Trailing-Edge
-
PDP-10 Archives
-
decus_20tap1_198111
-
decus/20-0004/a1lisp.doc
There are no other files named a1lisp.doc in the archive.
APPENDICES
Appendix 1
Transor
Introduction
transor is a LISP-to-LISP translator intended to help the user who has a
program coded in one dialect of LISP and wishes to carry it over to another.
The user loads transor along with a file of transformations. These
transformations describe the differences between the two LISPs, expressed in
terms of INTERLISP editor commands needed to convert the old to new, i.e. to
edit forms written in the source dialect to make them suitable for the target
dialect. transor then sweeps through the user's program and applies the edit
transformations, producing an object file for the target system. In addition,
transor produces a file of translation notes, which catalogs the major changes
made in the code as well as the forms that require further attention by the
user. Operationally, therefore, transor is a facility for conducting massive
edits, and may be used for any purpose which that may suggest.
Since the edit transformations are fundamental to this process, let us begin
with a definition and some examples. A transformation is a list of edit
commands associated with a literal atom, usually a function name. transor
conducts a sweep through the user's code, until it finds a form whose car is a
literal atom which has a transformation. The sweep then pauses to let the
editor execute the list of commands before going on. For example, suppose the
order of arguments for the function tconc must be reversed for the target
system. The transformation for tconc would then be: ((SW 2 3)). When the
sweep encounters the form (TCONC X (FOO)), this transformation would be
retrieved and executed, converting the expression to (TCONC (FOO) X). Then the
sweep would locate the next form, in this case (FOO), and any transformations
for foo would be executed, etc.
Most instances of tconc would be successfully translated by this
transformation. However, if there were no second argument to tconc, e.g. the
form to be translated was (TCONC X), the command (SW 2 3) would cause an error,
which transor would catch. The sweep would go on as before, but a note would
appear in the translation listing stating that the transformation for this
particular form failed to work. The user would then have to compare the form
and the commands, to figure out what caused the problem. One might, however,
anticipate this difficulty with a more sophisticated transformation:
((IF (## 3) ((SW 2 3)) ((-2 NIL)))), which tests for a third element and does
(SW 2 3) or (-2 NIL) as appropriate. It should be obvious that the translation
process is no more sophisticated than the transformations used.
A1.1
This documentation is divided into two main parts. The first describes how to
use transor assuming that the user already has a complete set of
transformations. The second documents transorset, an interactive routine
for building up such sets. transorset contains commands for writing and
editing transformations, saving one's work on a file, testing transformations
by translating sample forms, etc.
Two transformations files presently exist for translating programs into
INTERLISP. <LISP>SDS940.XFORMS is for old BBN LISP (SDS 940) programs, and
<LISP>LISP16.XFORMS is for Stanford AI LISP 1.6 programs. A set for LISP 1.5
is planned.
Using Transor
The first and most exasperating problem in carrying a program from one
implementation to another is simply to get it to read in. For example, SRI
LISP uses / exactly as INTERLISP uses %, i.e. as an escape character. The
function prescan exists to help with these problems: the user uses prescan
to perform an initial scan to dispose of these difficulties, rather than
attempting to transor the foreign sourcefiles directly.
prescan copies a file, performing character-for-character substitutions. It
is hand-coded and is much faster than either readc's or text-editors.
prescan[file;charlst] Makes a new version of file, performing
substitutions according to charlst. Each element
of charlst must be a dot-pair of two character
codes, (OLD . NEW).
For example, SRI files are prescan'ed with charlst = ((37 . 47) (47 . 37)),
which exchanges slash (47) and percent-sign (37).
The user should also make sure that the treatment of doublequotes by the source
and target systems is similar. In INTERLISP, an unmatched double-quote (unless
protected by the escape character) will cause the rest of the file to read in
as a string.
Finally, the lack of a STOP at the end of a file is harmless, since transor
will suppress END OF FILE errors and exit normally.
Translating
A1.2
transor is the top-level function of the translator itself, and takes one
argument, a file to be translated. The file is assumed to contain a sequence
of forms, which are read in, translated, and output to a file called file.TRAN.
The translation notes are meanwhile output to file.LSTRAN. Thus the usual
sequence for bring a foreign file to INTERLISP is as follows: prescan the
file; examine code and transformations, making changes to the transformations
if needed; transor the file; and clean up remaining problems, guided by the
notes. The user can now make a pretty file and proceed to exercise and check
out his program. To export a file, it is usually best to transor it, then
prescan it, and perform clean-up on the foreign system where the file can be
loaded.
transor[sourcefile] Translates sourcefile. Prettyprints translation
on file.TRAN: translation listing on file.LSTRAN.
transorform[form] Argument is a LISP form. Returns the
(destructively) translated form. The translation
listing is dumped to the primary output file.
transorfns[fnlst] Argument is a list of function names whose
interpreted definitions are destructively
translated. Listing to primary output file.
transform and transorfns can be used to translate expressions that are
already in core, whereas transor itself only works on files.
The Translation Notes
The translation notes are a catalog of changes made in the user's code, and
of problems which require, or may require, further attention from the user.
This catalog consists of two cross-indexed sections: an index of forms and an
index of notes. The first tabulates all the notes applicable to any form,
whereas the second tabulates all the forms to which any one note applies.
Forms appear in the index of forms in the order in which they were encountered,
i.e. the order in which they appear on the source and output files. The index
of notes shows the name of each note, the entry numbers where it was used, and
its text, and is alphabetical by name. The following sample was made by
translating a small test file written in SRI LISP.
A1.3
LISTING FROM TRANSORING OF FILE TESTFILE.;5
DONE ON 1-NOV-71 20:10:47
INDEX OF FORMS
1. APPLY/EVAL at
[DEFINEQ
(FSET (LAMBDA &
(PROG ...3...
(SETQ Z (COND
((ATOM (SETQ --))
(COND
((ATOM (SETQ Y (NLSETQ "(EVAL W)")))
--)
--))
--))
-- ]
2. APPLY/EVAL at
[DEFINEQ
(FSET (LAMBDA &
(PROG ...3...
(SETQ Z (COND
((ATOM (SETQ --))
(COND
((ATOM (SETQ --))
"(EVAL (NCONS W))")
--))
--))
-- ]
3. MACHINE-CODE at
[DEFINEQ
(LESS1 (LAMBDA &
(PROG ...3...
(COND
...2...
((NOT (EQUAL (SETQ X2 "(OPENR (MAKNUM & -))"
)
--))
--))
-- ]
4. MACHINE-CODE at
[DEFINEQ
(LESS1 (LAMBDA &
(PROG ...3...
(COND
...2...
((NOT (EQUAL & (SETQ Y2
"(OPENR (MAKNUM & --))")))
--))
-- ]
INDEX OF NOTES
APPLY/EVAL at 1, 2.
TRANSOR will translate the arguments of the APPLY or EVAL expression, but
the user must make sure that the run-time evaluation of the arguments returns a
BBN-compatible expression.
MACHINE-CODE at 3, 4.
Expression dependent on machine-code. User must recode.
A1.4
The translation notes are generated by the transformations used, and therefore
reflect the judgment of their author as to what should be included.
Straightforward conversions are usually made without comment; for example, the
DEFPROP's in this file were quietly changed to DEFINEQ's. transor found four
noteworthy forms on the file, and printed an entry for each in the index of
forms, consisting of an entry number, the name of the note, and a printout
showing the precise location of the form. The form appears in double-quotes
and is the last thing printed, except for closing parentheses and dashes. An
ampersand represents one non-atomic element not shown, and two or more elements
not shown are represented as ...n..., where n is the number of elements. Note
that the printouts describe expressions on the output file rather than the
source file; in the example, the DEFPROP's of SRI LISP have been replaced with
DEFINEQ's.
Errors and Messages
transor records its progress through the source file by teletype printouts
which identify each expression as it is read in. Progress within large
expressions, such as a long DEFINEQ, is reported every three minutes by a
printout showing the location of the sweep.
If a transformation fails, transor prints a diagnostic to the teletype which
identifies the faulty transformation, and resumes the sweep with the next form.
The translation notes will identify the form which caused this failure, and the
extent to which the form and its arguments were compromised by the error.
If the transformation for a common function fails repeatedly, the user can type
control-H. When the system goes into a break, he can use transorset to repair
the transformation, and even test it out (see TEST command, page A1.7). He
may then continue the main translation with OK.
Transorset
To use transorset, type transorset() to INTERLISP. transorset will
respond with a + sign, its prompt character, and await input. The user is now
in an executive loop which is like evalqt with some extra context and
capabilities intended to facilitate the writing of transformations.
transorset will thus progress apply and eval input, and execute history
commands just as evalqt would. Edit commands, however, are interpreted as
additions to the transformation on which the user is currently working.
transorset always saves on a variable named currentfn the name of the
last function whose transformation was altered or examined by the user.
currentfn thus represents the function whose transformation is currently being
worked on. Whenever edit commands are typed to the + sign, transorset will
add them to the transformation for currentfn. This is the basic mechanism for
writing a transformation. In addition, transorset contains commands for
printing out a transformation, editing a transformation, etc., which all assume
that the command applies to currentfn if no function is specified. The
following example illustrates this process.
A1.5
_TRANSORSET()
+FN TCONC [1]
TCONC
+(SW 2 3) [2]
+TEST (TCONC A B) [3]
P
(TCONC B A)
+TEST (TCONC X) [4]
TRANSLATION ERROR: FAULTY TRANSFORMATION
TRANSFORMATION: ((SW 2 3)) [5]
OBJECT FORM: (TCONC X)
1. TRANSFORMATION ERROR AT [6]
"(TCONC X)"
(TCONC X)
+(IF (## 3) ((SW 2 3)) ((-2 NIL] [7]
+SHOW
TCONC
[(SW 2 3)
(IF (## 3) [8]
((SW 2 3))
((-2 NIL]
TCONC
+ERASE [9]
TCONC
+REDO IF [10]
+SHOW
TCONC
[(IF (## 3)
((SW 2 3))
((-2 NIL]
TCONC
+TDST
=TEST [11]
(TCONC NIL X)
+
A1.6
In this example, the user begins by using the FN command to set currentfn to
TCONC [1]. He then adds to the (empty) transformation for tconc a command to
switch the order of the arguments [2] and tests the transformation [3]. His
second TEST [4] fails, causing an error diagnostic [5] and a translation note
[6]. He writes a better command [7] but forgets that the original SW command
is still in the way [8]. He therefore deletes the entire transformation [9]
and redoes the IF [10]. This time, the TEST works [11].
Transorset Commands
The following commands for manipulating transformations are all lispxmacros
which treat the rest of their input line as arguments. All are undoable.
FN Resets currentfn to its argument, and
returns the new value. In effect FN says you are
done with the old function (as least for the
moment) and wish to work on another. If the new
function already has a transformation, the message
(OLD TRANSFORMATIONS) is printed, and any
editcommands typed in will be added to the end of
the existing commands. FN followed by a carriage
return will return the value of currentfn without
changing it.
SHOW Command to prettyprint a transformation. SHOW
followed by a carriage return will show the
transformation for currentfn, and return
currentfn as its value. SHOW followed by one or
more function names will show each one in turn,
reset currentfn to the last one, and return the
new value of currentfn.
EDIT Command to edit a transformation. Similar to SHOW
except that instead of prettyprinting the
transformation, EDIT gives it to edite. The user
can then work on the transformation until he
leaves the editor with OK.
ERASE Command to delete a transformation.
Otherwise similar to SHOW.
TEST Command for checking out transformations. TEST
takes one argument, a form for translation. The
translation notes, if any, are printed to the
teletype, but in an abbreviated format which omits
the index of notes. The value returned is the
A1.7
translated form. TEST saves a copy of its
argument on the free variable testform, and if no
argument is given, it uses testform, i.e. tries
the previous test again.
DUMP Command to save your work on a file. DUMP takes
one argument, a filename. The argument is saved
on the variable dumpfile, so that if no argument
is provided, a new version of the previous file
will be created.
The DUMP command creates files by makefile. Normally fileFNS will be unbound,
but the user may set it himself; functions called from a transformation by the
E command may be saved in this way. DUMP makes sure that the necessary command
is included on the fileVARS to save the user's transformations. The user may
add anything else to his fileVARS that he wishes. When a transformation file
is loaded, all previous transformations are erased unless the variable merge
is set to T.
EXIT transorset returns NIL.
The REMARK Feature
The translation notes are generated by those transformations that are actually
executed via an editmacro called REMARK. REMARK takes one argument, the
name of a note. When the macro is executed, it saves the appropriate
information for the translation notes, and adds one entry to the index of
forms. The location that is printed in the index of forms is the editor's
location when the REMARK macro is executed.
To write a transformation which makes a new note, one must therefore do two
things: define the note, i.e. choose a new name and associate it with the
desired text; and call the new note with the REMARK macro, i.e. insert the edit
command (REMARK name) in some transformation. The NOTE command, described
below, is used to define a new note. The call to the note may be added to a
transformation like any other edit command. Once a note is defined, it may be
called from as many different transformations as desired.
The user can also specify a remark with a new text, without bothering to think
of a name and perform a separate defining operation, by calling REMARK with
more than one argument, e.g. (REMARK text-of-remark). This is interpreted to
mean that the arguments are the text. transorset notices all such expressions
as they are typed in, and handles naming automatically; a new name is
A1.8
1
generated and defined with the text provided, and the expression itself is
edited to be (REMARK generated-name). The following example illustrates the
use of REMARK.
**FOOTNOTES**
1
The name generated is the value of currentfn suffixed with a colon, or
with a number and a colon.
A1.9
_TRANSORSET()
+NOTE GREATERP/LESSP (BBN'S GREATERP AND LESSP ONLY [1]
TAKE TWO ARGUMENTS, WHEREAS SRI'S FUNCTIONS TAKE AN
INDEFINITE NUMBER. AT THE PLACES NOTED HERE, THE SRI CODE
USED MORE THAN TWO ARGUMENTS, AND THE USER MUST RECODE.]
GREATERP/LESSP
+FN GREATERP
GREATERP
+(IF (IGREATERP (LENGTH (##))3) NIL ((REMARK GREATERP/LESSP] [2]
+FN LESSP
LESSP
+REDO IF [3]
+SHOW
LESSP
[(IF (IGREATERP (LENGTH (##))
3)
NIL
((REMARK GREATERP/LESSP]
LESSP
+FN ASCII
(OLD TRANSFORMATIONS)
ASCII
+(REMARK ALTHOUGH THE SRI FUNCTION ASCII IS IDENTICAL [4]
TO THE BBN FUNCTION CHARACTER, THE USER MUST MAKE SURE THAT
THE CHARACTER BEING CREATED SERVES THE SAME PURPOSE ON BOTH
SYSTEMS, SINCE THE CONTROL CHARACTERS ARE ALL ASSIGNED
DIFFRENTLY.]
+SHOW [5]
ASCII
((1 CHARACTER)
(REMARK ASCII:))
ASCII
+NOTE ASCII: [6]
EDIT
*NTH -2
*P
... ASSIGNED DIFFRENTLY.)
*(2 DIFFERENTLY.)
OK
ASCII:
+
A1.10
In this example, the user defines a note named GREATERP/LESSP by using the NOTE
command [1], and writes transformations which call this note whenever the sweep
encounters a GREATERP or LESSP with more than two arguments [2-3]. Next, the
implicit naming feature is used [4] to add a REMARK command to the
transformation for ASCII, which has already been partly written. The user
realizes he mistyped part of the text, so he uses the SHOW command to find the
name chosen for the note [5]. Then he uses the NOTE command on this name,
ASCII:, to edit the note [6].
NOTE First argument is note name and must be a literal
atom. If already defined, NOTE edits the old
text; otherwise it defines the name, reading the
text either from the rest of the input line or
from the next line. The text may be given as a
line or as a list. Value is name of note.
2
The text is actually stored. as a comment, i.e. a * and %% are added in front
when the note is first defined. The text will therefore be lower-cased the
first time the user DUMPs (see Section 14).
DELNOTE Deletes a note completely (although any calls
to it remain in the transformations).
Controlling the Sweep
transor's sweep searches in print-order until it finds a form for which a
transformation exists. The location is marked, and the transformation is
executed. The sweep then takes over again, beginning from the marked location,
no matter where the last command of the transformation left the editor. User
transformations can therefore move around freely to examine the context,
without worrying about confusing the translator. However, there are many cases
where the user wants his transformation to guide the sweep, usually in order to
direct the processing of special forms and FEXPR's. For example, the
transformation for QUOTE has only one objective: to tell the sweep to skip over
the argument to QUOTE, which is (presumably) not a LISP form. NLAM is an
editmacro to permit this.
NLAM An atomic editmacro which sets a flag which causes
the sweep to skip the arguments of the current
form when the sweep resumes.
**FOOTNOTES**
2
On the global list usernotes.
A1.11
Special forms such as cond, prog, selectq, etc., present a more difficult
problem. For example, (COND (A B)) is processed just like (FOO (A B)): i.e.
after the transformation for cond finishes, the sweep will locate the "next
form," (A B), retrieve the transformation for the function A, if any, and
execute it. Therefore, special forms must have transformations that preempt
the sweep and direct the translation themselves. The following two atomic
editmacros permit such transformations to process their forms, translating or
skipping over arbitrary subexpressions as desired.
DOTHIS Translates the editor's current expression,
treating it as a single form.
DOTHESE Translates the editor's current expression,
treating it as a list of forms.
3
For example, a transformation for setq might be (3 DOTHIS). This translates
the second argument to a setq without translating the first. For cond, one
might write (1 (LPQ NX DOTHESE)), which locates each clause of the COND in
turn, and translates it as a list of forms, instead of as a single form.
The user who is starting a completely new set of transformations must begin by
writing transformations for all the special forms. To assist him in this and
prevent oversights, the file <LISP>SPECIAL.XFORMS contains a set of
transformations for LISP special forms, as well as some other transformations
which should also be included. The user will probably have to revise these
transformations substantially, since they merely perform sweep control for
INTERLISP, i.e. they make no changes in the object code. They are provided
chiefly as a checklist and tutorial device, since these transformations are
both the first to be written and the most difficult, especially for users new
to the INTERLISP editor.
* * *
When the sweep mechanism encounters a form which is not a list, or a form car
of which is not an atom, it retrieves one of the following special
transformations.
NLISTPCOMS Global value is used as a transformation for
any form which is not a list.
**FOOTNOTES**
3
Recall that a transformation is a list of edit commands. In this case,
there are two commands, 3 and DOTHIS.
A1.12
For example, if the user wished to make sure that all strings were quoted, he
might set nlistpcoms to
((IF (STRINGP (##)) ((ORR ((_ QUOTE))((MBD QUOTE)))) NIL)).
LAMBDACOMS Global value is used as a transformation for
any form, car of which is not an atom.
These variables are initialized by <LISP>SPECIAL.XFORMS and are saved by the
DUMP command. nlistpcoms is initially NIL, making it a NOP. lambdacoms is
initialized to check first for open LAMBDA expressions, processing them without
translation notes unless the expression is badly formed. Any other forms with
a non-atomic car are simply treated as lists of forms and are always mentioned
in the translation notes. The user can change or add to this algorithm simply
by editing or resetting lambdacoms.
A1.13