the q command to turn off stepping. (Also
see below for more info.)
(mbak) is a function to be used like the LISP system's (baklist). (mbak)
strips out from the result of (baklist) those functions that have to do with the
stepper.
The remainder of this section is a complete list of the Stepper commands,
which can be used for reference.
Commands which are not lists must be followed by a space. You can use rubout
before completing the command (and its space if necessary). Alternatively, you
may abort the command before completing it by doing a CTRL/x break.
Any S-expression that you type which is not recognized as a command will be
evaluated (within an errset to catch errors). Thus you can evaluate any atom or
do any function call simply by typing it following the prompting // as long as
it is not interpretable as one of the commands below (or nil). Note that you
can actually go to a tag within your prog simply by typing (go tag) after the
//. To evaluate a form which looks like a command, type (or form) to evaluate
it, e.g. (or a) evaluates the atom a. If you want you can even write functions
which know about the stepper and treat them as commands.
a (all) Automatically displays all forms and values seen by the stepper at
all levels. Typing a space at any time thereafter will cause the stepper
to leave this mode and prompt for a new command. If you want the stepper
to wait for a command after each form, you can use the d command.
Commands a ad (a -) c and cc pause after each new form is displayed if
%%ac-sleep is non-nil. Its value is used as the sleep time in seconds.
ad (all down) Automatically displays all forms and values encountered by the
December 6, 1977 ∪3-1.5.4 Page 3-47
**DRAFT** Maclisp Reference Manual **DRAFT**
stepper in evaluating the current form (i.e. at deeper levels). Typing a
space prior to completion will cause the stepper to leave this mode and
prompt for a new command. (Also see d.) Sleeps after each form, as
described under the a command.
(a lev) Automatically displays all forms and values at the indicated level and
lower (deeper) levels, turning itself off when evaluation pops to a level
with a smaller level number. Typing a space prior to completion will
cause the stepper to leave this mode and prompt for a new command. (Also
see d.) Sleeps after each form, as described under the a command.
b Sets a breakpoint to occur after evaluation of the current form. At the
break, the value to be returned is the value of %%value, and may be changed
by setq'ing this variable. The form that yielded this value is the value
of %%form. Type $P to proceed from the breakpoint. If you prefer that the
system wait rather than break see the wtif command. (b operates by adding
the current hooklevel to %%breaklist.) You can get automatic breaking at
all levels by using (retcond t) or conditional breaking as described below
for the (retcond ...) command.
c (current) Automatically displays all forms and values at just the current
level. Typing a space at any time during the display will cause the
stepper to leave this mode and prompt for a new command. The stepper does
not inspect the forms of lower levels - thus if a condition is to be tested
for at these levels, use cc. Sleeps after each form, as described under
the a command.
cc Like c, but inspects the lower levels.
ctog Flips the %%condnotallow toggle which is initially t, meaning do not allow
c, m, n, or u commands if a condition is being tested for. nil means allow
these anyway.
(cond ...) Tests for conditions prior to evaluation of each future form. For
pattern matching against the form using the matchf function and for other
information see the description of (cond ...) above.
special tests for (cond ...) :
To aid the specification of common tests, the following "flags" are
provided - the same effects could be obtained by inspecting %%form in your
own predicate given to cond. If the first argument to the cond is from the
set (form formq bind bindq atomval atomvalq fcn fcnq and andq) then the
Page 3-48 ∪3-1.5.4 December 6, 1977
**DRAFT** The System **DRAFT**
second argument is used to derive a test. This process is repeated with
the remaining arguments, if any. The resulting tests, together with any
remaining arguments not satisfying this process, are effectively or'ed
together to derive the overall condition (except for the and andq flag
special tests which are and'ed). The arguments are not evaluated when
typed but are evaluated each time the condition is tested. These flags
each may be used more than once.
The meanings of these flags are:
andq The next argument is and'ed with the remaining tests, and
must yield a non-nil value for the remainder of the
condition to succeed. (See the comments for cond in the
"complete list of commands" below regarding the use of side
effects)
atomvalq The next argument should be a list of two elements, the
first an (unquoted) name of an atom, and the second the
value of this atom for the test to succeed.
bindq Watch for the following (unquoted) atomic symbol to be bound
in a prog, or in either type of do, or an explicitly
evaluated lambda (as distinct from an applied lambda or
function call).
fcnq Watch for the following (unquoted) function name to be seen by eval as
the car of the form about to be evaluated. (This cannot
check for applied or mapped function calls).
formq The following (unquoted) S-expression is to be watched for. E.g. used
to check when a particular variable is about to be
evaluated.
These evaluate their argument each time the condition is tested
and
in order to get the desired S-expression or atom name, and then
bind
perform like their "q" counterparts. These are particularly
fcn
useful if the flag's argument is the value of a variable. (Be
form
sure not to change the variable's value accidentally while the
atomval
condition remains in effect.)
As a simple example, (cond fcnq rplacd) will check and stop when the function
rplacd is about to be used (i.e. when it is the the car of
the form to be evaluated).
December 6, 1977 ∪3-1.5.4 Page 3-49
**DRAFT** Maclisp Reference Manual **DRAFT**
The commands c, m, n, and u do not inspect all levels, and thus the
condition cannot be tested for at these levels. You can use cc, nn, mm, or
uu instead, or use the ctog command. Naturally, condition testing slows
the speed of execution at levels that are inspected by the stepper but
which you do not have displayed.
If you choose to, you can have your predicates produce side-effects such as
recording information of value to you or setting states for use by the
condition later. You can use the and, andq flags (more than once if you
like) to have the expressions executed even upon success, so long as these
flags appear first in the condition. Other conditions are evaluated in the
order of appearance until the first success is found.
d (down) Displays the next level down (as described above also). Note that
if the form is an atom, the effect is the same as the n command. Hence if
you want the stepper to display every form and value, but to wait for a
command after each form, just keep using the d command.
e (eval) Can be used to evaluate an arbitrary S-expression. It starts a new
line, waits for you to type the expression, evaluates it within an errset,
and prints the result. Comparable to just typing the expression or atom
after the //, but cannot be confused with a command, and the format is
nicer.
(= S-exp) Replaces the current form with the given S-expression, and then
prompts for another command, as described above. If two arguments are
given, then this expression will not be treated as a stepper command,
rather it will be evaluated (see comments at top of this section).
h CTRL/h break is executed. The current form is redisplayed
when $P is typed. The form about to be evaluated is the value of %%form.
Within the break, one can inspect the values of variables, etc., and even
reapply mev to any form.
k (kill) Does not evaluate the current form nor display any
value. This is good for avoiding side effects if restepping through a
program again. Equivalent to (= nil) followed by m command.
lr (last result) A complete rather than abbreviated printout of
the last result is given. (See (p - -) for further information.)
Page 3-50 ∪3-1.5.4 December 6, 1977
**DRAFT** The System **DRAFT**
m Next, like n but the result of the current form is not
displayed. If a condition is to be tested for at lower levels, use mm.
(matchf ...) is equivalent to (cond (matchf ...)), see the the description of
(cond ...) above.
mm Next, like nn but the result of the current form is not
displayed.
n (next) Displays the value of the current form and displays
the next form, then awaits the next command. Does not inspect the lower
levels. If a condition is to be tested for at lower levels, use nn
instead.
nn Like n but inspects the lower levels.
o (old) Does (mev 'last form). This is useful for seeing how
a form produced an unexpected value when you went over it with n or nn. If
reevaluating the form can produce side effects be careful. Can be exited
from by the xx command. (The old form is the value of %%oldform.)
ol (old, at current level) Does (mev 'last form at this
level). Behaves like o. Useful to see the form (at this level) which
produced the current value - rather than the last form printed out, as o
would yield. (The old form used here can be obtained by (get %%hooklevel
'oldform).)
po(in)(print) Redisplays the current form. This is useful if you wish to clear
the screen first with control-L. Gives typical abbreviated display (see (p
- -)), except has somewhat different effect if in display mode (see s
command). (For hackers of special data structures, e.g. "owl", printing
will be done with the function which is the value of the atom prin1 if non-
nil - as also applies to top-level in LISP. This value of prin1 is checked
only in the mev function. Moreover, unless you request LISP not to "snap
links" in compiled code, you may have to reload the stepper after changing
prin1.
December 6, 1977 ∪3-1.5.4 Page 3-51
**DRAFT** Maclisp Reference Manual **DRAFT**
pp (full print) Gives a complete printout of the current form.
ppp (even better printout) Pretty-prints the current form using the sprint
function. Uses a lot of screen in general, and so will turn on pagepause
for you.
(p - -) Resets the parameters for the abbreviated printout used for results,
forms and the p command. The first parameter is the prinlevel, the second
is the prinlength; both must be given. If nil is given instead of a number
no abbreviating is done with respect to that parameter; thus (p nil nil)
turns off abbreviation. (The current settings are the value of
%%hookprin.)
q (quit) Exits from the stepper. Previously requested breaks and conditions
are disabled, and any non-nil conditions are saved on the old property of
the condition name. (Control-G also exits as usual.)
s (show or display mode) For datapoints and other display terminals, this gives
a nice easily read output of selected levels that constitute the context of
the current evaluation. Specifically, it selects the current level for
sprinting as a "header", and as you go deeper, the local context is
abbreviate-printed under this header, and the current output will be
sprinted. s may be used as often as you like. Headers will automatically
be popped when you return. All sprinting is done with pagepause on. If
control-X is typed during sprinting, that expression will be redisplayed
using abbreviated-printing instead. When in this display mode, the p
command will clear the screen from the last form down, unless preceeded by
control-L (or if wrap-around occurred), in which case the screen is fully
redisplayed. Also see (s arg) for more information and options.
(s arg) If arg is positive, this selects the form at that level as the "header"
for s(how) mode. If negative, it uses the form at arg levels above the
current one. If arg is nil, display mode is turned off (headers are
remembered though). (s t) just turns display mode on if currently off
using the previously remembered headers if still applicable; but if it is
already on, this pops the stack of headers by one (normally headers are
automatically popped when the level is returned from). All sprinting is
done with pagepause on. If control-X is typed during sprinting, that
expression will be redisplayed using abbreviated-printing instead. Also
see the sn command.
Several parameters are user settable from their defaults. %%lowerdisplay and
Page 3-52 ∪3-1.5.4 December 6, 1977
**DRAFT** The System **DRAFT**
%%lowerdisplay-min control the maximum and minimum number of levels to
display below the header (defaults of 5 and 2). This is done in
abbreviate-printed form using %%shortprin which is a list of the prinlevel
and prinlength (defaults 3 and 3). Sprinting of forms and results will be
abbreviate-sprinted by the msprint function if the flatsize of the
expression exceeds %%flatsize-max (default about 450). The prinlevel and
prinlength for the latter are the list which is the value of %%sprintabbr
(default is (7 8)). If %%flatsize is nil, full sprinting will always be
used; (if negative, abbreviate-sprinting will always be used so that
infinite printing circular structures will sprint and abbreviate-print
finitely. To turn off sprinting of results setq %%result-sprint to nil
(default t). If %%mdistitle is neither nil nor a number, it will be
evaluated just after the screen is cleared, allowing printing of a title.
If it is a number, that number of blank lines will be left at the top of
the screen (also see sviewmsg function below). If the partial clearing of
the screen bothers your eyes, setq'ing %%eyestrain1 to a number of seconds
(e.g. 0.5 to 2.0) will slow down the new display depending on the number of
lines cleared.
sn Just for s(how) display mode. It prevents clearing of the screen after
prompting for another command, but only until the next prompting // after
that. Useful if you want a result to remain displayed a little longer. If
you want to prevent clearing of the screen for more than a couple of times,
use (s nil), then do (s t) when you want to resume display mode.
(retcond ...) Tests for conditions just after each form is evaluated, and
breaks when such condition is satisfied. At the break, the value to be
returned is the value of %%value, and may be changed by setq'ing this
variable. The form that yielded this value is the value of %%form. Type
$P to proceed from the breakpoint. The conditions are specified as for
(cond ...). Note that (retcond t) will give you a break as each level is
popped (returned from), including levels above the one where the request
was made. (retcond nil) disables the retcond. If you prefer waiting
rather than breaking see the wtif command.
Two additional flags are available:
valueq The test (equal %%value next-argument) is performed as if it were
and'ed with the remaining predicates in the condition.
value Like valueq but the test is (equal %%value (eval next-argument)). The
overall condition is maintained on the value of the atom %%retcond, and the
December 6, 1977 ∪3-1.5.4 Page 3-53
**DRAFT** Maclisp Reference Manual **DRAFT**
previous non-nil condition is on the old property of this atom. If you want
both cond and retcond conditions to be the same you can (setq %%retcond
%%cond). The value and valueq predicates will be ignored in a (cond ...).
u (up) Go up to next higher level. Current and lower levels are executed
without display. The lower levels are not inspected - thus if a condition
is to be tested for at these levels, use uu. This can be used to skip the
display of a function's internal evaluation after having seen the
arguments, as described in the previous section.
(u num) If num is positive (or zero), forms are not inspected nor displayed
until that level number is reached. If negative, it goes up this number
(absolute value) of levels relative to the current level. Thus (u -1) is
equivalent to u .
uu Like u, but also inspects lower levels. Use if you have a condition to be
tested.
(uu num) Like (u num) but slower. Use if testing for a condition. Note that
(uu -999) effectively means that you won't see any levels unless the
condition in a cond or retcond is satisfied.
wtal (wait-all) Flips a toggle which when on causes a pause after the
evaluation of every form, but before that value is returned. The system
waits for an input character. Typing y(es), b(reak), or h (for control-h)
followed by space will cause a break as would the b command. Typing just a
space, or any other character followed by a space, will proceed from the
pause. Default is off.
wtif (wait-if) Flips a toggle which when on causes requests by the b and
(retcond ...) commands to result in a pause rather than a break. The pause
is like that of the wtal command, and may be proceeded by a space; or a
break initiated by typing y, b, or h followed by a space. Default is off.
xx Does a control-X type of LISP quit. (A control-X typed after the // prompt
will be caught by an errset. The xx command is executed outside of that
errset.)
The following other facilities exist:
(gethklevel num) This function returns the S-expression that is on the
execution stack of the stepper at the given level number (see hkshow). Can
Page 3-54 ∪3-1.5.4 December 6, 1977
**DRAFT** The System **DRAFT**
be used to get an unsprinted unabbreviated display of the form or to record
or process the form as you desire, including reapplication of mev to it in
the current context.
(hkshow num) This function will display previous forms which are on the
execution stack, as seen by the stepper while it has been activated. The
previous num of levels are shown, with the current form last. If no
argument is given, then all levels are shown. The display is done under
the control of prinlevel and prinlength which are settable by the (p - -)
command. Of course this function can also be used as if it were a command
by typing it after the prompting //.
(hksprint num) This function will sprint the form on the level whose number is
given as the argument. Can also be used as a command.
(hkstart) Use this function to invoke or reinvoke the stepper from a breakpoint
or from a program as described above. If used within a break, type
(hkstart) by itself rather than within another S-expression or function, as
it has to climb the stack from the point of invocation. If an argument is
given to this fexpr, it will be evaluated just prior to establishing
stepping, with ↑w bound to nil, so that you can print out information if
called from a program.
(It is possible for the invocation of the stepper by this method to have limited
scope under some circumstances. Such a boundary would be a second
breakpoint higher on the stack or a previously terminated invocation of the
stepper that is still on the stack. Also if the program was initially
started without mev, and stepping is retained thoughout the rest of the
execution, stepping may also remain for forms typed at top level - to stop
this just do control-G (or use the q command) .)
(hkstop) This function turns off the stepper whenever executed - in the same
manner as the q command would.
hooklist is an atom whose value is inspected before each attempt to read a
command from the console. If hooklist is non-nil, it is assumed to be a
list of commands to the stepper - each is printed out when used and treated
as if it came from your typein. hooklist is also examined at each level
that is inspected by the stepper even if no command reading is done (e.g.
nn or uu modes).
(mbak) This function gives (baklist) but without the stepper functions, as
described above.
December 6, 1977 ∪3-1.5.4 Page 3-55
**DRAFT** Maclisp Reference Manual **DRAFT**
(mev top-form) This function initiates stepping and otherwise acts like eval of
one argument, a≥`described above.
(msprint form) Gives abbreviated sprinting of the form. A second and third
numeric argument specify the effective prinlevel and prinlength here, else
a list of two numbers found as the value of msprint are used. The current
implementation is somewhat slow as the regular sprint does not respond to
standard abbreviating.
(sviewmsg lineno toeval) Useful in conjunction with s(how) mode. Puts the
cursor at the lineno and evaluates the second argument, then returns the
cursor to its original position. lineno = 0 means top; if negative counts
from bottom, with -1 the bottom line. Typically have %%mdistitle (see (s
-) command) be a number to skip lines on top, and use sviewmsg to display
your debugging information up there.
If you really want specialized processing in particular situations, you can
inspect and/or change %%form in a (cond ...) predicate, and %%value in a
(retcond ...). If %%nohookflag is t, form and value printout and command
reading (except from a non-nil hooklist) is inhibited until it is reset to nil.
Normal command processing is invoked by (%%mhookcom) with %%nohookflag bound to
nil. Also described above are %%breaklist, %%cond, %%retcond, and %%hookprin.
1.5.5 The MAR Break Feature
This feature is currently available only in the ITS implementation.
The MAR break feature takes advantage of a hardware feature (the Memory
Address Register break) which interrupts whenever a given memory location is
accessed in a specified way. It allows the LISP user to specify an interrupt
function to run whenever a variable or list cell is modified. The user must
first "arm" the interrupt by saying (sstatus mar cond loc). cond is the
condition on which to interrupt:
0 Turn off the mar feature.
1 Interrupt on instruction fetch.
2 Interrupt on write (modification).
3 Interrupt on all references.
(numbers are octal)
On a KL-10 processor, additional conditions are available:
Page 3-56 ∪3-1.5.4 December 6, 1977
**DRAFT** The System **DRAFT**
10 Interrupt on data read.
11 Interrupt on data read or instruction fetch.
12 Interrupt on data read or write.
13 Interrupt on instruction fetch or write.
loc is any s-expression; that cell is the one monitored.
Example:
(setq foo (list 'a 'b))
(sstatus mar 2 foo)
will interrupt if the list cell in foo is ever rplaca'd or rplacd'd.
An example of the use of the mar-break interrupt:
December 6, 1977 ∪3-1.5.5 Page 3-57
**DRAFT** Maclisp Reference Manual **DRAFT**
(defun mar-tracer (x)
((lambda (val)
(sstatus mar 2
(get the-mar-variable 'value))
(nointerrupt nil) ;let endpagfn interrupts in
(terpri msgfiles)
(princ '|Now the variable | msgfiles)
(prin1 the-mar-variable msgfiles)
(princ '| has the value| msgfiles)
(prin1 val msgfiles))
(symeval the-mar-variable)))
(setq mar-break mar-tracer)
(defun mar fexpr (x)
(cond ((null x)(sstatus mar 0 nil))
(t (setq the-mar-variable (car x))
;make sure the variable has a value cell
(or (boundp the-mar-variable)
(set the-mar-variable nil))
(sstatus mar 2 (get the-mar-variable 'value)))))
(mar quux)
(setq quux 5)
;Now the variable quux has the value 5
5
(do ((quux 0 (+ quux 1))) ((= quux 2))
(hack quux))
;Now the variable quux has the value 0
;Now the variable quux has the value 1
;Now the variable quux has the value 2
;Now the variable quux has the value 5
nil
Notice that quux is altered by the do loop, and also by the restoration of the
old value 5. This example is for a KA-10 processor. On a KA-10, the user
interrupt is always run after the location has been changed. On a KL-10
processor, the interrupt occurs just before a modification or access rather than
just after.
The mar break feature is sometimes used by DDT to debug the LISP system. As
long as DDT and LISP do not both try to use the mar break feature on the same
Page 3-58 ∪3-1.5.5 December 6, 1977
**DRAFT** The System **DRAFT**
LISP at the same time, there should be no problem. (sstatus mar 0 nil) releases
the mar break feature entirely for use by DDT.
The suspend function will attempt to save and restore the state of the mar
break feature. If you don't want an armed mar break to persist beyond a call to
suspend, turn it off first with (sstatus mar 0 nil).
When a CTRL/g quit (or the ↑g function) forces a quit back to top level, it
disables the mar break before unwinding variable bindings and re-enables it
afterwards. This is because during a CTRL/g quit LISP may not be in a good
state for running user interrupt functions.
December 6, 1977 ∪3-1.5.5 Page 3-59
**DRAFT** Maclisp Reference Manual **DRAFT**
1.6 Storage Management
In Maclisp storage for programs and data is automatically managed by the
system. The casual user need not concern himself with storage management and
need not read this section. However, the user who is curious about the
implementation or who has to construct a subsystem on top of Maclisp may need to
be concerned with how the internal storage management routines work and how to
control their general functioning. In no case is it necessary to control the
exact step by step operations of storage management, but a variety of functions
are provided to set the general policy followed by the LISP storage management
procedures.
1.6.1 Garbage Collection
Garbage collection is the mechanism which LISP uses to control storage
allocation. Whenever LISP feels that too much storage is being used, a garbage
collection is initiated. The garbage collector traces through all the S-
expressions which can be reached by car'ing and cdr'ing from internal atomic
symbols' values and property lists, from forms and temporary results currently
being used by the evaluator, from data used by compiled code, and from the saved
values of bound variables. All the data which it finds in this way is "good"
data, in that it is possible for it to be used again. Everything else is
garbage, which can never again be used for anything because it cannot be
accessed, so the storage used by it is reclaimed and reused for creating new S-
expressions.
gc FSUBR
(gc) causes a garbage collection and returns nil.
gctwa FSUBR
gctwa is used to control the garbage collection of "truly worthless
atoms," which are atomic symbols which have no value and no properties, and
which are not referenced by any list structure, other than the obarray (the
current obarray if there is more than one).
Page 3-60 ∪3-1.6 December 6, 1977
**DRAFT** The System **DRAFT**
(gctwa) causes truly worthless atoms to be removed on the next garbage
collection.
(gctwa t) causes truly worthless atoms to be removed on each garbage
collection from now on. Note: gctwa does not evaluate its argument.
(gctwa nil) causes this continual removal of truly worthless atoms to be
shut off, but it does not affect whether the next garbage collection
removes twa's.
The value returned by gctwa is a fixnum which is 0 if no garbage
collection of truly worthless atoms will be done, 1 if twa's are to be
gc'ed on the next garbage collection, 10 if twa's are to be gc'ed on all
garbage collections, or 11 if both. (These numbers are octal; the decimal
values are 0, 1, 8, 9.)
↑d SWITCH
If the value of ↑d is non-nil, the garbage collector prints an
informative message each time garbage collection occurs. (In a Newio
implementation, this message is output to the files in msgfiles, see page
3-15) In the PDP-10 implementation, it also prints a message when a space
is expanded without first doing a complete garbage collection, or when a
file or inferior job is closed because a file object was garbage collected.
See also (status gcwho).
See also the user interrupts gc-daemon, gc-overflow, and gc-lossage.
1.6.2 Spaces
In Maclisp the storage used for LISP objects is divided into several
conceptual subdivisions, called spaces. Each space contains a different type of
object. Allocation proceeds separately in the different spaces, but garbage
collection of all spaces occurs together since an object in one space could
contain a pointer to an object in any other space.
For example, in the PDP-10 implementation, the spaces are as follows:
LIST Conses (dotted pairs) and lists.
December 6, 1977 ∪3-1.6.1 Page 3-61
**DRAFT** Maclisp Reference Manual **DRAFT**
FIXNUM Fixnums.
FLONUM Flonums.
BIGNUM Bignum headers. Bignums also occupy fixnum and list space.
SYMBOL Atomic symbols.
HUNK4 Hunks of various sizes. PDP-10 implementations without hunks do not
have these spaces.
HUNK8
HUNK16
...
ARRAY "Special array cells."
REGPDL The "regular" pushdown list.
SPECPDL The "special" pushdown list, used in binding.
FXPDL The fixnum pushdown list, used for temporary numeric values.
FLPDL The flonum pushdown list, used for temporary numeric values.
Binary Program Space used to hold arrays and compiled code.
pure LIST, pure FIXNUM, pure FLONUM, pure BIGNUM, pure HUNG4, ...
These spaces are used to store "pure" (read-only) data of the
indicated types. This is a feature used to make subsystems more
efficient. See page 3-69.
In the Multics implementation, the spaces are:
list Conses (dotted pairs), lists, atomic symbols, bignums, and
strings.
Static Storage Arrays, files, and linkage to compiled code.
markedpdl A pushdown list of LISP objects.
Page 3-62 ∪3-1.6.2 December 6, 1977
**DRAFT** The System **DRAFT**
unmarkedpdl A pushdown list of machine data, not LISP objects.
Note: in the Multics implementation there is no space for numbers because
numbers are stored in such a way that they do not take up any extra room.
The precise spaces available in a given implementation can be determined by
using (status spcnames), (status purspcnames), and (status pdlnames).
Associated with each space is information determining when an attempt to
allocate in that space should cause a garbage collection. The idea is that one
should allocate for quite a while in a space, and then decide that it is worth
the trouble of doing an expensive garbage collection in order to prevent the
space from using too many bits of actual storage.
The exact nature of this information varies with the space. In a pushdown
list (pdl) space, all information must be stored contiguously, so the only
parameter of interest is how big the pdl is. This can be measured in three
ways, so there are three parameters associated with a pdl:
pdlsize The number of words of valid data in the pdl at the moment.
pdlmax The size to which the pdl may grow before intervention is required.
This is used to detect infinite recursion.
pdlroom The size beyond which the pdl may not grow no matter what.
A space such as a list space has three parameters, called the gcsize, gcmax,
and gcmin. These are in machine-dependent units of "words". The gcsize is the
expected size of the space; as objects are allocated in the space, it will grow
without garbage collection until it reaches this size. When it gets above this
size garbage collection will occasionally be required, under control of the
other two parameters.
The gcmax is the maximum size to which the space should grow; if it gets this
big garbage collections may occur quite frequently in an attempt to prevent it
from growing bigger.
The gcmin specifies the minimum amount of free space after a garbage
collection. It may be either a fixnum, which specifies the number of words to
be free, or a flonum, which specifies the fraction of the space to be free. The
exact interpretation of this depends on the implementation. In the PDP-10
implementation, which uses free storage lists, the gcmin is the number of words
December 6, 1977 ∪3-1.6.2 Page 3-63
**DRAFT** Maclisp Reference Manual **DRAFT**
which must be on the free storage list after a garbage collection. If there are
not this many, the space is grown, except if its size approaches gcmax it may
not be grown by the full amount. In the Multics implementation, which uses a
compacting garbage collector, the criterion for garbage collection is not when a
free list is exhausted but when the space reaches a certain size. This size is
the maximum of gcsize and the sum of the size after compactification plus gcmin
(if it is a fixnum) or the size after compactification times 1/(1-gcmin) (if
gcmin is a flonum.) The effect of this is to allow the same amount of allocation
between garbage collections as there would be in the PDP-10 implementation with
the same gcmin.
Note that these controls over the sizes of spaces are somewhat inexact, since
there is rounding. For instance, the PDP-10 implementation presently allocates
memory to spaces in blocks of 512. words. The Multics implementation allocates
at least 16384. words between garbage collections and presently controls the
size of pushdown lists in blocks of 16. words.
Some spaces, such as Binary Program Space in the PDP-10 implementation or
Static storage in the Multics implementation are not subject to detailed control
by the user. The management of these spaces is entirely automatic. Generally
these are spaces where the rate of allocation is fairly placid and most objects,
once allocated, are used forever and never freed. Hence the exact policy used
for storage management in these spaces is not too important.
1.6.3 Storage Control Functions
alloc SUBR 1 arg
The alloc function is used to examine and set parameters of various
spaces having to do with storage management. To set parameters, the
argument to alloc should be a list containing an even number of elements.
The first element of a pair is the name of a space, and the second is
either a fixnum or a 3-list. A fixnum specifies the pdlmax (for a pdl
space) or the gcsize and gcmax (for other spaces.) A 3-list specifies, from
left to right, the gcsize, gcmax, and gcmin. nil means "don't change this
parameter." Otherwise a fixnum must be supplied, except in the third
element (the gcmin), where a flonum is acceptable. A 3-list cannot be used
with a pdl space.
An example of this use of alloc, in the PDP-10 implementation:
Page 3-64 ∪3-1.6.2 December 6, 1977
**DRAFT** The System **DRAFT**
(alloc '(list (30000. 5000. 0.25)
fixnum (4000. 7000. nil)
symbol 6000.
regpdl 2000.))
or, in the Multics implementation:
(alloc '(list (30000. nil 0.3)
markedpdl 5000.
unmarkedpdl 5000.))
alloc may also be called with an argument of t, which causes it to
return a list of all the spaces and their parameters. This list is in a
form such that it could be given back to alloc at some later time to set
the parameters back to what they are now.
See page 3-89 for some status functions which are related to the topic of
storage spaces.
1.6.4 Dynamic Space and Pdl Expansion
There are several user interrupts generated by the storage management. See
section page 3-16 for a description of user interrupts. The gc-daemon interrupt
occurs after each garbage collection. The argument passed to the gc-daemon
interrupt handler is a list of spaces and their sizes. In the PDP-10
implementation, the items on the list are of the form: (space-name free-before
free-after size-before size-after), where space is the name of a space, and
free-before indicates the number of cells free before the garbage collection and
free-after indicates the number of cells free afterwards. The last two numbers
are the size of the space (see (status spcsize)) before and after the GC. (The
sizes are in PDP-10 words.) In the Multics implementation, the items are of the
form: (space before . after). In the Multics implementation, where "free
cells" is a meaningless concept, only the difference of these two numbers is
significant; it represents the amount of compaction achieved.
The gc-lossage interrupt occurs if the garbage collector tries to expand a
space but fails because, for example, the operating system will not give it any
more storage. The argument passed to the interrupt service function is the name
December 6, 1977 ∪3-1.6.3 Page 3-65
**DRAFT** Maclisp Reference Manual **DRAFT**
of the space that lost. If the interrupt handler returns, the value is ignored,
and another garbage collection is attempted.
The pdl-overflow interrupt is signalled when some pushdown list exceeds its
pdlmax. The pdlmax is increased slightly so that the interrupt handler will
have room to run. The argument passed to the interrupt function is the name of
the pdl that overflowed. If the interrupt function uses too much pdl, this
interrupt will occur again. If this happens enough times, the pdlmax will reach
the pdlroom, there will be no room in the pdl to take a user interrupt, and an
uncorrectable error will occur.
The interrupt function can decide to terminate the computation that
overflowed the pdl, for example by doing (↑g) or a throw, or it can increase the
pdlmax by using alloc or (sstatus pdlmax) and then continue the computation by
returning. Note that, unlike most other user interrupts, if the pdl-overflow
interrupt function returns nil (or the ";bkpt pdl-overflow" is $P'ed), the
computation is continued as if the pdl overflow had not occurred.
The gc-overflow interrupt occurs when some space (other than a pdl) exceeds
its gcmax. This gives the user a chance to decide that the size of the space
should be increased and the computation continued, or that something is wrong
and the computation should be terminated. The argument passed to the interrupt
handler is the name of the space that overflowed. The interrupt handling
function will be able to run because the garbage collector makes sure that the
space is sufficiently large before signalling the interrupt, even if this makes
it become somewhat larger than its gcmax. This interrupt is similar to pdl-
overflow in that if the interrupt handler function returns at all, even if it
returns nil, the interrupted computation proceeds. To terminate the computation
an explicit (↑g) or throw must be done.
1.6.5 Initial Allocation
The PDP-10 implementations of Maclisp run on a machine with a limited-size
address space. Consequently the allocation of portions of this address space to
different uses, such as LISP storage spaces, becomes important. This is
particularly true of the DEC-10 implementations, which cannot take advantage of
paging.
When LISP is first entered, it goes through a dialogue with the user known as
"allocation." Normally the dialogue simply consists of the user declining to
Page 3-66 ∪3-1.6.4 December 6, 1977
**DRAFT** The System **DRAFT**
specify anything, in which case LISP chooses suitable defaults. If a large
problem is to be worked on, the defaults may be inappropriate and it may be
necessary to explicitly allocate a larger amount of storage. It is also
possible for the user's replies to come from a file.
If LISP is called with a command line from DDT, for example
:LISP INDEX LOADER COM:
it reads the indicated file in the same way that it would read .LISP. (INIT).
See below.
On the other hand, if LISP is called without a command line, it identifies
itself and asks
ALLOC?
Suitable responses are Y, N, and CTRL/q. There are other obscure characters
which can be used as replies to this question, but these three are sufficient
for most purposes. ("?" causes a list of suitable responses to be printed out.)
"N" means that you do not want to specify allocation. You will get the default.
CTRL/q means to read your initialization file (see below.) "Y" means that you
wish to go through the following sequence of questions and answers.
LISP types out the names of various spaces and their sizes. If the name of
the space is preceded by "#", then it cannot be expanded once allocated by this
dialogue. After each question you may enter altmode, which terminates the
dialogue and gives the remaining parameters default values, or space, which goes
on to the next question. Before your altmode or space you may put a number
which is the size you want that space to be, instead of the number that was
printed. CTRL/g restarts the dialogue with the "ALLOC?" question.
If you reply with a CTRL/q, it means to read your initialization file. In
the ITS implementation, this is either udir; .LISP. (INIT) or (INIT); udir
.LISP., where udir is your master sname. In the TOPS-10 implementation, this is
LISP.INI in the directory you are logged in to. In the Multics implementation,
this is hd>start←up.LISP, where hd is your home directory. Since the Multics
implementation doesn't have the allocation dialogue, this file is always read
when the LISP command is given with no arguments.
The first form in the file should be a comment which is used to answer the
questions. Note that supplying nonexistent space names in the comment doesn't
hurt, so you can use the same comment for different implementations. An example
of the form of this comment is:
December 6, 1977 ∪3-1.6.5 Page 3-67
**DRAFT** Maclisp Reference Manual **DRAFT**
(comment fixnum 5000 symbol 4000 flonum 2000
bignum 1400 regpdl 5000 hunk8 30000)
The remaining forms in the file are simply read and evaluated using the
standard read-eval-print loop.
Page 3-68 ∪3-1.6.5 December 6, 1977
**DRAFT** The System **DRAFT**
1.7 Implementing Subsystems with Maclisp
1.7.1 Entering LISP
A subsystem is an entity that exists in most time-sharing systems. It is
normally a complete world which the user enters by typing a command. He then
has whatever facilities the subsystem offers. A subsystem can be oriented
toward programming, as the Maclisp subsystem itself is, or it can be oriented
toward a particular application, for instance compiling LISP programs, operating
machinery, or solving differential equations.
Maclisp is frequently used as a base on which to build subsystems.
Consequently it has been equipped with a number of mechanisms which allow the
subsystem writer to gain complete control over the operation of Maclisp, make it
possible to hide the vagaries of Maclisp from the naive user of a different
subsystem, and provide increased efficiency in memory and processor usage for
heavily-used subsystems.
in the ITS implementation
LISP may be entered by the :LISP (or LISP↑K) command. The environment set up
by this command is the standard initial environment. LISP now goes through an
allocation dialogue and optionally reads your .LISP. (INIT) file. See page 3-64
for information on this.
LISP may be entered by the command :LISP name1 name2 dev dir (or :LISP
dev:dir;name1 name2). In this case the file dev:dir;name1 name2 is read in the
same way as a .LISP. (INIT) file. This can be used to start up a subsystem.
The device defaults to DSK, the directory defaults to your current DDT master
sname (..MSNAME), and the file names default to .LISP. (INIT).
It is also possible to build a subsystem in a LISP, then save it as TS FOO.
The :FOO (or FOO↑K) command will then enter the subsystem, bypassing the
allocation dialogue.
in the DEC-10 implementation
LISP may be entered by the monitor command R LISP or RUN LISP. The
allocation dialogue (see page 3-64) is entered. Optionally a LISP.INI in the
December 6, 1977 ∪3-1.7 Page 3-69
**DRAFT** Maclisp Reference Manual **DRAFT**
user's directory may be read. If the DEC-10 monitor allows the use of
additional arguments on the R command line, then the particular name of a file
may be specified (but not a directory). The extension defaults to INI. This
file is read in the same way as a LISP.INI file. This can be used to start up a
subsystem.
As in the ITS implementation, a subsystem can be saved and then invoked by
the appropriate R or RUN command.
in the Multics implementation
The Maclisp subsystem is entered by issuing the LISP command at Multics
command level. If LISP is called with no arguments, a copy of the standard
initial environment containing all the system functions and variables is made
the current environment. If the LISP command is issued with an argument, the
argument concatenated with ".sv.LISP" is the pathname of a saved environment
which is copied into the current environment. This saved environment would
contain some subsystem, which will receive control. Additional arguments to the
LISP command in this case are actually arguments to the subsystem.
Often one constructs a trivial command for getting into a subsystem, which
simply calls the LISP command with the right arguments.
For instance, the LISP compiler subsystem may be entered through the
LISP←compiler command, which calls LISP with the pathname of the saved
environment containing the LISP compiler as the first argument, and the
arguments to the LISP←compiler command as the remaining arguments.
When the standard initial environment (i.e. the ordinary Maclisp subsystem)
is entered, it checks for a segment named start←up.LISP in the user's home
directory. If such a segment exists, it is read in, using the load function.
This facility allows users to "customize" LISP.
1.7.2 Saving an Environment
A subsystem is constructed by the following procedure. One starts with the
ordinary Maclisp subsystem, and defines a number of function definitions and
variable values. This creates an environment which is capable of implementing
the desired subsystem. This environment is then saved in a file, and necessary
Page 3-70 ∪3-1.7.1 December 6, 1977
**DRAFT** The System **DRAFT**
mechanisms are set up so that an operating-system command can invoke LISP and
cause it to set up the environment saved in the file. When the saved
environment is invoked control is passed to the functions in it which then
proceed to do the business of the subsystem.
The exact way of saving the environment differs from implementation to
implementation. In the Multics implementation there is a function called save:
save FSUBR
(save foo) saves the current LISP environment in a file named
foo.sv.LISP in the working directory. foo is not evaluated. The saving
operation destroys the working copy of the environment, so when the save is
complete LISP returns to Multics command level.
All variable values, file objects, array contents, and function
definitions (and other properties) are saved, but the contents of the push
down lists, including previous values of bound variables, cannot be saved,
so save should only be used from top level. (See also the section on
Gaining and Keeping Control, below)
In the PDP-10 implementations there is a function called suspend:
suspend LSUBR 0 to 2 args
suspend puts LISP in a state such that it can be :PDUMP'ed (ITS) or
SSAVE'd (DEC-10) and later restarted. When the saved core image is
restarted, everything will be the same as it was when suspended, and
control will return from the invocation of suspend.
suspend may be used at any point in a computation, with the restriction
that no I/O devices other than the terminal may be in use. If an I/O
device other than the terminal is in use, a fail-act correctable error
occurs, indicating the offending device(s).
In the ITS implementation, care is taken so that all subsystems saved
with suspend from the same version of LISP will share the pure pages of
LISP. In addition, all invocations of a particular subsystem will share
the pure pages peculiar to that subsystem. Declaration of data to be
placed in pure pages is described in a later section.
December 6, 1977 ∪3-1.7.2 Page 3-71
**DRAFT** Maclisp Reference Manual **DRAFT**
In the TOPS-10 implementation, care is taken so that all LISPs and
subsystems saved from LISP will share the same high segment.
After suspend has prepared the LISP core-image for dumping, it returns
control to the operating system so that it can be dumped.
If suspend is given an argument, that argument is explodec'ed and the
resulting character string is passed back to the operating system as a
command. (See cline and valret.) (On ITS this is done with a .VALUE; on
SAIL this is done with a PTLOAD. On TOPS-10 the argument is ignored.)
The SAIL version uses the second argument, if present, as a file name.
The high segment is dumped under this name. The low segment should then be
saved with SAVE, not SSAVE. This is used to create subsystems that can
share the same high segment containing code specific to the subsystem.
Commonly one will write a setup routine for a subsystem like this:
(progn
(terpri)
(princ 'options:)
... read in options ...
(terpri)
(princ 'loading)
... load in files of functions ...
(gc)
(sstatus gctime 0)
(suspend)
(start-the-subsystem)
))
The subsystem's environment is now ready for dumping. Alternatively,
one might write (on ITS)
(suspend '|:PDUMP DSK:FOODIR; TS NSUBSYS/
:$ALL DONE$/
|)
which will do the dump itself and print a message when done. On SAIL one
might write
Page 3-72 ∪3-1.7.2 December 6, 1977
**DRAFT** The System **DRAFT**
(suspend '|SAVE SYS:FOO.SAV|
'(FOO SHR SYS))
See also the valret function.
1.7.3 Gaining and Keeping Control
In the Multics implementation, when a saved environment is restarted it looks
like an error that returns to top level. The forms in the errlist are
evaluated. These forms should do whatever is necessary to start up the
subsystem. The arguments to the command which invoked the subsystem may be
obtained via (status arg) or (status jcl).
In the PDP-10 implementations, when a saved environment is restarted
execution continues from the point where suspend was called. The next form
evaluated should do whatever is necessary to start up the subsystem. (It is
also possible to cause the same return to top level on startup as in the Multics
implementation by using valret instead of suspend.) The arguments of the
command line which invoked the subsystem may be obtained via (status jcl).
If the subsystem wants to hide the underlying Maclisp from the user, it has a
number of facilities available. By setting up its own user-interrupt handlers
it can handle any LISP errors which occur itself. In Newio implementations, it
can alter, augment, or abolish standard interrupt control characters. It can
replace the Maclisp interpretive interaction loop with its own by using (sstatus
toplevel) and (sstatus breaklevel). It can also provide a totally different
interaction loop by not returning control to the LISP top level when it is
started, but instead retaining control in its own functions which read and
respond to user input.
It is possible for a subsystem to retain the trappings of Maclisp but change
the way things read and print. Macro characters and the readtable can be used
to change the way input is parsed; alternatively, setq'ing the variable read
will redefine the system reader function (in the PDP-10 implementations). All
output by Maclisp (with the exception of character-string messages) is done
through the function prin1, and the subsystem may redefine this function. In
the Multics implementation one simply redefines it, but in the PDP-10
implementations the variable prin1 must be bound to the function which is to
substitute for prin1.
December 6, 1977 ∪3-1.7.2 Page 3-73
**DRAFT** Maclisp Reference Manual **DRAFT**
Some subsystems don't do any of this, but simply consist of standard Maclisp
augmented by some additional functions which may be used in forms typed in at
top level.
1.7.4 Purity
In the PDP-10 implementations, there are some facilities which allow
subsystems to put their non-changing data, function definitions, binary code,
etc. into pure pages. This decreases the load on memory by sharing pages
between multiple users of the same subsystem.
There are some extra storage spaces which are used to store pure (unchanging)
LISP objects. These are the pure list, pure fixnum, pure flonum, pure bignum,
and pure hunk spaces.
purcopy SUBR 1 arg
This function makes and returns a copy of its argument in pure storage.
This is primarily of use in the creation of large sharable systems like
MACSYMA. In implementations other than PDP-10 implementations with pure
spaces, purcopy simply returns its argument.
There are a number of features which control how binary code and
constants are purified when a compiled program is loaded into LISP.
bporg VARIABLE
The value of bporg should always be a fixnum, whose value is the address
of the first unused word of binary program space. This value generally
should not be altered by the user, but only examined. bporg is updated
whenever binary code is loaded by lap or fasload.
bpend VARIABLE
This variable should also always have a fixnum as its value; this
indicates the last available word of binary program space. This is updated
by many internal LISP routines, such as the garbage collector, the array
allocator, and lap and fasload.
Page 3-74 ∪3-1.7.3 December 6, 1977
**DRAFT** The System **DRAFT**
pagebporg SUBR no args
Causes the variable bporg to be adjusted upwards so as to lie on a page
boundary. This is principally useful on ITS in conjunction with the
function purify. pagebporg returns the new value of bporg.
getsp LSUBR 1 to 2 args
(getsp n flag) ensures that (- bpend bporg) is at least n, allocating
more memory if necessary. If flag is non-nil, then the memory added is
marked as being potentially purifiable by purify. If flag is omitted, the
value of pure is used. This is generally used by clever subsystems loaders
to expand binary program space quickly so that fasload will not require
several garbage collections to do the same thing. It can also be used by
programs which create and destroy many arrays. See also noret.
noret VARIABLE
Normally the garbage collector will return memory to the time-sharing
system if (- bpend bporg) is very large, but setting noret non-nil prevents
this. This is useful in conjunction with getsp.
purify SUBR 3 args
The first two arguments to purify should be fixnums, delimiting a range
of memory within the LISP system. The third argument is a flag. If it is
nil, then the pages covered by the specified range of memory are made
impure, i.e. writable. If it is t, then the pages are made pure, i.e.
read-only and sharable. If it is bporg, then the pages are made pure, but
in addition some work is done to make sure that no UUO on those pages may
ever be "clobbered". (See pure and purclobrl) This option should always be
used if the pages involved contain binary code loaded by lap or fasload.
Presently purify does nothing in the TOPS-10 implementation; it is intended
primarily for producing systems built on LISP, such as MACSYMA, in such a
way that pure pages can be shared between users. Example: the following
function might be used to produce a sharable system on ITS:
December 6, 1977 ∪3-1.7.4 Page 3-75
**DRAFT** Maclisp Reference Manual **DRAFT**
(defun superdump ()
(setq lopage (pagebporg)) ;save low page address
(setq pure 3) ;specifies pure code
(setq lopage (+ lopage 6000)) ;allow for area 1
; (ITS page = 2000 words)
(fasload funny fasl) ;load up system
(fasload weird fasl)
(uread some lap)
...
(sstatus toplevel ;set up top level for system
'(top-handler))
(setq hipage (pagebporg)) ;save high page address
(purify lopage (1- hipage) 'bporg) ;purify pages
(suspend '|:pdump sys:ts super↑M|) ;tell ddt to dump
(terpri) ;stuff for system startup
(princ 'welcome/ to/ supersystem/!)
(terpri))
pure VARIABLE
This variable, initially nil, should be made non-nil by the user before
loading binary code which is to be made pure. It signals lap and fasload
to be circumspect about any UUO's in the code, because pure UUO's cannot be
clobbered to be PUSHJ's or JRST's. lap solves this problem by clobbering
the UUO immediately if the referenced function is already defined and is
itself a subr rather than an expr; otherwise the UUO is made permanently
unclobberable (i.e. CALL is converted to CALLF, etc.).
fasload is somewhat more clever: it too tries to clobber each UUO
immediately, but if it can't it puts the address of the UUO on a list
called purclobrl, which is checked at the end of each call to fasload, and
each UUO on the list is clobbered at that time, if the appropriate function
had been loaded by that call to fasload. If the function never does get
defined, then purify will also check purclobrl and convert each UUO to its
permanently unclobberable form.
If pure has a fixnum as its value, then fasload (but not lap) behaves
somewhat differently. pure should be the desired size in words of the
"uuolinks area"; this is rounded up to a whole number of pages. (If pure
is between 1 and 8, it is the number of pages rather than the number of
Page 3-76 ∪3-1.7.4 December 6, 1977
**DRAFT** The System **DRAFT**
words.) Let the number of pages be n. First fasload calls pagebporg, and
then reserves 2n pages of binary program space, unless a previous call to
fasload has already reserved them (i.e. they are reserved only once). Thus
fasload has two sets of n pages to work with; we shall call the first set
"area 1" and the second set "area 2". Now whenever fasload has to load a
clobberable UUO, it does not place it in the code being loaded, but rather
hashes it and places it in area 1 if it was not there already; a copy is
placed in the same relative position in area 2. Then an XCT instruction
pointing to the UUO in area 1 is placed in the binary code. When all
loading has been done, area 2 may be purified, but area 1 may not.
Now when running the code, the UUO's pointed to by the XCT's may be
clobbered (the PDP-10 LISP UUO handler is clever about XCT), provided, of
course, that the value of nouuo is nil, and the code will run faster the
second time around because the XCT's will point to PUSHJ's. However, if
(sstatus uuolinks) is called, then area 2 is copied back into area 1,
effectively unclobbering all the UUO's. This allows the called functions
to be traced again, for example, or redefined as expr code. Naturally, an
area large enough to contain all the UUO's should be reserved; (status
uuolinks) (q.v.) yields information relevant to this. As a rule of thumb,
the area hsould be 20% larger than the number of functions called by the
loaded code (including LISP functions such as equal). For the DEC-10
version, pure may be set to a negative value. the magnitude is used as
above, while the sign controls which segment to load into (positive = low
segments, negative = high segment). A negative value also causes uuolinks
area 1 to go in the low segment, and area 2 in the high segment. For
compatibility, a negative value means the same as a positive value to the
ITS implementation.
purclobrl VARIABLE
Used by fasload to keep track of UUO's which are potentially but not
immediately clobberable.
*pure VARIABLE
This variable controls automatic purification of S-expressions and
atomic symbols. If it is set non-nil (the initial value is nil), then the
following are placed in pure storage spaces instead of regular storage
spaces: pnames of atomic symbols; list, fixnum, flonum, bignum, and hunk
December 6, 1977 ∪3-1.7.4 Page 3-77
**DRAFT** Maclisp Reference Manual **DRAFT**
constants used by code loaded with fasload; properties whose indicators are
in the list which is the value of the variable putprop (initially subr,
fsubr, and lsubr). In the SAIL implementation, if *pure is a fixnum, it
should be an estimate of the total number of pure data structures needed,
including all files previously loaded and the initial LISP system pure data
structures (currently about 6000. words). This causes purcopy to use the
high segment for pure data. Making the estimate in *pure too large merely
wastes space in the high segment; making it too small causes purcopy to
make copies in the low segment when it runs out of room in the high
segment. This whole feature only works if pure is a negative fixnum.
putprop VARIABLE
If the value of *pure is non-nil and the third argument to the putprop
function is in the list of indicators which is the value of the variable
putprop, then the second argument is passed through purcopy to purify the
structure. Furthermore, the two cells of the property list are cons'ed
from pure list space. Since impure cells must precede pure cells in the
property list, putprop may not put a new property at the front of the
property list in this case.
The putprop and remprop functions know about purified property lists.
If necessary, they will copy the property list (but not the properties
themselves) into non-pure storage so that it can be modified. This is true
regardless of the value of *pure. Recall also that defprop and defun use
putprop.
Page 3-78 ∪3-1.7.4 December 6, 1977
**DRAFT** The System **DRAFT**
1.8 Miscellaneous Functions
1.8.1 The Status Functions
status FSUBR
The status special form is used to get the value of various system
parameters. Its first argument, not evaluated, is an atomic symbol
indicating which of its many functions status should perform. The use of
additional arguments depends on what the first argument is. These
arguments may or may not be evaluated, depending on the first argument. If
certain additional arguments are omitted, a default value is supplied,
again depending on what the first argument is. The various status
functions are listed below.
sstatus FSUBR
The sstatus special form is used to set the value of various system
parameters. Its arguments are similar to those of status.
These are the things that you can do with status and sstatus:
STATUS FUNCTIONS FOR I/O
tabsize (status tabsize) returns the number of character positions assumed
between tab stops, which depends on the implementation. Currently
this is 8 in the PDP-10 implementation and 10. in the Multics
implementation.
newline (status newline) returns a fixnum which is the ascii code for the
character which marks the end of a line of input. For example, one
might say (= (setq ch (tyi)) (status newline)) . Cureently this is 15
(octal) in the PDP-10 implementation and 12 (octal) in the Multics
implementation.
charmode (status charmode f) returns the value of the character-mode switch for
the file f. If f is t or omitted the value of tyo (the default output
terminal) is assumed. If the character-mode switch is t (the normal
case for the terminal) output is sent to the device as soon as it is
December 6, 1977 ∪3-1.8 Page 3-79
**DRAFT** Maclisp Reference Manual **DRAFT**
generated. If the switch is nil (the normal case for files other than
the terminal) output is held until a newline is typed, an error
occurs, input is requested, or the buffer becomes full. (In the
Multics implementation, you can also cause the buffer to be sent by
using the force-output function.) This provides increased efficiency
at the cost of not immediately seeing all output in some cases.
(sstatus charmode x f) sets the character-mode switch of the file f (f
may be t or omitted to signify the output terminal in the tyo
variable) to x, which may be nil or t. x and f are evaluated.
Currently in the PDP-10 implementation it is not possible to set the
character-mode switch; one must specify it initially to open. See
also (status linmode) and (status filemode).
linmode (status linmode) reads the "line mode" and (sstatus linmode x) sets
the "line mode" to x (t or nil.) These functions take an optional
extra argument, which is the file whose line mode is being discussed.
This defaults to the value of tyi (the default input terminal). In
any case, this file must be a terminal. In some implementations the
"line mode" may not be changed. If the "line mode" is t, user input
is buffered up a line at a time before being sent to LISP. The input-
editing conventions of the host operating system are used. If the
"line mode" is nil, LISP sees each character as it is typed and
applies its own input editing conventions. This mode can provide
input facilities more suited to LISP and possibly better handling of
the terminal, if it is a type that LISP knows a great deal about.
However, it uses more machine resources. It is possible for a user
program to take direct control of the terminal when the "line mode" is
nil; however, this may require knowledge of the (sstatus tty). The
Multics implementation always operates with a "line mode" of t. See
also the (sstatus ttyscan) function below.
ttyint (sstatus ttyint char func file) turns on a tty interrupt character.
When the character char is typed on the input tty file, the LISP
program will be interrupted and the function func will be applied to
two arguments - file and char. If file is omitted, the value of tyi
(the default input file) is assumed. The char may be either a
character object or a fixnum (ascii code). Any of the 128. ascii
characters may be used. The func may be either an ordinary functional
form or a fixnum, which means the default system action for the
character with that ascii value. For instance a func of 7 means quit
back to the top level of LISP (control-G). If func is nil, the char
is made non-interrupting. All three arguments are evaluated.
Page 3-80 ∪3-1.8.1 December 6, 1977
**DRAFT** The System **DRAFT**
In the ITS implementation, it may be necessary to use (sstatus tty) to
inform ITS about non-standard interrupt characters.
In PDP-10 implementations which support fixnum tty input mode (more
than 128. characters), then when an interrupt occurs the character is
folded down to 7 bits before selecting an interrupt function.
However, the unfolded character is passed as the second argument to
the interrupt function, so that the function can filter out unwanted
characters. If a fixnum is used for the function, then the fixnum may
specify required or forbidden supra-ascii bits as follows: adding bit
n to the fixnum requires the bit, and adding (lsh n 18.) forbids it.
For example, in the ITS implementation using 400000207 (octal) as a
function allows CONTROL-G to perform a quit, but not the "pi"
character of CONTROL-META-G, since 200 requires the CONTROL bit and
400000000 forbids the META bit.
Example: a standard subsystem convention is to use ↑↑ (control-
uparrow, ascii 036) as a quit character which restarts the
subsystem.
(sstatus ttyint 36
(function (lambda (f c)
(sstatus toplevel '(restart-subsystem))
(do nil ((or (= (listen f) 0)
(= (tyi f) 36))))
(↑g))))
(defun restart-subsystem ()
(setq errlist nil)
(sstatus toplevel nil)
(nointerrupt nil)
(initialize-the-world)
(enter-subsystem))
(status ttyint char file) returns func, the interrupt function for the
character char on the tty file. file may be omitted. It defaults to
t.
ttycons (sstatus ttycons tty1 tty2) binds two tty files into a console. One
should be an input tty and the other an output tty. If tty1 or tty2
is t, it will be taken as the appropriate direction of the default
terminal (determined by the variables tyi and tyo). The binding into
a console is basically used for purposes of echoing. In addition,
December 6, 1977 ∪3-1.8.1 Page 3-81
**DRAFT** Maclisp Reference Manual **DRAFT**
interrupt characters typed on an input tty file should affect the
output on its corresponding output tty file, not on some other tty.
If one argument is a tty and the other is nil, a ttycons relationship
involving that tty will be broken.
To put echoing at the bottom of the screen, distinct from output,
(sstatus ttycons t
(open 'tty: '(echo out tty)))
which conses tty input with a new tty output channel which is set to
go to the echo area at the bottom of the screen.
(status ttycons tty1) returns the other tty file which is bound into a
console with tty1, or nil if there is none.
filemode (status filemode file) => (open-mode-list . internal-cruft). open-
mode-list is a suitable second argument for the open function.
internal-cruft is a list of implementation-dependent information which
may sometimes be needed by special programs.
The following symbols may appear in internal-cruft. (These are the
standardized ones; additional symbols may appear at the discretion of
the implementation.)
cursorpos This file (an output tty) has the ability to position
its cursor anywhere on the screen.
filepos The filepos function can be used to access randomly
within the file.
rubout This file (an output tty) has selective erase
capability. (cursorpos 'x) will work.
sail This file (an output tty) has the so-called SAIL
character set. This is an extension of ascii which is
related to but not necessarily identical to the actual
character set used at SAIL.
If file is a closed file object, (status filemode) merely returns nil
instead of giving an error.
Page 3-82 ∪3-1.8.1 December 6, 1977
**DRAFT** The System **DRAFT**
ttyscan (sstatus ttyscan func file) allows the user to supply a function which
performs initial processing of terminal input. func is a functional
form, and file must be a tty input file. If it is omitted, the value
of tyi (the default input terminal) is assumed. Both arguments are
evaluated.
When LISP wants to take input from file, it first calls the prescan
function, which is supposed to gobble down a complete unit of input
(for instance an S-expression) and return a list of characters. The
prescanner supplied automatically by LISP when a tty input file is
first opened counts parentheses and does fancy rubout processing on
display terminals. It also implements the CTRL/k, CTRL/l, and CTRL/u
characters (as appropriate to the implementation) which allow the
complete input to be redisplayed or cancelled, and takes care of
force-feed characters. A user-written prescanner might provide
additional features such as super-parentheses, name recognition and
completion, or fancy editing.
The prescan function func is applied to three arguments: the file, the
name of the input function on whose behalf it is acting (read, readch,
or readline), and a fixnum which, in the case of read, is the count of
the number of unmatched left-parentheses. It is supposed to return a
list of fixnums, which represent characters. The prescan function
should read the input with tyi, since it and tyipeek are the only
input functions which doesn't call the prescan.
If the pre-scan function returns nil, an eof condition occurs for the
input file. This is the standard way to signal over-rubout.
There is a function called rubout to assist the pre-scanner in
processing rubouts. It is described on page .
It is a good idea for the prescan function to lambda-bind echofiles to
nil so that characters do not appear in the echo files twice, and so
that the echo files reflect the "clean" input after rubout processing.
The system-supplied prescan function does this.
(status ttyscan file) returns the file's func.
This feature does not presently exist in the Multics implementation.
December 6, 1977 ∪3-1.8.1 Page 3-83
**DRAFT** Maclisp Reference Manual **DRAFT**
The following status functions exist only in the ITS implementation
ttysize (status ttysize f) returns the height and width of the terminal open
as the file f. If f is omitted, the value of tyo (the default output
terminal) is used. The result is a dotted pair of fixnums "(ttyheight
. ttywidth)" (cf. cursorpos). If the terminal is a printing console
instead of a display, then the height will be a very large number such
as 200000000000 or so.
ttytype (status ttytype f) returns the type of the terminal open as the file
f. If f is omitted, the value of tyo (the default output terminal) is
used. The result is a fixnum taken from the TCTYP (not TTYTYP!)
variable in ITS:
0 Printing terminal.
1 Good Datapoint.
2 Bad Datapoint ("loser").
3 Imlac.
4 Tektronix.
5 PDP-11 ("Knight") TV.
6 Memorex.
7 Software terminal.
10 Terminet.
11 Display using standard ascii display codes.
12 Datamedia.
This status call is meant only for esoteric purposes. To find out
whether you can position the cursor or erase characters, use (status
filemode).
tty (status tty f) returns a list of three fixnums which control the
behavior of the terminal open as file f. If f is omitted, the value
of tyi (the default input terminal) is used. (sstatus tty x y z f)
sets these three fixnums to x, y, and z. If z is omitted, it is not
changed. If f is omitted, the value of tyi (the default input
terminal) is used. The three fixnums are the ITS variables TTYST1,
TTYST2, and TTYSTS. Their meaning is as follows:
TTYST1 and TTYST2 are divided into twelve groups of six bits. Each
group controls a certain subset of the ascii characters. These
groups, in left-to-right order within TTYST1 and TTYST2, are:
Page 3-84 ∪3-1.8.1 December 6, 1977
**DRAFT** The System **DRAFT**
(0) ↑@-↑F ↑K ↑L ↑N-↑R ↑T-↑Z ↑\-↑←
(1) A-Z a-z (letters)
(2) 0-9 (digits)
(3) ! " α# $ % & ' , . : ; ? @ \ ` | }
(4) + * - / = ↑ ← (arithmetic operators)
(5) < > ( ) [ ] α (parentheses)
(6) ↑G ↑S
(7) ↑I ↑J (tab, linefeed)
(8) altmode (ascii 33)
(9) ↑M (carriage return)
(10) rubout (ascii 177)
(11) space, ↑H (backspace)
40 Echo the character when read by LISP (normal mode).
20 Echo the character when typed (this may disappear from ITS
soon!).
10 Echo in "image mode" rather than "ascii mode"
4 Convert lower case to upper case (applicable only to letters; not
used by LISP).
2 Activation characters. If this bit is not set, then ITS will not
schedule LISP to run when you type a character within the group
even if LISP was waiting for terminal input. For example,
letters and digits normally have this bit off, since typing one
cannot terminate an S-expression, and there is no point in
"activating" LISP until an S-expression terminator like space or
")" has been typed. An exception to this rule is that for tyi
and readch LISP asks ITS to activate on any character. Normally
only groups (0), (5), (6), (7), (9), (10), and (11) are
activators. If you alter the readtable, you may wish to change
the activator groups.
1 Interrupt characters. If this bit is not set, then even if you
have used (sstatus ttyint) to set up an interrupt function you
will not get the interrupt since ITS won't tell LISP about it.
Initially only groups (0) and (6) cause interrupts (these include
all control-characters except altmode and format effectors). If
you give @ an interrupt function, for example, you should set
this bit for group (3).
On a keyboard with supra-ascii characters, one may wish to use
meta-characters for interrupts. Suppose we want meta-D to run
the function foo. Then we would turn on the interrupt bit for
group 1 (letters) and say:
December 6, 1977 ∪3-1.8.1 Page 3-85
**DRAFT** Maclisp Reference Manual **DRAFT**
(sstatus ttyint 104 ;letter D
(function (lambda (f c)
(and (= c 504) ;meta bit
(foo)))) ; is 400
- filter out plain "D"
The third variable, TTYSTS, has these bits of interrest:
40000←22 (%TSFCO) If set, output on this terminal uses the supra-ascii
convention of an "alpha" prefix for "control" and "beta" for
"meta".
20000←22 (%TSALT) If set, ascii codes 175 and 176 are ∪not converted
to 33 on input.
10000←22 (%TSROL) If set, terminal is in scroll mode.
4000←22 (%TSSAI) If set, echoing and ascii output use the SAIL
character set; otherwise, control characters are output in
the form "↑X".
10←22 (%TSNOE) If set, defer echoing of typed characters.
2←22 (%TSSII) If set, terminal uses "super-image input" mode. Not
even ↑Z and ↑← can take their usual effect for ITS.
Notice that some of these bits affect output even though the argument
is an input file; they affect the output terminal which is logically
related by ITS to the input terminal. Normally LISP or the user will
arrange for (status ttycons) to reflect this relationship.
STATUS FUNCTIONS FOR THE OLD I/O SYSTEM
uread (status uread) returns a 4-list for the current uread input source, or
nil if uread is not being done.
(sstatus uread --args--) is the same as (uread --args--)
uwrite (status uwrite) returns the 2-list for the current uwrite output
destination.
(sstatus uwrite --args--) is the same as (uwrite --args--)
crunit (status crunit) returns a 2-list of the current unit; i.e. device and
directory.
Page 3-86 ∪3-1.8.1 December 6, 1977
**DRAFT** The System **DRAFT**
(sstatus crunit device directory) sets the current default device and
directory for uread, etc. The arguments are not evaluated.
crfile (status crfile) returns a 2-list giving the file names for the current
file in the "uread" I/O system.
(sstatus crfile name1 name2) sets the current default file names for
uread, etc. The arguments are not evaluated.
STATUS FUNCTIONS FOR THE READER
See section 13.6.2 for a description of how the parameters controlled by
these functions are used. All of these parameters are kept in the readtables,
and the status functions below deal with the readtable which is the value of the
variable readtable. Note: in the following, c represents an argument specifying
a character. If c is non-atomic it is evaluated, and the value must be a fixnum
which is the ascii code for a character. If c is atomic it is not evaluated,
and it may be a fixnum or a character object.
chtran (status chtran c) gets the character translation table entry for the
character c. This is the ascii code of a character substituted for c
when it appears in a pname being read in. This feature is used in the
PDP-10 implementations to translate lower-case input to upper case.
(sstatus chtran c k) sets c's character translation to k. f2k is
always evaluated. See the setsyntax function.
syntax (status syntax c) returns the syntax bits for the character c, as a
fixnum.
(sstatus syntax c m) sets c's syntax bits to m. m is evaluated and
returned. The setsyntax function is usually a better way to do this,
however.
Note that in the above two sstatus calls, if c is a macro character it is
changed back to its standard syntax and chtran before the requested operation is
performed. However, if in the standard readtable c is a macro (i.e. ' and ;),
instead of being changed to its standard syntax and chtran its syntax is set to
502 (slashified extended alphabetic) and its chtran is set to itself.
macro (status macro c) returns nil if c is not a macro character. If c is a
macro character it returns a list of the macro character function and
December 6, 1977 ∪3-1.8.1 Page 3-87
**DRAFT** Maclisp Reference Manual **DRAFT**
the type, which is nil for normal macros and splicing for splicing
macros.
(sstatus macro c f) makes c a macro character which calls the function
f with no arguments. f is evaluated. A fourth argument to sstatus
may be supplied. It is not evaluated. If it is an atomic symbol
whose pname begins with s, c is made a splicing macro. If f is nil,
instead of c being made a macro-character, c's macro abilities are
taken away and c becomes an ordinary extended-alphabetic character.
The setsyntax function is generally a better way to do this, however.
+ (status +) gets the value of the + switch (t or nil) in the readtable.
This switch is normally nil. If it is t, atomic symbols more than one
character long beginning with a + or a - are interpreted as numbers by
the reader even if they contain letters. This allows the use of input
bases greater than ten. See ibase.
(sstatus + x) sets the + switch to t or nil depending on x, which is
evaluated. The new value of the + switch is returned.
ttyread (status ttyread) returns the value of the ttyread switch in the
readtable. At present this is not used for anything in the Multics
implementation. In the PDP-10 implementation it controls whether tty
"force feed" characters are used. See section 13.6.2 for details of
force feed characters.
(sstatus ttyread x) sets the ttyread switch to t or nil depending on
x, which is evaluated. Again, file defaults to t. The new value of
the switch is returned.
STATUS FUNCTIONS FOR THE PRINTER
terpri (status terpri file) returns the value (t or nil) of the terpri switch
for the file, which defaults to t. This switch is normally nil. If
it is t, the output functions such as print and tyo will not output
any extra newlines when lines longer than linel are typed out. See
also the terpri variable.
(sstatus terpri x file) sets the terpri switch.
Page 3-88 ∪3-1.8.1 December 6, 1977
**DRAFT** The System **DRAFT**
← (status ← ) returns the value (t or nil) of the ← switch for the
current readtable, which defaults to t. If this switch is t, the ←
format for octal fixnums with lots of trailing zeroes is not used.
(sstatus ← x file) sets the ← switch to the value of x, t or nil.
abbreviate(status abbreviate) returns the value of the abbreviation control.
See section 13.7 for a description of the abbreviation control. Note
that the abbreviation control is kept in the current readtable.
(sstatus abbreviate n) sets the abbreviation control to n.
(sstatus abbreviate nil) turns off abbreviation.
(sstatus abbreviate t) turns on a maximal amount of abbreviation.
STATUS FUNCTIONS FOR THE GARBAGE COLLECTOR
gctime (status gctime) returns the number of microseconds spent garbage-
collecting.
(sstatus gctime n) resets the gctime counter to n and returns the
previous value of the gctime counter.
It is a good idea to do (gc) and then (sstatus gctime 0) just before
dumping out a subsystem. See also (status gcwho).
spcnames (status spcnames) returns a list of the names of all the spaces
available in the LISP being used. These are the names acceptable to
the alloc function and to the following status functions which require
a space argument.
spcsize (status spcsize space) returns the actual, current size of space, in
words. space is evaluated.
gcmax (status gcmax space) returns the gcmax parameter for space.
sstatus gcmax space n) sets the gcmax parameter for space to n.
See the alloc function.
gcmin (status gcmin space) returns the gcmin parameter for space.
December 6, 1977 ∪3-1.8.1 Page 3-89
**DRAFT** Maclisp Reference Manual **DRAFT**
sstatus gcmin space n) sets the gcmin parameter for space to n.
See the alloc function.
gcsize (status gcsize space) returns the gcsize parameter for space.
sstatus gcsize space n) sets the gcsize parameter for space to n.
See the alloc function.
purspcnames purspcnames) returns a list of names of spaces which have pure
(status
versions. See purcopy.
pursize (status pursize space) returns the size of the pure version of space.
Only spaces returned by (status purspcnames have pure versions.
pdlnames (status pdlnames) returns a list of the names of all the pdls
available in the LISP being used. These are the names acceptable to
the following status functions which require a pdl argument.
pdlsize (status pdlsize pdl) returns the current number of words on a pdl.
pdlroom (status pdlroom pdl) returns the "pdlroom" of a pdl, i.e. the maximum
size to which it may ever grow.
pdlmax (status pdlmax pdl) returns the current value of the pdlmax parameter
of a pdl.
(sstatus pdlmax pdl size) sets the pdlmax parameter for the pdl space
to size. Both arguments are evaluated. See also the alloc function.
The following status function exists only in the PDP-10 implementation
memfree (status memfree) returns the number of words of address space not yet
allocated for any purpose (i.e. still available for allocation to
various spaces).
ENVIRONMENT ENQUIRIES
Note: the various enquiries related to time may return nil in the rare
circumstance that LISP cannot determine what time it is.
Page 3-90 ∪3-1.8.1 December 6, 1977
**DRAFT** The System **DRAFT**
date (status date) returns a 3-list indicating the current date as (last-
two-digits-of-the-year month-number day).
dow (status dow) returns an interned atomic symbol which is the name of
the current day of the week.
daytime (status daytime) returns a 3-list of the 24-hour time of day as (hour
minute second).
system (status system x) returns a list of the system properties (that is,
properties whose values are supplied in the initial LISP system) of
the atomic symbol x, which is evaluated. This list may contain subr,
fsubr, lsubr, macro, autoload, or array if x is a system function, and
value if this atomic symbol is a system variable.
uname (status uname) returns an interned atomic symbol whose pname is the
user's login name. In the ITS implementation this is just the .UNAME
. In the Multics implementation this is in the format
User.Project.Instance; the dot will be slashified if print is used to
display this. In the TOPS-10 implementation this is actually a list
(proj prog) representing the project-programmer number rather than a
symbol. See also (status xuname).
udir (status udir) returns the name of the user's directory. In the ITS
implementation this is the user's "master sname", which is usually the
same as the user's name as returned by (status uname). In the Multics
implementation this is the user's default working directory. In the
TOPS-10 implementation this is a list (proj prog).
LISPversion
(status LISPversion) returns the version identification of LISP. This
is usually an atomic symbol.
jcl (status jcl) returns the "job command line" from DDT in the PDP-10
implementation. This is used to specify the init file, but
interpretation of the file name is terminated by altmode, and (in the
ITS implementation at least) additional information may follow the
altmode. The init file may wish to examine this information. In the
Multics implementation this returns the explodec'd second argument of
the LISP command, or else nil if the LISP command did not have two
arguments. If LISP was invoked by
LISP environment←name "foo bar"
December 6, 1977 ∪3-1.8.1 Page 3-91
**DRAFT** Maclisp Reference Manual **DRAFT**
then (status jcl) => (f o o / b a r)
This function is also used by subsystems implemented in Maclisp to
pick up the arguments from the command which invoked them.
The following status functions exist only in the PDP-10 implementation.
xuname (status xuname) in the ITS implementation returns the .XUNAME as an
interned atomic symbol. The xuname of a job is the name of the user
for whom it is being run. This is normally the same as .UNAME with
trailing digits stripped off. It is possible, however, to cause a job
to be run with a different xuname. System initializations should
check the xuname to determine which init files to use. In the DEC-10
implementation this is the same as (status uname).
jname (status jname) returns the job name as an interned atomic symbol. The
jname of a job is a unique identifier for it among all of the user's
jobs. In the ITS implementation this is the job's .JNAME . In the
DEC-10 implementation this is nnnLISP, where nnn is the job number in
decimal.
xjname (status xjname) in the ITS implementation returns the .XJNAME as an
interned atomic symbol. This is normally the name of the program that
the user invoked to create this job. Usually this will be the same as
the jname, but sa user can invoke several copies of a program at once,
each having a unique jname but all having the same xjname. In the
DEC-10 implementation this will return the job's "program name". On
Multics, it should return (status arg 0).
seglog (status seglog) returns the logarithm (base 2) of the size of a
"segment", the unit of space allocation. (The size of a segment need
not be the same as the size of a memory page.)
The following status functions only exist in the ITS implementation.
its (status its) returns a list of five items (obtained from the SSTATU
symbolic system call):
(1) The amount of time in seconds until ITS is going down, as a
flonum, -1.0 if ITS doesn't plan to go down, or -2.0 if ITS
already is down.
(2) a fixnum which is nonzero if ITS is being debugged.
Page 3-92 ∪3-1.8.1 December 6, 1977
**DRAFT** The System **DRAFT**
(3) the number of users logged in, as a fixnum.
(4) the number of memory errors the system has survived, as a fixnum.
(5) the time in seconds the system has been up, as a flonum.
Some of this information may be of use to a user handler for the sys-death
interrupt.
hactrn (status hactrn) returns an atomic symbol indicating what kind of job
is superior to the LISP. Current possibilities are ddt, LISP, t (a
superior of unknown type), and nil (no superior).
The following status functions exist only in the Multics implementation.
paging (status paging) returns a list of the paging-device page reads and
total page reads that have been caused by this process.
arg (status arg n) returns the n+1'th argument of the command which
invoked the subsystem, as an interned atomic symbol. (The first
argument, (status arg 0), is the name of the subsystem.) nil is
returned if n is greater than the number of arguments to the command.
STATUS FUNCTIONS FOR THE WHO-LINE
The following status functions exist only in the ITS implementation.
whol (sstatus whol a b c d) sets four user parameters for the "who-line"
which appears at the bottom of some display terminals. These control
the print format of information specified by (sstatus who2), (sstatus
who3), and (sstatus gcwho). Each of the four parameters specified
eight bits of information. Parameters a and c must be fixnums; b and
d may be fixnums or atomic symbols (which represent the ascii value of
the first character in their pnames). Their meaning is as follows:
a If 1, suppress display of entire who-line.
200
100 If 1, suppress space between halves of who2.
70 Mode for printing high 18. bits of who2.
7 Mode for printing low 18. bits of who2.
b If non-zero, print between halves of who2 as an ascii character.
177
200 If 1, print character twice.
c If 1, suppress space between who2 and who3.
200
December 6, 1977 ∪3-1.8.1 Page 3-93
**DRAFT** Maclisp Reference Manual **DRAFT**
177 As for a, but affects who3.
d As for b, but affects who3.
377
That is, if the who-line is displayed at all, the user information
appears in the form:
PPP##-QQQQ=RRRR@@+SSSS
where:
PPPP result of printing the high 18. bits of who2 as specified by
is the
a's 70 bits.
QQQQ bits of who2, by c's 7 bits.
low 18.
RRRR 18. bits of who3, by c's 70 bits.
high
SSSS bits of who3, by c's 7 bits.
low 18.
## to two characters, specified by d.
zero
- unless a's 100 bit is set.
space,
= unless c's 200 bit is set.
space,
+ unless c's 100 bit is set.
space,
The possible print modes are:
0 not print.
Do
1 in packed form:
Date
774000 Year mod 100.
3600 Month (January = 1)
174 Day of month.
Printed as mm/dd/yy.
2 in fortieths of a second. Printed as hh:mm:ss.t.
Time
3 in half-seconds, printed as hh:mm:ss.
Time
Page 3-94 ∪3-1.8.1 December 6, 1977
**DRAFT** The System **DRAFT**
4 number.
Octal
5 number (no "." is printed).
Decimal
6 sixbit characters.
Three
7
Unused.
(status who1) returns a list of four fixnums (a b c d).
who2 (sstatus who2 x) sets the who2 variable to x if it is a fixnum, or to
the sixbit value of the first six characters of the pname of x if it
is a symbol. (status who2 x) returns who2 as a fixnum.
who3 (sstatus who3 x) and (status who3) are analogous to those for who2.
Example:
(all numbers in octal)
(sstatus who1 166 0 144 254)
(sstatus who2 'FOOBAR)
(sstatus who3 (+ (LSH 1234 22) 3456))
would cause "FOOBAR 1234,,3456" to appear in the who-line. The general idea is
that the user should set up who1 just once and then keep updating who2 and who3
as appropriate. For example, a routing that processed functions in files might
do
(sstatus who1 166 0 166 0)
and on opening a new file do
(sstatus who2 (cadr (truname infile)))
(sstatus who3 0)
and on encountering a function do
(sstatus who3 fnname)
and so continuously display the filename and function name being processed. One
advantage of doing this over printing the information is that LISP can update
who-line information even though it doesn't have control of the console.
gcwho (sstatus gcwho n) sets the gcwho switch to n. Currently only the low
two bits are significant.
The 1 bit means that during a garbage collection the who-line should be
usurped to display the message "GC:XXXX", where "XXXX" is the reason for the
garbage collection. On completion of the garbage collection the original user
who-line information is restored.
December 6, 1977 ∪3-1.8.1 Page 3-95
**DRAFT** Maclisp Reference Manual **DRAFT**
The 2 bit means that at the end of the garbage collection the who2 word
should be altered so that the low 18. bits get the result of (status gctime),
converted to fortieths of a second, and the high 18 bits get the result of (//
(* (status gctime) 100.)(runtime)), which is the percentage of time spent in gc.
If the user gives 52 octal and '% as the first two arguments to (sstatus who1),
then these quantities will be displaced in the form "nn% hh:mm:ss.t", just like
the standard system statistics for the LISP job. The user can still use who3
for his own purposes.
Subsystems which use (sstatus gcwho) should perform
(gc)
(sstatus gctime 0)
before dumping out the subsystem, so that the gc time percentages will be
accurate .
(status gcwho) returns the value of the gcwho switch.
MISCELLANEOUS STATUS FUNCTIONS
evalhook (sstatus evalhook t) enables the evalhook feature; (sstatus evalhook
nil) disables it. (status evalhook) returns the state of the switch.
See page 3-32 for the details of this feature.
toplevel (status toplevel) returns the top-level form, which is continually
evaluated when LISP is at its top level. If this is nil, a normal
read-eval-print loop is used.
(sstatus toplevel x) evaluates and returns x and sets the top level
form to this value.
For example, to make Maclisp have an evalquote top level similar to
LISP 1.5:
(sstatus toplevel
'(progn (print *)
(apply (read) (read)) ))
See section 12.1 for further details.
breaklevel (status breaklevel) returns the break-loop form. (sstatus breaklevel
x) sets this form. See page 3-6 for how this is used.
Page 3-96 ∪3-1.8.1 December 6, 1977
**DRAFT** The System **DRAFT**
uuolinks (status uuolinks) returns a number which represents the number of
available slots for linking between compiled functions. (status
uuolinks) returns nil if no uuolinks pages have been set up yet (see
pure). Otherwise it returns a list of two items. The first is t if
area 2 has been purified (and so no new calls may be put in the area),
and nil otherwise. The second is the number of slots not yet used in
the uuolinks areas.
(sstatus uuolinks) causes all links between compiled functions to be
"unsnapped." This should be done whenever (nouuo t) is done to insure
that the interpreter always gets a chance to save debugging
information on every function call.
divov (status divov) returns the state of the "divide overflow" switch. If
this switch is nil an attempt to divide by zero causes an error. If
the switch is t the result of a division by zero is the numerator plus
1.
(sstatus divov x) sets the "divide overflow" switch to x.
In the PDP-10 implementation, divov applies only to quotient. // and
//$ do not detect division by zero.
mar (sstatus mar cond loc) arms the mar (memory address register)
interrupt (currently available only in the ITS implementation).
(status mar) returns a list of cond and loc, or nil if the mar feature
is not in use. See the mar break feature for more details (page 3-
56).
features (status features) returns a list of symbols representing the features
implemented in the LISP being used. The following symbols may appear
in this list:
bibop PDP-10 big-bag-of-pages memory management scheme
lap this LISP has a LISP Assembly Program loaded
sort the sorting functions described in chapter 11 are
present
edit the edit function described in chapter 18 is
present
December 6, 1977 ∪3-1.8.1 Page 3-97
**DRAFT** Maclisp Reference Manual **DRAFT**
fasload the fasload facility described in chapter 14 is
present
↑f the "moby I/O" facility is present
bignum the arbitrary-precision arithmetic package is
included in this LISP
hunk the hunks data type and its associated functions
are present.
funarg the "fake funarg" feature (second argument to eval,
third to apply, etc.) is present.
strings character strings and the functions on them
described in chapter 8 are present
newio the I/O functions described in chapter 13 are
included in this LISP; if this feature is not
present only some of those functions are
available.
roman this LISP can read and print roman numerals (see
base and ibase.
trace the trace package (chapter 15) is present.
grindef the function definition formatter (chapter 16) is
present.
grind the file formatter (chapter 16) is present.
compiler this is the LISP compiler (chapter 14) rather than
the interpreter (see also compiler-state).
fastarith the fast-arithmetic features of the compiler are
present
ml this LISP is on the MathLab machine at MIT
ai this LISP is on the AI machine at MIT
mc this LISP is on the MC machine at MIT
Page 3-98 ∪3-1.8.1 December 6, 1977
**DRAFT** The System **DRAFT**
SAIL this LISP is running at SAIL
H6180 this LISP is on an H6180 Multics machine or a
compatible machine such as a 68/60 or a 68/80.
its this LISP is on some ITS system
Multics this LISP is on some Multics system
TOPS-10 this LISP is on some DEC TOPS-10 system; or on some
TENEX system since the TENEX implementation runs
under a TOPS-10 emulator.
A package being "present" means that it has been loaded into the
environment. If (status features) claims it is not present, it may
still be available because it may be automatically loaded when
required. This does not apply to the compiler.
(car (last (status features))) is an implementation name, such as ITS
or TOPS-10 or Multics. The main idea behind this status call is that
an application package can be loaded into any Maclisp implementation
and can decide what to do on the basis of the features it finds
available.
feature (status feature foo) is roughly equivalent to (memq 'foo (status
features)), i.e. it determines whether this LISP has the foo-feature.
Note that foo is not evaluated.
(sstatus feature foo) makes foo a feature. foo is not evaluated. For
example, the trace package does (sstatus feature trace) when it is
loaded.
Example:
December 6, 1977 ∪3-1.8.1 Page 3-99
**DRAFT** Maclisp Reference Manual **DRAFT**
(cond ((status feature bignum)
(prog2 nil (eval (read)) (read))) ;use first
(t (read) (eval (read)) )) ;use second
(defun factorial (n) ;bignum version
(cond ((zerop n) 1)
((times n (factorial (sub1 n))))
))
(defun factorial (n) ;fixnum-only version
(do () ((not (> n 13.))) ;do until n <λ← 13.
(error "argument too big - factorial"
n
'wrng-type-arg))
(cond ((zerop n) 1)
((* n (factorial (1- n)))) ))
nofeature (sstatus nofeature foo) makes foo not be a feature. foo is not
evaluated.
(status nofeature foo) is equivalant to (not (status feature foo)).
status (status status foo) returns t if foo is a valid status function. If
it is not, nil is returned.
(status status) returns a list of valid status functions. The names
are truncated to some implementation-dependent number of characters,
such as 4 or 5.
sstatus (status sstatus foo) returns t if foo is a valid sstatus function. If
it is not, nil is returned.
(status sstatus) returns a list of valid sstatus functions. As with
(status status), the names are truncated to some implementation-
dependent number of characters, such as 4 or 5.
1.8.2 Time
Page 3-100 ∪3-1.8.1 December 6, 1977
**DRAFT** The System **DRAFT**
runtime SUBR no args
(runtime) returns as a fixnum the number of microseconds of cpu time
used so far by the process in which LISP is running. The difference
between two values of (runtime) indicates the amount of computation
that was done between the two calls to runtime.
time SUBR no args
(time) returns the time in seconds that the system has been up, as a
flonum. The difference between the results of two calls to time
indicates the amount of elapsed real time. (In the ITS
implementation, time that elapses while the system is stopped due to
memory errors is not considered "real" and not counted. Use (status
daytime) to measure true "real world" time.)
sleep SUBR 1 arg
(sleep n) causes a real-time delay of n seconds, then returns n. n
may be a fixnum or a flonum.
See also the alarmclock function, section 1.4.3, and the date, daytime, and dow
functions of the status special form, described in the preceding section.
1.8.3 Escaping from LISP
It is possible to escape temporarily from LISP to execute a command in the
host operating system. Of course, the program (or user) that supplies the
command has to know which operating system it is running under. It is also
possible for LISP to return permanently to the host operating system. This
discards the LISP environment and gives back whatever resources, such as memory,
it was using.
December 6, 1977 ∪3-1.8.2 Page 3-101
**DRAFT** Maclisp Reference Manual **DRAFT**
in the Multics implementation
cline SUBR 1 arg
(cline x), where x is a character string, executes the Multics command x
and returns nil. Example:
(cline "who -long")
quit SUBR no args
(quit) returns from the LISP command, freeing up the temporary segments
that were used to hold the LISP environment.
in the ITS implementation
valret LSUBR 0 or 1 args
(valret) does a .LOGOUT if LISP is a top level procedure, and otherwise
valrets ":VK " to DDT.
(valret x) effectively performs an explodec on x (in practice x is some
strange atomic symbol like |:PROCEED :DISOWN | , but it may be any S-
expression). If the string of characters is one of "$↑X.", ":KILL ", or
":KILL↑M" then valret performs a "silent kill" by executing a .BREAK
16,20000; otherwise valret performs a .VALUE, giving the character string
to DDT to evaluate as commands.
Examples:
(valret '|:PROCEED :DISOWN |)
starts the LISP running on its own without a terminal.
(valret '| :KILL :TECO↑M|)
kills the LISP and starts up a TECO.
(valret '0$N$P)
causes DDT to print out the contents of all non-zero locations in LISP
and then return to LISP.
Page 3-102 ∪3-1.8.3 December 6, 1977
**DRAFT** The System **DRAFT**
If the valret command leaves you in DDT (or anywhere other than the
LISP), a :CONTINUE or $P command to DDT will return to the LISP, which will
then return from the call to valret. However, valret differs from suspend
in two ways: it does not check for open files, and it does not alter the
starting address of the LISP. If you use valret to dump a subsystem, then
starting up the subsystem causes a return to top level and the consequent
evaluation of the top-level errlist (not the binding of errlist in effect
at the time of the valret!). Using suspend, the subsystem startup merely
causes a return from the suspend function.
in the TOPS-10 implementation
There is currently no way for LISP to return a command string to the
Monitor in the TOPS-10 implementation. However, (valret) will return
control to the monitor so that a command may be manually typed. Then type
CONTINUE to resume LISP.
The remarks above about valret and suspend hold for the TOPS-10
implementation also.
1.8.4 Additional Functions
subr SUBR 1 arg
The argument, a fixnum, is taken to be the address of a location within
a compiled or system function. subr attempts to determine the name of the
function by groveling through the current obarray looking at all the
compiled function properties. If it fails it returns "?". This is used by
baktrace, for example. This function may be useful to user handlers for
the machine-error interrupt, for example.
The following function exists only in the ITS implementation
syscall LSUBR 2 to 10 args
This function allows the LISP user to directly issue ITS symbolic system
calls. The first argument should be
(+ (lsh c 18.) n)
where n is the desired number of output values and c is the control bits
December 6, 1977 ∪3-1.8.3 Page 3-103
**DRAFT** Maclisp Reference Manual **DRAFT**
for the call. The second argument should be an atomic symbol whose pname
is the name of the ITS system call. The rest of the arguments should be
the input arguments for the call. An argument may be a fixnum, or a file
object (for which the file's channel number is used). If the call
succeeds, a list of n fixnums is returned. If an error occurs, the ITS
error code is returned as a fixnum.
Example: (syscall 0 'scml tyi 3)
sets the number of echo lines for the tty to 3 and returns nil on
success.
Page 3-104 ∪3-1.8.4 December 6, 1977
β