This document describes the geomview 1.3 lisp interpreter. This version is incompatible with previous versions in several ways. Since the previous one was used mostly just by Geometry Center staff, I am not going to write a document detailing the changes. The geomview lisp interpreter is not very well documented in general because I am strongly considering phasing it out and replacing it with a real lisp interpreter in a future version of geomview. If you have any questions about the current version or how to convert programs from an older version please contact me directly [ mbp@geom.umn.edu ].
#include "lisp.h"
void LInit();
Lake * LakeDefine(FILE *streamin, FILE *streamout, void *river);
void LakeFree(Lake *lake);
LObject * LNew(LType *type, LCell *cell);
LObject * LRefIncr(LObject *obj);
void LRefDecr(LObject *obj);
void LWrite(FILE *fp, LObject *obj);
void LFree(LObject *obj);
LObject * LCopy(LObject *obj);
LObject * LSexpr(Lake *lake);
LObject * LEval(LObject *obj);
LObject * LEvalSexpr(Lake *lake);
LList * LListNew();
LList * LListAppend(LList *list, LObject *obj);
void LListFree(LList *list);
LList * LListCopy(LList *list);
LObject * LListEntry(LList *list, int n);
int LListLength(LList *list);
int LParseArgs(char *name, Lake *lake, LList *args, ...);
int LDefun(char *name, LObjectFunc func, char *help);
void LListWrite(FILE *fp, LList *list);
LInterest * LInterestList(char *funcname);
LObject * LEvalFunc(char *name, ...);
int LArgClassValid(LType *type);
void LHelpDef(char *key, char *message);
LDEFINE(name, ltype, doc)
LDECLARE((name, LBEGIN,
...,
LEND));
This manual page assumes that you are familiar with the syntax of lisp. The first part describes the basics of using the interpreter. Some gory details that don't concern most users then follow.
The main steps in using the lisp interpreter are
1. call Linit() to initialize the interpreter
2. make calls to LDefun(), one for each lisp function you want the
interpreter to know about
3. define the "i/o lake"
4. parse input with calls to LSexpr() and evaluate the resulting
lisp objects with LEval() (or use LEvalSexpr() to combine both steps).
For example the following code defines a single function "f" and executes commands from standard input:
#include "lisp.h"
Lake *lake;
LObject *obj, *val;
LInit();
LDefun("f", f, NULL);
lake = LakeDefine(stdin, stdout, NULL);
while (!feof(stdin)) {
obj = LSexpr(lake);
val = LEval(obj);
LFree(obj);
LFree(val);
}
The second argument to LDefun() is a function pointer; LDefun() sets up a correspondence between the string "f" and the function f, which is assumed to have been previously declared. The section FUNCTION DEFINITIONS below gives the expected call syntax and behavior of such functions. (The third argument to LDefun() is a pointer to a string which documents the function and may be NULL if you don't care about documentation.) LakeDefine() defines an i/o lake; this is a generalization of the notion of an i/o stream. Most programs don't need to use the generalization, though, and can simply pass FILE pointers as LakeDefine()'s first two arguments and NULL as the third one. The section LAKES below gives the details for those who are interested. LSexpr() [which stands for Lisp Symbolic EXPRession] parses a single lisp expression from the lake, returning a lisp object which represents that expression. The lisp object is returned as an LObject pointer, which points to an opaque structure containing a representation of the expression. LEval() then evaluates the object; it is during the call to LEval() that the action of the expression takes place. Note that the last two lines of code in this example could have been replaced by the single line LEval(LSexpr(lake)) or, more efficiently, by LEvalSexpr(lake).
LDEFINE(f, LSTRING, "(f a b) returns a string representing the
sum of the integer a with the floating point number b.")
{
int a;
float b;
char buf[20], *s;
LDECLARE(("f", LBEGIN,
LINT, &a,
LFLOAT, &b,
LEND));
sprintf(buf,"%f",a+b);
s = strdup(buf);
return LNew(LSTRING, &s);
}
The important things about this function are:
1. It is declared with the LDEFINE macro, the general syntax
of which is LDEFINE(name, type, helpstr). name should be
a valid C identifer and will be used to construct the actual
name of the C function by prepending an 'L' and the name of
the help string by prepending an 'H'. type should be a lisp
object type identifier (see below) and determines the type
of object that the function returns. helpstr is a documentation
string.
2. The use of the LDECLARE macro. More about this below.
3. It returns an LObject *. All lisp functions must actually
return a value. If you don't care what value they return
you can return one of the pre-defined values Lnil or Lt
(and specify LVOID as the type in the LDEFINE header).
This particular example is a function which takes two arguments, an int and a float, and returns a string object representing their sum. A lisp call to this function might look like "(f 1 3.4)".
The LDECLARE macro, defined in lisp.h, sets up the correspondence between variables in the C code and arguments in the lisp call to the function. Note that the arguments to LDECLARE are delimited by *two* pairs of parentheses (this is because C does not allow macros with a variable number of arguments; LDECLARE thus actually takes one argument which is a parenthesized list of an arbitrary number of items). The general usage of LDECLARE is
LDECLARE(( name, LBEGIN,
<argspec>,
...,
LEND ));
where name is the name of the function (as specified to LDefun()). <argspec> is an argument specification, which in general consists of a lisp type identifier followed by an address. The identifier indicates the data type of the argument. The builtin type identifiers are LINT (integer), LFLOAT (float), LSTRING (string), LLOBJECT (lisp object), and LLIST (lisp list). Applications may define additional types whose identifiers may also be used here; see the section CUSTOM LISP TYPES below for details. There may be any number of <argspec>'s; the last must be followed by the special keyword LEND.
LHOLD, LLIST, &list,
specifies an unevalutated list argument. This feature is really useful only for LLIST, LLOBJECT, and array types (see below) since the other types evalutate to themselves.
In general an <argspec> in the LDECLARE call consists of a keyword followed by the address of a scalar data type. Since it is relatively common to use a lisp list to represent an array of values, however, the special <argspec> keyword LARRAY is provided for dealing with them. It has a different syntax: it should be followed by a lisp type identifier which specifies the type of the elements of the array and then by two addresses --- the address of an array and the address of an integer count. Upon entry to LDECLARE the count specifies how many elements may be written into the array. LDECLARE then modifies this number to indicate the number of entries actually parsed. For example:
LDEFINE(myfunc, ...)
{
float f[2];
int fn = 2;
LDECLARE(("myfunc", LEBGIN
LHOLD, LARRAY, f, &fn,
LEND));
/* at this point the value of fn has been modified to
be the number of entries actually appearing in the
list argument; and this number of values have been
written into the array f. */
...
}
defines a function "myfunc" which takes a list of up to 2 floats as its only argument. Valid calls to "myfunc" would be "(myfunc ())", "(myfunc (7))", and "(myfunc (7 8))".
Note the use of LHOLD; this is necessary because otherwise the lisp system would attempt to evaluate the list as a function call before passing it off to myfunc.
Normally excess arguments also elicit an error message. The LREST keyword allows control over this situation. If LREST is followed by a pointer to an LList * variable, then trailing arguments are parsed, evaluated (unless LHOLD was used), and the list of them is stored in the given variable. (Note that the value is an LList, not an LObject of type LLIST -- if there are no excess arguments, the value is NULL, not an empty LLIST.) If LREST is followed by a NULL pointer, excess arguments are silently ignored. LREST might be useful when a function's argument types are not known. It's not necessary to specify LEND after LREST.
LNew(): creates a new lisp object of the given type with
the given value. The "type" argument is one of the
values LSTRING or LLIST, or a type pointer defining
a custom object type (see CUSTOM OBJECT TYPES
below).
LRefIncr(): increments the reference count of a lisp
object. The lisp interpreter uses the convention
that when a procedure returns a lisp object, the
caller owns the object and thus has responsibility
for freeing it. LRefIncr() can be used to increment
the reference count of an existing object about to
be returned. New objects created by LNew() have
their reference count initialized to 1 and hence do
not need to be LRefIncr()'ed.
LRefDecr(): decrements the reference count of a lisp
object. This should probably not be called by
application programs; it is used internally.
LWrite(): writes a formatted string representation of a
lisp object to a stream.
LFree(): free the space assoicated with a lisp object
LCopy(): construct a copy of a lisp object
In addition to the predefined lisp object types you may define your own custom types. This is done by constructing a structure containing various function pointers for manipulating objects of your new type. The address of this structure is then the type identifier for this type and may be used in LDECLARE <argspec>'s and in LNew() calls. (The type names LINT, LSTRING and the other builtin types are actually pointers to predefined structures.) The structure is of type LType as defined in lisp.h:
struct LType {
/* name of type */
char *name;
/* size of corresponding C type */
int size;
/* extract cell value from obj */
int (*fromobj)(/* LObject *obj, void *x */);
/* create a new LObject of this type */
LObject *(*toobj)(/* void *x */);
/* free a cell of this type */
void (*free)(/* void *x */);
/* write a cell value to a stream */
void (*write)(/* FILE *fp, void *x */);
/* test equality of two cells of this type */
int (*match)(/* void *a, void *b */);
/* pull a cell value from a va_list */
void (*pull)(/* va_list *a_list, void *x */);
/* parse an object of this type */
LObject *(*parse)(/* Lake *lake */);
/* magic number; always set to LTypeMagic */
int magic;
};
The void * pointers in the above point to objects of the type you are defining. For examples of how to define new types see the code in lisp.c that defines the string and list types. See also the file TYPES.DOC in the lisp source code directory for further details.
The term "Lake" is supposed to connote something more general than a stream; it also seemed particularly appropriate since this interpreter was written in the City of Lakes.
This section is X rated. Don't read it unless you are really serious.
The lisp interpreter works by first parsing (LSexpr()) an expression then evaluating it (LEval()). The LDECLARE macro is a mechanism which allows both the syntax (for parsing) and the sematics (for evaluation) of an expression to be specified in the same convenient place --- at the top of the C function which implements the function. The call syntax of all such C functions is
LObject *func(Lake *lake, LList *args)
When parsing a call to the corresponding lisp function, LSexpr() calls func with that lake pointer, and with args pointing to the head of the list in the parse tree corresponding to this function call. LDECLARE parses the arguments in the call (by reading them from the lake) and appends them to this list. (Note: the head of this list is the function itself, so the first argument becomes entry #2 in the list.)
When evaluating the function call, LEval() calls func with lake=NULL and with args pointing to the call's argument list. (In this case the first entry of the list is the first argument.) LDECLARE then converts the arguments in the list into the appropriate C data types, writing their values into the addresses in the <argspec>s.
One side-effect of using lake=NULL as the signal to evaluate rather than to parse is that the value of the lake pointer is not available at evaluation time. Some functions, however, may want to do something with the lake they were parsed from. For example, the "write" function in geomview writes data to the output stream associated with its input stream. (In geomview these streams are all stored in a general "Pool" structure which is retained as the "river" member of the lake.) The special token LLAKE may be used to cause the lake pointer to be saved in the args list at parse time and written into a variable at evaluation time. It is used exactly like the other (scalar) argument keywords:
LObject *func(Lake *lake, LList *args)
Lake *mylake;
LDECLARE(("myfunc", LBEGIN
LARG_LAKE, &mylake,
...
LARG_END));
At evaluation time LDECLARE will set mylake to have the value that lake had at parse time. This looks just like a specification for an argument to the lisp function but it is not --- it is just a way to tell LDECLARE to remember the lake pointer between parse- and evaluation-time.