File perq-pascal-lisp-project/pas0.perq from the latest check-in



 (* PreProcessor Version - Run  through Filter *)
      (* PERQ version *)
(*%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%                                     
% 		PASCAL BASED MINI-LISP
%
% File: 	PAS0.PAS - PASCAL/LISP KERNEL
% ChangeDate: 	11:00pm Monday, 28 September 1981
% By: 		Ralph Ottenheimer big -> fix, END comment FOR #pta
%		COMPRESS & EXPLODE
% 
% 	    All RIGHTS RESERVED
%           COPYRIGHT (C) - 1981 - M. L. GRISS
%           Computer Science Department
%           University of Utah
%
%           Do Not distribute with out written consent of M. L. Griss
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%*)


    PROGRAM pas0 (input,output, symin,finput,foutput);
    (************************************************************)
    (* support routines for a "lisp" machine.  uses a register  *)
    (* model with a stack for holding frames.  stack also used  *)
    (* to hold compiler generated constants.                    *)
    (* written by william f. galway and martin l. griss         *)
    (* modified by ralph ottenheimer may 81                     *)
    (* append pas1...pasn at  end                               *)
    (* -------------------------------------------------------- *)
    (* symin is input channel one--used to initialize "symbol   *)
    (* table".  input is input channel two--standard input.     *)
    (* output is output channel one--the standard output.       *)
    (* finput is file input channel three.                      *)
    (* foutput is file output channel four.                      *)
    (************************************************************)


imports Stream from Stream;
imports system from system;
imports io_others from io_others;
imports io_unit from io_unit;


CONST
    (* for terak, perq, Apollo, vax  *)
 sp = ' '; 
 ht = 9;          (* ascii codes *)
 lf = 10;
 cr = 13; 
 nul = 0;

   eos = chr(0);        (* KLUDGE: null string *)
    inchns = 3;       (* number of input channels.  *)
    outchns = 4;      (* number of output channels. *)
    begin_comment = '%';

(* Initial symbols, needed in Kernel *)
    xtoktype  =  129; (* slot in idspace for toktype. *)
    xbstack   = 130;  (* Bstack Pointer *)
    xthrowing = 131;  (* If throw mode *)
    xinitform = 132;  (* for restart *)
    xraise    = 133;  (* for RAISE of lc in ids *)

    chartype  =  3;   (* various token types *)
    inttype  =  1;
    idtype  =  2;

   (* no shift const *)
    (* assumed to be at least 16 bits long.  low order 13 bits  *)
    (* are the info, top 3 are the tag.                         *)
     int_offset = 32767; (* PERQ item is record *   )

    (* the various tags - can't use a defined scalar type *)
    (* because of the lack of convertion functions.       *)
    inttag = 0;    (* info is an integer                  *)
    chartag = 1;   (* info is a character code            *)
    pairtag = 2;   (* info points to pair                 *)
    idtag = 3;     (* info points to identifier           *)
    codetag = 4;   (* info is index into a case statement *)
    (*                that calls appropriate function.    *)
    errtag = 5;    (* info is an error code - see below.  *)
    fixtag = 6;    (* info points to a full word (or      *)
    (*                longer) integer.                    *)
    flotag = 7;    (* info points to a float number.      *)

    (* error codes.  corresponding to tag = errtag.  *)
    noprspace = 1;    (* no more "pair space"--can't cons. *)
    notpair = 2;      (* a pair operation attempted on a non-pair. *)
    noidspace = 3;    (* no more free identifiers *)
    undefined = 4;    (* used to mark undefined function cells (etc?) *)

    maxpair = 3700;   (* max number of pairs allowed. *)
    maxident = 800;   (* max number of identifiers *)
    maxstrsp = 4500;  (* size of string (literal) storage space. *)
    maxintsp = 200;   (* max number of long integers allowed *)
     maxflosp = 50;     (* max number of floating numbers allowed *)

    hidmax = 50;      (* number of hash values for identifiers *)
    maxgcstk = 100;   (* size of garbage collection stack.    *)
    stksize = 500;    (* stack size *)
    maxreg = 15;      (* number of registers in lisp machine. *)

    eofcode = 26;     (* magic character code for eof, ascii for *)
    (*  cntrl-z.  kludge, see note in xrdtok.  *)
    choffset = 1;     (* add choffset to ascii code to get address  *)
    (* in id space for corresponding identifier.  *)
    nillnk = 0;       (* when integers are used as pointers.  *)


TYPE
    onechar = char;     (* for terak,perq,Apollo*)

    (* note we allow zero for id_ptr, allowing a "nil" link. *)

    stringp = 1..maxstrsp;        (* pointer into string space. *)
    id_ptr = 0..maxident;            (* pointer into id space. *)

       itemref = RECORD
	           tag:integer;
                 info:integer;
                 END;
    itemtype = 0..7;    (* the tags *)


    pair = PACKED RECORD
		      prcar: itemref;
		      prcdr: itemref;
	(* OLD        markflag:boolean , but wastes space *)
		  END;


   ascfile = PACKED FILE OF onechar;

    ident = PACKED RECORD           (* identifier *)
		       idname: stringp;
		       val: itemref;       (* value *)
		       plist: itemref;     (* property list *)
		       funcell: itemref;   (* function cell *)
		       idhlink: id_ptr;    (* hash link *)
		   END;
    longint = integer;
VAR
    (* global information *)
    nilref,trueref: itemref;    (* refers to identifiers "nil", and "t". *)
    initphase: integer;                (* Start up *)
    r: ARRAY[1..maxreg] OF itemref;
    rxx,ryy: itemref;



    (* "st" is the stack pointer into "stk".  it counts the number of  *)
    (* items on the stack, so it runs from zero while the stack starts *)
    (* at one.                                                         *)
    st: 0..stksize;
    stk: ARRAY[1..stksize] OF itemref;

    (* pair space *)
    prspace: PACKED ARRAY[1..maxpair] OF pair; (* all pairs stored here. *)
    freepair: integer;          (* pointer to next free pair in prspace. *)

    (* identifier space *)
    idhead: ARRAY[0..hidmax] OF id_ptr;
    idspace: PACKED ARRAY[1..maxident] OF ident;
    freeident: integer;

    (* string space *)
    strspace: PACKED ARRAY[1..maxstrsp] OF onechar;
    freestr: stringp;

    (* large integer space *)
    intspace: ARRAY[1..maxintsp] OF longint;    (* use long int on terak *)
    freeint: 1..maxintsp;

    (* floating point number space *)
    flospace: ARRAY[1..maxflosp] OF real;
    freefloat: 1..maxflosp;

    (* i/o channels *)
  (* files declared on header *)
    symin : ascfile;
    finput : ascfile;
    foutput : ascfile;
    inchnl: 1..inchns;      (* current input channel number  *)
    outchnl: 1..outchns;    (* current output channel number *)

    (* "current character" for each input channel.                    *)
    (* may want to include more than one character at some later date *)
    (* (for more lookahead).                                          *)
    ichrbuf: ARRAY[1..inchns] OF onechar;

    (* for collecting statistics. *)
    gccount: integer;           (* counts garbage collections *)
    (* counts from last garbage collection. *)
    consknt: integer;           (* number of times "cons" called *)



(* ........ Everything nested inside CATCH *)

Procedure Xcatch;  (* ----------- Outermost Procedure ----------- *)
 var catch_stk:0..stksize;
     catch_Bstk:itemref;

PROCEDURE xread;
    FORWARD;

PROCEDURE xprint;
    FORWARD;

PROCEDURE xunbindto;
    FORWARD;

PROCEDURE xeval;
    FORWARD;

 Procedure Xthrow;
    begin (* throw value *)
        idspace[Xthrowing].val := trueref;
      exit(xeval)
    end (* throw *);
    
 Handler CtlC;  (* ------- handle runaway aborts ------- *)
 
 begin
    write('^C');
    IOKeyClear;
    IObeep;
    if initphase > 1 then Xthrow;
 end;

    (********************************************************)
    (*                                                      *)
    (*             item selectors & constructors            *)
    (*                                                      *)
    (********************************************************)


FUNCTION tag_of(item: itemref): itemtype;

    BEGIN (* tag_of *)
       tag_of := item.tag;
    END;
    (* tag_of *)

FUNCTION info_of(item: itemref): integer;

 BEGIN (* info_of *)
      info_of := item.info
    END;
    (* info_of *)

PROCEDURE mkitem(tag: itemtype; info: longint; VAR item: itemref);
    (* do range checking on info. ints run from -4096 to +4095 *)
    (* everything else runs from 0 to 8191. ints & chars       *)
    (* contain their info, all others points into an           *)
    (* appropriate space.                                      *)

    PROCEDURE mkfixint;
	BEGIN (* mkfixint *)
	IF freeint < maxintsp THEN     (* convert to fixnum *)
	    BEGIN
	    tag := fixtag;
	    intspace[freeint] := info;
	    info := freeint;        (* since we want the pointer *)
	    freeint := freeint + 1;
	    END
	ELSE BEGIN
                 writeln('*****FIXNUM SPACE EXHAUSTED');
                 (* should do gc *)
		 exit(pas0);
             END;
	END;
	(* mkfixint *)


    BEGIN (* mkitem *)
    IF tag = inttag THEN
      BEGIN
	IF (info < -int_offset) OR (info > int_offset - 1) THEN mkfixint
 	END
    ELSE IF tag = fixtag THEN mkfixint

	 ELSE IF info < 0 THEN
		  BEGIN
		  writeln('*****MKITEM: BAD NEG');
                exit(pas0);
		  END;
    (* nothing special to do for other types *)

    (* pack tag and info into 16-bit item.   *)
       item.tag := tag;
       item.info := info
    END;
    (* mkitem *)

PROCEDURE set_info(VAR item: itemref; newinfo: longint);
    BEGIN (* set_info *)
    mkitem(tag_of(item), newinfo, item)
    END;
    (* set_info *)

PROCEDURE set_tag(VAR item: itemref; newtag: itemtype);
    BEGIN (* set_tag *)
    mkitem(newtag, info_of(item), item)
    END;
    (* set_tag *)

PROCEDURE mkident(id: integer; reg: integer);
    (* make identifier "id" in register "reg" *)
    BEGIN       (* mkident *)
    mkitem(idtag, id, r[reg]);
    END;
    (* mkident *)

PROCEDURE mkint(int: longint; reg: integer);
    BEGIN       (* mkint *)
    mkitem(inttag, int, r[reg]);
    END;
    (* mkint *)

PROCEDURE mkpair(pr: integer; reg: integer);
    BEGIN (* mkpair *)
    mkitem(pairtag, pr, r[reg])
    END;
    (* mkpair *)

PROCEDURE int_val(item: itemref; VAR number: longint);
    (* returns integer value of item (int or fixnum). *)
    (* must return 'number' in var parameter instead  *)
    (* of function value since long integers are not  *)
    (* a legal function type in ucsd pascal.          *)
    BEGIN (* int_val *)
    IF tag_of(item) = inttag THEN
	number := info_of(item)
    ELSE IF tag_of(item) = fixtag THEN
	     number := intspace[info_of(item)]
  ELSE writeln('***** ILLEGAL DATA TYPE FOR NUMERIC OPERATION')
	(* halt or fatal error *)
    END;
    (* int_val *)


    (********************************************************)
    (*                                                      *)
    (*                  stack allocation                    *)
    (*                                                      *)
    (********************************************************)

PROCEDURE alloc(n: integer);
    BEGIN
    IF n + st <= stksize THEN
	st := n+st
    ELSE
	BEGIN
   writeln('*****LISP STACK OVERFLOW');
	writeln('     TRIED TO ALLOCATE ',n);
	writeln('     CURRENT STACK TOP IS ',st);
	END;
    END;

PROCEDURE dealloc(n: integer);
    BEGIN
    IF st - n >= 0 THEN
	st := st - n
    ELSE
	writeln('*****LISP STACK UNDERFLOW');
    END;

    (* optimized allocs *)

PROCEDURE alloc1;
    BEGIN alloc(1) END;

PROCEDURE dealloc1;
    BEGIN dealloc(1) END;

PROCEDURE alloc2;
    BEGIN alloc(2) END;

PROCEDURE dealloc2;
    BEGIN dealloc(2) END;

PROCEDURE alloc3;
    BEGIN alloc(3) END;

PROCEDURE dealloc3;
    BEGIN dealloc(3) END;


    (********************************************************)
    (*                                                      *)
    (*              support for register model              *)
    (*                                                      *)
    (********************************************************)

PROCEDURE load(reg: integer; sloc: integer);
    BEGIN
    IF sloc < 0 THEN r[reg] := r[-sloc]
    ELSE  r[reg] := stk[st-sloc];
    (* will, fix for load (pos,pos) *)
    END;

PROCEDURE store(reg: integer; sloc: integer);
    BEGIN
    stk[st-sloc] := r[reg];
    END;

    (* optimized load/store. *)
PROCEDURE load10;
    BEGIN
    load(1,0);
    END;

PROCEDURE store10;
    BEGIN
    store(1,0);
    END;

PROCEDURE storenil(sloc: integer);
    BEGIN
    stk[st-sloc] := nilref;
    END;

(* Other primitives ?? *)

    (********************************************************)
    (*                                                      *)
    (*              identifier lookup & entry               *)
    (*                                                      *)
    (********************************************************)

function nmhash(nm: stringp): integer;
    CONST
	hashc = 256;
    VAR
	i,tmp: integer;
    BEGIN
    tmp := 0;
    i := 1;     (* get hash code from first three chars of string. *)
    WHILE (i <= 3) AND (strspace[nm+i] <> eos) DO
	BEGIN
	tmp := ord(strspace[nm+i]) + hashc*tmp;
	i := i + 1;
	END;
    nmhash := abs(tmp) MOD hidmax;      (* abs because mod is screwy. *)
    END;

FUNCTION eqstr(s1,s2: stringp): boolean;
    BEGIN
    WHILE (strspace[s1] = strspace[s2]) AND (strspace[s1] <> eos) DO
	BEGIN
	s1 := s1 + 1;
	s2 := s2 + 1;
	END;
    eqstr := (strspace[s1] = strspace[s2]);
    END;

PROCEDURE nmlookup(nm: stringp; VAR found: boolean; VAR hash: integer;
		   VAR loc: itemref);
    (* lookup a name in "identifier space".                                 *)
    (* "hash" returns the hash value for the name.                          *)
    (* "loc" returns the location in the space for the (possibly new)       *)
    (* identifier.                                                          *)
    BEGIN
    hash := nmhash(nm);
    mkitem(idtag, idhead[hash], loc);
    (* default is identifier, but may be "error". *)
    (* start at appropriate hash chain. *)

    found := false;
    WHILE (info_of(loc) <> nillnk) AND (NOT found) DO
	BEGIN
	found := eqstr(nm, idspace[info_of(loc)].idname);
	IF NOT found THEN
	    set_info(loc, idspace[info_of(loc)].idhlink);
	(* next id in chain *)
	END;
    IF NOT found THEN               (* find spot for new identifier *)
	BEGIN
	IF freeident=nillnk THEN    (* no more free identifiers. *)
	    mkitem(errtag, noidspace, loc)
	ELSE
	    BEGIN
	    set_info(loc, freeident);
	    freeident := idspace[freeident].idhlink;
	    END;
	END;
    END;

PROCEDURE putnm(nm: stringp; VAR z: itemref; VAR found: boolean);
    (* put a new name into identifier space, or return old location *)
    (* if it's already there.                                       *)
    VAR
	tmp: ident;
	hash: integer;
    BEGIN
    nmlookup(nm, found, hash, z);
    IF (NOT found) AND (tag_of(z) = idtag) THEN
	BEGIN
	tmp.idname := nm;
	tmp.idhlink := idhead[hash];   (* put new ident at head of chain     *)
	tmp.val := nilref;             (* initialize value and property list *)
	tmp.plist := nilref;
	tmp.funcell := nilref;         (* also, the function cell *)
	idhead[hash] := info_of(z);
	idspace[info_of(z)] := tmp;
	END;
    END;

PROCEDURE xfaststat;
    (* give quick summary of statistics gathered *)
    BEGIN
    writeln('CONSES:',consknt);
    writeln('ST    :',st);
    END;


    (********************************************************)
    (*                                                      *)
    (*              the garbage collector                   *)
    (*                                                      *)
    (********************************************************)

PROCEDURE xgcollect;
    VAR
	i: integer;
	markedk: integer;   (* counts the number of pairs marked *)
	freedk: integer;    (* counts the number of pairs freed. *)
	gcstkp: 0..maxgcstk; (* note the garbage collection stack   *)
	mxgcstk: 0..maxgcstk;           (* is local to this procedure. *)
	gcstk: ARRAY[1..maxgcstk] OF integer;
        markflag: PACKED ARRAY[1..maxpair] OF boolean;
(* used not to have array here *)
        
    PROCEDURE pushref(pr: itemref);
	(* push the address of an unmarked pair, if that's what it is. *)
	BEGIN
	IF tag_of(pr) = pairtag THEN
	    IF NOT markflag[info_of(pr)] THEN  (* was .markflag *)
		BEGIN
		IF gcstkp < maxgcstk THEN
		    BEGIN
		    gcstkp := gcstkp + 1;
		    gcstk[gcstkp] := info_of(pr);
		    IF gcstkp > mxgcstk THEN
			mxgcstk := gcstkp;
		    END
		ELSE
		    BEGIN
		    writeln('*****GARBAGE STACK OVERFLOW');
                  exit(pas0);
		    END;
		END;
	END;

    PROCEDURE mark;
	(* "recursively" mark pairs referred to from gcstk. gcstk is used to *)
	(* simulate recursion.                                               *)
	VAR
	    prloc: integer;
	BEGIN
	WHILE gcstkp > 0 DO
	    BEGIN
	    prloc := gcstk[gcstkp];
	    gcstkp := gcstkp - 1;
            markflag[prloc] := true;
(* OLD      prspace[prloc].markflag := true;  *)
	    pushref(prspace[prloc].prcdr);
	    pushref(prspace[prloc].prcar);  (* trace the car first. *)
	    END;
	END;

    BEGIN       (* xgcollect *)
    writeln('***GARBAGE COLLECTOR CALLED');
    gccount := gccount + 1;          (* count garbage collections. *)
    xfaststat;   (* give summary of statistics collected *)
    consknt := 0;       (* clear out the cons counter *)
    gcstkp := 0;        (* initialize the garbage stack pointer. *)
    mxgcstk := 0;       (* keeps track of max stack depth. *)

    (* clear markflags *)
    FOR i := 1 TO maxpair DO markflag[i] := false;
    (* OLD: wasnt needed *)
    (* mark things from the "computation" stack. *)
    FOR i := 1 TO st DO
	BEGIN
	pushref(stk[i]);
	mark;
	END;
    (* mark things from identifier space. *)
    FOR i := 1 TO maxident DO
	BEGIN
	pushref(idspace[i].val);
	mark;
	pushref(idspace[i].plist);
	mark;
	pushref(idspace[i].funcell);
	mark;
	END;

    (* reconstruct free list by adding things to the head. *)
    freedk := 0;
    markedk := 0;
    FOR i:= 1 TO maxpair - 1 DO
	BEGIN
	IF markflag[i] THEN
	(* OLD:	IF prspace[i].markflag THEN  *)
	    BEGIN
	    markedk := markedk + 1;
	    markflag[i] := false
	    (* OLD: prspace[i].markflag := false *)
	    END
	ELSE
	    BEGIN
	    prspace[i].prcar := nilref;
	    mkitem(pairtag, freepair, prspace[i].prcdr);
	    freepair := i;
	    freedk := freedk + 1
	    END
	END;
    writeln(freedk,' PAIRS FREED.');
    writeln(markedk,' PAIRS IN USE.');
    writeln('MAX GC STACK WAS ',mxgcstk);
    END;
    (* xgcollect *)

    (********************************************************)
    (*                                                      *)
    (*                  lisp primitives                     *)
    (*                                                      *)
    (********************************************************)

    (* return r[1].r[2] in r[1] *)
PROCEDURE xcons;
    VAR p: integer;

    BEGIN
    (* push args onto stack, in case we need to garbage collect the *)
    (* references will be detected.                                 *)
    alloc(2);
    stk[st] := r[1];
    stk[st-1] := r[2];

    IF prspace[freepair].prcdr = nilref THEN xgcollect;

    p := freepair;
    freepair := info_of(prspace[p].prcdr);
    prspace[p].prcar := stk[st];
    prspace[p].prcdr := stk[st - 1];
    mkpair(p, 1);       (* leave r[1] pointing at new pair. *)

    consknt := consknt + 1;
    dealloc(2);
    END;

PROCEDURE xncons;
    BEGIN r[2] := nilref;
    xcons;
    END;

PROCEDURE xxcons;
    BEGIN rxx := r[1];
    r[1] := r[2];
    r[2] := rxx;
    xcons;
    END;

    (* return car of r[1] in r[1] *)
PROCEDURE xcar;
    BEGIN
    IF tag_of(r[1]) = pairtag THEN
	r[1] := prspace[info_of(r[1])].prcar
    ELSE
	mkitem(errtag, notpair, r[1]);
    END;

PROCEDURE xcdr;
    BEGIN
    IF tag_of(r[1]) = pairtag THEN
	r[1] := prspace[info_of(r[1])].prcdr
    ELSE
	mkitem(errtag, notpair, r[1]);
    END;

PROCEDURE xrplaca;
    BEGIN
    IF tag_of(r[1]) = pairtag THEN
	prspace[info_of(r[1])].prcar:=r[2]
    ELSE
	mkitem(errtag, notpair, r[1]);
    END;

PROCEDURE xrplacd;
    BEGIN
    IF tag_of(r[1]) = pairtag THEN
	prspace[info_of(r[1])].prcdr :=r[2]
    ELSE
	mkitem(errtag, notpair, r[1]);
    END;

    (* anyreg car and cdr *)
PROCEDURE anycar(VAR a, b: itemref);
    BEGIN
    IF tag_of(a) = pairtag THEN
	b := prspace[info_of(a)].prcar
    ELSE
	mkitem(errtag, notpair, b);
    END;

PROCEDURE anycdr(VAR a, b: itemref);
    BEGIN
    IF tag_of(a) = pairtag THEN
	b := prspace[info_of(a)].prcdr
    ELSE
	mkitem(errtag, notpair, b);
    END;


    (********************************************************)
    (*                                                      *)
    (*              compress & explode                      *)
    (*                                                      *)
    (********************************************************)
PROCEDURE compress;     (* returns new id from list of chars *)
      VAR i: stringp;
          clist, c: itemref;
          found: boolean;
          int: integer;

    FUNCTION is_int(i: stringp; VAR int: longint): boolean;
        VAR negative, could_be: boolean;

        BEGIN   (* is_int *)
        int := 0;
        could_be := true;
        negative := strspace[i] = '-';
        IF negative OR (strspace[i] = '+') THEN i := i + 1;

        WHILE could_be AND (strspace[i] <> eos) DO
            BEGIN
            IF (strspace[i] >= '0') AND (strspace[i] <= '9') THEN
                 int := int * 10 + (ord(strspace[i]) - ord('0'))
            ELSE could_be := false;
            i := i + 1
            END;

        IF negative THEN int := -int;
        is_int := could_be
        END     (* is_int *);

    BEGIN     (* compress *)
    clist := r[1];        (* list of chars *)
    i := freestr; (* point to possible new string *)

    WHILE (i < maxstrsp) AND (clist <> nilref) DO
        BEGIN
        IF tag_of(clist) = PAIRTAG THEN
            BEGIN
            c := prspace[info_of(clist)].prcar;
            clist := prspace[info_of(clist)].prcdr;
            IF tag_of(c) = IDTAG THEN
                IF (info_of(c) > choffset) AND
                   (info_of(c) < choffset + 128) THEN 
                    BEGIN 
                    strspace[i] := chr(info_of(c) - choffset);
                    i := i + 1
                    END
                ELSE 
               writeln('*****COMPRESS: LIST ID NOT SINGLE CHAR')
            ELSE 
           writeln('*****COMPRESS: LIST ITEM NOT ID');
            END 
        ELSE 
       writeln('*****COMPRESS: ITEM NOT LIST')
        END (* WHILE *);

    strspace[i] := eos;   (* terminate string *)

    IF (i >= maxstrsp) THEN
	writeln('*****STRING SPACE EXHAUSTED')
    ELSE IF is_int(freestr, int) THEN
         mkint(int, 1)
    ELSE (* look the name up, return itemref for it *)
        BEGIN
        putnm(freestr, r[1], found);
	IF NOT found THEN
	    freestr := i + 1;
	END
    END       (* compress *);

PROCEDURE explode;      (* returns list of chars from id or int *)
      
      FUNCTION id_explode(i: stringp): itemref;
          BEGIN (* id_explode *)
          IF strspace[i] = eos THEN id_explode := nilref
          ELSE 
              BEGIN
              r[2] := id_explode(i + 1);
              mkident(ord(strspace[i]) + choffset, 1);
              xcons;
              id_explode := r[1]
              END
          END   (* id_explode *);

     FUNCTION int_explode(i: integer): itemref;
          VAR negative: boolean;

          BEGIN (* int_explode *)
          r[1] := nilref;
          IF i < 0 THEN
              BEGIN negative := true;
              i := -i
              END
          ELSE negative := false;

          WHILE i > 0 DO
              BEGIN
              r[2] := r[1];
              mkident(i MOD 10 + ord('0') + choffset, 1);
              xcons;
              i := i DIV 10
              END;

          IF negative THEN
              BEGIN 
              r[2] := r[1];
              mkident(ord('-') + choffset, 1);
              xcons
              END;
          int_explode := r[1]
          END (* int_explode *);

      BEGIN     (* explode *)
      IF tag_of(r[1]) = IDTAG THEN 
          r[1] := id_explode(idspace[info_of(r[1])].idname)
      ELSE IF tag_of(r[1]) = INTTAG THEN 
          r[1] := int_explode(info_of(r[1]))
      ELSE IF tag_of(r[1]) = FIXTAG THEN
          r[1] := int_explode(intspace[info_of(r[1])])
      ELSE 
     writeln('***** EXPLODE: ARG BAD TYPE')
      END       (* explode *);


    (********************************************************)
    (*                                                      *)
    (*                    i/o primitives                    *)
    (*                                                      *)
    (********************************************************)

procedure xopen;

var s1: string; i,j : integer;

  handler ResetError(name: PathName);
  begin
    writeln('**** Could not open file -  ',name,' for read');
    exit(xopen);
  end;

  handler RewriteError(name: PathName);
  begin
    writeln('**** Could not open file -  ',name,' for write');
    exit(xopen);
  end;
  
begin
      IF tag_of(r[1]) = IDTAG THEN 
      begin
        i := idspace[info_of(r[1])].idname;
        s1[0] := chr(255);
        j:= 0;
        WHILE (i <= maxstrsp) AND (strspace[i] <> eos) do
        begin
          j:= j + 1;
          s1[j] := strspace[i];
          i:= i + 1;
        end;
        s1[0]:= chr(j);
        
        IF tag_of(r[2]) = IDTAG THEN 
         case  strspace[idspace[info_of(r[2])].idname] of
         'i', 'I': begin reset(finput,s1); mkint(3,1) end;
         'o', 'O': begin rewrite(foutput,s1); mkint(4,1) end;
         otherwise: writeln('**** OPEN: ARG2 NOT INPUT/OUTPUT');
         end  else writeln('**** OPEN: ARG2 BAD TYPE')
     end else writeln('**** OPEN: ARG1 BAD TYPE');
end;

procedure xclose;

begin
  case info_of(r[1]) of
  1,2: ;
  3: close(finput);
  4: close(foutput);
  end;
end;

PROCEDURE xrds;
  (* Select channel for input *)
  VAR tmp:longint;
    BEGIN
        tmp:=inchnl;
        inchnl := info_of(r[1]);
         mkint(tmp,1)
    END;

PROCEDURE xwrs;
  (* Select channel for output *)
  VAR tmp:longint;
    BEGIN
        tmp:=outchnl;
        outchnl := info_of(r[1]);
        mkint(tmp,1)
    END;

PROCEDURE xterpri;
(* need to change for multiple output channels.  *)
    BEGIN
     case outchnl of
     1: writeln(' ');

     2: writeln(foutput,' ')
     end;
   END;


PROCEDURE xwrtok;
    (* doesn't expand escaped characters in identifier names *)
    VAR
      temp_real: real; (* KLUDGE: for bug *)
	i: integer;
    BEGIN
      case outchnl of
      1: BEGIN
    IF tag_of(r[1]) = inttag THEN
	BEGIN
	IF info_of(r[1]) = 0 THEN
	    write('0')
	ELSE
	    write('  ', info_of(r[1]):0);
	END

    ELSE IF tag_of(r[1]) = fixtag THEN
	     write(intspace[info_of(r[1])])

	 ELSE IF tag_of(r[1]) = flotag THEN
                BEGIN temp_real:= flospace[info_of(r[1])];
                  write( '* Real number bug *', trunc (temp_real))
                END
	      ELSE IF tag_of(r[1]) = idtag THEN
		       BEGIN
		       i := idspace[info_of(r[1])].idname;
		       WHILE (i <= maxstrsp) AND (strspace[i] <> eos) DO
			   BEGIN
			   write(strspace[i]);
			   i:= i + 1;
			   END;
		       END

		   ELSE IF tag_of(r[1]) = chartag THEN
		    write(chr(info_of(r[1]) - choffset))

		   ELSE IF tag_of(r[1]) = errtag THEN
		         writeln(' *** Error # ', '  ',info_of(r[1]):0)
		   ELSE IF tag_of(r[1]) = codetag THEN
			   write(' ## ','  ', info_of(r[1]):0)

		   ELSE  write(' ? ','  ' ,tag_of(r[1]):0,' /  ' ,info_of(r[1]):0,' ? ');
    END;
    
    4: BEGIN
    IF tag_of(r[1]) = inttag THEN
	BEGIN
	IF info_of(r[1]) = 0 THEN
	    write(foutput,'0')
	ELSE
	    write(foutput,'  ', info_of(r[1]):0);
	END

    ELSE IF tag_of(r[1]) = fixtag THEN
	     write(foutput,intspace[info_of(r[1])])

	 ELSE IF tag_of(r[1]) = flotag THEN
                BEGIN temp_real:= flospace[info_of(r[1])];
                  write(foutput, '* Real number bug *', trunc (temp_real))
                END
	      ELSE IF tag_of(r[1]) = idtag THEN
		       BEGIN
		       i := idspace[info_of(r[1])].idname;
		       WHILE (i <= maxstrsp) AND (strspace[i] <> eos) DO
			   BEGIN
			   write(foutput,strspace[i]);
			   i:= i + 1;
			   END;
		       END

		   ELSE IF tag_of(r[1]) = chartag THEN
		    write(foutput,chr(info_of(r[1]) - choffset))

		   ELSE IF tag_of(r[1]) = errtag THEN
		         writeln(foutput,' *** Error # ', '  ',info_of(r[1]):0)
		   ELSE IF tag_of(r[1]) = codetag THEN
			   write(foutput,' ## ','  ', info_of(r[1]):0)

		   ELSE  write(foutput,' ? ','  ' ,tag_of(r[1]):0,' /  ' ,info_of(r[1]):0,' ? ');
    END;
    END; (*case*)
  end; (*wrtoken*)
  
  
PROCEDURE rdchnl(chnlnum: integer; VAR ch: onechar);
    BEGIN
    IF (chnlnum < 1) OR (chnlnum > inchns) THEN
	writeln('*****BAD INPUT CHANNEL FOR RDCHNL',chnlnum)
    ELSE
	CASE chnlnum OF
	    1:  BEGIN
	    ch := symin^;  (* a little strange, but avoids  *)
	    get(symin);              (* initialization problems *)
	    ichrbuf[inchnl] := symin^; (* Peek ahead *)
	    END;

	    2:  BEGIN
	    ch := input^;
	    get(input);
	    ichrbuf[inchnl] := input^;
	    END;

	    3:  BEGIN
	    ch := finput^;
	    get(finput);
	    ichrbuf[inchnl] := finput^;
	    END;

    END;
    (* case *)
    END;
    (* rdchnl *)

FUNCTION eofchnl: boolean;
    BEGIN
	CASE inchnl OF
	    1:  eofchnl := eof(symin);
	    2:  eofchnl := eof(input);
	    3:  eofchnl := eof(finput);
	    END;
    END;


 FUNCTION eol: boolean;
    BEGIN
	CASE inchnl OF
	    1:  eol := eoln(symin);
	    2:  eol := eoln(input);
	    3:  eol := eoln(finput);
	    END;
    END;

    (********************************************************)
    (*                                                      *)
    (*                   token scanner                      *)
    (*                                                      *)
    (********************************************************)

PROCEDURE xrdtok;
LABEL 1;

    VAR
	ch,ch1,ChangedCh: onechar;
	i: integer;
	anint: longint;
	moreid: boolean;
	found: boolean;

    FUNCTION digit(ch: onechar): boolean;
	BEGIN
	digit := ( '0' <= ch ) AND ( ch <= '9');
	END;


    FUNCTION escalpha(VAR ch: onechar): boolean;
	(* test for alphabetic or escaped character.                 *)
	(* note side effect in ChangedCh.                                *)
	BEGIN
        ChangedCh := Ch;
	IF ( 'A' <= ch ) AND ( ch <= 'Z') THEN
	    escalpha := true
	ELSE IF ( ord('A')+32 <= ord(ch)) AND ( ord(ch) <= ord('Z')+32) THEN
               BEGIN
                 IF idspace[xraise].val=trueref
                  THEN Changedch := chr(ord(ch)-32);
		 escalpha := true;    (* lower case alphabetics *)
               END
	     ELSE IF ch='!' THEN
		      BEGIN
		      rdchnl(inchnl,ch);
                      ChangedCh:=Ch;
                      escalpha := true;
		      END
		  ELSE
		      escalpha := false;
	END;

    FUNCTION alphanum(VAR ch: onechar): boolean;
	(* test if escalfa or digit *)
	VAR b: boolean;
	BEGIN
        ChangedCh:=Ch;
	b := digit(ch);
	IF NOT b THEN b := escalpha(ch);
	alphanum := b;
	END;

    FUNCTION whitesp(ch: onechar): boolean;
    VAR ascode:integer
    BEGIN
        ascode:=ord(ch);
        WHITESP := (CH = SP) OR (ascode = CR) OR (ascode = LF)
	 OR (ascode = ht) or (ascode = nul);        (* null?? *)
    END; 

	(* reads fixnums...need to read flonums too *)

    var negflag : integer;
    BEGIN       (* xrdtok *)
1:
    IF NOT eofchnl THEN
        REPEAT                          (* skip leading white space. *)
            rdchnl(inchnl,ch)
        UNTIL (NOT whitesp(ch)) OR eofchnl;
    IF eofchnl THEN
        mkitem(chartag, eofcode + choffset, r[1])
        (* should really return !$eof!$  *)
    ELSE
        BEGIN
        IF digit(ch) or  (ch = '-') THEN
            set_tag(r[1], inttag)
        ELSE IF escalpha(ch) THEN
                 set_tag(r[1], idtag)
             ELSE
                 set_tag(r[1], chartag);

        CASE tag_of(r[1]) OF
            chartag:  BEGIN
                  if ch = begin_comment then
                      BEGIN
                      while not eol do rdchnl(inchnl, ch);
                      (*REPEAT rdchnl(inchnl, ch)*) 
                      (*UNTIL eol;        (of selected input *)
                      rdchnl(inchnl, ch);
                      GOTO 1
                      END;
                  set_tag(r[1], idtag);
                  mkitem(inttag, chartype, idspace[xtoktype].val);
                  set_info(r[1], ord(ch) + choffset);
                  END;
            inttag:   BEGIN
                 mkitem(inttag, inttype, idspace[xtoktype].val);
                 negflag := 1;
                 if ch = '-' then 
                     begin
                       anint := 0;
                       negflag := -1
                     end  else
                     anint := ord(ch) - ord('0');
                 WHILE digit(ichrbuf[inchnl]) DO
                     BEGIN
                     rdchnl(inchnl,ch);
                     anint := 10 * anint + (ord(ch) - ord('0'))
                     END;
                 anint := anint * negflag;
                 set_info(r[1], anint)
                 END;
            idtag:    BEGIN
                mkitem(inttag, idtype, idspace[xtoktype].val);
                i := freestr; (* point to possible new string *)
                moreid := true;
                WHILE (i < maxstrsp) AND moreid DO
                    BEGIN
                    strspace[i] := ChangedCh; (* May have Case Change, etc *)
                    i:= i + 1;
                    moreid :=alphanum(ichrbuf[inchnl]);  (* PEEK ahead char *)
                    IF moreid THEN rdchnl(inchnl,ch) (* Advance readch *)
                    END;
                strspace[i] := eos;   (* terminate string *)
                IF (i >= maxstrsp) THEN
                    writeln('*****STRING SPACE EXHAUSTED')
                ELSE  (* look the name up, return itemref for it *)
                    BEGIN
                    putnm(freestr, r[1], found);
                    IF NOT found THEN
                        freestr := i + 1;
                    END;
                END;
                (* of case idtag *)
            END;
        (* of case *)
        END;
    END;
    (* xrdtok *)
    (* for DEBUG *)
    (********************************************************)
    (*                                                      *)
    (*                    initialization                    *)
    (*                                                      *)
    (********************************************************)


PROCEDURE init;
    (* initialization procedure depends on  *)
    (* ability to load stack with constants *)
    (* from a file.                         *)
    VAR
	strptr: stringp;
	nam: PACKED ARRAY[1..3] OF onechar;
	(* holds 'nil', other strings? *)
	i, n: integer;
	idref: itemref;
	found: boolean;

	(* init is divided into two parts so it can compile on terak *)
    PROCEDURE init1;
	BEGIN

	(* initialize top of stack *)
	st := 0;

	freefloat := 1;
	freeint := 1;

	(* define nilref - the id, nil, is defined a little later. *)
	freeident := 1;
	mkitem(idtag, freeident, nilref);

	(* initialize pair space. *)
	FOR i := 1 TO maxpair - 1 DO      (* initialize free list. *)
	    BEGIN
          (* OLD: prspace[i].MarkFlag := false; *)
	    prspace[i].prcar := nilref;         (* just for fun *)
	    mkitem(pairtag, i + 1, prspace[i].prcdr);
	    END;
	prspace[maxpair].prcar := nilref;
	prspace[maxpair].prcdr := nilref;       (* end flag *)
	freepair := 1;                  (* point to first free pair *)


	(* initialize identifier space and string space. *)
	freestr := 1;
	FOR i := 0 TO hidmax - 1 DO
	    idhead[i] := nillnk;
	FOR i := 1 TO maxident DO
	    BEGIN
	    IF i < maxident THEN
		idspace[i].idhlink := i + 1
	    ELSE    (* nil to mark the final identifier in the table. *)
		idspace[i].idhlink := nillnk;
	    (* set function cells to undefined *)
	    mkitem(errtag, undefined, idspace[i].funcell);
	    mkitem(errtag, undefined, idspace[i].val);
	    mkitem(errtag, undefined, idspace[i].plist);
	    
	    END;

	(* nil must be the first identifier in the table--id #1 *)
	(* must fill in fields by hand for nilref.*)
	(* putnm can handle any later additions.  *)
	nam := 'NIL';
	strptr := freestr;
	FOR i := 1 TO 3 DO
	    BEGIN
	    strspace[strptr] := nam[i];
	    strptr:= strptr + 1;
	    END;
	strspace[strptr] := eos;
        putnm(freestr, nilref, found);
	IF NOT found THEN
	    freestr := strptr + 1;

	(* make the single character ascii identifiers, except nul(=eos). *)
	FOR i := 1 TO 127  DO
	    BEGIN
	    strspace[freestr] := chr(i);
	    strspace[freestr + 1] := eos;
	    putnm(freestr, idref, found);
	    IF NOT found THEN
		freestr := freestr + 2;
	    IF i = ord('T') THEN
		trueref := idref;
	    (* returns location for 't. *)
	    END;

	(* clear the counters *)
        idspace[xraise].val := trueref;
	gccount := 0;
	consknt := 0;
	END;
	(* init1 *)

    PROCEDURE init2;
	BEGIN
	(* load "symbol table" with identifiers, constants, and functions.  *)
	inchnl := 1;        (* select symbol input file. *)
	outchnl := 1;        (* select standard output file. *)
	reset(symin,'paslsp.ini');
	reset(input);
	rewrite(output);
	xrdtok;     (* get count of identifiers. *)
	IF tag_of(r[1]) <> inttag THEN
	    writeln('*****BAD SYMBOL TABLE, INTEGER EXPECTED AT START');
	n := info_of(r[1]);
	FOR i := 1 TO n DO
	    xrdtok;
	(* reading token magically loads it into id space. *)
	xrdtok;         (* look for zero terminator. *)
	IF (tag_of(r[1]) <> inttag) OR (info_of(r[1]) <> 0) THEN
	    writeln('*****BAD SYMBOL TABLE, ZERO EXPECTED AFTER IDENTIFIERS');

	xrdtok;         (* count of constants  *)
	IF tag_of(r[1]) <> inttag THEN
    writeln('*****BAD SYMBOL TABLE, INTEGER EXPECTED BEFORE CONSTANTS');
	n := info_of(r[1]);
	alloc(n);       (* space for constants on the stack *)
	FOR i := 1 TO n DO
	    BEGIN
	    xread;
	    stk[i] := r[1];
	    END;
	xrdtok;
	IF (tag_of(r[1]) <> inttag) OR (info_of(r[1]) <> 0) THEN
    writeln('*****BAD SYMBOL TABLE, ZERO EXPECTED AFTER CONSTANTS');

	xrdtok;     (* count of functions. *)
	IF tag_of(r[1]) <> inttag THEN
	   writeln('*****BAD SYMBOL TABLE, INTEGER EXPECTED BEFORE FUNCTIONS');
	n := info_of(r[1]);
	FOR i := 1 TO n DO
	    (* for each function *)
	    (* store associated code *)
	    BEGIN
	    xrdtok;
	    mkitem(codetag, i, idspace[info_of(r[1])].funcell);
	    END;
	xrdtok;
	IF (tag_of(r[1]) <> inttag) OR (info_of(r[1]) <> 0) THEN
    writeln('*****BAD SYMBOL TABLE, ZERO EXPECTED AFTER FUNCTIONS');

	END;
	(* init2 *)
    BEGIN       (* init *)
    init1;
    init2;
    END;
    (* init *)

    (********************************************************)
    (*                                                      *)
    (*                 arithmetic functions                 *)
    (*                                                      *)
    (********************************************************)

PROCEDURE xadd1;
    VAR i: longint;

    BEGIN
    int_val(r[1], i);
    mkint(i + 1, 1)
    END;

PROCEDURE xdifference;
    VAR i1, i2: longint;

    BEGIN
    int_val(r[1], i1);
    int_val(r[2], i2);
    mkint(i1 - i2, 1)
    END;

PROCEDURE xdivide;      (* returns dotted pair (quotient . remainder). *)
    VAR quot, rem: integer;
	i1, i2: longint;

    BEGIN
    int_val(r[1], i1);
    int_val(r[2], i2);

    mkint(i1 DIV i2, 1);
    mkint(i1 MOD i2, 2);
    xcons
    END;

PROCEDURE xgreaterp;
    VAR i1, i2: longint;

    BEGIN
    int_val(r[1], i1);
    int_val(r[2], i2);

    IF i1 > i2 THEN
	r[1] := trueref
    ELSE
	r[1] := nilref;
    END;

PROCEDURE xlessp;
    VAR i1, i2: longint;

    BEGIN
    int_val(r[1], i1);
    int_val(r[2], i2);

    IF i1 < i2 THEN
	r[1] := trueref
    ELSE
	r[1] := nilref;
    END;

PROCEDURE xminus;
    VAR i: longint;

    BEGIN
    int_val(r[1], i);
    mkint(-i, 1)
    END;

PROCEDURE xplus2;
    VAR i1, i2: longint;

    BEGIN
    int_val(r[1], i1);
    int_val(r[2], i2);
    mkint(i1 + i2, 1)
    END;

PROCEDURE xquotient;
    VAR i1, i2: longint;

    BEGIN
    int_val(r[1], i1);
    int_val(r[2], i2);
    mkint(i1 DIV i2, 1)
    END;

PROCEDURE xremainder;
    VAR i1, i2: longint;

    BEGIN
    int_val(r[1], i1);
    int_val(r[2], i2);
    mkint(i1 MOD i2, 1)
    END;

PROCEDURE xtimes2;
    VAR i1, i2: longint;

    BEGIN
    int_val(r[1], i1);
    int_val(r[2], i2);
    mkint(i1 * i2, 1)
    END;
    (* xtimes2 *)


    (********************************************************)
    (*                                                      *)
    (*                    support for eval                  *)
    (*                                                      *)
    (********************************************************)


PROCEDURE execute(code: integer);
    FORWARD;

    (* Xapply(fn,arglist)-- "fn" is an operation code. *)
PROCEDURE xxapply;
    VAR
	i: integer;
	code: integer;
	tmp: itemref;
	tmpreg: ARRAY[1..maxreg] OF itemref;
    BEGIN
    code := info_of(r[1]);
    r[1] := r[2];
    i := 1;
    (* spread the arguments  *)
    WHILE (r[1] <> nilref) AND (i <= maxreg) DO
	BEGIN
	tmp := r[1];
	xcar;
	tmpreg[i] := r[1];
	i := i + 1;
	r[1] := tmp;
	xcdr;
	END;
    WHILE i > 1 DO
	BEGIN
	i := i - 1;
	r[i] := tmpreg[i];
	END;
    execute(code);
    END;



REDUCE Historical
REDUCE Sourceforge Project | Historical SVN Repository | GitHub Mirror | SourceHut Mirror | NotABug Mirror | Chisel Mirror | Chisel RSS ]