Google
 

Trailing-Edge - PDP-10 Archives - decuslib20-01 - decus/20-0004/22lisp.tty
There are no other files named 22lisp.tty in the archive.











                               SECTION 22

                                                     1
                 THE PROGRAMMER'S ASSISTANT AND LISPX



22.1 Introduction

This  chapter describes  one of  the newer  additions to  INTERLISP: the
programmer's assistant.  The central idea of the  programmer's assistant
is that the user, rather  than talking to a passive system  which merely
responds to each input and waits for the next, is instead  addressing an
active intermediary, namely  his assistant.  Normally, the  assistant is
invisible  to the  user,  and simply  carries out  the  user's requests.
However, since the assistant remembers  what the user has told  him, the
user can instruct  him to repeat a  particular operation or  sequence of
operations,  with  possible  modifications, or  to  undo  the  effect of
certain specified operations.  Like DWIM, the programmer's  assistant is
not  implemented as  a single  function or  group of  functions,  but is
                                                    2
instead  dispersed  throughout  much  of  INTERLISP.   Like   DWIM,  the
programmer's  assistant embodies  a  philosophy and  approach  to system
design whose  ultimate goal  is to  construct a  programming environment
which  would  "cooperate"  with  the  user  in  the  development  of his
programs,  and free  him  to concentrate  more fully  on  the conceptual
difficulties and creative aspects of the problem he is trying to solve.


Example

The following  dialogue, taken  from an actual  session at  the console,
gives the  flavor of the  programmer's assistant facility  in INTERLISP.
The  user is  about to  edit a  function loadf,  which  contains several
constructs of the form (PUTD FN2 (GETD FN1)).  The user plans to replace
each of these by equivalent MOVD expressions.





------------------------------------------------------------------------
1
    The  programmer's  assistant  was  designed  and  implemented  by W.
    Teitelman. It is discussed in [Tei4].

2
    Some  of  the  features  of  the  programmer's  assistant  have been
    described elsewhere, e.g., the UNDO command in the editor,  the file
    package, etc.




                                  22.1



     _EDITF(LOADFF]                                                  [1]
     =LOADF
     EDIT
     *PP
       [LAMBDA (X Y)
         [COND
           ((NULL (GETD (QUOTE READSAVE)))
             (PUTD (QUOTE READSAVE)
                   (GETD (QUOTE READ]
         (PUTD (QUOTE READ)
                (GETD (QUOTE REED)))
         (NLSETQ (SETQ X (LOAD X Y)))
         (@UTD (QUOTE READ)
               (GETD (QUOTE READSAVE)))
         X]

     *F PUTD (1 MOVD)                                                [2]
     *3 (XTRR 2)                                                     [3]
     =XTR
     *0P                                                             [4]
     =0 P
     (MOVD (QUOTE READSAVE) (QUOTE READ))
     *(SW 2 3)                                                       [5]
     *

                                       3
At [1], the user begins to  edit loadf.  At [2] the user finds  PUTD and
replaces it by MOVD.  He then shifts context to the third subexpression,
[3], extracts  its second  subexpression, and ascends  one level  [4] to
print  and  result.   The  user  now  switches  the  second   and  third
subexpression [5], thereby completing the operation for this PUTD.  Note
that  up  to  this  point,  the  user  has  not  directly  addressed the
assistant.   The user  now  requests that  the assistant  print  out the
operations that the user has performed, [6], and the user then instructs
the assistant to REDO FROM F, [7], meaning repeat the entire sequence of
operations 15 through 20.  The user then prints the  current expression,
and observes that the second PUTD has now been successfully transformed.

     *?? FROM F                                                      [6]

     15.  *F PUTD
     16.  *(1 MOVD)
     17.  *3
     18.  *(XTR 2)
     19.  *0
     20.  *(SW 2 3)

     *REDO FROM F                                                    [7]


------------------------------------------------------------------------
3
    We prefer to consider the programmer's assistant as the moving force
    behind this  type of  spelling correction  (even though  the program
    that does the work is part of the DWIM package).  Whereas correcting
    @PRINT to  PRINT, or XTRR  to XTR does  not require  any information
    about what this  user is doing,  correcting LOADFF to  LOADF clearly
    required noticing when this user defined loadf.




                                  22.2



     *P
     (MOVD (QUOTE REED) (QUOTE READ))
     *

The user now asks the assistant  to replay the last three steps  to him,
[8].  Note that the entire REDO FROM F operation is now grouped together
as a single unit, [9],  since it corresponded to a single  user request.
Therefore, the  user can instruct  the assistant to  carry out  the same
operation  again  by  simply  saying  REDO.   This  time  a  problem  is
encountered [10], so the user  asks the assistant what it was  trying to
do [11].

     *?? FROM -3                                                     [8]

     19.  *0
     20.  *(SW 2 3)
     21.  REDO FROM F                                                [9]
          *F PUTD
          *(1 MOVD)
          *3
          *(XTR 2)
          *0
          *(SW 2 3)

     *REDO

     PUTD ?                                                         [10]

     *?? -1                                                         [11]

     22.  REDO
          *F PUTD
          *(1 MOVD)
          *3
          *(XTR 2)
          *0

The user then realizes the problem is that the third PUTD  is misspelled
in the definition of LOADF (see page 22.2).  He therefore  instructs the
assistant to USE  @UTD FOR PUTD, [12],  and the operation  now concludes
successfully.





















                                  22.3



     *USE @UTD FOR PUTD                                             [12]
     *P
     (MOVD (QUOTE READSAVE) (QUOTE READ))
     *^ PP
        [LAMBDA (X Y)
          [COND
            ((NULL (GETD (QUOTE READSAVE)))
              (MOVD (QUOTE READ)
                    (QUOTE READSAVE]
          (MOVD (QUOTE REED)
                (QUOTE READ))
          (NLSETQ (SETQ X (LOAD X Y)))
          (MOVD (QUOTE READSAVE)
                (QUOTE READ))
          X]
     *OK
     LOADF
     _

An  important point  to note  here  is that  while the  user  could have
defined a macro to execute this operation, the operation is sufficiently
complicated that he  would want to try  out the individual  steps before
attempting  to  combine them.   At  this point,  he  would  already have
executed the operation  once.  Then he would  have to type in  the steps
again to define them as a macro, at which point the operation would only
be  repeated once  more before  failing.  Then  the user  would  have to
repair the macro, or else change @UTD to PUTD by hand so that  his macro
would work correctly.  It is  far more natural to decide after  trying a
series  of  operations  whether  or  not  one  wants  them  repeated  or
forgotten.   In  addition,  frequently  the  user  will  think  that the
operation(s) in question will never need be repeated, and  only discover
afterwards  that  he  is  mistaken, as  occurs  when  the  operation was
incorrect, but salvageable:

     *P
     (LAMBDA (STR FLGCQ VRB) **COMMENT** (PROG & & LP1 & LP2 & &))
     *-1 -1 P
     (RETURN (COND &))
     *(-2 ((EQ BB (QUOTE OUT)) BB]                                   [1]
     *P
     (RETURN (& BB) (COND &))
                                                                     [2]
     *UNDO
     (-2 --) UNDONE
     *2 P
     (COND (EXPANS & & T))
     *REDO EQ
     *P
     (COND (& BB) (EXPANS & & T)
     *

Here the  operation was correct,  [1], but the  context in which  it was
executed, [2], was wrong.

This example also  illustrates one of the  most useful functions  of the
programmer's assistant: its UNDO capability.  In most systems, if a user
suspected  that a  disaster might  result from  a  particular operation,
e.g., an  untested program running  wild and chewing  up a  complex data




                                  22.4



structure, he would prepare for this contingency by saving the  state of
part  or all  of his  environment before  attempting the  operation.  If
anything went  wrong, he would  then back up  and start  over.  However,
saving/dumping  operations  are usually  expensive  and  time consuming,
especially  compared  to  a short  computation,  and  are  therefore not
performed that frequently, and of  course there is always the  case when
diaster  strikes as  a  result of  a  "debugged" or  at  least innocuous
operation, as shown in the following example:

     _(MAPC ELTS (FUNCTION (LAMBDA (X) (REMPROP X (QUOTE MORPH]      [1]
     NIL
     _UNDO                                                           [2]
     MAPC UNDONE.
     _USE ELEMENTS FOR ELTS
                                                                     [3]
     NIL
     _

The user types an expression which removes the property MORPH from every
member of the list ELTS [1],  and then realizes that he meant  to remove
that  property only  from those  members of  the list  ELEMENTS,  a much
shorter list.  In other words, he has deleted a lot of  information that
he actually wants saved.  He therefore simply reverses the effect of the
MAPC by  typing UNDO [2],  and then  does what he  intended via  the USE
command [3].


22.2 Overview

The programmer's assistant facility  is built around a  memory structure
called the "history list." The history list is a list of the information
associated with each  of the individual  "events" that have  occurred in
                                                                   4
the  system,  where  each  event corresponds  to  one  user  input.  For
example, (XTR 2) ([3] on page 22.2) is a single event, while REDO FROM F
([7] on page 22.2) is also a single event, although the  latter includes
executing the operation (XTR 2), as well as several others.

Associated with  each event  on the history  list is  its input  and its
value, plus other optional information such as  side-effects, formatting
information, etc.  If the event corresponds to a history  command, e.g.,
REDO FROM F, the  input corresponds to what  the user would have  had to
type to execute the same operation(s), although the user's actual input,
i.e., the history command, is saved in order to clarify the  printout of
that event  ([9] on page  22.3).  Note that  if a history  command event
combines several events, it will have more than one value:







------------------------------------------------------------------------
4
    For  various  reasons, there  are  two history  lists:  one  for the
    editor,  and one  for lispx,  which processes  inputs to  evalqt and
    break, see page 22.36.




                                  22.5



     _(LOG (ANTILOG 4))
     4.0
     _USE 4.0 40 400 FOR 4
     4.0
     40.0
     ARG NOT IN RANGE
     400

     _USE -40.0 -4.00007 -19.
     -40.0
     -4.00007
     -19.0
     _USE LOG ANTILOG FOR ANTILOG LOG IN -2 AND -1
     4.0
     40.0
     400.0
     4.00007
     19.0
     _??

     4.  USE LOG ANTILOG FOR ANTILOG LOG IN -2 -1
         _(ANTILOG (LOG 4.0))
         4.0
         _(ANTILOG (LOG 40))
         40.0
         _(ANTILOG (LOG 400))
         400.0
         _(ANTILOG (LOG -40.0))
         40.0
         _(ANTILOG (LOG -4.00007))
         4.00007
         _(ANTILOG (LOG -19.0))
         19.0
     3.  USE -40.0 -4.00007 -19.0
         _(LOG (ANTILOG -40.0))
         -40.0
         _(LOG (ANTILOG -4.00007))
         -4.00007
         _(LOG (ANTILOG -19.0))
         -19.0
     2.  USE 4.0 40 400 FOR 4
         _(LOG (ANTILOG 4.0))
         4.0
         (LOG (ANTILOG 40.0)
         40.0
         _(LOG (ANTILOG 400))

     1.  _(LOG (ANTILOG 4))
         4.0

As new events occur, existing  events are aged, and the oldest  event is
"forgotten." For efficiency, the storage used to represent the forgotten
event is cannibalized and reused in the representation of the new event,
so the history list  is actually a ring  buffer.  The size of  this ring








                                  22.6


                                                         5
buffer  is a  system parameter  called the  'time-slice.'   Larger time-
slices enable longer "memory spans," but tie up  correspondingly greater
amounts  of  storage.   Since  the  user  seldom  needs  really "ancient
history," and a  NAME and RETRIEVE facility  is provided for  saving and
remembering selected events,  a relatively small  time slice such  as 30
events is more than adequate, although some users prefer to set the time
slice as large as 100 events.

Events on the history list can  be referenced in a number of  ways.  The
output on page 22.8 shows  a printout of a history list  with time-slice
16.  The numbers printed at the left of the page are the  event numbers.
More recent events have higher  numbers; the most recent event  is event
                                                                    6
number 52, the oldest and about-to-be-forgotten event is number  37.  At
this  point  in   time,  the  user   can  reference  event   number  51,
RECOMPILE(EDIT),  by its  event number,  51; its  relative  position, -2
(because it  occurred two events  back from the  current time), or  by a
"description" of its input, e.g., (RECOMPILE (EDIT)), or (&  (EDIT)), or
even  just EDIT.   As  new events  occur, existing  events  retain their
absolute event numbers, although their relative positions change.

Similarly, descriptor references may require more precision to  refer to
an  older  event.  For  example,  the description  RECOMPILE  would have
sufficed to refer to event 51 had event 52, also containing a RECOMPILE,
not intervened.  Event specification will be described in detail later.






















------------------------------------------------------------------------
5
    Initially 30 events. The time-slice can be changed with the function
    changeslice, page 22.43.

6
    When the event  number of the current  event is 100, the  next event
    will be given number 1. (If the time slice is greater than  100, the
    "roll-over" occurs at the next  highest hundred, so that at  no time
    will two events ever have the same event number. For example, if the
    time slice is 150, event number 1 follows event number 200.)




                                  22.7



     _??

     52. ... HIST UNDO
         _RECOMPILE(HIST)
         HIST.COM
         _RECOMPILE(UNDO)
         UNDO.COM
     51. _RECOMPILE(EDIT)
         EDIT.COM
     50. _LOGOUT]

     49. _MAKEFILES]
         (EDIT UNDO HIST)
     48. _EDITF(UNDOLISPX)
         UNDOLISPX
     47. REDO GETD
         _GETD(FIE)
         (LAMBDA (X) (MAPC X (F/L (PRINT X))))
     46. _UNDO
         FIE
     45. _GETD(FIE)
         (LAMBDA (X) (MAPC X (FUNCTION (LAMBDA (X) (PRINT X)))))
     44. _FIE]
         NIL
     43. _DEFINEQ((FIE (LAMBDA (X) (MAPC X (F/L (PRINT X))))))
         (FIE)
     42. REDO GETD
         _GETD(FIE)
         (LAMBDA (Y) Y)
     41. _UNDO
         MOVD
     40. REDO GETD
         _GETD(FIE)
         (LAMBDA (X) X)
     39. _MOVD(FOO FIE)
         FIE
     38. _DEFINEQ((FOO (LAMBDA (X) X)))
         (FOO)
     37. _GETD(FIE)
         (LAMBDA (Y) Y)

The most  common interaction with  the programmer's assistant  occurs at
the top level evalqt, or in a break, where the user types in expressions
for evaluation,  and sees  the values  printed out.   In this  mode, the
assistant  acts much  like a  standard LISP  evalqt, except  that before
attempting to evaluate an input, the assistant first stores it in  a new
entry on the history list.   Thus if the operation is aborted  or causes
an error, the input is still saved and available for modification and/or
reexecution.  The assistant also notes new functions and variables to be
added  to its  spelling lists  to enable  future corrections.   Then the
assistant executes the computation (i.e., evaluates the form  or applies
the function  to its  arguments), saves the  value in  the entry  on the
history list corresponding to the input, and prints the result, followed









                                  22.8


                                                              7
by a prompt character to indicate it is again ready for input.

If the input typed by the  user is recognized as a history  command, the
assistant takes special  action.  Commands such  as UNDO, ??,  NAME, and
RETRIEVE are immediately performed.  Commands that  involved reexecution
of previous inputs,  e.g., REDO and USE,  are achieved by  computing the
corresponding input expression(s)  and then unreading them.   The effect
of this unreading operation  is to cause the assistant's  input routine,
lispxread, to act  exactly as though these  expression were typed  in by
the user.  Except for  the fact that these  inputs are not saved  on new
and  separate  entries on  the  history list,  but  associated  with the
history  command  that generated  them,  they are  processed  exactly as
though they had been typed.

The advantage of this  implementation is that it makes  the programmer's
assistant a callable facility for  other system packages as well  as for
users with  their own  private executives.   For example,  break1 accept
user inputs, recognizes and executes certain break commands  and macros,
and interprets  anything else as  INTERLISP expressions  for evaluation.
To interface break1 with the programmer's assistant required three small
modifications  to break1:  (1) input  was to  be obtained  via lispxread
instead of read; (2) instead  of calling eval or apply  directly, break1
was to give those  inputs it could not  interpret to lispx, and  (3) any
commands or macros handled by break1, i.e., not given to lispx,  were to
be  stored  on  the  history list  by  break1  by  calling  the function
historysave, a part of the assistant package.

Thus when the user typed in a break command, the command would be stored
on  the history  list as  a  result of  (3).  If  the user  typed  in an
expression for  evaluation, it  would be evaluated  as before,  with the
expression and its value both saved  on the history list as a  result of
(2).  Now if the user entered a break and typed three inputs: EVAL, (CAR
!VALUE), and OK, at the next break, he could achieve the same  effect by
typing REDO  FROM EVAL.  This  would cause the  assistant to  unread the
three expressions EVAL, (CAR !VALUE), and OK.  Because of (1),  the next
"input" seen by break1 would then be EVAL, which break1 would interpret.
Next would come (CAR !VALUE), which would be given to lispx to evaluate,
and then  would come  OK, which  break1 would  again process.   Thus, by
virtue of unreading, history operations will work even for  those inputs
not interpretable by lispx, in this case, EVAL and OK.




------------------------------------------------------------------------
7
    The  function that  accepts a  user input,  saves the  input  on the
    history list, performs the indicated computation or history command,
    and  prints the  result, is  lispx. lispx  is called  by  evalqt and
    break1,  and  in  most  cases,  is  synonymous   with  "programmer's
    assistant." However, for various  reasons, the editor saves  its own
    inputs  on a  history  list, carries  out the  requests,  i.e., edit
    commands,  and  even  handles undoing  independently  of  lispx. The
    editor  only  calls lispx  to  execute a  history  command,  such as
    REDO, USE, etc. Therefore we  use the term assistant  (loosely) when
    the discussion applies to  features shared by evalqt, break  and the
    editor,  and the  term  lispx when  we are  discussing  the specific
    function.




                                  22.9



The net effect of  this implementation of the programmer's  assistant is
to  provide a  facility which  is easily  inserted at  many  levels, and
embodies a consistent set of commands and conventions for  talking about
past events.  This gives the  user the subjective feeling that  a single
agent is watching everything he  does and says, and is  always available
to help.


22.3 Event Specification

All history commands use the same conventions and syntax  for indicating
which event or  events on the history  list the command refers  to, even
though different commands may be concerned with different aspects of the
corresponding   event(s),   e.g.,  side-effects,   value,   input,  etc.
Therefore, before  discussing the various  history commands in  the next
section,  this  section  describes  the  types  of  event specifications
currently implemented.  All examples  refer to the history list  on page
22.8.

An event address identifies one event on the history list.   It consists
of a sequence  of "commands" for moving  an imaginary cursor up  or down
the history list, much in the  manner of the arguments to the  @ command
in break (see Section 15).  The event identified is the one  "under" the
imaginary  cursor when  there  are no  more commands.   (If  any command
fails, an error is generated and the history command is aborted.)

The commands are interpreted as follows:


n (n > 1)               move  forward n  events, i.e.,  in  direction of
                        increasing event number.  If given as  the first
                        "command,"  n  specifies  the  event  with event
                        number n.


n (n < -1)              move backward -n events.


_atom                   specifies an  event whose function  matches atom
                        (i.e., for apply format only), e.g., whereas FIE
                        would  refer to  event 47,  _FIE would  refer to
                                                  8
                        event 44.   Similarly, ED$  would  specify event
                        51, whereas _ED$ event 48.


_                       next  search  is   to  go  forward   instead  of
                        backward, (if given as the first "command", next
                        search begins with last, i.e., oldest,  event on
                        history list),  e.g., _  LAMBDA refers  to event
                        38; MAKEFILES _ RECOMPILE refers to event 51.




------------------------------------------------------------------------
8
    i.e., EDalt-mode.




                                 22.10



F                       next object is to be searched for, regardless of
                        what  it  is,  e.g.,  F -2  looks  for  an event
                        containing a -2.


=                       next  object  (presumably a  pattern)  is  to be
                        matched against values, instead of inputs, e.g.,
                        = UNDO refers  to event 49;  45 = FIE  refers to
                        event 43; _ = LAMBDA refers to event 37.


\                       specifies the event last located.


SUCHTHAT pred           specifies an event for which pred, a function of
                        two arguments, when  given the input  portion of
                        the event as  its first argument, and  the event
                        itself  as  its second  argument,  returns true.
                        E.g.,      SUCHTHAT      (LAMBDA (X Y)     (MEMB
                        (QUOTE *ERROR*) Y)) specifies an event  in which
                                          9
                        an error occurred.


pat                     anything  else  specifies an  event  whose input
                        contains  an  expression  that  matches  pat  as
                        described in Section 9.


Note: each  search skips  the current event,  i.e., each  command always
moves the cursor.  For example, if  FOO refers to event n, FOO  FIE will
refer to some event before event n, even if there is a FIE in event n.

An event specification specifies one or more events:

FROM #1 THRU #2         the sequence of events from the
#1 THRU #2              event with address #1 through event with address
                           10
                        #2,   e.g., FROM  GETD THRU 49  specifies events
                        47, 48, and 49.  #1 can be more recent  than #2,
                        e.g., FROM 49 THRU GETP specifies events 49, 48,
                        and 47 (note reversal of order).


FROM #1 TO #2           Same as THRU but does not include event #2.
#1 TO #2


------------------------------------------------------------------------
9
    See  page  22.36 for  discussion  of  the format  of  events  on the
    history list.

10
    i.e., the symbol #1 corresponds  to all words between FROM  and THRU
    in the event specification, and #2 to all words from THRU to the end
    of the  event specification.  For example,  in FROM FOO 2 THRU FIE -
    1, #1 is (FOO 2), and #2 is (FIE -1).




                                 22.11



FROM #1                 Same as FROM #1 THRU -1, e.g., FROM 49 specifies
                        events 49, 50, 51, and 52.


THRU #2                 Same as FROM -1 THRU #2, e.g., THRU 49 specifies
                        events  52, 51,  50, and  49.  Note  reversal of
                        order.


TO                      Same as FROM -1 TO #2.


#1 AND #2 AND ... AND #n
                        i.e.,   a  sequence   of   event  specifications
                        separated  by  AND's, e.g.,  FROM  47  TO LOGOUT
                        would be equivalent to 47 AND 48 AND MAKEFILES.


ALL #1                  specifies  all events  satisfying #1,  e.g., ALL
                        LOAD, ALL SUCHTHAT FOO.


empty                   i.e., nothing specified, same as -1, unless last
                                                                    11
                        event was an UNDO, in which case same as -2.


@ atom                  refers to the events named by atom, via the NAME
                        command, page  22.21 e.g., if  the user  names a
                        particular event or events FOO, @  FOO specifies
                        those events.


@@ B                    B is an  event specification and  interpreted as
                        above, but with respect to the  archived history
                        list, as specified on page 22.23.


If no events can be found that satisfy the event specification, spelling
correction on each  word in the  event specification is  performed using
lispxfindsplst  as a  spelling  list, e.g.,  REDO  3 THRUU  6  will work
correctly. If the event specification still fails to specify  any events
after spelling correction, an error is generated.


22.4 History Commands

All history  commands can  be input as  either lists,  or as  lines (see
readline Section 14, and also page 22.39).

B is used to denote an event specification.  Unless specified otherwise,
B omitted is the same as B =-1, e.g., REDO and REDO -1 are the same.


------------------------------------------------------------------------
11
    For example,  if the  user types (NCONC FOO FIE),  he can  then type
    UNDO, followed by USE NCONC1.




                                 22.12



REDO B                  redoes the event or events specified by B, e.g.,
                        REDO FROM -3 redoes the last three events.


REDO B N TIMES          redoes  the event  or  events specified  by  B N
                        times, e.g., REDO 10 TIMES redoes the last event
                        ten times.  If n is not a positive number, e.g.,
                        REDO  MANY  TIMES,  the effect  is  the  same as
                        though  n  were  infinite:  the   events(s)  are
                        repeated until  an error  occurs, or  user types
                        control-d.


USE exprs FOR args IN B substitutes exprs for args in B, and  redoes the
                        result, e.g.,
                        USE LOG ANTILOG FOR ANTILOG LOG IN -2 AND -1.
                        Substitution is  done by esubst,  Section 9, and
                        is carried  out as  described below.   exprs and
                        args can include non-atomic members.


USE exprs1 FOR args1 AND ... AND exprsn FOR argsn IN B
                        More   general   form  of   USE   command.   See
                        description of substitution algorithm below.


Every USE command involves three pieces of information:  the expressions
to be  substituted, the arguments  to be substituted  for, and  an event
specification,  which  defines  the  expression  (input)  in  which  the
                          12
substitution  takes place.    Any expression  to be  substituted  can be
preceded by a !, meaning that  the expression is to be substituted  as a
segment, e.g., LIST(A B C) followed by USE ! (X Y Z) FOR B  will produce
(A X Y Z C), and USE ! NIL FOR B will produce (A C).

If args are omitted, i.e., the form of the command is USE exprs IN B, or
just USE exprs (which is equivalent  to USE exprs IN -1), and  the event
referred  to was  itself  a USE  command, the  arguments  and expression
substituted into  are the  same as  for the  indicated USE  command.  In
effect, this  USE command  is thus  a continuation  of the  previous USE
command.    For   example,   on  page   22.6,   when   the   user  types
(LOG (ANTILOG 4)), followed by USE 4.0 40 400 FOR 4, followed by
USE -40.0 -4.00007 -19., the latter command is equivalent to
USE -40.0 -4.00007 -19. FOR 4 IN -2.

If args are  omitted and the  event referred to  was not a  USE command,
substitution  is for  the operator  in that  command, i.e.,  if  a lispx
input, the name  of the function,  if an edit  command, the name  of the
command.  For example ARGLIST(FF) followed by USE CALLS is equivalent to
USE CALLS FOR ARGLIST.



------------------------------------------------------------------------
12
    The  USE  command  is  parsed by  a  small  finite  state  parser to
    distinguish   the   expressions   and   arguments.    For   example,
    USE FOR FOR AND AND AND FOR FOR will be parsed correctly.




                                 22.13



If B is  omitted, but args  are specified, the  first member of  args is
used for B, e.g., USE PUTD  FOR @UTD is equivalent to USE PUTD  FOR @UTD
          13
IN F @UTD.

If the USE command has the same number of expressions as  arguments, the
                                          14
substitution procedure is straightforward,   i.e., USE X Y FOR U V means
substitute   X   for   U   and  Y   for   V,   and   is   equivalent  to
USE X FOR U AND Y FOR V.    However,  the   USE  command   also  permits
distributive substitutions,  i.e., substituting several  expressions for
the same argument.  For example, USE A B C FOR X means  first substitute
A for X then substitute B for X (in a new copy of the  expression), then
substitute  C for  X.  The  effect  is the  same as  three  separate USE
commands.  Similarly,  USE A B C FOR D AND X Y Z FOR W is  equivalent to
USE A FOR D AND X FOR W,  followed by  USE B FOR D AND Y FOR W, followed
                                                              15
by    USE C FOR D AND Z FOR W.     USE A B C FOR D AND X FOR Y      also
corresponds to three substitions,  the first with A  for D and X  for Y,
the second with B for  D, and X for Y, and  the third with C for  D, and
again X for Y.  However, USE A B C FOR D AND X Y FOR Z is  ambiguous and
will  cause  an  error.   Essentially,  the  USE  command   operates  by
proceeding from left to right handling each "AND"  separately.  Whenever
the number of expressions  exceeds the number of  expressions available,
                         16
the expressions multiply.


FIX B                   puts the user in the editor looking at a copy of
                        the input(s) for B, Whenever the user exits via,
                        OK, the result is unread and  reexecuted exactly
                        as with REDO.




------------------------------------------------------------------------
13
    The  F is  inserted to  handle correctly  the case  where  the first
    member of args  is a number, e.g.,  USE 4.0 4.0 400 FOR 4. Obviously
    the  user  means find  the  event  containing a  4  and  perform the
    indicated  substitutions,  whereas  USE 4.0 40 400 FOR 4 IN 4  would
    mean perform the substitutions in event number 4.

14
    Except when one of the arguments and one of the expressions  are the
    same,  e.g.,   USE X Y FOR Y X,  or   USE X FOR Y AND Y FOR X.  This
    situation  is  noticed   when  parsing  the  command,   and  handled
    correctly.

15
    or USE X FOR Y AND A B C FOR D.

16
    Thus USE A B C D FOR E F means substitute  A for E at the  same time
    as  substituting B  for F,  then in  another copy  of  the indicated
    expression, substitute C for E and  D for F. Note that this  is also
    equivalent to USE A C FOR E AND B D FOR F.




                                 22.14



FIX is provided for those  cases when the modifications to  the input(s)
are  not  of  the  type  that  can  be  specified  by  USE,   i.e.,  not
substitutions.  For example:

     _(DEFINEQ FOO (LAMBDA (X) (FIXSPELL SPELLINGS2 X 70]

     INCORRECT DEFINING FORM
     FOO

     _FIX
     EDIT
     *P
     (DEFINEQ FOO (LAMBDA & &))
     *(LI 2)
     OK
     (FOO)
     _

The user  can also  specify the edit  command(s) to  lispx, by  typing -
followed by  the command(s) after  the event specification,  e.g., FIX -
 (LI 2).  In this case, the editor will not type EDIT, or wait for an OK
after executing the commands.


Implementation of REDO, USE, and FIX

The input portion of an  event is represented internally on  the history
list simply  as a linear  sequence of the  expressions which  were read.
For  example, an  input in  apply  format is  a list  consisting  of two
expressions,  and  an  input  in  eval format  is  a  list  of  just one
           17
expression.    Thus if  the user  wishes to  convert an  input  in apply
format to eval format, he  simply moves the function name inside  of the
argument list:

















------------------------------------------------------------------------
17
    For inputs in eval  format, i.e., single expressions, FIX  calls the
    editor so that the current expression is that input, rather than the
    list consisting  of that  input - see the  example on  the preceding
    page. However, the entire list is actually being edited. Thus if the
    user typed ^ P in that example, he would see ((DEFINEQ FOO &)).




                                 22.15



     _MAPC(FOOFNS (F/L (AND (EXPRP X) (PRINT X]
     NIL
     _EXPRP(FOO1)
     T
     _FIX MAPC
     EDIT
     *P
     (MAPC (FOOFNS &))
     *(BO 2)
     *(LI 1)
     *P
     ((MAPC FOOFNS &))
     OK
     FOO1
     FIE2
     FUM
     NIL
     _

By simply converting the  input from two expressions to  one expression,
the desired effect, that of mapping down the list that was the  value of
foofns, was achieved.

REDO, USE,  and FIX all  operate by obtaining  the input portion  of the
corresponding event,  processing the input  (except for REDO),  and then
storing it on the history list as the input portion of a new event.  The
history command completes operating by simply unreading the input.  When
the input is subsequently "reread", the event which already contains the
input  will  be  retrieved  and used  for  recording  the  value  of the
operation, saving side-effects, etc.,  instead of creating a  new event.
Otherwise the input is treated exactly the same as if it had  been typed
in directly.

If B  specifies more than  one event, the  inputs for  the corresponding
events  are simply  concatenated into  a linear  sequence,  with special
                                                                      18
markers (called pseudo-carriage returns) representing carriage returns
inserted  between each  input to  indicate where  new lines  start.  The
result of this concatenation is then treated as the input referred to by
B.  For example, when the user typed REDO FROM F ([7] on page  22.2) the
inputs for the corresponding six events were concatenated to produce:
(F PUTD "<c.r.>" (1 MOVD) "<c.r.>" 3 "<c.r.>" (XTR 2) "<c.r.>"
0 "<c.r.>" (SW 2 3)).     Similarly,    if    the    user    had   typed
USE PUTD FOR @UTD IN 15 THRU 20,   the  above   list  would   have  been
constructed, and then PUTD substituted for @UTD throughout it.

The same convention is used for representing multiple inputs when  a USE
command  involves sequential  substitutions.  For  example, if  the user
types GETD(FOO)  and then USE FIE FUM FOR FOO,  the input  sequence that



------------------------------------------------------------------------
18
    The value of the variable  histstr0 is used to represent  a carriage
    return.  For readability, this  value is the string  "<c.r.>".  Note
    that since the comparison is  made using eq, this marker  will never
    be confused with a string that was typed in by the user.




                                 22.16



will be  constructed is  (GETD (FIE) "<c.r.>" GETD (FUM)), which  is the
result of substituting FIE for FOO in (GETD (FOO)) concatenated with the
result of substituting FUM for FOO in (GETD (FOO)).

Once such  a multiple input  is constructed, it  is treated  exactly the
same as a single  input, i.e., the input  sequence is recorded in  a new
event, and then unread, exactly as described above.  When the inputs are
"reread,"  the "pseudo-carriage-returns"  are treated  by  lispxread and
readline  exactly  as  real  carriage  returns,  i.e.,  they   serve  to
distinguish between apply  and eval formats on  inputs to lispx,  and to
delimit line commands to the editor.  Note that once this multiple input
has been entered as the input portion of a new event, that event  can be
treated exactly the same as one resulting from type in.  In other words,
no special checks have to be  made when referencing an event, to  see if
it is  simple or  multiple.  Thus,  when the  user types  REDO following
REDO FROM F, ([10] page 22.3) REDO  does not even notice that  the input
retrieved  from  the  previous  event  is (F PUTD "<c.r.>" ... (SW 2 3))
i.e., a  multiple input, it  simply records this  input and  unreads it.
Similarly, when the user  then types USE @UTD FOR PUTD on  this multiple
input,  the USE  command simply  carries out  the substitution,  and the
result    is    the    same   as    though    the    user    had   typed
USE @UTD FOR PUTD IN 15 THRU 20.

In sum, this implementation permits B to refer to a single simple event,
or to several events, or  to a single event originally  constructed from
several events  (which may themselves  have been multiple  input events,
etc.) without having to treat each case separately.


History Commands Applied to History Commands

Since history commands themselves do not appear in the input  portion of
events (although they  are stored elsewhere in  the event), they  do not
interfere   with   or   affect  the   searching   operations   of  event
specifications.   In effect,  history  commands are  invisible  to event
               19
specifications.    As a  result, history  commands themselves  cannot be
recovered for  execution in the  normal way.  For  example, if  the user
types USE A B C FOR D  and follows  this with  USE E FOR D, he  will not
produce the effect of  USE A B C FOR E (but instead will simply  cause E
to be substituted for D in  the last event containing a D).   To produce
this   effect,   i.e.,    USE A B C FOR E,   the   user    should   type
USE D FOR E IN USE.  The appearance of the  word REDO, USE or FIX  in an
event address specifies a search for the corresponding  history command.
(For example, the user can also type UNDO REDO.) It also  specifies that
the text of the history command itself be treated as though it  were the
input.  However,  the user  must remember  that the  context in  which a
history command is  reexecuted is that of  the current history,  not the
original context.  For example, if the user types USE FOO FOR FIE IN -1,
and then later types REDO USE, the -1 will refer to the event before the
REDO, not before the USE.



------------------------------------------------------------------------
19
    With  the exception  described  below under  "History  Commands that
    Fail".




                                 22.17



History Commands that Fail

The one exception to the statement that "history commands  are invisible
to event specifications" occurs when a history command fails  to produce
any     input.     For     example,    suppose     the     user    types
USE LOG FOR ANTILOG AND ANTILOG FOR LOGG,   causing  lispx   to  respond
LOGG ?.  Since the USE command  did not produce any input, the  user can
repair it by typing USE LOG FOR LOGG (i.e., does not have to  specify IN
USE).  This latter USE command will invoke a search for LOGG, which will
find  the   bad  USE  command.    lispx  then  performs   the  indicated
substitution,  and unreads  USE LOG FOR ANTILOG AND ANTILOG FOR LOG.  In
turn, this USE command invokes  a search for ANTILOG, which,  because it
was not typed in but reread, ignores the bad USE command which was found
by the earlier search for LOGG, and which is still on the  history list.
In other words, history commands that fail to produce input  are visible
to searches arising from event specifications typed in by the  user, but
not to secondary event specifications.

In addition, if the most recent event is a history command  which failed
to produce input, a secondary event specification will  effectively back
up the history  list one event so  that relative event numbers  for that
event  specification  will  not  count  the  bad  history  command.  For
example,           suppose           the            user           types
USE LOG FOR ANTILOG AND ANTILOG FOR LOGG IN -2 AND -1,  and  after lispx
types  LOGG ?,  the user  types  USE LOG FOR LOGG.  He  thus  causes the
command   USE LOG FOR ANTILOG AND ANTILOG FOR LOG IN -2 AND -1   to   be
constructed and unread.  In the normal case, -1 would refer to  the last
event,  i.e., the  "bad" USE  command, and  -2 to  the event  before it.
However,  in  this case,  -1  refers to  the  event before  the  bad USE
command, and the -2 to the event before that.  In short, the caveat that
"the user must remember that  the context in which a history  command is
reexecuted is  that of  the current history,  not the  original context"
does not apply if the correction is performed immediately.


More History Commands

RETRY B                 similar to  REDO except  sets helpclock  so that
                        any  errors that  occur while  executing  B will
                        cause breaks.


... vars                similar  to  USE  except  substitutes   for  the
                        (first) operand.

For  example,  EXPRP(FOO)  followed  by ... FIE FUM  is   equivalent  to
USE FIE FUM FOR FOO.  See also event 52 on page 22.8.


?? B                    prints history list.  If B is omitted, ?? prints
                        the  entire  history list,  beginning  with most
                        recent events.   Otherwise ?? prints  only those
                        events  specified  in   B  (and  in   the  order
                        specified), e.g., ?? -1, ?? 10 THRU 15, etc.

?? commands are not  entered on the history  list, and so do  not affect
relative event numbers.   In other words,  an event specification  of -1
typed  following  a  ??  command will  refer  to  the  event immediately
preceding the ?? command.



                                 22.18



?? will print the history command, if any, associated with each event as
shown  at [9]  on page  22.3  and page  22.6.  Note  that  these history
commands are not preceded by prompt characters, indicating they  are not
                20
stored as input.

?? prints multiple input events under one event number (see page 22.6).

Since events are initially stored  on the history list with  their value
field equal to bell (control-G),  if an operation fails to  complete for
any reason, e.g., causes an error, is aborted, etc., its "value" will be
bell.  This is the explanation for the blank line in event 2, page 22.6,
and event 50, page 22.8.

?? is implemented via  the function printhistory, page 22.48,  which can
also be called directly by the user.


UNDO B                  undoes the side effects of the specified events.
                        For each  event undone,  UNDO prints  a message:
                        e.g.,   RPLACA UNDONE, REDO UNDONE    etc.    If
                        nothing  is  undone because  nothing  was saved,
                        UNDO types NOTHING SAVED. If nothing  was undone
                        because the  event(s) were already  undone, UNDO
                        types  ALREADY UNDONE.  If  B  is   empty,  UNDO
                        searches back for the last event  that contained
                        side effects, was not undone, and itself was not
                                        21 22
                        an UNDO command.   


UNDO B : x1 ... xn      Each xi refers to  a message printed by  DWIM in



------------------------------------------------------------------------
20
    REDO, RETRY, USE, ..., and FIX  commands, i.e., those  commands that
    reexecute previous  events, are  not stored  as inputs,  because the
    input portion for these  events are the expressions to  be "reread".
    The  history  commands UNDO, NAME, RETRIEVE, BEFORE,  and  AFTER are
    recorded as inputs, and ?? prints them exactly as they were typed.

21
    Note that the user  can undo UNDO commands themselves  by specifying
    the corresponding event address, e.g., UNDO -7 or UNDO UNDO.

22
    UNDOing events in the reverse order from which they were executed is
    guaranteed  to restore  all pointers  correctly, e.g.,  to  undo all
    effects of last five events, perform UNDO THRU -5, not UNDO FROM -5.
    Undoing out of  order may have  unforseen effects if  the operations
    are dependent. For example, if the user  performed (NCONC1 FOO FIE),
    followed by (NCONC1 FOO FUM), and then undoes  the (NCONC1 FOO FIE),
    he will also have undone the (NCONC1 FOO FUM). If he then undoes the
    (NCONC1 FOO FUM), he will  cause the FIE  to reappear, by  virtue of
    restoring FOO to its state before the execution of (NCONC1 FOO FUM).
    For more details, see page 22.34.




                                 22.19



                        the event(s) specified  by B.  The  side effects
                        of the corresponding DWIM corrections,  and only
                        those side effects, are undone.


For  example,  if  the  message  PRINTT [IN FOO] -> PRINT  were printed,
                                                        23
UNDO : PRINTT or UNDO : PRINT would undo the correction.

$  is a  special form  of the  USE command  for  conveniently specifying
character substitutions.  In addition, it has a number of useful default
options in connection with events that involve errors.


$ x FOR y               equivalent to USE $x$ FOR $y$

For  example,  the  user types  MOVD(FOO FOOSAVE T),  he  can  then type
$ FIE FOR FOO to perform MOVD(FIE FIESAVE T).  Note that USE FIE FOR FOO
would perform MOVD(FIE FOOSAVE T).

An abbreviated form of $ is available:


$ y x                   same as $ x FOR y, i.e., y's are changed to x's.
                        can  also  be written  as  $ y TO x, $ y = x, or
                        $ y -> x.

$ does event location the same as the USE command, i.e., if IN -- is not
                             24
specified, it searches for y.

After $ finds  the event, it  looks to see if  an error was  involved in
            25
that  event,    and  if  the  indicated  character  substitution  can be
performed in the offender.  If so, $ assumes the substitution  refers to
the offender, performs the substitution, and then substitutes the result
for   the   offender   throughout.    For   example,   the   user  types
(PRETTYDEF FOOFNS 'FOO FOOVARS) causing a U.B.A. FOOOVARS error message.



------------------------------------------------------------------------
23
    Some portions of the messages printed by DWIM are strings, e.g., the
    message FOO UNSAVED is printed by printing FOO and  then " UNSAVED".
    Therefore,  if the  user types  UNDO : UNSAVED, the  DWIM correction
    will  not   be  found.   He  should   instead  type   UNDO : FOO  or
    UNDO : $UNSAVED$ (alt-modeUNSAVEDalt-mode, see R command  in editor,
    Section 9).

24
    However, unlike USE, $ can only be used to specify  one substitution
    at a time.

25
    Whenever an error  occurs, the object  of the error  message, called
    the offender, is  automatically saved on  that event's entry  in the
    history list, under the property *ERROR*.




                                 22.20



The user can now type $ OO O, which will change FOOOVARS to FOOVARS, but
not change FOOFNS or FOO.

If an error  did occur in  the specified event,  the user can  also omit
specifying y, in which case the offender is used.  Thus, the  user could
have corrected the above example by simply typing $ FOOVARS.  Similarly,
if   the    user   types    LOAD(PRSTRUC PROP),   causing    the   error
FILE NOT FOUND PRSTRUC, he can request the file to be loaded from LISP's
directory  by  simply  typing  $ <LISP>$.   Since  esubst  is  used  for
substituting, this  is equivalent  to performing  (R PRSTRUC <LISP>$) on
the  event,  and   therefore  replaces  PRSTRUC   by  <LISP>PRSTRUC (see
Section 9).  Note also the usefulness of $ '$, meaning: put a ' in front
of the offender.

$ also works for events in  the editor.  For example, if the  user types
(MOVE COND 33 2 TO BEFORE HERE),  and editor  types 33 ?,  the  user can
type $ 3, causing 3 to be substituted for 33 in the MOVE command.

Finally, the user can omit both  x and y.  This specifies that  two alt-
modes be packed onto the end of the offender, and the result substituted
throughout the specified event.  For example, suppose the user  types to
the  editor  (MOVE 3 2 TO AFTER CONDD 1),  and  gets  the  error message
CONDD ?.  because the find command  failed to find CONDD.  $  will cause
the  edit command  (MOVE 3 2 TO AFTER CONDD$$ 1) to  be  executed, which
will search for an  atom that is "close" to  CONDD in the sense  used by
                                                        26
the spelling corrector (see pattern type 6b, Section 9).

Note  that $  never searches  for  an error.   Thus, if  the  user types
LOAD(PRSTRUC PROP) causing a FILE NOT FOUND error, types CLOSEALL(), and
then types  $ <LISP>$, lispx  will complain  that there  is no  error in
CLOSEALL().    In   this   case,   the   user   would   have   to   type
$ <LISP>$ IN LOAD, or  $ PRS <LISP>PRS (which would  cause a  search for
PRS).

Note also that $ operates on input, not on programs.  If the  user types
FOO(), and within the call  to FOO gets a U.D.F. CONDD error,  he cannot
repair this by $ COND. lispx will type CONDD NOT FOUND IN FOO().


NAME atom B             saves  the  event(s)  (including  side  effects)
                        specified  by B  on  the property  list  of atom
                        (under the  property HISTORY)  e.g., NAME FOO 10
                        THRU 15.  NAME commands are undoable.

Commands defined by  NAME can also be  typed in directly as  though they





------------------------------------------------------------------------
26
    The  same  effect  could  be  achieved  by  $ COND,  which specifies
    substituting COND for CONDD, but not by $ CONDD$$, since  the latter
    is equivalent  to performing (R CONDD CONDD$$)  on the  event, which
    would  result in  CONDDCONDDCONDD  being substituted  for  CONDD (as
    described in Section 9).




                                 22.21



                                                               27
were built-in commands, e.g., FOOC is equivalent to REDO @ FOO.

Commands defined by NAME can also be parameterized, i.e., be  defined to
take arguments:


NAME name (args) : B    args are interpreted the same as the arguments
or                      for a USE command.  See page 22.13.  When name
NAME name ... args ... : B
                        is invoked, the argument values  are substituted
                        for args  using the same  substitution algorithm
                        as for USE.

For example,  following the  event (PUTD 'FOO (COPY (GETP 'FIE 'EXPR))),
the user types NAME MOVE FOO  FIE : PUTD.  Then typing MOVE  TEST1 TEST2
would  cause  (PUTD 'TEST1 (COPY (GETP 'TEST2 'EXPR)))  to  be executed,
i.e., would be  equivalent to typing  USE TEST1 TEST2  FOR FOO FIE  IN @
MOVE.  Typing MOVE A B C D would cause two PUTD's to be  executed.  Note
that  !'s and  $'s  can also  be employed  the  same as  with  USE.  For
example, if following
         _PREPINDEX(<MANUAL>14LISP.XGP)
         _FIXFILE(<MANUAL>14LISP.XGPIDX)
the  user performed  NAME FOO  $14$ :  -2 AND  -1, then  FOO  $15$ would
                                                            28
perform the indicated two operations with 14 replaced by 15.


RETRIEVE atom           Retrieves and reenters  on the history  list the
                        events named by  atom.  Causes an error  if atom
                        was not named by a NAME command.


For example, if the user performs NAME FOO 10 THRU 15, and at  some time
later types RETRIEVE FOO, 6 new  events will be recorded on  the history
list (whether or not the corresponding events have been  forgotten yet).
Note that RETRIEVE  does not reexecute  the events, it  simply retrieves
them.  The  user can  then REDO, UNDO, FIX,  etc.  any  or all  of these
events.  Note that the user can combine the effects of a RETRIEVE  and a
subsequent  history command  in  a single  operation by  using  an event
specification of  the form  @ atom,  as described  on page  22.12, e.g.,
REDO @ FOO  is equivalent  to RETRIEVE FOO,  followed by  an appropriate




------------------------------------------------------------------------
27
    However, if FOO  is the name of  a variable, it would  be evaluated,
    i.e., FOOC would return the value of FOO.

28
    NAME FOO B is equivalent to NAME FOO : B.  In either case, if FOO is
    invoked with arguments, an error is generated.








                                 22.22


     29
REDO.   Note that UNDO @ FOO and ?? @ FOO are permitted.


BEFORE atom             undoes the effects of the events named by atom.


AFTER atom              undoes a BEFORE atom.


BEFORE/AFTER provide a convenient way of flipping back and forth between
two states, namely  that state before a  specified event or  events were
executed, and that state after execution.  For example, if the  user has
a complex data structure which he wants to be able to interrogate before
and after certain modifications, he can execute the  modifications, name
the corresponding events with the NAME command, and then can  turn these
                                                      30
modifications off and on via BEFORE or AFTER commands.   Both BEFORE and
AFTER are NOPs if the atom was already in the corresponding  state; both
generate errors if atom was not named by a NAME command.

Note:  since  UNDO, NAME, RETRIEVE, BEFORE, and AFTER  are  recorded  as
inputs they  can be referenced  by REDO, USE, etc.   in the  normal way.
However, the  user must  again remember  that the  context in  which the
command  is  reexecuted is  different  than the  original  context.  For
example,   if   the  user   types   NAME FOO DEFINEQ THRU COMPILE,  then
types ... FIE,    the   input    that   will    be   reread    will   be
NAME FIE DEFINEQ THRU COMPILE  as  was intended,  but  both  DEFINEQ and
COMPILE, will  refer to  the most recent  event containing  those atoms,
namely the event consisting of NAME FOO DEFINEQ THRU COMPILE!


ARCHIVE B               records the events specified by B on a permanent
                        history   list.   This   history  list   can  be
                        referenced   by  preceding   a   standard  event
                        specification  with @@,  e.g., ?? @@  prints the
                        archived history  list, REDO @@ -1  will recover
                        the  corresponding   event  from   the  archived
                        history list and redo it, etc.

The user can also provide for automatic archiving of selected  events by
appropriately defining archivefn, or by putting the  property *ARCHIVE*,
value T, on the event.   Events that are referenced by  history commands
are  automatically  marked  for archiving  in  this  fashion.   For more
details, see page 22.28.


------------------------------------------------------------------------
29
    Actually, REDO @ FOO is better than RETRIEVE followed by  REDO since
    in the latter case, the corresponding events would be entered on the
    history list twice, once for the RETRIEVE and once for the REDO.

30
    The  alternative to  BEFORE/AFTER  for repeated  switching  back and
    forth involves  UNDO, UNDO of  the UNDO, UNDO of  that etc.  At each
    stage, the user would have to locate the correct event to  undo, and
    furthermore would run the risk of that event being "forgotten" if he
    did not switch at least once per time-slice.




                                 22.23



FORGET B                permanently  erases  the  record  of   the  side
                        effects for the events specified by B.  If  B is
                        omitted, forgets side effects for entire history
                        list.

FORGET is provided for users  with space problems.  For example,  if the
user has just performed sets, rplacas, rplacds, putd, remprops, etc.  to
release storage, the old  pointers would not be garbage  collected until
the corresponding  events age sufficiently  to drop off  the end  of the
history list and  be forgotten.  FORGET can  be used to  force immediate
forgetting  (of  the   side-effects  only).   FORGET  is   not  undoable
(obviously).


22.5 Miscellaneous Features and Commands

TYPE-AHEAD              is a command that allows the user  to type-ahead
                        an indefinite number of inputs.

The assistant responds to TYPE-AHEAD with a prompt character of  >.  The
user  can now  type in  an indefinite  number of  lines of  input, under
errorset protection.  The input lines are saved and unread when the user
exits the type-ahead loop  with the command $GO (alt-modeGO).   While in
the type-ahead loop, ?? can be used to print the type-ahead, FIX to edit
the type-ahead, and $Q to erase the last input (may be used repeatedly).
For example:




































                                 22.24



     _TYPE-AHEAD
     >SYSOUT(TEM)
     >MAKEFILE(EDIT)
     >BRECOMPILE((EDIT WEDIT))
     >F
     >$Q
     \\F
     >$Q
     \\BRECOMPILE
     >LOAD(WEDIT PROP)
     >BRECOMPILE((EDIT WEDIT))
     >F
     >MAKEFILE(BREAK)
     >LISTFILES(EDIT BREAK)
     >SYSOUT(CURRENT)
     >LOGOUT]
     >??
          >SYSOUT(TEM)
          >MAKEFILE(EDIT)
          >LOAD(WEDIT PROP)
          >BRECOMPILE((EDIT WEDIT))
                             31
          >F
          >MAKEFILE(BREAK)
          >LISTFILES(EDIT BREAK)
          >SYSOUT(CURRENT)
          >LOGOUT]
     >FIX
     EDIT
     *(R BRECOMPILE BCOMPL)
     *P
     ((LOGOUT) (SYSOUT &) (LISTFILES &) (MAKEFILE &) (F) (BCOMPL &)
     (LOAD &) (MAKEFILE &) (SYSOUT &))
     *(DELETE LOAD)
     *OK
     >$GO

The TYPE-AHEAD command may be aborted by $STOP; control-E  simply aborts
the current line of input.


$BUFS (alt-modeBUFS)    is a command for recovering the input buffers.

Whenever an error occurs in executing a lispx input or edit  command, or
a  control-E or  control-D is  typed, the  input buffers  are  saved and
cleared.  The $BUFS command is used to restore the input  buffers, i.e.,
its effect is exactly the same  as though the user had retyped  what was
"lost." For example:





------------------------------------------------------------------------
31
    Note that type-ahead can be addressed to the compiler, since it uses
    lispxread for input. Type-ahead can also be directed to  the editor,
    but type-ahead to the editor and to lispx cannot be intermixed.




                                 22.25



     *(-2 (SETQ X (COND ((NULL Z) (CONS     (user typed control-E)
     *P
     (COND (& &) (T &))
     *2
     *$BUFS
     (-2 (SETQ X (COND ((NULL Z) (CONS

and user can now finish typing the (-2 ..) command.

Note: the type-ahead does not have to have been seen by INTERLISP, i.e.,
echoed, since the system buffer is also saved.

Input buffers are not saved on the history list, but on a free variable.
Thus, only the contents of the input buffer as of the last  clearbuf can
ever be recovered.  However,  input buffers cleared at evalqt  are saved
independently from those cleared by break or the editor.   The procedure
followed when the  user types $BUFS is  to recover first from  the local
                                             32
buffer, otherwise from  the top level buffer.    Thus the user  can lose
input in the editor, go back  to evalqt, lose input there, then  go back
into  the  editor, recover  the  editor's buffer,  etc.   Furthermore, a
buffer cleared at the top can be recovered in a break, and vice versa.

The following four commands, DO, !F, !E, and !N, are only  recognized in
the editor:


DO com                  allows the user to supply the command  name when
                        it  was omitted.   (USE is  used when  a command
                        name is incorrect).

For example, suppose the user wants to perform
(-2 (SETQ X (LIST Y Z)))  but  instead  types  just (SETQ X (LIST Y Z)).
The editor  will type SETQ ?,  whereupon the user  can type  DO -2.  The
effect  is  the same  as  though the  user  had typed  FIX,  followed by
(LI 1), (-1 -2), and OK,  i.e., the command  (-2 (SETQ X (LIST Y Z))) is
executed.  DO also works if the last command is a line command.


!F                      same as DO F.

In the case of !F, the  previous command is always treated as  though it
were a line command, e.g., if the user types (SETQ X &) and then !F, the
effect  is  the   same  as  though   he  had  typed   F (SETQ X &),  not
(F (SETQ X &)).





------------------------------------------------------------------------
32
    The local  buffer is stored  on lispxbufs; the  top level  buffer on
    toplispxbufs. The forms of both buffers are (CONS (LINBUF) (SYSBUF))
    (see Section 14). Recovery of  a buffer is destructive,  i.e., $BUFS
    sets the corresponding variable to NIL. If the user types $BUFS when
    both lispxbufs and  toplispxbufs are NIL, the  message NOTHING SAVED
    is typed, and an error generated.




                                 22.26



!E                      same  as  DO  E.  Note  !E  works  correctly for
                        "commands" typed in eval or apply format.


!N                      same as DO N.


control-U               when typed in at any point during an input being
                        read by lispxread, permits the user to  edit the
                        input  before  it  is  returned  to  the calling
                        function.

This feature is useful for correcting mistakes noticed in  typing before
the input is executed, instead of waiting till after execution  and then
performing  an  UNDO  and  a  FIX.   For  example,  if  the  user  types
(DEFINEQ (FOO (LAMBDA (X) (FIXSPELL X  and  at  that  point  notices the
missing left parenthesis, instead  of completing the input  and allowing
the  error to  occur, and  then  fixing the  input, he  can  simply type
          33
control-U,   finish typing normally,  whereupon the editor is  called on
(FOO (LAMBDA (X) (FIXSPELL X -- ], which the user can then repair, e.g.,
by  typing  (LI 1).  If  the  user exits  from  the editor  via  OK, the
(corrected)  expression will  be  returned to  whoever  called lispxread
                                     34
exactly as though it had been  typed.   If the user exits via  STOP, the
expression is  returned so that  it can be  stored on the  history list.
However it will not be executed.  In other words, the effect is the same
as though the user had typed control-E at exactly the right instant.


valueof[x]              is an nlambda  function for obtaining  the value
                                                                      35
                        of  a  particular  event,  e.g., (VALUEOF -1),
                        (VALUEOF _FOO -2).

                        The  value  of an  event  consisting  of several
                        operations is a list  of the values for  each of
                        the individual operations.





------------------------------------------------------------------------
33
    Control-U can be typed at any point, even in the middle of  an atom;
    it simply sets a variable checked by lispxread.

34
    Control-U also works for calls to readline, i.e., for line commands.

35
    Although the input for valueof is entered on the history list before
    valueof  is called,  valueof[-1] still  refers to  the value  of the
    expression  immediately before  the valueof  input,  because valueof
    effectively backs the  history list up  one entry when  it retrieves
    the specified  event. Similarly, (VALUEOF FOO)  will find  the first
    event before this one that contains a FOO.




                                 22.27



Note:  the  value  field  of a  history  entry  is  initialized  to bell
(control-G).   Thus a  value of  bell indicates  that  the corresponding
operation did  not complete, i.e.,  was aborted or  caused an  error (or
else returned bell).


prompt#flg              is a flag which when set to T causes the current
                        event number to be printed before each _,: and *
                        prompt    characters.    See    description   of
                        promptchar, page 22.42.

                        prompt#flg is initially NIL.


archivefn               allows  the  user   to  specify  events   to  be
                        automatically archived.

When archivefn is set to T, and an event is about to drop off the end of
the history list and be forgotten, archivefn is called giving it  as its
first  argument  the input  portion  of  the event,  and  as  its second
                             36
argument,  the entire  event.    If archivefn  returns T,  the  event is
archived.  For example, some users like to keep a record of all calls to
load.  Defining archivefn as:   (LAMBDA (X Y) (EQ (CAR X) (QUOTE LOAD)))
will accomplish this.  Note that archivefn must be both set and defined.
archivefn is initially NIL and undefined.

The user can also specify that a particular event be archived when it is
about to drop off  the end of the  history list by putting  the property
*ARCHIVE*, value  T, on the  event, e.g., by  means of  an appropriately
defined lispxuserfn (see  below).  One use of  this feature is  that the
system automatically adds the *ARCHIVE* property to all events  that are
                                37
referenced by history  commands.   Thus once an  event is redone,  it is
guaranteed to be saved.


lispxmacros             provides a macro facility for lispx.

lispxmacros allows the user to  define his own lispx commands.  It  is a
list of elements of the form (command def).  Whenever command appears as
the first expression on a line in a lispx input, the  variable lispxline
is bound to the rest of  the line, the event is recorded on  the history
list, and def is evaluated.  Similarly, whenever command appears  as car
of a form in  a lispx input, the variable  lispxline is bound to  cdr of
the form,  the event recorded,  and def is  evaluated.  (See  page 22.48
for  an  example of  a  lispxmacro).  RETRIEVE,  BEFORE,  and  AFTER are



------------------------------------------------------------------------
36
    In case archivefn needs to examine the value of the event,  its side
    effects,  etc.  See  page  22.36 for  discussion  of  the  format of
    history lists.

37
    unless archiveflg=NIL.  archiveflg is initially T.




                                 22.28



implemented as lispxmacros.  In addition in INTERLISP-10,  LISP, SNDMSG,
TECO, and EXEC are lispxmacros which perform the corresponding  calls to
subsys  (Section  21),  and  CONTIN  is  a  lispxmacro   which  performs
(SUBSYS T).  SY  is a  lispxmacro which performs  the SYSTAT  command.  
Finally, DIR is a  lispxmacro which calls the function  directory, e.g.,
DIR *.COM;* lists  all compiled files.   (For more details,  see Section
21.)


lispxhistorymacros      provides a macro facility for history commands.

lispxhistorymacros allows the user  to define his own  history commands.
The format  of lispxhistorymacros  is the same  as that  of lispxmacros,
except  that  the result  of  evaluating def  is  treated as  a  list of
expressions to  be unread,  exactly as though  the expressions  had been
                                                          38
retrieved by a REDO command, or computed by a USE command.


lispxuserfn             provides a  way for a  user function  to process
                        selected inputs.

                                                 39
When  lispxuserfn  is  set to  T,  it  is applied    to  all  inputs not
recognized  as  one of  the  commands described  above.   If lispxuserfn
decides to  handle this  input, it  simply processes  it (the  event was
already stored on the history list before lispxuserfn was  called), sets
lispxvalue to the value for  the event, and returns T.  lispx  will then
know not to  call eval or apply,  and will simply store  lispxvalue into
the value slot for the event, and print it.  If lispxuserfn returns NIL,
lispx  proceeds by  calling eval  or apply  in the  usual way.   Thus by
appropriately defining  (and setting) lispxuserfn,  the user can  with a
minimum of effort incorporate the features of the programmer's assistant
into his own executive (actually it is the other way around).












------------------------------------------------------------------------
38
    See page 22.15  for discussion of  implementation of REDO,  USE, and
    FIX.

39
    Like archivefn, lispxuserfn must be both set and defined.









                                 22.29



                                                 40
The following output illustrates such a coupling.

**SETQ(ALTFORM (MAPCONC NASDIC (F/L (GETP X 'ALTFORMS]               [1]
=NASDICT
(AL26 BE7 CO56 CO57 CO60 C13 H3 MN54 NA22 SC46 S34 TI44)
**(GIVE ME LINES CONTAINING COBALT)                                  [2]
SAMPLE   PHASE     CONSTIT.  CONTENT   UNIT      CITATION  TAG
S10002   OVERALL   C056      40.0      DPM/KG    D70-237   0
                   C13       8.8       DEL       D70-228   0
                   H3        314.0     DPM/KG
                   MN54      28

**GETP(COBALT ALTFORMS)                                              [3]
(CO56 CO57 CO60 C13 H3 MN54 NA22 SC46 S34 T144)
**UNDO MAPCONC                                                       [4]
SETQ UNDONE.
**REDO GETP                                                          [5]
(CO56 CO57 CO60)
**REDO COBALT                                                        [6]
SAMPLE   PHASE     CONSTIT.  CONTENT   UNIT      CITATION  TAG
S10002   OVERALL   CO57      40.0      DPM/KG    D70-237   0
S10003   OVERALL   CO        15.0                D70-203   0
                             14.1                D70-216
                   CO56      43.0      DPM/KG    D70-237   0
                   CO57      43.0                D70-241   0
                   CO60      1.0
**USE MANGANESE FOR COBALT

The  user  is running  under  his own  executive  program  which accepts
requests  in the  form  of sentences,  which  it first  parses  and then
executes.   The  user  first   "innocently"  computes  a  list   of  all
ALTERNATIVE-FORMS for the elements in his system [1].  He then  inputs a
request  in  sentence  format  [2] expecting  to  see  under  the column
CONSTIT. only cobalt, CO, or  its alternate forms, CO56, CO57,  or CO60.
Seeing C13, H3, and MN54, he aborts the output, and checks  the property
ALTFORMS for COBALT [3].  The appearance of C13, H3, MN54, he aborts the
output, and checks the property ALTFORMS for COBALT [3].  The appearance
of C13, H3, MN54 et al, remind him that the mapconc is  destructive, and
that  in  the  process  of  making  a  list  of  the  ALTFORMS,  he  has
inadvertently strung them all together.  Recovering from  this situation
would require him to  individually examine and correct the  ALTFORMs for
each  element in  his dictionary,  a tedious  process.  Instead,  he can
simply  UNDO MAPCONC,  [4]  check  to make  sure  the  ALTFORM  has been
corrected [5],  then redo  his original request  [6] and  continue.  The
UNDO is  possible because  the first  input was  executed by  lispx; the
(GIVE ME LINES CONTAINING  COBALT) is possible because the  user defined
lispxuserfn appropriately; and the REDO and USE are possible because the
(GIVE ME LINES CONTAINING COBALT) was stored on the history  list before
it was transmitted to lispxuserfn and the user's parsing program.



------------------------------------------------------------------------
40
    The output is from  the Lunar Sciences Natural  Language Information
    System  being developed  for the  NASA Manned  Spacecraft  Center by
    William A. Woods of Bolt Beranek and Newman Inc., Cambridge, Mass.




                                 22.30



lispxuserfn is a function of two  arguments, x and line, where x  is the
first  expression typed,  and line  the  rest of  the line,  as  read by
readline (see page 22.39).   For example, if the user  types FOO(A B C),
x=FOO, and line=((A B C)); if the user types (FOO A B C), x=(FOO A B C),
and line=NIL; and if the user types FOO A B C, x=FOO and line=(A B C).

Thus in the above example, lispxuserfn would be defined as:

         [LAMBDA (X LINE)
             (COND
               ((AND (NULL LINE)
                     (LISTP X))
                  (SETQ LISPXVALUE (PARSE X))
                  T]

Note that since  lispxuserfn is called for  each input (except  for p.a.
commands),  it can  also be  used to  monitor some  condition  or gather
statistics.

In  addition  to  saving  inputs and  values,  lispx  saves  most system
messages     on    the     history    list,     e.g.,    FILE CREATED --
, (fn REDEFINED), (var RESET), output                                 of
TIME, BREAKDOWN, STORAGE, DWIM messages, etc.  When  printhistory prints
the event, this output is replicated.  This facility is  implemented via
the   functions   lispxprint,   lispxprin1,   lispxprin2,   lispxspaces,
                          41
lispxterpri, and lispxtab.   In addition to performing the corresponding
output operation, these functions store an appropriate expression on the
                                              42
history event under the property *LISPXPRINT*.   This expression is used
by printhistory to reproduce the output.

In addition to the above features, lispx checks to see if car or  cdr of
NIL or car of T have been clobbered, and if so, restores them and prints
a message.  Lispx also performs spelling corrections using  lispxcoms, a
list of its commands, as a spelling list whenever it is given an unbound
atom  or undefined  function, i.e.,  before attempting  to  evaluate the





------------------------------------------------------------------------
41
    The function userlispxprint is available to aid the user in defining
    additional  lispxprinting  functions for  already  existing printing
    functions.  The user can  define a lispxprinting function  by simply
    giving    it    the    definition    of     userlispxprint,    e.g.,
    MOVD(USERLISPXPRINT LISPXPRINTDEF), as long as the new function name
    is formed by adding "LISPX" to the front of the name of  an existing
    printing  function,  and that  this  function takes  three  or fewer
    arguments. userlispxprint is defined to look back on the stack, find
    the name  of the  calling function, strip  off the  leading "LISPX",
    perform  the  appropriate  saving  information,  and  then  call the
    function to do the actual printing.

42
    unless lispxprintflg is NIL.




                                 22.31



      43
input.


22.6 Undoing

The  UNDO capability  of the  programmer's assistant  is  implemented by
requiring  that each  operation that  is to  be undoable  be responsible
itself  for saving  on  the history  list enough  information  to enable
reversal of its  side effects.  In other  words, the assistant  does not
"know" when it is about to perform a destructive operation, i.e.,  it is
not constantly  checking or anticipating.   Instead, it  simply executes
operations, and any undoable changes that occur are  automatically saved
                                                    44
on  the history  list by  the responsible  function.   The  operation of
UNDOing, which involves recovering the saved information  and performing
the corresponding  inverses, works the  same way, so  that the  user can
UNDO an UNDO, and UNDO that etc.

At each point, until the  user specifically requests an operation  to be
undone, the assistant  does not know,  or care, whether  information has
been saved to enable the  undoing.  Only when the user attempts  to undo
an operation does the assistant check to see whether any information has
been saved.  If none has been saved, and the user has specifically named
the event he wants undone, the assistant types NOTHING SAVED.  (When the
user simply  types UNDO,  the assistant searches  for the  last undoable
event,  ingnoring  events  already undone  as  well  as  UNDO operations
themselves.)

This  implementation minimizes  the  overhead for  undoing.   Only those
operations which actually make changes are affected, and the overhead is
small: two or three cells of storage for saving the information,  and an
extra  function  call.   However,  even  this  small  price  may  be too
expensive  if the  operation is  sufficiently primitive  and repetitive,
i.e.,  if  the  extra   overhead  may  seriously  degrade   the  overall





------------------------------------------------------------------------
43
    lispx  is  also   responsible  for  rebinding  helpclock,   used  by
    breakcheck, Section 16, for computing the amount of time spent  in a
    computation, in order to determine whether to go into a break if and
    when an error occurs.

44
    When the number of changes that have been saved exceeds the value of
    #undosaves (initially set to 50),  the user is asked if he  wants to
    continue saving the undo information for this event. The  purpose of
    this feature is  to avoid tying up  large quantities of  storage for
    operations that  will never  need to be  undone. The  interaction is
    handled by the same routines used by DWIM, so that the input buffers
    are  first saved  and cleared,  the message  typed, then  the system
    waits dwimwait  seconds, and  if there is  no response,  assumes the
    default answer, which in this case is NO. Finally the  input buffers
    are restored. See page 22.45 for details.




                                 22.32



                           45
performance of the program.   Hence not every destructive operation in a
program should necessarily be  undoable; the programmer must  be allowed
to decide each case individually.

Therefore for each primitive destructive operation, we  have implemented
two separate  functions, one  which always  saves information,  i.e., is
always undoable, and one which does not, e.g., /rplaca and  rplaca, /put
        46
and put.   In the  various system packages, the appropriate  function is
used.  For example, break uses /putd and /remprop so as to  be undoable,
                                                                      47
and  DWIM  uses  /rplaca  and /rplacd,  when  it  makes  a correction.
Similarly the  user can simply  use the corresponding  / function  if he
wants to make a destructive operation in his own program undoable.  When
the  / function  is called,  it will  save the  undo information  in the
current event on the history list.

However, all operations  that are typed in  to lispx are  made undoable,
                                                           48
simply  by  substituting  the  corresponding  /  function       for  any
                                          49
destructive function throughout the input.   For example, on  page 22.7,



------------------------------------------------------------------------
45
    The rest of the discussion applies only to lispx; the editor handles
    undoing itself in a slightly different fashion, as described on page
    22.49.

46
    The "slash" functions  currently implemented are  /addprop, /attach,
    /closer,  /control,  /deletecontrol,  /dremove,  /dreverse, /dsubst,
    /echocontrol,  /echomode,  /lconc,  /listput,   /listput1,  /mapcon,
    /mapconc,   /movd,  /nconc,   /nconc1,  /putassoc,   /putd,  /putdq,
    /puthash, /putprop, /remprop, /rplaca, /rplacd, /rplnode, /rplnode2,
    /set,  /seta,  /setbrk, /setd,  /setproplist,  /setsepr, /setsyntax,
    /settopval, and /tconc. Note that /setq and /setqq are not included.
    If the user wants a  set operation undoable in his program,  he must
    see /set, or /rplaca.

47
    The  effects  of   the  following  functions  are   always  undoable
    (regardless of whether or  not they are typed in):  define, defineq,
    defc (used to give a function a compiled code  definition), deflist,
    load, savedef, unsavedef,  break, unbreak, rebreak,  trace, breakin,
    unbreakin, changename, editfns,  editf, editv, editp,  edite, editl,
    esubst, advise, unadvise, readvise, plus any changes caused by DWIM.

48
    Since there  is no /setq,  setq's appearing  in type-in  are handled
    specially by substituting a call to saveset, page 22.35.

49
    The substitution is performed  by the function lispx/,  described on
    page 22.46.




                                 22.33



when    the    user    typed    (MAPCONC NASDIC (F/L ...))     it    was
(/MAPCONC NASDIC (F/L ...)) that was evaluated.  Since the system cannot
know whether efficiency and overhead are serious considerations  for the
execution of  an expression  in a  user program,  the user  must decide,
e.g.,  call  /mapconc  if he  wants  the  operation  undoable.  However,
expressions  that  are  typed-in rarely  involve  iterations  or lengthy
computations   directly.   Therefore,   if  all   primitive  destructive
functions that are immediately contained in a type-in are made undoable,
there will rarely be a significant loss of efficiency.  Thus lispx scans
all user input before  evaluating it, and substitutes  the corresponding
undoable function  for all  primitive destructive  functions.  Obviously
with a more sophisticated analysis of both user input and user programs,
the  decision  concerning which  operations  to make  undoable  could be
better advised.  However, we have found the configuration described here
to be a  very satisfactory one.   The user pays  a very small  price for
being able to undo what he types in, and if he wishes to protect himself
from  malfunctioning  in  his  own programs,  he  can  have  his program
specifically call undoable functions,  or go into testmode  as described
next.


Testmode

Because  of efficiency  considerations, the  user may  not  want certain
functions  undoable  after his  program  becomes  operational.  However,
while debugging he  may find it desirable  to protect himself  against a
program  running  wild,  by  making  primitive   destructive  operations
undoable.  The function testmode provides this capability by temporarily
making everything undoable.


testmode[flg]           testmode[T] redefines all  primitive destructive
                                 50
                        functions    with  their  corresponding undoable
                        versions and sets testmodeflg to  T.  testmode[]
                        restores  the  original  definitions,  and  sets
                                           51
                        testmodeflg to NIL.

Note  that setq's  are  not undoable,  even  in testmode.   To  make the
corresponding operation  undoable in testmode,  set or rplaca  should be
used.


Undoing Out of Order

/rplaca and /rplacd operate by saving the pointer that is to  be changed
and its  original contents  (i.e., /rplaca saves  car and  /rplacd saves



------------------------------------------------------------------------
50
    i.e., the "slash" functions; see footnote on page 22.33.

51
    testmode  will  have no  effect  on compiled  mapconc's,  since they
    compile open with frplacd's.




                                 22.34



cdr).  Undoing /rplaca and  /rplacd simply restores the  pointer.  Thus,
if  the  user  types (RPLACA FOO 1),  followed  by  (RPLACA FOO 2), then
undoes both events by undoing the most recent event first,  then undoing
the older event, FOO will be restored to its state before  either rplaca
operated.  However if the user  undoes the first event, then  the second
event, (CAR FOO) will be 1, since this is what was in car of  FOO before
(RPLACA FOO 2)   was  executed.    Similarly,  if   the   user  performs
(NCONC1 FOO 1)  then  (NCONC1 FOO 2), undoing  just  (NCONC1 FOO 1) will
remove both 1 and 2 from FOO.  The problem in both cases is that the two
operations  are not  "independent."  In general,  operations  are always
independent if they affect different lists or different sublists  of the
           52
same  list.    Undoing  in  reverse  order  of  execution,   or  undoing
independent operations,  is always guaranteed  to do the  "right" thing.
However, undoing dependent operations  out of order may not  always have
the predicted effect.


Saveset

Setq's are made  undoable on type in  by substituting a call  to saveset
(described in detail  on page 22.44), whenever  setq is the name  of the
function to be applied, or car of the form to be evaluated.  In addition
to  saving enough  information on  the history  list to  enable undoing,
saveset operates in a manner  analogous to savedef when it resets  a top
level value,  i.e., when  it changes a  top level  binding from  a value
other than NOBIND to a new value  that is not equal to the old  one.  In
this case, saveset saves the old value of the variable being set  on the
variable's  property  list  under the  property  VALUE,  and  prints the
message  (variable  RESET).   The  old value  can  be  restored  via the
               53
function unset,   which also saves the current value (but does not print
a message).  Thus unset can be  used to flip back and forth  between two
values.

rpaq and rpaqq  are implemented via calls  to saveset.  Thus  old values
will be saved and messages  printed for any variables that are  reset as





------------------------------------------------------------------------
52
    Property  list  operations,  (i.e., put,  addprop  and  remprop) are
    handled specially  so that  they are  always independent,  even when
    they affect the same property  list. For example, if the  user types
    PUT(FOO FIE1 FUM1)  then PUT(FOO FIE2 FUM2),  then undoes  the first
    event, the FIE2 property will remain, even though CDR(FOO)  may have
    been NIL at the time the first event was executed.

53
    Of course,  UNDO can be  used as long  as the event  containing this
    call to  saveset is still  active. Note however  that the  old value
    will remain on the  property list, and therefore be  recoverable via
    unset, even after the original event has been forgotten.






                                 22.35


                             54
the result of loading a file.   Calls to set and setqq appearing in type
in are also converted to appropriate calls to saveset.

For  top  level  variables,  saveset  also  adds  the  variable  to  the
appropriate spelling list, thereby  noticing variables set in  files via
rpaq or rpaqq, as well as those set via type in.


Undonlsetq and Resetundo

The function undonlsetq provides  a limited form of backtracking:  if an
error occurs  under the undonlsetq,  all undoable side  effects executed
under the undonlsetq are undone.  resetundo, for use in conjunction with
resetlst  and  resetsave  (Section  5),  provides  a  more  general undo
capability in that the user can specify that the side effects  be undone
after the specified computation finishes, is aborted by an error,  or by
a control-D.  undonlsetq and  resetundo are described in detail  on page
22.47.


22.7 Format and Use of the History List

There  are currently  two history  lists, lispxhistory  and edithistory.
Both history lists have the same format, and in fact, each use  the same
function,  historysave,  for  recording  events,  and  the  same  set of
functions  for implementing  commands that  refer to  the  history list,
                                               55
e.g., historyfind, printhistory, undosave, etc.

Each history list is a list of the form (l event# size mod), where  l is
the list of events with the most recent event first, event# is the event
number for the  most recent event  on l, size is  the size of  the time-
slice, i.e., the  maximum length of l,  and mod is the  highest possible
event number (see footnote on page 22.7).  lispxhistory  and edithistory
are  both  initialized  to  (NIL 0 30 100).   Setting   lispxhistory  or
edithistory  to  NIL  is  permitted,  and  simply  disables  all history
features, i.e., lispxhistory and  edithistory act like flags as  well as
repositories of events.

Each   individual    event   on    l   is   a    list   of    the   form
(input id value . props),  where input  is  the input  sequence  for the
event,  as  described  on  page  22.15-17,  id  the   prompt  character,




------------------------------------------------------------------------
54
    To  complete the  analogy  with define,  saveset will  not  save old
    values on property lists if dfnflg=T, e.g., when load is called with
    second  argument T,  (however,  the call  to saveset  will  still be
    undoable,) and when dfnflg=ALLPROP, the value is stored  directly on
    the property list under  property VALUE (the latter applies  only to
    calls from rpaqq and rpaq).

55
    A third history list, archivelst, is used when events  are archived,
    as described on page 22.23. It too uses the same format.




                                 22.36



              56
e.g., _, :, *,   and value is the value of the event, and is initialized
        57
to bell.

props is  a property list,  i.e., of the  form (property  value property
value --).  props can be used to associate arbitrary information  with a
particular event.  Currently,  the properties SIDE,  *ARCHIVE*, *GROUP*,
*HISTORY*,   *PRINT*,   USE-ARGS,   ...ARGS,   *ERROR*,   *CONTEXT*  and
*LISPXPRINT* are being  used.  The value of  property SIDE is a  list of
the  side  effects of  the  event.  (See  discussion  of  undosave, page
22.45,   and  undolispx,   page  22.47.)   The  *HISTORY*   and  *GROUP*
properties are used for  commands that reexecute previous  events, i.e.,
REDO, RETRY, USE, ..., and FIX.  The value of the *HISTORY*  property is
the history command  itself, i.e., what  the user actually  typed, e.g.,
REDO FROM F, and is used by the ?? command for printing the  event.  The
value of the property  *PRINT* is also for  use by the ??  command, when
special  formatting  is  required,  for  example,  in   printing  events
corresponding to the break commands OK, GO, EVAL, and ?=.   USE-ARGS and
...ARGS  are  used  to  save  the  arguments  and  expression   for  the
corresponding history command.  *ERROR*  and *CONTEXT* are used  to save
information  when  errors  occur  for subsequent  use  by  the  $  and ?
commands.  The  property *ARCHIVE* on  an event causes  the event  to be
automatically archived when it "falls  off the end" of the  history list
(see page 22.28).  *LISPXPRINT*  is used to record calls  to lispxprint,
lispxprin1, et al, (see page 22.31).

When lispx is given an  input, it calls historysave to record  the input
                58
in a  new event.   Normally,  historysave returns as  its value  the new
event.  lispx binds lispxhist to the value of historysave, so  that when
the  operation has  completed,  lispx knows  where to  store  the value,
                              59
namely in caddr  of lispxhist.   lispxhist  also provides access  to the



------------------------------------------------------------------------
56
    id is one of the arguments  to lispx and to historysave. A  user can
    call lispx giving it any  prompt character he wishes (except  for *,
    since  in certain  cases, lispx  must use  the value  of id  to tell
    whether or not it was called from the editor.) For example,  on page
    22.30, the user's prompt character was **.

57
    On edithistory, this field is used to save the side effects  of each
    command.

58
    The commands ??, FORGET, TYPE-AHEAD, $BUFS, and ARCHIVE are executed
    immediately, and are not recorded on the history list.

59
    Note that  by the  time it  completes, the  operation may  no longer
    correspond  to  the  most  recent event  on  the  history  list. For
    example, all inputs typed to a lower break will appear later  on the
    history list.




                                 22.37



property list for the current  event.  For example, the /  functions are
all implemented  to call undosave,  which simply adds  the corresponding
information to  lispxhist under  the property  SIDE, or  if there  is no
property SIDE, creates one, and then adds the information.

After binding lispxhist, lispx  executes the input, stores its  value in
caddr of lispxhist, prints the value, and returns.

When the input is a REDO, RETRY, USE, ..., or FIX command, the procedure
is similar,  except that  the event  is also  given a  *GROUP* property,
initially NIL, and  a *HISTORY* property,  and lispx simply  unreads the
input and returns.  When the  input is "reread", it is  historysave, not
lispx, that notices this fact, and finds the event from which  the input
                60
originally came.   historysave then adds a new (input id value  . props)
entry to the *GROUP* property for this event, and returns this  entry as
the "new event." lispx then proceeds exactly as when its input was typed
directly, i.e., it binds lispxhist to the value of historysave, executes
the input, stores the value in caddr of lispxhist, prints the value, and
returns.  In fact, lispx never notices whether it is working  on freshly
typed input, or input  that was reread.  Similarly, undosave  will store
undo  information  on lispxhist  under  the property  SIDE  the  same as
always,  and does  not know  or care  that lispxhist  is not  the entire
event, but one of the  elements of the *GROUP* property.  Thus  when the
event is finished, its entry will look like:
(input id value *HISTORY* command *GROUP* ((input1                   id1
value1 SIDE side1)
                                           (input2                   id2
value2 SIDE side2)
                                                                61
                                            ...))

This  implementation  removes  the  burden  from  the  function  calling
historysave of distinguishing between new input and reexecution of input
                                            62
whose history entry has already been set up.  




------------------------------------------------------------------------
60
    If historysave cannot find the event, for example if a  user program
    unreads  the  input  directly,  and  not  via  a   history  command,
    historysave proceeds as though the input were typed.

61
    In this  case, the value  field is not  being used;  valueof instead
    collects each of the values from the *GROUP* property, i.e., returns
    mapcar[get[event;*GROUP*];CADDR].   Similarly,   undo   operates  by
    collecting  the SIDE  properties from  each of  the elements  of the
    *GROUP* property, and then undoing them in reverse order.

62
    Although we have not yet done so, this implementation, i.e., keeping
    the  various  "sub-events"  separate  with  respect  to  values  and
    properties, also permits constructing commands for operating on just
    one of the sub-events.




                                 22.38



22.8 Lispx and readline

lispx is called with the first  expression typed on a line as  its first
argument, lispxx.

If this is not a list,  lispx always does a readline, and  treats lispxx
plus the line as the input  for the event, and stores it  accordingly on
                 63
the history list.   Then it decides what to do with the input,  i.e., if
it is  not recognized  as a command,  a lispxmacro,  or is  processed by
                                   64
lispxuserfn,  call  eval  or apply.    readline  normally  is terminated
either by (1) a carriage return that is not preceded by a space,  or (2)
a list that is terminated by a  ], or (3) an unmatched ) or ],  which is
not included  in the  line.  However, when  called from  lispx, readline
operates differently in two respects:

    (1)  If the line consists of a single ) or ], readline returns (NIL)
         instead of NIL, i.e., the ) or ] is included in the line.  This
         permits  the  user  to  type FOO)  or  FOO],  meaning  call the
         function'   FOO  with   no  arguments,   as  opposed   to  FOOC
         (FOOcarriage-return), meaning evaluate the variable FOO.

    (2)  If  the first  expression on  the line  is a  list that  is not
         preceded by any spaces, the list terminates the line regardless
         of whether or not it is terminated by ].  This permits the user
         to type EDITF(FOO) as a single input.

Note  that if  any spaces  are inserted  between the  atom and  the left
parentheses  or bracket,  readline will  assume that  the list  does not
terminate the line.  This is to  enable the user to type a  line command
such  as USE (FOO) FOR FOO.   In this  case, a  carriage return  will be
typed  after  (FOO)  followed  by  "..."  as  described  in  Section 14.
Therefore,  if the  user  accidentially puts  an extra  space  between a
function and  its arguments,  he will  have to  complete the  input with
another carriage return, e.g.,

                        _EDITF(FOO)
                        ...C
                        EDIT
                        *








------------------------------------------------------------------------
63
    If lispxx is a list car  of which is LAMBDA or NLAMBDA,  lispx calls
    lispxread to obtain the arguments.

64
    If the  input consists of  one expression, eval  is called;  if two,
    apply; if more than two, the entire line is treated as a single form
    and eval is called.




                                 22.39



22.9 Functions

                                               65
lispx[lispxx;lispxid;lispxxmacros;lispxxuserfn]
                        lispx  is  like eval/apply.   It  carries  out a
                        single computation, and returns its  value.  The
                        first argument, lispxx is the result of a single
                        call to lispxread.  lispx will call readline, if
                        necessary  as  described on  page  22.39.  lispx
                        prints the value of the computation, as  well as
                                                                   66
                        saving the input and value on lispxhistory.

                        If  lispxx  is a  history  command,  lispx calls
                        historysave, executes  the command,  and returns
                        the value of historysave.

                        If   the   value   of   the   fourth   argument,
                        lispxxmacros,  is not  NIL,  it is  used  as the
                        lispx macros, otherwise  the top level  value of
                        lispxmacros is used.  If the value of  the fifth
                        argument, lispxxuserfn, is  not NIL, it  is used
                        as  lispxuserfn.   In  this  case,  it   is  not
                        necessary to both set and define  lispxuserfn as
                        described on page 22.29.

The overhead for a call  to lispx (in INTERLISP-10) is  approximately 17
milliseconds,  of which  12 milliseconds  are spent  in  maintaining the
spelling lists.   In other words,  in INTERLISP, the  user pays  17 more
milliseconds  for each  eval  or apply  input over  a  conventional LISP
executive, in order to enable the features described in this chapter.


userexec[lispxid;lispxxmacros;lispxxuserfn]
                        repeatedly calls lispx under errorset protection
                        specifying  lispxxmacros  and  lispxxuserfn, and
                        using lispxid (or _ if lispxid=NIL) as  a prompt
                        character.    Userexec   is   exited   via   the
                        lispxmacro OK, or else with a retfrom.


lispxread[file;rdtbl]   is   a   generalized   read.    If  readbuf=NIL,
                        lispxread  performs  read[file;rdtbl],  which it



------------------------------------------------------------------------
65
    lispxid corresponds  to id  on page  22.36. Lispx  also has  a fifth
    argument, lispxflg, which is used by the E command in the editor.

66
    Note that the  history is not one  of the arguments to  lispx, i.e.,
    the  editor must  bind  (reset) lispxhistory  to  edithistory before
    calling lispx to carry out a history command.
    Lispx  will  continue  to  operate  as  an  eval/apply  function  if
    lispxhistory is NIL. Only those functions and commands  that involve
    the history list will be affected.




                                 22.40



                        returns  as  its  value.   (If  the  user  types
                        control-U  during  the call  to  read, lispxread
                        calls the editor and returns the edited value.)

                        If  readbuf is  not NIL,  lispxread  "reads" the
                        next  expression on  readbuf,  i.e., essentially
                        returns
                        (PROG1 (CAR READBUF)
                                                             67
                               (SETQ READBUF (CDR READBUF))).


readline, described in Section 14, also uses this generalized  notion of
reading.  When  readbuf is  not NIL,  readline "reads"  expressions from
readbuf until it either reaches the end of readbuf, or until it  reads a
pseudo-carriage return (see  page 22.16).  In  both cases, it  returns a
list of the expressions  it has "read".  (The pseudo-carriage  return is
not included in the list.)

When readbuf  is not  NIL, both lispxread  and readline  actually obtain
their input  by performing (APPLY* LISPXREADFN FILE),  where lispxreadfn
is initially set  to READ.  Thus, if  the user wants lispx,  the editor,
break, et al to do  their reading via a different input  function, e.g.,
uread,  he  simply  sets  lispxreadfn  to  the  name  of  that  function
(or an appropriate LAMBDA expression).


lispxreadp[flg]         is  a generalized  readp.  If  flg=T, lispxreadp
                        returns T  if there is  any input waiting  to be
                        "read", a la lispxread.  If  flg=NIL, lispxreadp
                        returns T only if there is any input  waiting to
                        be "read" on this line.  In both  cases, leading
                        spaces  are  ignored,  i.e.,  skipped  over with
                        readc, so that  if only spaces have  been typed,
                        lispxreadp will return NIL.


lispxunread[lst]        unreads lst, a  list of expressions to  be read.
                        If readbuf is not NIL, lispxunread  attaches lst
                        at the front of readbuf, separating it  from the
                        rest of readbuf with a histstr0.  The definition
                        of lispxunread is:

                        (LISPXUNREAD
                          [LAMBDA (LST)
                            (SETQ READBUF (COND
                                ((NULL READBUF)
                                  LST)
                                (T (APPEND LST (CONS HISTSTR0
                                                      READBUF])


------------------------------------------------------------------------
67
    Except that pseudo-carriage returns, as represented by the  value of
    histstr0, are ignored, i.e., skipped.  Lispxread also sets rereadflg
    to NIL when it  reads via read, and  sets rereadflg to the  value of
    readbuf when rereading.




                                 22.41



promptchar[id;flg;hist] prints the prompt character id.

                        promptchar will not print anything when the next
                        input  will be  "reread", i.e.,  readbuf  is not
                        NIL.   promptchar  will  also  not   print  when
                        readp[]=T, unless flg is T.

Thus the editor calls promptchar with flg=NIL so that extra *'s  are not
printed  when the  user types  several commands  on one  line.  However,
evalqt calls promptchar with flg=T, since it always wants the  _ printed
(except when "rereading").

                        Finally, if prompt#flg is T and hist is not NIL,
                        promptchar prints  the current event  number (of
                        hist) before printing id.


lispxeval[lispxform;lispxid]
                        evaluates  lispxform  (using eval)  the  same as
                        though  it were  typed  in to  lispx,  i.e., the
                        event is  recorded, and  the evaluation  is made
                        undoable by substituting the slash functions for
                        the  corresponding  destructive   functions,  as
                        described on page 22.33.  lispxeval  returns the
                        value of the form, but does not print it.


historysave[history;id;input1;input2;input3;props]
                        records one event on history.  If input1  is not
                        NIL,    the    input    is    of     the    form
                        (input1 input2 . input3).  If input1 is NIL, and
                        input2  is not  NIL, the  input is  of  the form
                        (input2 . input3).  Otherwise, the input is just
                        input3.

                        historysave  creates   a  new  event   with  the
                        corresponding input, id, value field initialized
                        to bell, and props.  If the history  has reached
                        its  full size,  the last  event is  removed and
                        cannibalized.

                        The  value  of  historysave  is  the  new event.
                        However, if rereadflg  is not NIL, and  the most
                        recent event  on the  history list  contains the
                        history  command   that  produced   this  input,
                        historysave  does not  create a  new  event, but
                        simply adds an (input id bell . props)  entry to
                        the *GROUP* property for that event  and returns
                        that entry.  See discussion on page 22.38.


lispxfind[history;line;type;backup]
                        line is  an event specification,  type specifies
                        the  format  of  the  value  to  be  returned by
                        lispxfind,  and  can be  either  ENTRY, ENTRIES,
                        COPY, COPIES, INPUT, or REDO.   lispxfind parses
                        line,   and   uses  historyfind   to   find  the
                        corresponding events.  lispxfind  then assembles
                        and returns the appropriate structure.



                                 22.42



lispxfind incorporates the following special features:

1)  if backup=T, lispxfind interprets line in the context of the history
    list before the current event was added.  This feature is  used, for
    example,  by valueof,  so that  (VALUEOF -1) will  not refer  to the
    valueof event itself;

2)  if line=NIL  and the last  event is  an UNDO, the  next to  the last
    event is taken.  This permits the user to type UNDO followed by REDO
    or USE;

3)  lispxfind recognizes @@, and substitutes archivelst for history (see
    page 22.12); and

4)  lispxfind  recognizes @,  and retrieves  the  corresponding event(s)
    from the property list of the atom following @ (see page 22.12).


historyfind[lst;index;mod;x;y]
                        searches  lst  and  returns  the  tails  of  lst
                        beginning  with  the event  corresponding  to x.
                        lst,  index, and  mod are  as described  on page
                        22.36.
                        x  is an  event  address, as  described  on page
                        22.10-12,    e.g.,     (43), (-1),    (FOO FIE),
                                          68
                        (LOAD _ FOO), etc.   If historyfind  cannot find
                        x, it generates an error.


entry#[hist;x]          hist  is  a  history  list,  i.e.,  of  the form
                        described on page 22.36.  x is one of the events
                        on hist, i.e., (MEMB X (CAR HIST)) is true.  The
                        value of entry# is the event number for x.


valueof[x]              is an  nlambda, nospread function  for obtaining
                        the  value of  the event  specified by  x, e.g.,
                        (VALUEOF -1),  (VALUEOF LOAD 1),  etc.   valueof
                        returns a list of the corresponding values  if x
                        specifies a multiple event.


                      69
changeslice[n;history]  changes time-slice for history to n.  If history
                        is   NIL,    changes   both    edithistory   and
                        lispxhistory.


------------------------------------------------------------------------
68
    If y is  given, the event address  is the list difference  between x
    and  y,  e.g., x=(FOO FIE AND \ -1), y=(AND \ -1)  is  equivalent to
    x=(FOO FIE), y=NIL.

69
    changeslice has a  third argument used by  the system for  undoing a
    changeslice.




                                 22.43



Note: the effect of increasing a time-slice is gradual: the history list
is simply allowed to grow to the corresponding length before  any events
are  forgotten.   Decreasing  a  time-slice  will  immediately  remove a
sufficient number of the older events to bring the history list  down to
the proper size.  However, changeslice is undoable, so that these events
are (temporarily) recoverable.   Thus if the  user wants to  recover the
storage associated with these  events without waiting n more  events for
the changeslice event to be forgotten, he must perform a FORGET command.


saveset[name;value;topflg;flg]
                        an  undoable  set.  (see  page  22.35).  saveset
                        scans  the pushdown  list looking  for  the last
                        binding of name, sets name to value, and returns
                        value.

                        If the binding changed was a top  level binding,
                        name  is added  to spellings3  (see Section 17).
                        Furthermore, if  the old  value was  not NOBIND,
                        and was also not equal to the new value, saveset
                        calls the file  package to update  the necessary
                        file records.  Then,  if dfnflg is not  equal to
                        T,  saveset prints  (name RESET), and  saves the
                        old value  on the property  list of  name, under
                        the  property  VALUE.   If  flg=NOPRINT, saveset
                        saves  the old  value,  but does  not  print the
                        message.  This option is used by unset.

                        If  topflg=T, saveset  operates as  above except
                        that it does not scan the pushdown list but goes
                        right to name's value cell, e.g.,  rpaqq[x;y] is
                        simply  saveset[x;y;T].  When  topflg is  T, and
                        dfnflg  is ALLPROP  and  the old  value  was not
                        NOBIND,  saveset  simply  stores  value  on  the
                        property list of name under the  property VALUE,
                        and  returns  value.  This  option  is  used for
                        loading  files  without  disturbing  the current
                        value of variables (see Section 14).

                        If  flg=NOSAVE, saveset  does not  save  the old
                        value on the property list, nor does it add name
                        to spellings3.  However, the call to  saveset is
                        still undoable.  This option is used by /set.


unset[name]             if name does not contain a property VALUE, unset
                        generates  an   error.  Otherwise   unset  calls
                        saveset with name, the property value, topflg=T,
                        and flg=NOPRINT.













                                 22.44


                  70
undosave[undoform]      if lispxhist is not NIL (see discussion  on page
                        22.37), and get[lispxhist;SIDE] is not  equal to
                        NOSAVE, undosave adds  undoform to the  value of
                        the property SIDE on lispxhist, creating  a SIDE
                        property  if one  does not  already  exist.  The
                                                               71
                        form   of  undoform   is   (fn . args),    i.e.,
                        undoform     is     undone     by     performing
                        apply[car[undoform];cdr[undoform]].
                        For example, if  the definition  of FOO  is def,
                        /putd[FOO;newdef]  will  cause  a  call undosave
                        with undoform =(/PUTD FOO def).

                        car  of  the  SIDE  property  is  the  number of
                        "undosaves",  i.e., length  of cdr  of  the SIDE
                        property, which is the list of  undoforms.  Each
                        call to undosave increments this count, and adds
                        undoform to  the front of  the list,  i.e., just
                        after  the count.   When the  count  reaches the
                                                             72
                        value  of #undosaves  (initially 50),   undosave
                        prints a message asking the user if he  wants to
                        continue  saving.   If the  user  answers  NO or
                        defaults, undosave makes NOSAVE be the  value of
                        the  property SIDE,  which disables  any further
                        saving for this event.  If the user answers YES,
                        undosave changes the count to -1, which  is then
                                                                73
                        never incremented, and continues saving.


/rplnode[x;a;d]         Undoably  performs rplaca[x;a]  and rplacd[x;d].
                        Value is x.  Generates an error, ILLEGAL ARG, if
                        x  is not  a list.   The principle  advantage of
                        /rplnode  is that  when  x is  a  list, /rplnode
                        saves      its      undo      information     as
                        cons[x;cons[car[x];cdr[x]]],               i.e.,
                        (x originalcar . originalcdr),   and   therefore
                        requires only 3 cells of storage, instead of the



------------------------------------------------------------------------
70
    Undosave has  a second  optional argument,  histentry, which  can be
    used to specify lispxhist directly,  saving the time to look  it up.
    If both histentry and lispxhist are NIL, undosave is a NOP.

71
    Except for /rplnode, as described below.

72
    #undosaves=NIL is equivalent to #undosaves=infinity.

73
    load initializes the count on SIDE to -1, so that regardless  of the
    value of #undosaves, no message  will be printed, and the  load will
    be undoable.




                                 22.45



                        8 that  would be  required for  a /rplaca  and a
                        /rplacd   that   saved   their   information  as
                                          74
                        described earlier.

                        /rplnode has a BLKLIBRARYDEF.


/rplnode2[x;y]          same as /rplnode[x;car[y];cdr[y]].


Note:  for  consistency,  there are  definitions  for  both  rplnode and
rplnode2, although  there primary reason  for existence is  the undoable
versions.


new/fn[fn]              After the user has defined /fn,  new/fn performs
                        the necessary housekeeping operations to make fn
                        be undoable.

For example, the user could define /radix as
(LAMBDA (X) (UNDOSAVE (LIST (QUOTE /RADIX) (RADIX X))) and  then perform
new/fn[radix], and  radix would  then be  undoable when  typed in  or in
testmode.


lispx/[x;fn;vars]       performs  the  substitution of  /  functions for
                        destructive functions.  If fn is not NIL,  it is
                        the name  of a function,  and x is  its argument
                        list.   If fn  is  NIL, x  is a  form.   In both
                        cases,  lispx/  returns x  with  the appropriate
                        substitutions.    Vars  is   a  list   of  bound
                        variables (optional).

                        lispx/ incorporates information about the syntax
                        and  semantics  of  INTERLISP  expressions.  For
                        example,  it does  not bother  to  make undoable
                        operations involving  variables bound in  x.  It
                        does   not   perform   substitution   inside  of
                        expressions car of  which is NLAMBDA,  i.e., has
                        argtype 1 or 3  (unless car of the form  has the
                        property  INFO  value  EVAL,  as   described  in
                        Section 20).  For example, (BREAK PUTD) typed to
                        lispx,   will   break   on   putd,   not  /putd.
                        Similarly, substitution  should be  performed in
                        the  arguments  for functions  like  mapc, rptq,
                        etc., since these contain expressions  that will
                        be evaluated  or applied.   For example,  if the
                        user types mapc[(FOO1 FOO2 FOO3);PUTD]  the putd
                        must be replaced by /putd.


------------------------------------------------------------------------
74
    Actually, /rplaca and /rplacd also use this format for  saving their
    undo information when their  first arguments are lists.  However, if
    both a  /rplaca and /rplacd  are to be  performed, it is  still more
    efficient to use /rplnode (3 cells versus 6 cells).




                                 22.46



undolispx[line]         line  is an  event specification.   undolispx is
                        the  function  that  executes  UNDO  commands by
                        calling undolispx1 on the appropriate entry(s).

undolispx1[event;flg]   undoes one  event.  The  value of  undolispx1 is
                        NIL if  there is nothing  to be undone.   If the
                        event  is  already  undone,   undolispx1  prints
                                                          75
                        ALREADY UNDONE   and  returns   T.    Otherwise,
                        undolispx1 undoes  the event, prints  a message,
                        e.g., SETQ UNDONE, and returns T.

Undoing an event  consists of mapping down  (cdr of) the  property value
for SIDE, and  for each element, applying  car to cdr, and  then marking
the event undone by attaching  (with /attach) a NIL to the front  of its
SIDE  property.  Note  that  the undoing  of  each element  on  the SIDE
property  will  usually  cause  undosaves to  be  added  to  the current
lispxhist, thereby enabling the effects of undolispx1 to be undone.


undonlsetq[form]        is  an  nlambda  function  similar   to  nlsetq.
                        undonlsetq  evaluates  form,  and  if  no  error
                        occurs    during    the    evaluation,   returns
                        list[eval[form]] and passes the undo information
                                                    76
                        from form (if  any) upwards.   If an  error does
                        occur, the value  of undonlsetq is NIL,  and any
                        changes   made   by  /   functions   during  the
                        evaluation of form are undone.

                        undonlsetq compiles open.

                        undonlsetq will operate even if  lispxhistory or
                        lispxhist are  NIL, or if  #undosaves is  or has
                        been exceeded for this event.


resetundo[x;stopflg]    For  use in  conjunction with  resetlst (Section



------------------------------------------------------------------------
75
    If flg=T  and the event  is already undone,  or is an  undo command,
    undolispx1  takes no  action and  returns NIL.  Undolispx  uses this
    option to  search for the  last event to  undo. Thus  when line=NIL,
    undolispx simply searches history until it finds an event  for which
    undolispx1     returns      T,     i.e.,      undolispx     performs
    (SOME (CDAR LISPXHISTORY) (F/L (UNDOLISPX1 X T)))

76
    Actually, undonlsetq  does not  rebind lispxhist,  so that  any undo
    information  is stored  directly on  the history  event,  exactly as
    though there  were no undonlsetq.  Instead, undonlsetq  simply marks
    the state  of the  undo information when  it starts,  so that  if an
    error occurs, it can then know how much to undo. The purpose of this
    is so that if the user control-D's out of the undonlsetq,  the event
    is still undoable.




                                 22.47



                        5).  resetundo[] initializes the saving  of undo
                        information and returns a value which when given
                        back  to resetundo  undoes the  intervening side
                        effects.

For example,  (RESETLST (RESETSAVE (RESETUNDO))  . forms) will  undo the
side  effects of  forms  on normal  exit, or  if  an error  occurs  or a
control-D is typed.  Note that (UNDOLSETQ form) could be written as:

    (RESETLST (RESETSAVE  (RESETUNDO)
                          (AND (EQ RESETSTATE 'ERROR)
                               (RESETUNDO OLDVALUE)))
              . forms)

                        If stopflg=T, resetundo stops  accumulating undo
                                                      77
                        information it is saving on x.

For example,

    (RESETLST (SETQ FOO (RESETUNDO))
              (RESETSAVE NIL (LIST 'RESETUNDO FOO))
              (ADVISE --)
              (RESETUNDO FOO T)
              . forms)

would cause the advice to be undone, but not any of the side  effects in
forms.


printhistory[history;line;skipfn;novalues]
                        line  is an  event  specification.  printhistory
                        prints the events on history specified  by line,
                        e.g.,  (-1 THRU -10).  skipfn  is  an (optional)
                        functional  argument  that  is  applied  to each
                        event before  printing.  If  its value  is true,
                        the  event is  skipped, i.e.,  not  printed.  If
                        novalues=T,   or   novalues   applied   to   the
                        corresponding event  is true,  the value  is not
                                78
                        printed.

For example, the following lispxmacro  will define ??' as a  command for
printing  the history  list while  skipping all  "large events"  and not
printing any values.

              (??' (PRINTHISTORY LISPXHISTORY LISPXLINE
                       (FUNCTION (LAMBDA (X)
                          (IGREATERP (COUNT (CAR X)) 5)))


------------------------------------------------------------------------
77
    Note that this has no  bearing on the saving of undo  information on
    higher resetundo's, or on being able to undo the entire event.

78
    For example, novalues is T when printing events on edithistory.




                                 22.48



                        T))


22.10 The Editor and the Assistant

As  mentioned earlier,  all of  the remarks  concerning  "the assistant"
apply  equally  well to  user  interactions with  evalqt,  break  or the
editor.  The  differences between the  editor's implementation  of these
features  and  that  of lispx  are  mostly  obvious  or inconsequential.
However,  for   completeness,  this   section  discusses   the  editor's
implementation of the programmer's assistant.

The editor uses promptchar to print its prompt character, and lispxread,
lispxreadp, and readline for obtaining inputs.  When the editor is given
an input, it calls historysave to record the input in a new event on its
                           79
history list,  edithistory.   Edithistory  follows the  same conventions
and format as lispxhistory.  However, since edit commands have no value,
the editor  uses the value  field for saving  side effects,  rather than
storing them under the property SIDE.

The editor  processes DO, !E, !F,  and !N  commands itself,  since lispx
does  not  recognize these  commands.   The editor  also  processes UNDO
                                                          80
itself, as  described below.   All other  history commands    are simply
given  to   lispx  for  execution,   after  first   binding  (resetting)
lispxhistory to edithistory.  The editor also calls lispx when  given an
                                    81
E command as described in Section 9.

The major implementation difference between the editor and  lispx occurs



------------------------------------------------------------------------
79
    Except that the  atomic commands OK, STOP, SAVE, P, ?, PP and  E are
    not recorded. In addition, number commands are grouped together in a
    single event. For example,  3 3 -1 is considered as one  command for
    changing position.

80
    as  indicated by  their  appearance on  historycoms, a  list  of the
    history  commands.   editdefault  interrogates   historycoms  before
    attempting spelling correction. (All of the commands  on historycoms
    are also on editcomsa and editcomsl so that they can be corrected if
    misspelled in the editor.) Thus if the user defines a lispxmacro and
    wishes it to operate in the editor as well, he need simply add it to
    historycoms. For  example, RETRIEVE is  implemented as  a lispxmacro
    and works equally well in lispx and the editor.

81
    In this case, the editor uses the fifth argument to lispx, lispxflg,
    to  specify  that any  history  commands  are to  be  executed  by a
    recursive call to lispx,  rather than by unreading. For  example, if
    the user  types E REDO  in the editor,  he wants  the last  event on
    lispxhistory  processed as  lispx input,  and not  to be  unread and
    processed by the editor.




                                 22.49



in undoing.  Edithistory is a list of only the last n commands,  where n
is the value of the time-slice.  However the editor provides for undoing
all  changes made  in a  single editing  session, even  if  that session
consisted of  more than  n edit commands.   Therefore, the  editor saves
undo  information  independently  of  the  edithistory  on  a  list call
undolst, (although it also stores each entry on undolst in the  field of
the corresponding event on edithistory.) Thus, the commands UNDO, !UNDO,
                                                82
and UNBLOCK,  are not dependent  on edithistory,   i.e.,  UNDO specifies
undoing  the last  command  on undolst,  even  if that  event  no longer
appears  on  edithistory.  The  only  interaction between  UNDO  and the
history  list occurs  when  the user  types  UNDO followed  by  an event
specification.  In  this case,  the editor calls  lispxfind to  find the
event, and  then undoes  the corresponding entry  on undolst.   Thus the
user  can  only  undo  a  specified  command  within  the  scope  of the
edithistory.   (Note  that  this  is also  the  only  way  UNDO commands
themselves can  be undone,  that is,  by using  the history  feature, to
specify the corresponding event, e.g., UNDO UNDO.)

The implementation  of the actual  undoing is similar  to the way  it is
done in lispx: each command  that makes a change in the  structure being
edited does  so via a  function that records  the change on  a variable.
After the command  has completed, this variable  contains a list  of all
the  pointers  that  have  been  changed  and  their  original contents.
Undoing  that  command  simply  involves  mapping  down  that  list  and
restoring the pointers.


22.11 Statistics

The programmer's assistant keeps various statistics about  system usage,
e.g., number of lispx inputs,  number of undo saves, number of  calls to
editor,  number of  edit commands,  number of  p.a. commands,  cpu time,
console time, et al.  These can be viewed via the function lispxstats.


lispxstats[]            prints statistics.

The user  can add  his own statistics  to the  lispx statistics  via the
function addstats.


addstats[statlst]       no  spread,  nlambda.   Statlst  is  a  list  of
                        elements of the form (statistic-name . message),
                        e.g.,  (EDITCALLS  CALLS  TO  EDITOR) (UNDOSTATS
                        CHANGES UNDONE), etc.  statistic-name is  set to
                        the  cell  in   an  unboxed  array,   where  the
                        corresponding  statistic will  be  stored.  This
                        statistic can then be incremented by lispxwatch.





------------------------------------------------------------------------
82
    and in fact will work if edithistory=NIL, or even in a  system which
    does not contain lispx at all.




                                 22.50



lispxwatch[stat;n]      increments   stat  by   n  (or   1   if  n=NIL).
                        lispxwatch has a BLKLIBRARYDEF.

The  user can  save his  statistics  for loading  into a  new  system by
performing MAKEFILE(DUMPSTATS).  After the file DUMPSTATS is loaded, the
statistics printed by lispxstats will be the same as those that would be
printed following the makefile.


22.12 Greeting and User Initialization

Many of the features of INTERLISP are parameterized to allow the user to
adjust the  system to his  or her own  tastes.  Among the  more commonly
adjusted  parameters  are  prompt#flg,  dwimwait,  changeslice,  #rpars,
lowercase, archivefn,  #undosaves, fltfmt, etc.   In addition,  the user
can  modify the  action  of system  functions in  ways  not specifically
provided for by using advise (Section 19).

In order  to encourage this  procedure, and to  make it as  painless and
automatic as possible, the  p.a. includes a facility for  a user-defined
profile.  When INTERLISP is first  run, it obtains and evaluates  a list
                                                             83
of  user-specified expressions  for initializing  the system,    and the
p.a.  prints  a  greeting, e.g.,  "HELLO, WARREN."  or  "GOOD AFTERNOON,
DANNY.", etc.

Greeting  (i.e., the  initialization) is  undoable, and  is stored  as a
separate  event on  the history  list.  The  user can  also specifically
invoke the greeting operation via the function greet, for example, if he
wishes to effect another user's initialization.


greet[name;flg]         performs  greeting  for user  whose  username is
                        name,  or  if  name=NIL,  for  login  name  (see
                        username and usernumber, Section 21), i.e., when
                        INTERLISP first starts up, it  performs greet[].
                        Before    greet    performs     the    indicated
                        initialization, it  first undoes the  effects of




------------------------------------------------------------------------
83
    In INTERLISP-10,  a specially formatted  file on the  LISP directory
    contains the  initializations for  all users.  This file  is indexed
    into using the user's usernumber as a key. The expressions  (if any)
    found  there  are  then   evaluated.  The  user  can   also  specify
    initializations  by maintaining  a file  INIT.LISP on  hs directory.
    See discussion below concerning implementation.












                                 22.51



                                              84 85
                        the previous greeting.      If flg=T, greet also
                        resets the  counters for the  various statistics
                        reported by lispxstats (page 22.50).

greet also sets the variable username to the name for which the greeting
was performed.   Sysin is advised  to compare username  with username[].
If  they  are  the  same, sysin  prints  heraldstring,  followed  by the
                                                                      86
greeting message. Ohterwise, sysin prints a message alerting the user.
For example, if user HARTLEY performs  a sysin of a sysout made  by user
GOODWIN, the following message is printed:

              ****ATTENTION USER HARTLEY:
              THIS SYSOUT IS INITIALIZED FOR USER GOODWIN.
              TO REINITIALIZE, TYPE GREET()


INTERLISP-10 Implementation of Greeting

greet operates off the  file <LISP>USERNAMEFILE.  To change  an existing
initialization, or create  a new one,  a new <LISP>USERNAMEFILE  must be
written.   This is  accomplished  by loading  the  file <LISP>USERNAMES,
editing  usernamelst, and  then performing  makeusernames[],  which will
create new versions of both USERNAMEFILE and USERNAMES.  (Note  that the
person performing this operation  must therefore either be  connected to
the LISP directory, or have write access to it.)

usernamelst    is     a    list    of     elements    of     the    form
(username firstname T . forms), e.g.,  (TEITELMAN WARREN  T (CHANGESLICE
100)  (SETQ DWIMWAIT  5)).  cadr  of the  list is  used in  the greeting
message.  cdddr is a list of forms that are evaluated.

usernamelst can be  edited just like any  other list, e.g.,  with editv.
The file  USERNAMEFILE, created  by makeusernames,  contains usernamelst
along with an  index block which contains  for each user  on usernamelst
the address in the file (i.e., byte position) of the start of his entry.
greet then simply does an sfptr and a read.




------------------------------------------------------------------------
84
    The side effects  of the greeting operation  are stored on  a global
    variable as  well as  the history list,  thus enabling  the previous
    greeting to be undone even if it is no longer on the history list.

85
    In addition, makesys is advised to undo the effects of  the previous
    greeting, thereby returning the system to a pristine state.

86
    sysout first checks the  value of the variable  sysoutgag, initially
    NIL. If sysoutgag is T, no message is printed following a sysin.  If
    sysoutgag is  a list,  it will be  evaluated in  lieu of  printing a
    message. For example, the user can use this option to print  his own
    message.




                                 22.52



If usernamelst contains an element for which the username is  NIL, i.e.,
an element of the form  (NIL . forms), this is interpreted to  mean that
forms are evaluated regardless of  user name.  This feature can  be used
to "patch" an INTERLISP  system when a bug  is found, or to  change some
default for INTERLISP at a particular site, e.g., turn off DWIM, perform
lowercase[T],  etc.   Individual  user  initialization  will   still  be
performed following this system initialization.

If a  user wishes  to be  able to  change or  add to  his initialization
frequently without the  necessity of making  a new USERNAMEFILE,  he can
accomplish  this  by  putting  initializing forms  on  to  a  file named
INIT.LISP  in his  login directory.   greet checks  to see  if  the file
INIT.LISP  exists,  and  if  so,  reads  and  evaluates  (undoably)  the
expressions in the file.
















































                                 22.53