summaryrefslogtreecommitdiff
path: root/new/lex.c
diff options
context:
space:
mode:
authorJakob Kaivo <jkk@ung.org>2022-03-04 12:32:20 -0500
committerJakob Kaivo <jkk@ung.org>2022-03-04 12:32:20 -0500
commit55f277e77428d7423ae906a8e1f1324d35b07a7d (patch)
tree5c1c04703dff89c46b349025d2d3ec88ea9b3819 /new/lex.c
import Miranda 2.066 from upstream
Diffstat (limited to 'new/lex.c')
-rw-r--r--new/lex.c1213
1 files changed, 1213 insertions, 0 deletions
diff --git a/new/lex.c b/new/lex.c
new file mode 100644
index 0000000..a4e9d09
--- /dev/null
+++ b/new/lex.c
@@ -0,0 +1,1213 @@
+/* MIRANDA LEX ANALYSER */
+
+/**************************************************************************
+ * Copyright (C) Research Software Limited 1985-90. All rights reserved. *
+ * The Miranda system is distributed as free software under the terms in *
+ * the file "COPYING" which is included in the distribution. *
+ *------------------------------------------------------------------------*/
+
+#include "data.h"
+#include "lex.h"
+#include <errno.h>
+
+extern word DICSPACE; /* see steer.c for default value */
+/* capacity in chars of dictionary space for storing identifiers and file names
+ to get a larger name space just increase this number */
+extern FILE *s_in;
+extern word echoing,listing,verbosity,magic,inbnf,inlex;
+word fileq=NIL; /* list of currently open-for-input files, of form
+ cons(strcons(stream,<ptr to element of 'files'>),...)*/
+word insertdepth= -1,margstack=NIL,col=0,lmargin=0;
+word echostack=NIL;
+word lverge=0,vergstack=NIL;
+char *prefixbase; /* stores prefixes for pathnames, to get static resolution */
+word prefixlimit=1024; /* initial size of space for prefixes */
+word prefix,prefixstack=NIL; /* current prefix, stack of old prefixes */
+word atnl=1,line_no=0;
+word lastline;
+word litstack=NIL,linostack=NIL;
+word c=' ', lastc;
+word commandmode;
+word common_stdin,common_stdinb,cook_stdin;
+word litmain=0,literate=0; /* flags "literate" comment convention */
+char *dic,*dicp,*dicq;
+char *pathname();
+
+setupdic()
+{ dicp=dicq=dic=malloc(DICSPACE);
+ if(dic==NULL)mallocfail("dictionary");
+ /* it is not permissible to realloc dic, because at the moment identifiers
+ etc. contain absolute pointers into the dictionary space - so we must
+ choose fairly large initial value for DICSPACE. Fix this later */
+ prefixbase=malloc(prefixlimit);
+ prefixbase[0]='\0';
+ prefix=0;
+}
+
+/* this allows ~login convention in filenames */
+/* #define okgetpwnam
+/* suppress 26.5.06 getpwnam cases runtime error when statically linked (Linux) */
+
+#ifdef okgetpwnam
+#include <pwd.h>
+struct passwd *getpwnam();
+#endif
+char *getenv();
+
+char *gethome(n) /* for expanding leading `~' in tokens and pathnames */
+char *n;
+{ struct passwd *pw;
+ if(n[0]=='\0')return(getenv("HOME"));
+#ifdef okgetpwnam
+ if(pw=getpwnam(n))return(pw->pw_dir);
+#endif
+ return(NULL);
+}
+
+#define ovflocheck if(dicq-dic>DICSPACE)dicovflo()
+
+dicovflo() /* is this called everywhere it should be? Check later */
+{ fprintf(stderr,"\npanic: dictionary overflow\n"); exit(1); }
+
+char *token() /* lex analyser for command language (very simple) */
+{ extern char *current_script;
+ word ch=getchar();
+ dicq = dicp; /* uses top of dictionary as temporary work space */
+ while(ch==' '||ch=='\t')ch=getchar();
+ if(ch=='~')
+ { char *h;
+ *dicq++ = ch;
+ ch=getchar();
+ while(isalnum(ch)||ch=='-'||ch=='_'||ch=='.')
+ *dicq++ = ch,ch=getchar();
+ /* NB csh does not allow `.' in user ids when expanding `~'
+ but this may be a mistake */
+ *dicq='\0';
+ if(h=gethome(dicp+1))
+ (void)strcpy(dicp,h),dicq=dicp+strlen(dicp);
+ }
+#ifdef SPACEINFILENAMES
+ if(ch!='"'&&ch!='<') /* test added 9.5.06 see else part */
+#endif
+ while(!isspace(ch)&&ch!=EOF)
+ { *dicq++ = ch;
+ if(ch=='%')
+ if(dicq[-2]=='\\')(--dicq)[-1]='%';
+ else dicq--,(void)strcpy(dicq,current_script),dicq+=strlen(dicq);
+ ch=getchar(); }
+#ifdef SPACEINFILENAMES
+ else { word closeq= ch=='<'?'>':'"'; /* this branch added 9.5.06 */
+ *dicq++ = ch; /* to allow spaces in "tok" or <tok> */
+ ch=getchar();
+ while(ch!=closeq&&ch!='\n'&&ch!=EOF)
+ *dicq++ = ch, ch=getchar();
+ if(ch==closeq)*dicq++ = ch, ch=getchar(); }
+#endif
+ *dicq++ = '\0';
+ ovflocheck;
+ while(ch==' '||ch=='\t')ch=getchar();
+ ungetc(ch,stdin);
+ return(*dicp=='\0'?(char *)NULL:dicp);
+} /* NB - if no token returns NULL rather than pointer to empty string */
+
+char *addextn(b,s) /* if(b)force s to end in ".m", and resolve <quotes> */
+word b;
+char *s;
+{ extern char *miralib;
+ extern char linebuf[];
+ word n=strlen(s);
+ /* printf("addextn(%s)\n",s); /* DEBUG */
+ if(s[0]=='<'&&s[n-1]=='>')
+ { static miralen=0; /* code to handle quotes added 21/1/87 */
+ if(!miralen)miralen=strlen(miralib);
+ strcpy(linebuf,miralib);
+ linebuf[miralen]= '/';
+ strcpy(linebuf+miralen+1,s+1);
+ strcpy(dicp,linebuf);
+ s=dicp;
+ n=n+miralen-1;
+ dicq=dicp+n+1;
+ dicq[-1] = '\0'; /* overwrites '>' */
+ ovflocheck; } else
+ if(s[0]=='\"'&&s[n-1]=='\"')
+ { /*strip quotes */
+ dicq=dicp; s++;
+ while(*s)*dicq++ = *s++;
+ dicq[-1]='\0'; /* overwrites '"' */
+ s=dicp; n=n-2;
+ }
+ if(!b||strcmp(s+n-2,".m")==0)return(s);
+ if(s==dicp)dicq--;/*if s in scratch area at top of dic, extend in situ*/
+ else { /* otherwise build new copy at top of dic */
+ dicq=dicp;
+ while(*s)*dicq++ = *s++;
+ *dicq = '\0'; }
+ if(strcmp(dicq-2,".x")==0)dicq -= 2; else
+ if(dicq[-1]=='.')dicq -= 1;
+ (void)strcpy(dicq,".m");
+ dicq += 3;
+ ovflocheck;
+ /* printf("return(%s)\n",dicp); /* DEBUG */
+ return(dicp);
+} /* NB - call keep(dicp) if the result is to be retained */
+
+word brct=0;
+
+spaces(n)
+word n;
+{ while(n-- >0)putchar(' ');
+}
+
+litname(s)
+char *s;
+{ word n=strlen(s);
+ return(n>=6 && strcmp(s+n-6,".lit.m")==0);
+}
+
+word getch() /* keeps track of current position in the variable "col"(column) */
+{ word ch= getc(s_in);
+ if(ch==EOF&&!atnl&&tl[fileq]==NIL) /* badly terminated top level file */
+ { atnl=1; return('\n'); }
+ if(atnl)
+ { if((line_no==0&&!commandmode||magic&&line_no==1)&&litstack==NIL)
+ litmain=literate= (ch=='>')||litname(get_fil(current_file));
+ if(literate)
+ { word i=0;
+ while(ch!=EOF&&ch!='>')
+ { ungetc(ch,s_in);
+ line_no++;
+ (void)fgets(dicp,250,s_in);
+ if(i==0&&line_no>1)chblank(dicp); i++;
+ if(echoing)spaces(lverge),fputs(dicp,stdout);
+ ch=getc(s_in); }
+ if((i>1||line_no==1&&i==1)&&ch!=EOF)chblank(dicp);
+ if(ch=='>')
+ { if(echoing)putchar(ch),spaces(lverge);ch=getc(s_in); }
+ } /* supports alternative `literate' comment convention */
+ atnl=0; col= lverge+literate;
+ if(!commandmode&&ch!=EOF)line_no++; }
+ if(echoing&&ch!=EOF)
+ { putchar(ch);
+ if(ch=='\n'&&!literate)
+ if(litmain)putchar('>'),spaces(lverge);
+ else spaces(lverge);
+ }
+ if(ch=='\t')col= ((col-lverge)/8 + 1)*8+lverge;
+ else col++;
+ if(ch=='\n')atnl= 1;
+ return(ch); }
+
+word blankerr=0;
+
+chblank(s)
+char *s;
+{ while(*s==' '||*s=='\t')s++;
+ if(*s=='\n')return;
+ syntax("formal text not delimited by blank line\n");
+ blankerr=1;
+ reset(); /* easiest way to recover is to pretend it was an interrupt */
+}
+
+/* getlitch gets a character from input like getch, but using C escaping
+ conventions if the char is backslash -- for use in reading character
+ and string constants */
+
+word rawch;
+/* it is often important to know, when certain characters are returned (e.g.
+ quotes and newlines) whether they were escaped or literal */
+
+word errch; /* for reporting unrecognised \escape */
+
+word getlitch()
+{ extern word UTF8;
+ word ch=c;
+ rawch = ch;
+ if(ch=='\n')return(ch); /* always an error */
+ if(UTF8&&ch>127)
+ { /* UTF-8 uses 2 or 3 bytes for unicode points to 0xffff */
+ word ch1=c=getch();
+ if((ch&0xe0)==0xc0) /* 2 bytes */
+ { if((ch1&0xc0)!=0x80)
+ return -5; /* not valid UTF8 */
+ c=getch();
+ return sto_char((ch&0x1f)<<6|ch1&0x3f); }
+ word ch2=c=getch();
+ if((ch&0xf0)==0xe0) /* 3 bytes */
+ { if((ch1&0xc0)!=0x80||(ch2&0xc0)!=0x80)
+ return -5; /* not valid UTF8 */
+ c=getch();
+ return sto_char((ch&0xf)<<12|(ch1&0x3f)<<6|ch2&0x3f); }
+ word ch3=c=getch();
+ if((ch&0xf8)==0xf0) /* 4 bytes, beyond basic multiligual plane */
+ { if((ch1&0xc0)!=0x80||(ch2&0xc0)!=0x80||(ch3&0xc0)!=0x80)
+ return -5; /* not valid UTF8 */
+ c=getch();
+ return((ch&7)<<18|(ch1&0x3f)<<12|(ch2&0x3f)<<6|ch3&0x3f); }
+ return(-5);
+ /* not UTF8 */
+ }
+ if(ch!='\\')
+ { c=getch(); return(ch); }
+ ch = getch();
+ c = getch();
+ switch(ch)
+ { case '\n': return(getlitch()); /* escaped nl was handled in 'getch()' */
+ case 'a': return('\a');
+ case 'b': return('\b');
+ case 'f': return('\f'); /* form feed */
+ case 'n': return('\n'); /* newline, == linefeed */
+ case 'r': return('\r'); /* carriage return */
+ case 't': return('\t');
+ case 'v': return('\v');
+ case 'X': /* omit for Haskell escape rules, see also lines marked H */
+ case 'x': if(isxdigit(c))
+ { word value, N=ch=='x'?4:6; /* N=7 for Haskell escape rules */
+ char hold[8];
+ ch = c;
+ word count=0;
+ /* while(ch=='0'&&isxdigit(peekch()))ch=getch(); /* H-lose leading 0s */
+ while(isxdigit(ch)&&count<N)
+ hold[count++]=ch,ch=getch();
+ /* read upto N hex digits */
+ hold[count] = '\0';
+ sscanf(hold,"%x",&value);
+ c = ch;
+ return value>UMAX?-3 /* \x out of range */
+ :sto_char(value); }
+ else return -2; /* \x with no hex digits */
+ default: if('0'<=ch&&ch<='9')
+ { word n=ch-'0',count=1,N=3; /* N=8 for Haskell escape rules */
+ ch = c;
+ /* while(ch=='0'&&isdigit(peekch()))ch=getch(); /* H-lose leading 0s */
+ while(isdigit(ch)&&count<N)
+ /* read upto N digits */
+ { n = 10*n+ch-'0';
+ count++;
+ ch = getch(); }
+ c = ch;
+ return /* n>UMAX?-4: /* H \decimal out of range */
+ sto_char(n); }
+ if(ch=='\''||ch=='"'||ch=='\\'||ch=='`')return(ch); /* see note */
+ if(ch=='&')return -7; /* Haskell null escape, accept silently */
+ errch=ch<=255?ch:'?';
+ return -6; /* unrecognised \something */
+ }
+} /* note: we accept \` for ` because getlitch() is used by charclass() */
+
+char *rdline() /* used by the "!" command -- see RULES */
+{ extern char *current_script;
+ static char linebuf[BUFSIZE];
+ char *p=linebuf;
+ word ch=getchar(),expansion=0;
+ while(ch==' '||ch=='\t')ch=getchar();
+ if(ch=='\n'||ch=='!'&&!(*linebuf))
+ { /* "!!" or "!" on its own means repeat last !command */
+ if(*linebuf)printf("!%s",linebuf);
+ while(ch!='\n'&&ch!=EOF)ch=getchar();
+ return(linebuf); }
+ if(ch=='!')
+ expansion=1,p=linebuf+strlen(linebuf)-1; /* p now points at old '\n' */
+ else ungetc(ch,stdin);
+ while((*p++ =ch=getchar())!='\n'&&ch!=EOF)
+ if(p-linebuf>=BUFSIZE)
+ { *p='\0';
+ fprintf(stderr,"sorry, !command too long (limit=%d chars): %s...\n",
+ BUFSIZE,linebuf);
+ while((ch=getchar())!='\n'&&ch!=EOF);
+ return(NULL);
+ } else
+ if(p[-1]=='%')
+ if(p>linebuf+1&&p[-2]=='\\')(--p)[-1]='%'; else
+ { (void)strncpy(p-1,current_script,linebuf+BUFSIZE-p);
+ p = linebuf+strlen(linebuf);
+ expansion = 1;
+ }
+ *p = '\0';
+ if(expansion)printf("!%s",linebuf);
+ return(linebuf); }
+
+setlmargin() /* this and the next routine are used to enforce the offside
+ rule ("yylex" refuses to read a symbol if col<lmargin) */
+{ margstack= cons(lmargin,margstack);
+ if(lmargin<col)lmargin= col; } /* inner scope region cannot "protrude" */
+
+unsetlmargin()
+{ if(margstack==NIL)return; /* in case called after `syntax("..")' */
+ lmargin= hd[margstack];
+ margstack= tl[margstack]; }
+
+word okid();
+word okulid();
+word PREL=1;
+
+#define isletter(c) ('a'<=c&&c<='z'||'A'<=c&&c<='Z')
+
+errclass(word val, word string)
+/* diagnose error in charclass, string or char const */
+{ char *s = string==2?"char class":string?"string":"char const";
+ if(val==-2)printf("\\x with no xdigits in %s\n",s); else
+ if(val==-3)printf("\\hexadecimal escape out of range in %s\n",s); else
+ if(val==-4)printf("\\decimal escape out of range in %s\n",s); else
+ if(val==-5)printf("unrecognised character in %s"
+ "(UTF8 error)\n",s); else
+ if(val==-6)printf("unrecognised escape \\%c in %s\n",errch,s); else
+ if(val==-7)printf("illegal use of \\& in char const\n"); else
+ printf("unknown error in %s\n",s);
+ acterror(); }
+
+yylex() /* called by YACC to get the next symbol */
+{ extern word SYNERR,exportfiles,inexplist,sreds;
+ /* SYNERR flags context sensitive syntax error detected in actions */
+ if(SYNERR)return(END); /* tell YACC to go home */
+ layout();
+ if(c=='\n') /* can only occur in command mode */
+/* if(magic){ commandmode=0; /* expression just read, now script */
+/* line_no=2;
+/* return(c); } else /* no longer relevant 26.11.2019 */
+ return(END);
+ if(col<lmargin)
+ if(c=='='&&(margstack==NIL||col>=hd[margstack]))/* && part fixes utah.bug*/
+ { c = getch();
+ return(ELSEQ); /* ELSEQ means "OFFSIDE =" */
+ }
+ else return(OFFSIDE);
+ if(c==';') /* fixes utah2.bug */
+ { c=getch(); layout();
+ if(c=='='&&(margstack==NIL||col>=hd[margstack]))
+ { c = getch();
+ return(ELSEQ); /* ELSEQ means "OFFSIDE =" */
+ }
+ else return(';');
+ }
+ if(
+ /* c=='_'&&okid(peekch()) || /* _id/_ID as lowercase id */
+ isletter(c)){ kollect(okid);
+ if(inlex==1){ layout();
+ yylval=name();
+ return(c=='='?LEXDEF:
+ isconstructor(yylval)?CNAME:
+ NAME); }
+ if(inbnf==1)
+ /* add trailing space to nonterminal to avoid clash
+ with ordinary names */
+ dicq[-1] = ' ',
+ *dicq++ = '\0';
+ return(identifier(0)); }
+ if('0'<=c&&c<='9'||c=='.'&&peekdig())
+ { if(c=='0'&&tolower(peekch())=='x')
+ hexnumeral(); else /* added 21.11.2013 */
+ if(c=='0'&&tolower(peekch())=='o')
+ getch(),c=getch(),octnumeral(); /* added 21.11.2013 */
+ else numeral();
+ return(CONST); }
+ if(c=='%'&&!commandmode)return(directive());
+ if(c=='\'')
+ { c = getch();
+ yylval= getlitch();
+ if(yylval<0){ errclass(yylval,0); return CONST; }
+ if(!is_char(yylval))
+ printf("%simpossible event while reading char const ('\\%u\')\n",
+ echoing?"\n":"",yylval),
+ acterror();
+ if(rawch=='\n'||c!='\'')syntax("improperly terminated char const\n");
+ else c= getch();
+ return(CONST); }
+ if(inexplist&&(c=='\"'||c=='<'))
+ { if(!pathname())syntax("badly formed pathname in %export list\n");
+ else exportfiles=strcons(addextn(1,dicp),exportfiles),
+ keep(dicp);
+ return(PATHNAME); }
+ if(inlex==1&&c=='`')
+ { return(charclass()?ANTICHARCLASS:CHARCLASS); }
+ if(c=='\"')
+ { string();
+ if(yylval==NIL)yylval=NILS; /* to help typechecker! */
+ return(CONST); }
+ if(inbnf==2) /* fiddle to offside rule in grammars */
+ if(c=='[')brct++; else if(c==']')brct--; else
+ if(c=='|'&&brct==0)
+ return(OFFSIDE);
+ if(c==EOF)
+ { if(tl[fileq]==NIL&&margstack!=NIL)return(OFFSIDE); /* to fix dtbug */
+ fclose((FILE *)hd[hd[fileq]]);
+ fileq= tl[fileq]; insertdepth--;
+ if(fileq!=NIL&&hd[echostack])
+ { if(literate)putchar('>'),spaces(lverge);
+ printf("<end of insert>"); }
+ s_in= fileq==NIL?stdin:(FILE *)hd[hd[fileq]];
+ c= ' ';
+ if(fileq==NIL)
+ { lverge=c=col=lmargin=0;
+ /* c=0; necessary because YACC sometimes reads 1 token past END */
+ atnl=1;
+ echoing=verbosity&listing;
+ lastline=line_no;
+ /* hack so errline can be set right if err at end of file */
+ line_no=0;
+ litmain=literate=0;
+ return(END); }
+ else { current_file = tl[hd[fileq]];
+ prefix=hd[prefixstack];
+ prefixstack=tl[prefixstack];
+ echoing=hd[echostack];
+ echostack=tl[echostack];
+ lverge=hd[vergstack];
+ vergstack=tl[vergstack];
+ literate=hd[litstack];
+ litstack=tl[litstack];
+ line_no=hd[linostack];
+ linostack=tl[linostack]; }
+ return(yylex()); }
+ lastc= c;
+ c= getch();
+#define try(x,y) if(c==x){ c=getch(); return(y); }
+ switch(lastc) {
+ case '_': if(c=='') /* underlined something */
+ { c=getch();
+ if(c=='<'){ c=getch(); return(LE); }
+ if(c=='>'){ c=getch(); return(GE); }
+ if(c=='%'&&!commandmode)return(directive());
+ if(isletter(c)) /* underlined reserved word */
+ { kollect(okulid);
+ if(dicp[1]=='_'&&dicp[2]=='')
+ return(identifier(1)); }
+ syntax("illegal use of underlining\n");
+ return('_'); }
+ return(lastc);
+ case '-': try('>',ARROW) try('-',MINUSMINUS) return(lastc);
+ case '<': try('-',LEFTARROW) try('=',LE) return(lastc);
+ case '=': if(c=='>'){ syntax("unexpected symbol =>\n"); return '='; }
+ try('=',EQEQ) return(lastc);
+ case '+': try('+',PLUSPLUS) return(lastc);
+ case '.': if(c=='.')
+ { c=getch();
+ return(DOTDOT);
+ }
+ return(lastc);
+ case '\\': try('/',VEL) return(lastc);
+ case '>': try('=',GE) return(lastc);
+ case '~': try('=',NE) return(lastc);
+ case '&': if(c=='>')
+ { c=getch();
+ if(c=='>')yylval=1;
+ else yylval=0,ungetc(c,s_in);
+ c=' ';
+ return(TO); }
+ return(lastc);
+ case '/': try('/',DIAG) return(lastc);
+ case '*': try('*',collectstars()) return(lastc);
+ case ':': if(c==':')
+ { c=getch();
+ if(c=='='){ c=getch(); return(COLON2EQ); }
+ else return(COLONCOLON);
+ }
+ return(lastc);
+ case '$': if(
+ /* c=='_'&&okid(peekch())|| /* _id/_ID as id */
+ isletter(c))
+ { word t;
+ kollect(okid);
+ t=identifier(0);
+ return(t==NAME?INFIXNAME:t==CNAME?INFIXCNAME:'$'); }
+ /* the last alternative is an error - caveat */
+ if('1'<=c&&c<='9')
+ { word n=0;
+ while(isdigit(c)&&n<1e6)n=10*n+c-'0',c=getch();
+ if(n>sreds)
+ /* sreds==0 everywhere except in semantic redn clause */
+ printf("%ssyntax error: illegal symbol $%d%s\n",
+ echoing?"\n":"",n,n>=1e6?"...":""),
+ acterror();
+ else { yylval=mkgvar(n); return(NAME); }
+ }
+ if(c=='-')
+ { if(!compiling)
+ syntax("unexpected symbol $-\n"); else
+ {c=getch(); yylval=common_stdin; return(CONST); }}
+ /* NB we disallow recursive use of $($/+/-) inside $+ data
+ whence addition of `compiling' to premises */
+ if(c==':')
+ { c=getch();
+ if(c!='-')syntax("unexpected symbol $:\n"); else
+ { if(!compiling)
+ syntax("unexpected symbol $:-\n"); else
+ {c=getch(); yylval=common_stdinb; return(CONST); }}} /* $:- */
+ if(c=='+')
+ { /* if(!(commandmode&&compiling||magic))
+ syntax("unexpected symbol $+\n"); else /* disallow in scripts */
+ if(!compiling)
+ syntax("unexpected symbol $+\n"); else
+ { c=getch();
+ if(commandmode)
+ yylval=cook_stdin;
+ else yylval=ap(readvals(0,0),OFFSIDE);
+ return(CONST); }}
+ if(c=='$')
+ { if(!(inlex==2||commandmode&&compiling))
+ syntax("unexpected symbol $$\n"); else
+ { c=getch();
+ if(inlex) { yylval=mklexvar(0); return(NAME); }
+ else return(DOLLAR2); }}
+ if(c=='#')
+ { if(inlex!=2)syntax("unexpected symbol $#\n"); else
+ { c=getch(); yylval=mklexvar(1); return(NAME); }}
+ if(c=='*')
+ { c=getch(); yylval=ap(GETARGS,0); return(CONST); }
+ if(c=='0')
+ syntax("illegal symbol $0\n");
+ default: return(lastc);
+}}
+
+layout()
+{L:while(c==' '||c=='\n'&&!commandmode||c=='\t') c= getch();
+ if(c==EOF&&commandmode){ c='\n'; return; }
+ if(c=='|'&&peekch()=='|' /* ||comments */
+ || col==1&&line_no==1 /* added 19.11.2013 */
+ &&c=='#'&&peekch()=='!') /* UNIX magic string */
+ { while((c=getch())!='\n'&&c!=EOF);
+ if(c==EOF&&!commandmode)return;
+ c= '\n';
+ goto L; }
+}
+
+collectstars()
+{ word n=2;
+ while(c=='*')c=getch(),n++;
+ yylval= mktvar(n);
+ return(TYPEVAR);
+}
+
+word gvars=NIL; /* list of grammar variables - no need to reset */
+
+mkgvar(i) /* make bound variable (corresponding to $i in bnf rule) */
+word i;
+{ word *p= &gvars;
+ while(--i)
+ { if(*p==NIL)*p=cons(sto_id("gvar"),NIL);
+ p= &tl[*p]; }
+ if(*p==NIL)*p=cons(sto_id("gvar"),NIL);
+ return(hd[*p]);
+} /* all these variables have the same name, and are not in hashbucket */
+
+word lexvar=0;
+
+mklexvar(i) /* similar - corresponds to $$, $# on rhs of %lex rule */
+word i; /* i=0 or 1 */
+{ extern word ltchar;
+ if(!lexvar)
+ lexvar=cons(sto_id("lexvar"),sto_id("lexvar")),
+ id_type(hd[lexvar])=ltchar,
+ id_type(tl[lexvar])=genlstat_t();
+ return(i?tl[lexvar]:hd[lexvar]);
+}
+
+word ARGC;
+char **ARGV; /* initialised in main(), see steer.c */
+
+conv_args() /* used to give access to command line args
+ see case GETARGS in reduce.c */
+{ word i=ARGC,x=NIL;
+ if(i==0)return(NIL); /* possible only if not invoked from a magic script */
+ { while(--i)x=cons(str_conv(ARGV[i]),x);
+ x=cons(str_conv(ARGV[0]),x); }
+ return(x);
+}
+
+str_conv(s) /* convert C string to Miranda form */
+char *s;
+{ word x=NIL,i=strlen(s);
+ while(i--)x=cons(s[i],x);
+ return(x);
+} /* opposite of getstring() - see reduce.c */
+
+okpath(ch)
+word ch;
+{ return(ch!='\"'&&ch!='\n'&&ch!='>'); }
+
+char *pathname() /* returns NULL if not valid pathname (in string quotes) */
+{ layout();
+ if(c=='<') /* alternative quotes <..> for system libraries */
+ { extern char *miralib;
+ char *hold=dicp;
+ c=getch();
+ (void)strcpy(dicp,miralib);
+ dicp+=strlen(miralib);
+ *dicp++ = '/';
+ kollect(okpath);
+ dicp=hold;
+ if(c!='>')return(NULL);
+ c=' ';
+ return(dicp); }
+ if(c!='\"')return(NULL);
+ c=getch();
+ if(c=='~')
+ { char *h,*hold=dicp;
+ extern char linebuf[];
+ *dicp++ = c;
+ c=getch();
+ while(isalnum(c)||c=='-'||c=='_'||c=='.')
+ *dicp++ = c, c=getch();
+ *dicp='\0';
+ if(h=gethome(hold+1))
+ (void)strcpy(hold,h),dicp=hold+strlen(hold);
+ else (void)strcpy(&linebuf[0],hold),
+ (void)strcpy(hold,prefixbase+prefix),
+ dicp=hold+strlen(prefixbase+prefix),
+ (void)strcpy(dicp,&linebuf[0]),
+ dicp+=strlen(dicp);
+ kollect(okpath);
+ dicp=hold;
+ } else
+ if(c=='/') /* absolute pathname */
+ kollect(okpath);
+ else { /* relative pathname */
+ char *hold=dicp;
+ (void)strcpy(dicp,prefixbase+prefix);
+ dicp+=strlen(prefixbase+prefix);
+ kollect(okpath);
+ dicp=hold; }
+ if(c!='\"')return(NULL);
+ c = ' ';
+ return(dicp);
+} /* result is volatile - call keep(dicp) to retain */
+
+adjust_prefix(f) /* called at %insert and at loadfile, to get static pathname
+ resolution */
+char *f;
+{ /* the directory part of the pathname f becomes the new
+ prefix for pathnames, and we stack the current prefix */
+ char *g;
+ prefixstack=strcons(prefix,prefixstack);
+ prefix += strlen(prefixbase+prefix)+1;
+ while(prefix+strlen(f)>=prefixlimit) /* check and fix overflow */
+ prefixlimit += 1024, prefixbase=realloc(prefixbase,prefixlimit);
+ (void)strcpy(prefixbase+prefix,f);
+ g=rindex(prefixbase+prefix,'/');
+ if(g)g[1]='\0';
+ else prefixbase[prefix]='\0';
+}
+
+/* NOTES on how static pathname resolution is achieved:
+(the specification is that pathnames must always be resolved relative to the
+file in which they are encountered)
+Definition -- the 'prefix' of a pathname is the initial segment up to but not
+including the last occurrence of '/' (null if no '/' present).
+Keep the wd constant during compilation. Have a global char* prefix, initially
+null.
+1) Whenever you read a relative pathname(), insert 'prefix' on the front of it.
+2) On entering a new level of insert, stack old prefix and prefix becomes that
+ of new file name. Done by calling adjust_prefix().
+3) On quitting a level of insert, unstack old prefix.
+*/
+
+peekdig()
+{ word ch = getc(s_in);
+ ungetc(ch,s_in);
+ return('0'<=ch&&ch<='9');
+}
+
+peekch()
+{ word ch = getc(s_in);
+ ungetc(ch,s_in);
+ return(ch);
+}
+
+openfile(n) /* returns 0 or 1 as indication of success - puts file on fileq
+ if successful */
+char *n;
+{ FILE *f;
+ f= fopen(n,"r");
+ if(f==NULL)return(0);
+ fileq= cons(strcons(f,NIL),fileq);
+ insertdepth++;
+ return(1);
+}
+
+identifier(s) /* recognises reserved words */
+word s; /* flags looking for ul reserved words only */
+{ extern word lastid,initialising;
+ if(inbnf==1)
+ { /* only reserved nonterminals are `empty', `end', `error', `where' */
+ if(is("empty ")||is("e_m_p_t_y"))return(EMPTYSY); else
+ if(is("end ")||is("e_n_d"))return(ENDSY); else
+ if(is("error ")||is("e_r_r_o_r"))return(ERRORSY); else
+ if(is("where ")||is("w_h_e_r_e"))return(WHERE); }
+ else
+ switch(dicp[0])
+ { case 'a': if(is("abstype")||is("a_b_s_t_y_p_e"))
+ return(ABSTYPE);
+ break;
+ case 'd': if(is("div")||is("d_i_v"))
+ return(DIV);
+ break;
+ case 'F': if(is("False")) /* True, False alleged to be predefined, not
+ reserved (??) */
+ { yylval = False;
+ return(CONST); }
+ break;
+ case 'i': if(is("if")||is("i_f"))
+ return(IF);
+ break;
+ case 'm': if(is("mod")||is("m_o_d"))
+ return(REM);
+ break;
+ case 'o': if(is("otherwise")||is("o_t_h_e_r_w_i_s_e"))
+ return(OTHERWISE);
+ break;
+ case 'r': if(is("readvals")||is("r_e_a_d_v_a_l_s"))
+ return(READVALSY);
+ break;
+ case 's': if(is("show")||is("s_h_o_w"))
+ return(SHOWSYM);
+ break;
+ case 'T': if(is("True"))
+ { yylval = True;
+ return(CONST); }
+ case 't': if(is("type")||is("t_y_p_e"))
+ return(TYPE);
+ break;
+ case 'w': if(is("where")||is("w_h_e_r_e"))
+ return(WHERE);
+ if(is("with")||is("w_i_t_h"))
+ return(WITH);
+ break;
+ }
+ if(s){ syntax("illegal use of underlining\n"); return('_'); }
+ yylval=name(); /* not a reserved word */
+ if(commandmode&&lastid==0&&id_type(yylval)!=undef_t)lastid=yylval;
+ return(isconstructor(yylval)?CNAME:NAME);
+}
+
+word disgusting=0; /* flag to turn off typecheck, temporary hack for jrc */
+
+directive() /* these are of the form "%identifier" */
+{ extern word SYNERR,magic;
+ word holdcol=col-1,holdlin=line_no;
+ c = getch();
+ if(c=='%'){ c=getch(); return(ENDIR); }
+ kollect(okulid);
+ switch(dicp[0]=='_'&&dicp[1]==''?dicp[2]:dicp[0])
+ { case 'b': if(is("begin")||is("_^Hb_^He_^Hg_^Hi_^Hn"))
+ if(inlex)
+ return(LBEGIN);
+ if(is("bnf")||is("_^Hb_^Hn_^Hf"))
+ { setlmargin(); col=holdcol+4;
+ /* `indent' to right hand end of directive */
+ return(BNF); }
+ break;
+ case 'e': if(is("export")||is("_e_x_p_o_r_t"))
+ { if(magic)syntax(
+ "%export directive not permitted in \"-exp\" script\n");
+ return(EXPORT); }
+ break;
+ case 'f': if(is("free")||is("_f_r_e_e"))
+ { if(magic)syntax(
+ "%free directive not permitted in \"-exp\" script\n");
+ return(FREE); }
+ break;
+ case 'i': if(is("include")||is("_i_n_c_l_u_d_e"))
+ { if(!SYNERR){ layout(); setlmargin(); }
+ /* does `indent' for grammar */
+ if(!pathname())
+ syntax("bad pathname after %include\n");
+ else yylval=strcons(addextn(1,dicp),
+ fileinfo(get_fil(current_file),holdlin)),
+ /* (includee,hereinfo) */
+ keep(dicp);
+ return(INCLUDE); }
+ if(is("insert")||is("_i_n_s_e_r_t"))
+ { char *f=pathname();
+ if(!f)syntax("bad pathname after %insert\n"); else
+ if(insertdepth<12&&openfile(f))
+ { adjust_prefix(f);
+ vergstack=cons(lverge,vergstack);
+ echostack=cons(echoing,echostack);
+ litstack=cons(literate,litstack);
+ linostack=strcons(line_no,linostack);
+ line_no=0; atnl=1; /* was line_no=1; */
+ keep(dicp);
+ current_file = make_fil(f,fm_time(f),0,NIL);
+ files = append1(files,cons(current_file,NIL));
+ tl[hd[fileq]] = current_file;
+ s_in = (FILE *)hd[hd[fileq]];
+ literate= peekch()=='>'||litname(f);
+ col=lverge=holdcol;
+ if(echoing)
+ { putchar('\n');
+ if(!literate)
+ if(litmain)putchar('>'),spaces(holdcol);
+ else spaces(holdcol); }
+ c = getch(); } /* used to precede previous cmd when echo
+ was delayed by one char, see getch() */
+ else { word toomany=(insertdepth>=12);
+ printf("%s%%insert error - cannot open \"%s\"\n",
+ echoing?"\n":"",f);
+ keep(dicp);
+ if(toomany)printf(
+ "too many nested %%insert directives (limit=%d)\n",
+ insertdepth);
+ else
+ files = append1(files,cons(make_fil(f,0,0,NIL),NIL));
+ /* line above for benefit of `oldfiles' */
+ acterror(); }
+ return(yylex()); }
+ break;
+ case 'l': if(is("lex")||is("_^Hl_^He_^Hx"))
+ { if(inlex)syntax("nested %lex not permitted\n");
+ /* due to use of global vars inlex, lexdefs */
+ return(LEX); }
+ if(is("list")||is("_l_i_s_t"))
+ { echoing=verbosity; return(yylex()); }
+ break;
+ case 'n': if(is("nolist")||is("_n_o_l_i_s_t"))
+ { echoing=0; return(yylex()); }
+ break;
+ }
+ if(echoing)putchar('\n');
+ printf("syntax error: unknown directive \"%%%s\"\n",dicp),
+ acterror();
+ return(END);
+}
+
+okid(ch)
+word ch;
+{ return('a'<=ch&&ch<='z'||'A'<=ch&&ch<='Z'||'0'<=ch&&ch<='9'
+ ||ch=='_'||ch=='\''); }
+
+okulid(ch)
+word ch;
+{ return('a'<=ch&&ch<='z'||'A'<=ch&&ch<='Z'||'0'<=ch&&ch<='9'
+ ||ch=='_'||ch==''||ch=='\''); }
+
+kollect(f)
+/* note top of dictionary used as work space to collect current token */
+word (*f)();
+{ dicq= dicp;
+ while((*f)(c)){ *dicq++ = c; c= getch(); }
+ *dicq++ = '\0';
+ ovflocheck;
+}
+
+char *keep(p) /* call this to retain volatile string for later use */
+char *p;
+{ if(p==dicp)dicp= dicq;
+ else (void)strcpy(dicp,p),
+ p=dicp,
+ dicp=dicq=dicp+strlen(dicp)+1,
+ dic_check();
+ return(p);
+}
+
+dic_check() /* called from REDUCE */
+{ ovflocheck; }
+
+numeral()
+{ word nflag=1;
+ dicq= dicp;
+ while(isdigit(c))
+ *dicq++ = c, c=getch();
+ if(c=='.'&&peekdig())
+ { *dicq++ = c, c=getch(); nflag=0;
+ while(isdigit(c))
+ *dicq++ = c, c=getch(); }
+ if(c=='e')
+ { word np=0;
+ *dicq++ = c, c=getch(); nflag=0;
+ if(c=='+')c=getch(); else /* ignore + before exponent */
+ if(c=='-')*dicq++ = c, c=getch();
+ if(!isdigit(c)) /* e must be followed by some digits */
+ syntax("badly formed floating point number\n");
+ while(c=='0')
+ *dicq++ = c, c=getch();
+ while(isdigit(c))
+ np++, *dicq++ = c, c=getch();
+ if(!nflag&&np>3) /* scanf falls over with silly exponents */
+ { syntax("floating point number out of range\n");
+ return; }
+ }
+ ovflocheck;
+ if(nflag) /* `.' or `e' makes fractional */
+ *dicq = '\0',
+ yylval= bigscan(dicp); else
+ { double r=0.0;
+ if(dicq-dicp>60) /* this allows 59 chars */
+ /* scanf crashes, on VAX, gives wrong answers, on ORION 1/05 */
+ { syntax("illegal floating point constant (too many digits)\n");
+ return; }
+ errno=0;
+ *dicq = '\n';
+ sscanf(dicp,"%lf",&r);
+ if(errno)fpe_error(); else
+ yylval= sto_dbl((double)r); }
+}
+
+hexnumeral() /* added 21.11.2013 */
+{ extern word errno;
+ word nflag=1;
+ dicq= dicp;
+ *dicq++ = c, c=getch(); /* 0 */
+ *dicq++ = c, c=getch(); /* x */
+ if(!isxdigit(c)&&c!='.')syntax("malformed hex number\n");
+ while(c=='0'&&isxdigit(peekch()))c=getch(); /* skip zeros before first nonzero digit */
+ while(isxdigit(c))
+ *dicq++ = c, c=getch();
+ ovflocheck;
+ if(c=='.'||tolower(c)=='p') /* hex float, added 20.11.19 */
+ { double d;
+ if(c=='.')
+ { *dicq++ = c, c=getch();
+ while(isxdigit(c))
+ *dicq++ = c, c=getch(); }
+ if(c=='p')
+ { *dicq++ = c, c=getch();
+ if(c=='+'||c=='-')*dicq++ = c, c=getch();
+ if(!isdigit(c))syntax("malformed hex float\n");
+ while(isdigit(c))
+ *dicq++ = c, c=getch(); }
+ ovflocheck;
+ *dicq='\0';
+ if(dicq-dicp>60||sscanf(dicp,"%lf",&d)!=1)
+ syntax("malformed hex float\n");
+ else yylval= sto_dbl(d);
+ return; }
+ *dicq = '\0';
+ yylval= bigxscan(dicp+2,dicq);
+}
+
+octnumeral() /* added 21.11.2013 */
+{ extern word errno;
+ word nflag=1;
+ dicq= dicp;
+ if(!isdigit(c))syntax("malformed octal number\n");
+ while(c=='0'&&isdigit(peekch()))c=getch(); /* skip zeros before first nonzero digit */
+ while(isdigit(c)&&c<='7')
+ *dicq++ = c, c=getch();
+ if(isdigit(c))syntax("illegal digit in octal number\n");
+ ovflocheck;
+ *dicq = '\0';
+ yylval= bigoscan(dicp,dicq);
+}
+
+word namebucket[128]; /* each namebucket has a list terminated by 0, not NIL */
+
+hash(s) /* returns a value in {0..127} */
+char *s;
+{ word h = *s;
+ if(h)while(*++s)h ^= *s; /* guard necessary to deal with s empty */
+ return(h&127);
+}
+
+isconstrname(s)
+char *s;
+{ if(s[0]=='$')s++;
+ return isupper(*s); /* formerly !islower */
+}
+
+getfname(x)
+/* nonterminals have an added ' ', getfname returns the corresponding
+ function name */
+word x;
+{ char *p = get_id(x);
+ dicq= dicp;
+ while(*dicq++ = *p++);
+ if(dicq-dicp<3)fprintf(stderr,"impossible event in getfname\n"),exit(1);
+ dicq[-2] = '\0'; /* overwrite last char */
+ ovflocheck;
+ return(name());
+}
+
+isnonterminal(x)
+word x;
+{ char *n;
+ if(tag[x]!=ID)return(0);
+ n = get_id(x);
+ return(n[strlen(n)-1]==' ');
+}
+
+name()
+{ word q,h;
+ q= namebucket[h=hash(dicp)];
+ while(q&&!is(get_id(hd[q])))q= tl[q];
+ if(q==0)
+ { q = sto_id(dicp);
+ namebucket[h] = cons(q,namebucket[h]);
+ keep(dicp); }
+ else q= hd[q];
+ return(q); }
+/* note - keeping buckets sorted didn't seem to help (if anything slightly
+ slower) probably because ordering only relevant if name not present, and
+ outweighed by increased complexity of loop */
+
+static word inprelude=1;
+
+make_id(n) /* used in mira_setup(), primdef(), predef(), all in steer.c */
+char *n;
+{ word x,h;
+ h=hash(n);
+ x = sto_id(inprelude?keep(n):n);
+ namebucket[h] = cons(x,namebucket[h]);
+ return(x); }
+
+findid(n) /* like name() but returns NIL rather than create new id */
+char *n;
+{ word q;
+ q= namebucket[hash(n)];
+ while(q&&!strcmp(n,get_id(hd[q]))==0)q= tl[q];
+ return(q?hd[q]:NIL); }
+
+word *pnvec=0,nextpn,pn_lim=200; /* private name vector */
+
+reset_pns() /* (re)initialise private name space */
+{ nextpn=0;
+ if(!pnvec)
+ { pnvec=(word *)malloc(pn_lim*sizeof(word));
+ if(pnvec==NULL)mallocfail("pnvec"); }
+}
+
+make_pn(val) /* create new private name with value val */
+word val;
+{ if(nextpn==pn_lim)
+ { pn_lim+=400;
+ pnvec=(word *)realloc(pnvec,pn_lim*sizeof(word));
+ if(pnvec==NULL)mallocfail("pnvec"); }
+ pnvec[nextpn]=strcons(nextpn,val);
+ return(pnvec[nextpn++]);
+}
+
+sto_pn(n) /* return n'th private name, extending pnvec if necessary */
+word n;
+{ if(n>=pn_lim)
+ { while(pn_lim<=n)pn_lim+=400;
+ pnvec=(word *)realloc(pnvec,pn_lim*sizeof(word));
+ if(pnvec==NULL)mallocfail("pnvec"); }
+ while(nextpn<=n) /* NB allocates all missing names upto and including nth*/
+ pnvec[nextpn]=strcons(nextpn,UNDEF),nextpn++;
+ return(pnvec[n]);
+}
+
+mkprivate(x) /* disguise identifiers prior to removal from environment */
+word x; /* used in setting up prelude - see main() in steer.c */
+{ while(x!=NIL)
+ { char *s = get_id(hd[x]);
+ get_id(hd[x])[0] += 128; /* hack to make private internal name */
+ x = tl[x]; } /* NB - doesn't change hashbucket */
+ inprelude=0;
+}
+
+word sl=100;
+
+string()
+{ word p;
+ word ch,badch=0;
+ c = getch();
+ ch= getlitch();
+ p= yylval= cons(NIL,NIL);
+ while(ch!=EOF&&rawch!='\"'&&rawch!='\n')
+ if(ch==-7) ch=getlitch(); else /* skip \& */
+ if(ch<0){ badch=ch; break; }
+ else { p= tl[p]= cons(ch,NIL);
+ ch= getlitch(); }
+ yylval= tl[yylval];
+ if(badch)errclass(badch,1);
+ if(rawch=='\n')
+ syntax("non-escaped newline encountered inside string quotes\n"); else
+ if(ch==EOF)
+ { if(echoing)putchar('\n');
+ printf("syntax error: script ends inside unclosed string quotes - \n");
+ printf(" \"");
+ while(yylval!=NIL&& sl-- )
+ { putchar(hd[yylval]);
+ yylval= tl[yylval]; }
+ printf("...\"\n");
+ acterror(); }
+}
+
+charclass()
+{ word p;
+ word ch,badch=0,anti=0;
+ c = getch();
+ if(c=='^')anti=1,c=getch();
+ ch= getlitch();
+ p= yylval= cons(NIL,NIL);
+ while(ch!=EOF&&rawch!='`'&&rawch!='\n')
+ if(ch==-7)ch=getlitch(); else /* skip \& */
+ if(ch<0){ badch=ch; break; }
+ else { if(rawch=='-'&&hd[p]!=NIL&&hd[p]!=DOTDOT)
+ ch=DOTDOT; /* non-initial, non-escaped '-' */
+ p= tl[p]= cons(ch,NIL);
+ ch= getlitch(); }
+ if(hd[p]==DOTDOT)hd[p]='-'; /* naturalise a trailing '-' */
+ for(p=yylval;tl[p]!=NIL;p=tl[p]) /* move each DOTDOT to front of range */
+ if(hd[tl[p]]==DOTDOT)
+ { hd[tl[p]]=hd[p],hd[p]=DOTDOT;
+ if(hd[tl[p]]>=hd[tl[tl[p]]])
+ syntax("illegal use of '-' in [charclass]\n");
+ }
+ yylval= tl[yylval];
+ if(badch)errclass(badch,2);
+ if(rawch=='\n')
+ syntax("non-escaped newline encountered in char class\n"); else
+ if(ch==EOF)
+ { if(echoing)putchar('\n');
+ printf(
+ "syntax error: script ends inside unclosed char class brackets - \n");
+ printf(" [");
+ while(yylval!=NIL&& sl-- )
+ { putchar(hd[yylval]);
+ yylval= tl[yylval]; }
+ printf("...]\n");
+ acterror(); }
+ return(anti);
+}
+
+reset_lex() /* called after an error */
+{ extern word errs,errline;
+ extern char *current_script;
+ /*printf("reset_lex()\n"); /* DEBUG */
+ if(!commandmode)
+ { if(!errs)errs=fileinfo(get_fil(current_file),line_no);
+ /* convention, if errs set contains location of error, otherwise pick up
+ from current_file and line_no */
+ if(tl[errs]==0&&(char *)hd[errs]==current_script)
+ /* at end of file, so line_no has been reset to 0 */
+ printf("error occurs at end of ");
+ else printf("error found near line %d of ",tl[errs]);
+ printf("%sfile \"%s\"\ncompilation abandoned\n",
+ (char *)hd[errs]==current_script?"":"%insert ",
+ (char *)hd[errs]);
+ if((char *)hd[errs]==current_script)
+ errline=tl[errs]==0?lastline:tl[errs],errs=0;
+ else { while(tl[linostack]!=NIL)linostack=tl[linostack];
+ errline=hd[linostack]; }
+ /* tells editor where to find error - errline contains location of 1st
+ error in main script, errs is hereinfo of upto one error in %insert
+ script (each is 0 if not set) - some errors can set both */
+ }
+ reset_state();
+}
+
+reset_state() /* reset all global variables used by compiler */
+{ extern word TABSTRS,SGC,newtyps,algshfns,showchain,inexplist,sreds,
+ rv_script,idsused;
+ /* printf("reset_state()\n"); /* DEBUG */
+ if(commandmode)
+ while(c!='\n'&&c!=EOF)c=getc(s_in); /* no echo */
+ while(fileq!=NIL)fclose((FILE *)hd[hd[fileq]]),fileq=tl[fileq];
+ insertdepth= -1;
+ s_in=stdin;
+ echostack=idsused=prefixstack=litstack=linostack=vergstack
+ =margstack=NIL;
+ prefix=0; prefixbase[0]='\0';
+ echoing=verbosity&listing;
+ brct=inbnf=sreds=inlex=inexplist=commandmode=lverge=col=lmargin=0;
+ atnl=1;
+ rv_script=0;
+ algshfns=newtyps=showchain=SGC=TABSTRS=NIL;
+ c=' ';
+ line_no=0;
+ litmain=literate=0;
+ /* printf("exit reset_state()\n"); /* DEBUG */
+}
+
+/* end of MIRANDA LEX ANALYSER */
+