sl@0: /*
sl@0:  * re_*comp and friends - compile REs
sl@0:  * This file #includes several others (see the bottom).
sl@0:  *
sl@0:  * Copyright (c) 1998, 1999 Henry Spencer.  All rights reserved.
sl@0:  * 
sl@0:  * Development of this software was funded, in part, by Cray Research Inc.,
sl@0:  * UUNET Communications Services Inc., Sun Microsystems Inc., and Scriptics
sl@0:  * Corporation, none of whom are responsible for the results.  The author
sl@0:  * thanks all of them. 
sl@0:  * 
sl@0:  * Redistribution and use in source and binary forms -- with or without
sl@0:  * modification -- are permitted for any purpose, provided that
sl@0:  * redistributions in source form retain this entire copyright notice and
sl@0:  * indicate the origin and nature of any modifications.
sl@0:  * 
sl@0:  * I'd appreciate being given credit for this package in the documentation
sl@0:  * of software which uses it, but that is not a requirement.
sl@0:  * 
sl@0:  * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES,
sl@0:  * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY
sl@0:  * AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL
sl@0:  * HENRY SPENCER BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
sl@0:  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
sl@0:  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
sl@0:  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
sl@0:  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
sl@0:  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
sl@0:  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
sl@0:  *
sl@0:  */
sl@0: 
sl@0: #include "regguts.h"
sl@0: 
sl@0: /*
sl@0:  * forward declarations, up here so forward datatypes etc. are defined early
sl@0:  */
sl@0: /* =====^!^===== begin forwards =====^!^===== */
sl@0: /* automatically gathered by fwd; do not hand-edit */
sl@0: /* === regcomp.c === */
sl@0: int compile _ANSI_ARGS_((regex_t *, CONST chr *, size_t, int));
sl@0: static VOID moresubs _ANSI_ARGS_((struct vars *, int));
sl@0: static int freev _ANSI_ARGS_((struct vars *, int));
sl@0: static VOID makesearch _ANSI_ARGS_((struct vars *, struct nfa *));
sl@0: static struct subre *parse _ANSI_ARGS_((struct vars *, int, int, struct state *, struct state *));
sl@0: static struct subre *parsebranch _ANSI_ARGS_((struct vars *, int, int, struct state *, struct state *, int));
sl@0: static VOID parseqatom _ANSI_ARGS_((struct vars *, int, int, struct state *, struct state *, struct subre *));
sl@0: static VOID nonword _ANSI_ARGS_((struct vars *, int, struct state *, struct state *));
sl@0: static VOID word _ANSI_ARGS_((struct vars *, int, struct state *, struct state *));
sl@0: static int scannum _ANSI_ARGS_((struct vars *));
sl@0: static VOID repeat _ANSI_ARGS_((struct vars *, struct state *, struct state *, int, int));
sl@0: static VOID bracket _ANSI_ARGS_((struct vars *, struct state *, struct state *));
sl@0: static VOID cbracket _ANSI_ARGS_((struct vars *, struct state *, struct state *));
sl@0: static VOID brackpart _ANSI_ARGS_((struct vars *, struct state *, struct state *));
sl@0: static chr *scanplain _ANSI_ARGS_((struct vars *));
sl@0: static VOID leaders _ANSI_ARGS_((struct vars *, struct cvec *));
sl@0: static VOID onechr _ANSI_ARGS_((struct vars *, pchr, struct state *, struct state *));
sl@0: static VOID dovec _ANSI_ARGS_((struct vars *, struct cvec *, struct state *, struct state *));
sl@0: static celt nextleader _ANSI_ARGS_((struct vars *, pchr, pchr));
sl@0: static VOID wordchrs _ANSI_ARGS_((struct vars *));
sl@0: static struct subre *subre _ANSI_ARGS_((struct vars *, int, int, struct state *, struct state *));
sl@0: static VOID freesubre _ANSI_ARGS_((struct vars *, struct subre *));
sl@0: static VOID freesrnode _ANSI_ARGS_((struct vars *, struct subre *));
sl@0: static VOID optst _ANSI_ARGS_((struct vars *, struct subre *));
sl@0: static int numst _ANSI_ARGS_((struct subre *, int));
sl@0: static VOID markst _ANSI_ARGS_((struct subre *));
sl@0: static VOID cleanst _ANSI_ARGS_((struct vars *));
sl@0: static long nfatree _ANSI_ARGS_((struct vars *, struct subre *, FILE *));
sl@0: static long nfanode _ANSI_ARGS_((struct vars *, struct subre *, FILE *));
sl@0: static int newlacon _ANSI_ARGS_((struct vars *, struct state *, struct state *, int));
sl@0: static VOID freelacons _ANSI_ARGS_((struct subre *, int));
sl@0: static VOID rfree _ANSI_ARGS_((regex_t *));
sl@0: static VOID dump _ANSI_ARGS_((regex_t *, FILE *));
sl@0: static VOID dumpst _ANSI_ARGS_((struct subre *, FILE *, int));
sl@0: static VOID stdump _ANSI_ARGS_((struct subre *, FILE *, int));
sl@0: static char *stid _ANSI_ARGS_((struct subre *, char *, size_t));
sl@0: /* === regc_lex.c === */
sl@0: static VOID lexstart _ANSI_ARGS_((struct vars *));
sl@0: static VOID prefixes _ANSI_ARGS_((struct vars *));
sl@0: static VOID lexnest _ANSI_ARGS_((struct vars *, chr *, chr *));
sl@0: static VOID lexword _ANSI_ARGS_((struct vars *));
sl@0: static int next _ANSI_ARGS_((struct vars *));
sl@0: static int lexescape _ANSI_ARGS_((struct vars *));
sl@0: static chr lexdigits _ANSI_ARGS_((struct vars *, int, int, int));
sl@0: static int brenext _ANSI_ARGS_((struct vars *, pchr));
sl@0: static VOID skip _ANSI_ARGS_((struct vars *));
sl@0: static chr newline _ANSI_ARGS_((NOPARMS));
sl@0: #ifdef REG_DEBUG
sl@0: static chr *ch _ANSI_ARGS_((NOPARMS));
sl@0: #endif
sl@0: static chr chrnamed _ANSI_ARGS_((struct vars *, chr *, chr *, pchr));
sl@0: /* === regc_color.c === */
sl@0: static VOID initcm _ANSI_ARGS_((struct vars *, struct colormap *));
sl@0: static VOID freecm _ANSI_ARGS_((struct colormap *));
sl@0: static VOID cmtreefree _ANSI_ARGS_((struct colormap *, union tree *, int));
sl@0: static color setcolor _ANSI_ARGS_((struct colormap *, pchr, pcolor));
sl@0: static color maxcolor _ANSI_ARGS_((struct colormap *));
sl@0: static color newcolor _ANSI_ARGS_((struct colormap *));
sl@0: static VOID freecolor _ANSI_ARGS_((struct colormap *, pcolor));
sl@0: static color pseudocolor _ANSI_ARGS_((struct colormap *));
sl@0: static color subcolor _ANSI_ARGS_((struct colormap *, pchr c));
sl@0: static color newsub _ANSI_ARGS_((struct colormap *, pcolor));
sl@0: static VOID subrange _ANSI_ARGS_((struct vars *, pchr, pchr, struct state *, struct state *));
sl@0: static VOID subblock _ANSI_ARGS_((struct vars *, pchr, struct state *, struct state *));
sl@0: static VOID okcolors _ANSI_ARGS_((struct nfa *, struct colormap *));
sl@0: static VOID colorchain _ANSI_ARGS_((struct colormap *, struct arc *));
sl@0: static VOID uncolorchain _ANSI_ARGS_((struct colormap *, struct arc *));
sl@0: static int singleton _ANSI_ARGS_((struct colormap *, pchr c));
sl@0: static VOID rainbow _ANSI_ARGS_((struct nfa *, struct colormap *, int, pcolor, struct state *, struct state *));
sl@0: static VOID colorcomplement _ANSI_ARGS_((struct nfa *, struct colormap *, int, struct state *, struct state *, struct state *));
sl@0: #ifdef REG_DEBUG
sl@0: static VOID dumpcolors _ANSI_ARGS_((struct colormap *, FILE *));
sl@0: static VOID fillcheck _ANSI_ARGS_((struct colormap *, union tree *, int, FILE *));
sl@0: static VOID dumpchr _ANSI_ARGS_((pchr, FILE *));
sl@0: #endif
sl@0: /* === regc_nfa.c === */
sl@0: static struct nfa *newnfa _ANSI_ARGS_((struct vars *, struct colormap *, struct nfa *));
sl@0: static VOID freenfa _ANSI_ARGS_((struct nfa *));
sl@0: static struct state *newstate _ANSI_ARGS_((struct nfa *));
sl@0: static struct state *newfstate _ANSI_ARGS_((struct nfa *, int flag));
sl@0: static VOID dropstate _ANSI_ARGS_((struct nfa *, struct state *));
sl@0: static VOID freestate _ANSI_ARGS_((struct nfa *, struct state *));
sl@0: static VOID destroystate _ANSI_ARGS_((struct nfa *, struct state *));
sl@0: static VOID newarc _ANSI_ARGS_((struct nfa *, int, pcolor, struct state *, struct state *));
sl@0: static struct arc *allocarc _ANSI_ARGS_((struct nfa *, struct state *));
sl@0: static VOID freearc _ANSI_ARGS_((struct nfa *, struct arc *));
sl@0: static struct arc *findarc _ANSI_ARGS_((struct state *, int, pcolor));
sl@0: static VOID cparc _ANSI_ARGS_((struct nfa *, struct arc *, struct state *, struct state *));
sl@0: static VOID moveins _ANSI_ARGS_((struct nfa *, struct state *, struct state *));
sl@0: static VOID copyins _ANSI_ARGS_((struct nfa *, struct state *, struct state *));
sl@0: static VOID moveouts _ANSI_ARGS_((struct nfa *, struct state *, struct state *));
sl@0: static VOID copyouts _ANSI_ARGS_((struct nfa *, struct state *, struct state *));
sl@0: static VOID cloneouts _ANSI_ARGS_((struct nfa *, struct state *, struct state *, struct state *, int));
sl@0: static VOID delsub _ANSI_ARGS_((struct nfa *, struct state *, struct state *));
sl@0: static VOID deltraverse _ANSI_ARGS_((struct nfa *, struct state *, struct state *));
sl@0: static VOID dupnfa _ANSI_ARGS_((struct nfa *, struct state *, struct state *, struct state *, struct state *));
sl@0: static VOID duptraverse _ANSI_ARGS_((struct nfa *, struct state *, struct state *));
sl@0: static VOID cleartraverse _ANSI_ARGS_((struct nfa *, struct state *));
sl@0: static VOID specialcolors _ANSI_ARGS_((struct nfa *));
sl@0: static long optimize _ANSI_ARGS_((struct nfa *, FILE *));
sl@0: static VOID pullback _ANSI_ARGS_((struct nfa *, FILE *));
sl@0: static int pull _ANSI_ARGS_((struct nfa *, struct arc *));
sl@0: static VOID pushfwd _ANSI_ARGS_((struct nfa *, FILE *));
sl@0: static int push _ANSI_ARGS_((struct nfa *, struct arc *));
sl@0: #define	INCOMPATIBLE	1	/* destroys arc */
sl@0: #define	SATISFIED	2	/* constraint satisfied */
sl@0: #define	COMPATIBLE	3	/* compatible but not satisfied yet */
sl@0: static int combine _ANSI_ARGS_((struct arc *, struct arc *));
sl@0: static VOID fixempties _ANSI_ARGS_((struct nfa *, FILE *));
sl@0: static int unempty _ANSI_ARGS_((struct nfa *, struct arc *));
sl@0: static VOID cleanup _ANSI_ARGS_((struct nfa *));
sl@0: static VOID markreachable _ANSI_ARGS_((struct nfa *, struct state *, struct state *, struct state *));
sl@0: static VOID markcanreach _ANSI_ARGS_((struct nfa *, struct state *, struct state *, struct state *));
sl@0: static long analyze _ANSI_ARGS_((struct nfa *));
sl@0: static VOID compact _ANSI_ARGS_((struct nfa *, struct cnfa *));
sl@0: static VOID carcsort _ANSI_ARGS_((struct carc *, struct carc *));
sl@0: static VOID freecnfa _ANSI_ARGS_((struct cnfa *));
sl@0: static VOID dumpnfa _ANSI_ARGS_((struct nfa *, FILE *));
sl@0: #ifdef REG_DEBUG
sl@0: static VOID dumpstate _ANSI_ARGS_((struct state *, FILE *));
sl@0: static VOID dumparcs _ANSI_ARGS_((struct state *, FILE *));
sl@0: static int dumprarcs _ANSI_ARGS_((struct arc *, struct state *, FILE *, int));
sl@0: static VOID dumparc _ANSI_ARGS_((struct arc *, struct state *, FILE *));
sl@0: #endif
sl@0: static VOID dumpcnfa _ANSI_ARGS_((struct cnfa *, FILE *));
sl@0: #ifdef REG_DEBUG
sl@0: static VOID dumpcstate _ANSI_ARGS_((int, struct carc *, struct cnfa *, FILE *));
sl@0: #endif
sl@0: /* === regc_cvec.c === */
sl@0: static struct cvec *newcvec _ANSI_ARGS_((int, int, int));
sl@0: static struct cvec *clearcvec _ANSI_ARGS_((struct cvec *));
sl@0: static VOID addchr _ANSI_ARGS_((struct cvec *, pchr));
sl@0: static VOID addrange _ANSI_ARGS_((struct cvec *, pchr, pchr));
sl@0: static VOID addmcce _ANSI_ARGS_((struct cvec *, chr *, chr *));
sl@0: static int haschr _ANSI_ARGS_((struct cvec *, pchr));
sl@0: static struct cvec *getcvec _ANSI_ARGS_((struct vars *, int, int, int));
sl@0: static VOID freecvec _ANSI_ARGS_((struct cvec *));
sl@0: /* === regc_locale.c === */
sl@0: static int nmcces _ANSI_ARGS_((struct vars *));
sl@0: static int nleaders _ANSI_ARGS_((struct vars *));
sl@0: static struct cvec *allmcces _ANSI_ARGS_((struct vars *, struct cvec *));
sl@0: static celt element _ANSI_ARGS_((struct vars *, chr *, chr *));
sl@0: static struct cvec *range _ANSI_ARGS_((struct vars *, celt, celt, int));
sl@0: static int before _ANSI_ARGS_((celt, celt));
sl@0: static struct cvec *eclass _ANSI_ARGS_((struct vars *, celt, int));
sl@0: static struct cvec *cclass _ANSI_ARGS_((struct vars *, chr *, chr *, int));
sl@0: static struct cvec *allcases _ANSI_ARGS_((struct vars *, pchr));
sl@0: static int cmp _ANSI_ARGS_((CONST chr *, CONST chr *, size_t));
sl@0: static int casecmp _ANSI_ARGS_((CONST chr *, CONST chr *, size_t));
sl@0: /* automatically gathered by fwd; do not hand-edit */
sl@0: /* =====^!^===== end forwards =====^!^===== */
sl@0: 
sl@0: 
sl@0: 
sl@0: /* internal variables, bundled for easy passing around */
sl@0: struct vars {
sl@0: 	regex_t *re;
sl@0: 	chr *now;		/* scan pointer into string */
sl@0: 	chr *stop;		/* end of string */
sl@0: 	chr *savenow;		/* saved now and stop for "subroutine call" */
sl@0: 	chr *savestop;
sl@0: 	int err;		/* error code (0 if none) */
sl@0: 	int cflags;		/* copy of compile flags */
sl@0: 	int lasttype;		/* type of previous token */
sl@0: 	int nexttype;		/* type of next token */
sl@0: 	chr nextvalue;		/* value (if any) of next token */
sl@0: 	int lexcon;		/* lexical context type (see lex.c) */
sl@0: 	int nsubexp;		/* subexpression count */
sl@0: 	struct subre **subs;	/* subRE pointer vector */
sl@0: 	size_t nsubs;		/* length of vector */
sl@0: 	struct subre *sub10[10];	/* initial vector, enough for most */
sl@0: 	struct nfa *nfa;	/* the NFA */
sl@0: 	struct colormap *cm;	/* character color map */
sl@0: 	color nlcolor;		/* color of newline */
sl@0: 	struct state *wordchrs;	/* state in nfa holding word-char outarcs */
sl@0: 	struct subre *tree;	/* subexpression tree */
sl@0: 	struct subre *treechain;	/* all tree nodes allocated */
sl@0: 	struct subre *treefree;		/* any free tree nodes */
sl@0: 	int ntree;		/* number of tree nodes */
sl@0: 	struct cvec *cv;	/* interface cvec */
sl@0: 	struct cvec *cv2;	/* utility cvec */
sl@0: 	struct cvec *mcces;	/* collating-element information */
sl@0: #		define	ISCELEADER(v,c)	(v->mcces != NULL && haschr(v->mcces, (c)))
sl@0: 	struct state *mccepbegin;	/* in nfa, start of MCCE prototypes */
sl@0: 	struct state *mccepend;	/* in nfa, end of MCCE prototypes */
sl@0: 	struct subre *lacons;	/* lookahead-constraint vector */
sl@0: 	int nlacons;		/* size of lacons */
sl@0: };
sl@0: 
sl@0: /* parsing macros; most know that `v' is the struct vars pointer */
sl@0: #define	NEXT()	(next(v))		/* advance by one token */
sl@0: #define	SEE(t)	(v->nexttype == (t))	/* is next token this? */
sl@0: #define	EAT(t)	(SEE(t) && next(v))	/* if next is this, swallow it */
sl@0: #define	VISERR(vv)	((vv)->err != 0)	/* have we seen an error yet? */
sl@0: #define	ISERR()	VISERR(v)
sl@0: #define	VERR(vv,e)	((vv)->nexttype = EOS, ((vv)->err) ? (vv)->err :\
sl@0: 							((vv)->err = (e)))
sl@0: #define	ERR(e)	VERR(v, e)		/* record an error */
sl@0: #define	NOERR()	{if (ISERR()) return;}	/* if error seen, return */
sl@0: #define	NOERRN()	{if (ISERR()) return NULL;}	/* NOERR with retval */
sl@0: #define	NOERRZ()	{if (ISERR()) return 0;}	/* NOERR with retval */
sl@0: #define	INSIST(c, e)	((c) ? 0 : ERR(e))	/* if condition false, error */
sl@0: #define	NOTE(b)	(v->re->re_info |= (b))		/* note visible condition */
sl@0: #define	EMPTYARC(x, y)	newarc(v->nfa, EMPTY, 0, x, y)
sl@0: 
sl@0: /* token type codes, some also used as NFA arc types */
sl@0: #define	EMPTY	'n'		/* no token present */
sl@0: #define	EOS	'e'		/* end of string */
sl@0: #define	PLAIN	'p'		/* ordinary character */
sl@0: #define	DIGIT	'd'		/* digit (in bound) */
sl@0: #define	BACKREF	'b'		/* back reference */
sl@0: #define	COLLEL	'I'		/* start of [. */
sl@0: #define	ECLASS	'E'		/* start of [= */
sl@0: #define	CCLASS	'C'		/* start of [: */
sl@0: #define	END	'X'		/* end of [. [= [: */
sl@0: #define	RANGE	'R'		/* - within [] which might be range delim. */
sl@0: #define	LACON	'L'		/* lookahead constraint subRE */
sl@0: #define	AHEAD	'a'		/* color-lookahead arc */
sl@0: #define	BEHIND	'r'		/* color-lookbehind arc */
sl@0: #define	WBDRY	'w'		/* word boundary constraint */
sl@0: #define	NWBDRY	'W'		/* non-word-boundary constraint */
sl@0: #define	SBEGIN	'A'		/* beginning of string (even if not BOL) */
sl@0: #define	SEND	'Z'		/* end of string (even if not EOL) */
sl@0: #define	PREFER	'P'		/* length preference */
sl@0: 
sl@0: /* is an arc colored, and hence on a color chain? */
sl@0: #define	COLORED(a)	((a)->type == PLAIN || (a)->type == AHEAD || \
sl@0: 							(a)->type == BEHIND)
sl@0: 
sl@0: 
sl@0: 
sl@0: /* static function list */
sl@0: static struct fns functions = {
sl@0: 	rfree,			/* regfree insides */
sl@0: };
sl@0: 
sl@0: 
sl@0: 
sl@0: /*
sl@0:  - compile - compile regular expression
sl@0:  ^ int compile(regex_t *, CONST chr *, size_t, int);
sl@0:  */
sl@0: int
sl@0: compile(re, string, len, flags)
sl@0: regex_t *re;
sl@0: CONST chr *string;
sl@0: size_t len;
sl@0: int flags;
sl@0: {
sl@0: 	struct vars var;
sl@0: 	struct vars *v = &var;
sl@0: 	struct guts *g;
sl@0: 	int i;
sl@0: 	size_t j;
sl@0: 	FILE *debug = (flags&REG_PROGRESS) ? stdout : (FILE *)NULL;
sl@0: #	define	CNOERR()	{ if (ISERR()) return freev(v, v->err); }
sl@0: 
sl@0: 	/* sanity checks */
sl@0: 
sl@0: 	if (re == NULL || string == NULL)
sl@0: 		return REG_INVARG;
sl@0: 	if ((flags&REG_QUOTE) &&
sl@0: 			(flags&(REG_ADVANCED|REG_EXPANDED|REG_NEWLINE)))
sl@0: 		return REG_INVARG;
sl@0: 	if (!(flags&REG_EXTENDED) && (flags&REG_ADVF))
sl@0: 		return REG_INVARG;
sl@0: 
sl@0: 	/* initial setup (after which freev() is callable) */
sl@0: 	v->re = re;
sl@0: 	v->now = (chr *)string;
sl@0: 	v->stop = v->now + len;
sl@0: 	v->savenow = v->savestop = NULL;
sl@0: 	v->err = 0;
sl@0: 	v->cflags = flags;
sl@0: 	v->nsubexp = 0;
sl@0: 	v->subs = v->sub10;
sl@0: 	v->nsubs = 10;
sl@0: 	for (j = 0; j < v->nsubs; j++)
sl@0: 		v->subs[j] = NULL;
sl@0: 	v->nfa = NULL;
sl@0: 	v->cm = NULL;
sl@0: 	v->nlcolor = COLORLESS;
sl@0: 	v->wordchrs = NULL;
sl@0: 	v->tree = NULL;
sl@0: 	v->treechain = NULL;
sl@0: 	v->treefree = NULL;
sl@0: 	v->cv = NULL;
sl@0: 	v->cv2 = NULL;
sl@0: 	v->mcces = NULL;
sl@0: 	v->lacons = NULL;
sl@0: 	v->nlacons = 0;
sl@0: 	re->re_magic = REMAGIC;
sl@0: 	re->re_info = 0;		/* bits get set during parse */
sl@0: 	re->re_csize = sizeof(chr);
sl@0: 	re->re_guts = NULL;
sl@0: 	re->re_fns = VS(&functions);
sl@0: 
sl@0: 	/* more complex setup, malloced things */
sl@0: 	re->re_guts = VS(MALLOC(sizeof(struct guts)));
sl@0: 	if (re->re_guts == NULL)
sl@0: 		return freev(v, REG_ESPACE);
sl@0: 	g = (struct guts *)re->re_guts;
sl@0: 	g->tree = NULL;
sl@0: 	initcm(v, &g->cmap);
sl@0: 	v->cm = &g->cmap;
sl@0: 	g->lacons = NULL;
sl@0: 	g->nlacons = 0;
sl@0: 	ZAPCNFA(g->search);
sl@0: 	v->nfa = newnfa(v, v->cm, (struct nfa *)NULL);
sl@0: 	CNOERR();
sl@0: 	v->cv = newcvec(100, 20, 10);
sl@0: 	if (v->cv == NULL)
sl@0: 		return freev(v, REG_ESPACE);
sl@0: 	i = nmcces(v);
sl@0: 	if (i > 0) {
sl@0: 		v->mcces = newcvec(nleaders(v), 0, i);
sl@0: 		CNOERR();
sl@0: 		v->mcces = allmcces(v, v->mcces);
sl@0: 		leaders(v, v->mcces);
sl@0: 		addmcce(v->mcces, (chr *)NULL, (chr *)NULL);	/* dummy */
sl@0: 	}
sl@0: 	CNOERR();
sl@0: 
sl@0: 	/* parsing */
sl@0: 	lexstart(v);			/* also handles prefixes */
sl@0: 	if ((v->cflags&REG_NLSTOP) || (v->cflags&REG_NLANCH)) {
sl@0: 		/* assign newline a unique color */
sl@0: 		v->nlcolor = subcolor(v->cm, newline());
sl@0: 		okcolors(v->nfa, v->cm);
sl@0: 	}
sl@0: 	CNOERR();
sl@0: 	v->tree = parse(v, EOS, PLAIN, v->nfa->init, v->nfa->final);
sl@0: 	assert(SEE(EOS));		/* even if error; ISERR() => SEE(EOS) */
sl@0: 	CNOERR();
sl@0: 	assert(v->tree != NULL);
sl@0: 
sl@0: 	/* finish setup of nfa and its subre tree */
sl@0: 	specialcolors(v->nfa);
sl@0: 	CNOERR();
sl@0: 	if (debug != NULL) {
sl@0: 		fprintf(debug, "\n\n\n========= RAW ==========\n");
sl@0: 		dumpnfa(v->nfa, debug);
sl@0: 		dumpst(v->tree, debug, 1);
sl@0: 	}
sl@0: 	optst(v, v->tree);
sl@0: 	v->ntree = numst(v->tree, 1);
sl@0: 	markst(v->tree);
sl@0: 	cleanst(v);
sl@0: 	if (debug != NULL) {
sl@0: 		fprintf(debug, "\n\n\n========= TREE FIXED ==========\n");
sl@0: 		dumpst(v->tree, debug, 1);
sl@0: 	}
sl@0: 
sl@0: 	/* build compacted NFAs for tree and lacons */
sl@0: 	re->re_info |= nfatree(v, v->tree, debug);
sl@0: 	CNOERR();
sl@0: 	assert(v->nlacons == 0 || v->lacons != NULL);
sl@0: 	for (i = 1; i < v->nlacons; i++) {
sl@0: 		if (debug != NULL)
sl@0: 			fprintf(debug, "\n\n\n========= LA%d ==========\n", i);
sl@0: 		nfanode(v, &v->lacons[i], debug);
sl@0: 	}
sl@0: 	CNOERR();
sl@0: 	if (v->tree->flags&SHORTER)
sl@0: 		NOTE(REG_USHORTEST);
sl@0: 
sl@0: 	/* build compacted NFAs for tree, lacons, fast search */
sl@0: 	if (debug != NULL)
sl@0: 		fprintf(debug, "\n\n\n========= SEARCH ==========\n");
sl@0: 	/* can sacrifice main NFA now, so use it as work area */
sl@0: 	(DISCARD)optimize(v->nfa, debug);
sl@0: 	CNOERR();
sl@0: 	makesearch(v, v->nfa);
sl@0: 	CNOERR();
sl@0: 	compact(v->nfa, &g->search);
sl@0: 	CNOERR();
sl@0: 
sl@0: 	/* looks okay, package it up */
sl@0: 	re->re_nsub = v->nsubexp;
sl@0: 	v->re = NULL;			/* freev no longer frees re */
sl@0: 	g->magic = GUTSMAGIC;
sl@0: 	g->cflags = v->cflags;
sl@0: 	g->info = re->re_info;
sl@0: 	g->nsub = re->re_nsub;
sl@0: 	g->tree = v->tree;
sl@0: 	v->tree = NULL;
sl@0: 	g->ntree = v->ntree;
sl@0: 	g->compare = (v->cflags&REG_ICASE) ? casecmp : cmp;
sl@0: 	g->lacons = v->lacons;
sl@0: 	v->lacons = NULL;
sl@0: 	g->nlacons = v->nlacons;
sl@0: 
sl@0: 	if (flags&REG_DUMP)
sl@0: 		dump(re, stdout);
sl@0: 
sl@0: 	assert(v->err == 0);
sl@0: 	return freev(v, 0);
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - moresubs - enlarge subRE vector
sl@0:  ^ static VOID moresubs(struct vars *, int);
sl@0:  */
sl@0: static VOID
sl@0: moresubs(v, wanted)
sl@0: struct vars *v;
sl@0: int wanted;			/* want enough room for this one */
sl@0: {
sl@0: 	struct subre **p;
sl@0: 	size_t n;
sl@0: 
sl@0: 	assert(wanted > 0 && (size_t)wanted >= v->nsubs);
sl@0: 	n = (size_t)wanted * 3 / 2 + 1;
sl@0: 	if (v->subs == v->sub10) {
sl@0: 		p = (struct subre **)MALLOC(n * sizeof(struct subre *));
sl@0: 		if (p != NULL)
sl@0: 			memcpy(VS(p), VS(v->subs),
sl@0: 					v->nsubs * sizeof(struct subre *));
sl@0: 	} else
sl@0: 		p = (struct subre **)REALLOC(v->subs, n*sizeof(struct subre *));
sl@0: 	if (p == NULL) {
sl@0: 		ERR(REG_ESPACE);
sl@0: 		return;
sl@0: 	}
sl@0: 	v->subs = p;
sl@0: 	for (p = &v->subs[v->nsubs]; v->nsubs < n; p++, v->nsubs++)
sl@0: 		*p = NULL;
sl@0: 	assert(v->nsubs == n);
sl@0: 	assert((size_t)wanted < v->nsubs);
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - freev - free vars struct's substructures where necessary
sl@0:  * Optionally does error-number setting, and always returns error code
sl@0:  * (if any), to make error-handling code terser.
sl@0:  ^ static int freev(struct vars *, int);
sl@0:  */
sl@0: static int
sl@0: freev(v, err)
sl@0: struct vars *v;
sl@0: int err;
sl@0: {
sl@0: 	if (v->re != NULL)
sl@0: 		rfree(v->re);
sl@0: 	if (v->subs != v->sub10)
sl@0: 		FREE(v->subs);
sl@0: 	if (v->nfa != NULL)
sl@0: 		freenfa(v->nfa);
sl@0: 	if (v->tree != NULL)
sl@0: 		freesubre(v, v->tree);
sl@0: 	if (v->treechain != NULL)
sl@0: 		cleanst(v);
sl@0: 	if (v->cv != NULL)
sl@0: 		freecvec(v->cv);
sl@0: 	if (v->cv2 != NULL)
sl@0: 		freecvec(v->cv2);
sl@0: 	if (v->mcces != NULL)
sl@0: 		freecvec(v->mcces);
sl@0: 	if (v->lacons != NULL)
sl@0: 		freelacons(v->lacons, v->nlacons);
sl@0: 	ERR(err);			/* nop if err==0 */
sl@0: 
sl@0: 	return v->err;
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - makesearch - turn an NFA into a search NFA (implicit prepend of .*?)
sl@0:  * NFA must have been optimize()d already.
sl@0:  ^ static VOID makesearch(struct vars *, struct nfa *);
sl@0:  */
sl@0: static VOID
sl@0: makesearch(v, nfa)
sl@0: struct vars *v;
sl@0: struct nfa *nfa;
sl@0: {
sl@0: 	struct arc *a;
sl@0: 	struct arc *b;
sl@0: 	struct state *pre = nfa->pre;
sl@0: 	struct state *s;
sl@0: 	struct state *s2;
sl@0: 	struct state *slist;
sl@0: 
sl@0: 	/* no loops are needed if it's anchored */
sl@0: 	for (a = pre->outs; a != NULL; a = a->outchain) {
sl@0: 		assert(a->type == PLAIN);
sl@0: 		if (a->co != nfa->bos[0] && a->co != nfa->bos[1])
sl@0: 			break;
sl@0: 	}
sl@0: 	if (a != NULL) {
sl@0: 		/* add implicit .* in front */
sl@0: 		rainbow(nfa, v->cm, PLAIN, COLORLESS, pre, pre);
sl@0: 
sl@0: 		/* and ^* and \A* too -- not always necessary, but harmless */
sl@0: 		newarc(nfa, PLAIN, nfa->bos[0], pre, pre);
sl@0: 		newarc(nfa, PLAIN, nfa->bos[1], pre, pre);
sl@0: 	}
sl@0: 
sl@0: 	/*
sl@0: 	 * Now here's the subtle part.  Because many REs have no lookback
sl@0: 	 * constraints, often knowing when you were in the pre state tells
sl@0: 	 * you little; it's the next state(s) that are informative.  But
sl@0: 	 * some of them may have other inarcs, i.e. it may be possible to
sl@0: 	 * make actual progress and then return to one of them.  We must
sl@0: 	 * de-optimize such cases, splitting each such state into progress
sl@0: 	 * and no-progress states.
sl@0: 	 */
sl@0: 
sl@0: 	/* first, make a list of the states */
sl@0: 	slist = NULL;
sl@0: 	for (a = pre->outs; a != NULL; a = a->outchain) {
sl@0: 		s = a->to;
sl@0: 		for (b = s->ins; b != NULL; b = b->inchain)
sl@0: 			if (b->from != pre)
sl@0: 				break;
sl@0: 		if (b != NULL) {		/* must be split */
sl@0: 			if (s->tmp == NULL) {  /* if not already in the list */
sl@0: 			                       /* (fixes bugs 505048, 230589, */
sl@0: 			                       /* 840258, 504785) */
sl@0: 				s->tmp = slist;
sl@0: 				slist = s;
sl@0: 			}
sl@0: 		}
sl@0: 	}
sl@0: 
sl@0: 	/* do the splits */
sl@0: 	for (s = slist; s != NULL; s = s2) {
sl@0: 		s2 = newstate(nfa);
sl@0: 		copyouts(nfa, s, s2);
sl@0: 		for (a = s->ins; a != NULL; a = b) {
sl@0: 			b = a->inchain;
sl@0: 			if (a->from != pre) {
sl@0: 				cparc(nfa, a, a->from, s2);
sl@0: 				freearc(nfa, a);
sl@0: 			}
sl@0: 		}
sl@0: 		s2 = s->tmp;
sl@0: 		s->tmp = NULL;		/* clean up while we're at it */
sl@0: 	}
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - parse - parse an RE
sl@0:  * This is actually just the top level, which parses a bunch of branches
sl@0:  * tied together with '|'.  They appear in the tree as the left children
sl@0:  * of a chain of '|' subres.
sl@0:  ^ static struct subre *parse(struct vars *, int, int, struct state *,
sl@0:  ^ 	struct state *);
sl@0:  */
sl@0: static struct subre *
sl@0: parse(v, stopper, type, init, final)
sl@0: struct vars *v;
sl@0: int stopper;			/* EOS or ')' */
sl@0: int type;			/* LACON (lookahead subRE) or PLAIN */
sl@0: struct state *init;		/* initial state */
sl@0: struct state *final;		/* final state */
sl@0: {
sl@0: 	struct state *left;	/* scaffolding for branch */
sl@0: 	struct state *right;
sl@0: 	struct subre *branches;	/* top level */
sl@0: 	struct subre *branch;	/* current branch */
sl@0: 	struct subre *t;	/* temporary */
sl@0: 	int firstbranch;	/* is this the first branch? */
sl@0: 
sl@0: 	assert(stopper == ')' || stopper == EOS);
sl@0: 
sl@0: 	branches = subre(v, '|', LONGER, init, final);
sl@0: 	NOERRN();
sl@0: 	branch = branches;
sl@0: 	firstbranch = 1;
sl@0: 	do {	/* a branch */
sl@0: 		if (!firstbranch) {
sl@0: 			/* need a place to hang it */
sl@0: 			branch->right = subre(v, '|', LONGER, init, final);
sl@0: 			NOERRN();
sl@0: 			branch = branch->right;
sl@0: 		}
sl@0: 		firstbranch = 0;
sl@0: 		left = newstate(v->nfa);
sl@0: 		right = newstate(v->nfa);
sl@0: 		NOERRN();
sl@0: 		EMPTYARC(init, left);
sl@0: 		EMPTYARC(right, final);
sl@0: 		NOERRN();
sl@0: 		branch->left = parsebranch(v, stopper, type, left, right, 0);
sl@0: 		NOERRN();
sl@0: 		branch->flags |= UP(branch->flags | branch->left->flags);
sl@0: 		if ((branch->flags &~ branches->flags) != 0)	/* new flags */
sl@0: 			for (t = branches; t != branch; t = t->right)
sl@0: 				t->flags |= branch->flags;
sl@0: 	} while (EAT('|'));
sl@0: 	assert(SEE(stopper) || SEE(EOS));
sl@0: 
sl@0: 	if (!SEE(stopper)) {
sl@0: 		assert(stopper == ')' && SEE(EOS));
sl@0: 		ERR(REG_EPAREN);
sl@0: 	}
sl@0: 
sl@0: 	/* optimize out simple cases */
sl@0: 	if (branch == branches) {	/* only one branch */
sl@0: 		assert(branch->right == NULL);
sl@0: 		t = branch->left;
sl@0: 		branch->left = NULL;
sl@0: 		freesubre(v, branches);
sl@0: 		branches = t;
sl@0: 	} else if (!MESSY(branches->flags)) {	/* no interesting innards */
sl@0: 		freesubre(v, branches->left);
sl@0: 		branches->left = NULL;
sl@0: 		freesubre(v, branches->right);
sl@0: 		branches->right = NULL;
sl@0: 		branches->op = '=';
sl@0: 	}
sl@0: 
sl@0: 	return branches;
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - parsebranch - parse one branch of an RE
sl@0:  * This mostly manages concatenation, working closely with parseqatom().
sl@0:  * Concatenated things are bundled up as much as possible, with separate
sl@0:  * ',' nodes introduced only when necessary due to substructure.
sl@0:  ^ static struct subre *parsebranch(struct vars *, int, int, struct state *,
sl@0:  ^ 	struct state *, int);
sl@0:  */
sl@0: static struct subre *
sl@0: parsebranch(v, stopper, type, left, right, partial)
sl@0: struct vars *v;
sl@0: int stopper;			/* EOS or ')' */
sl@0: int type;			/* LACON (lookahead subRE) or PLAIN */
sl@0: struct state *left;		/* leftmost state */
sl@0: struct state *right;		/* rightmost state */
sl@0: int partial;			/* is this only part of a branch? */
sl@0: {
sl@0: 	struct state *lp;	/* left end of current construct */
sl@0: 	int seencontent;	/* is there anything in this branch yet? */
sl@0: 	struct subre *t;
sl@0: 
sl@0: 	lp = left;
sl@0: 	seencontent = 0;
sl@0: 	t = subre(v, '=', 0, left, right);	/* op '=' is tentative */
sl@0: 	NOERRN();
sl@0: 	while (!SEE('|') && !SEE(stopper) && !SEE(EOS)) {
sl@0: 		if (seencontent) {	/* implicit concat operator */
sl@0: 			lp = newstate(v->nfa);
sl@0: 			NOERRN();
sl@0: 			moveins(v->nfa, right, lp);
sl@0: 		}
sl@0: 		seencontent = 1;
sl@0: 
sl@0: 		/* NB, recursion in parseqatom() may swallow rest of branch */
sl@0: 		parseqatom(v, stopper, type, lp, right, t);
sl@0: 	}
sl@0: 
sl@0: 	if (!seencontent) {		/* empty branch */
sl@0: 		if (!partial)
sl@0: 			NOTE(REG_UUNSPEC);
sl@0: 		assert(lp == left);
sl@0: 		EMPTYARC(left, right);
sl@0: 	}
sl@0: 
sl@0: 	return t;
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - parseqatom - parse one quantified atom or constraint of an RE
sl@0:  * The bookkeeping near the end cooperates very closely with parsebranch();
sl@0:  * in particular, it contains a recursion that can involve parsing the rest
sl@0:  * of the branch, making this function's name somewhat inaccurate.
sl@0:  ^ static VOID parseqatom(struct vars *, int, int, struct state *,
sl@0:  ^ 	struct state *, struct subre *);
sl@0:  */
sl@0: static VOID
sl@0: parseqatom(v, stopper, type, lp, rp, top)
sl@0: struct vars *v;
sl@0: int stopper;			/* EOS or ')' */
sl@0: int type;			/* LACON (lookahead subRE) or PLAIN */
sl@0: struct state *lp;		/* left state to hang it on */
sl@0: struct state *rp;		/* right state to hang it on */
sl@0: struct subre *top;		/* subtree top */
sl@0: {
sl@0: 	struct state *s;	/* temporaries for new states */
sl@0: 	struct state *s2;
sl@0: #	define	ARCV(t, val)	newarc(v->nfa, t, val, lp, rp)
sl@0: 	int m, n;
sl@0: 	struct subre *atom;	/* atom's subtree */
sl@0: 	struct subre *t;
sl@0: 	int cap;		/* capturing parens? */
sl@0: 	int pos;		/* positive lookahead? */
sl@0: 	int subno;		/* capturing-parens or backref number */
sl@0: 	int atomtype;
sl@0: 	int qprefer;		/* quantifier short/long preference */
sl@0: 	int f;
sl@0: 	struct subre **atomp;	/* where the pointer to atom is */
sl@0: 
sl@0: 	/* initial bookkeeping */
sl@0: 	atom = NULL;
sl@0: 	assert(lp->nouts == 0);	/* must string new code */
sl@0: 	assert(rp->nins == 0);	/*  between lp and rp */
sl@0: 	subno = 0;		/* just to shut lint up */
sl@0: 
sl@0: 	/* an atom or constraint... */
sl@0: 	atomtype = v->nexttype;
sl@0: 	switch (atomtype) {
sl@0: 	/* first, constraints, which end by returning */
sl@0: 	case '^':
sl@0: 		ARCV('^', 1);
sl@0: 		if (v->cflags&REG_NLANCH)
sl@0: 			ARCV(BEHIND, v->nlcolor);
sl@0: 		NEXT();
sl@0: 		return;
sl@0: 		break;
sl@0: 	case '$':
sl@0: 		ARCV('$', 1);
sl@0: 		if (v->cflags&REG_NLANCH)
sl@0: 			ARCV(AHEAD, v->nlcolor);
sl@0: 		NEXT();
sl@0: 		return;
sl@0: 		break;
sl@0: 	case SBEGIN:
sl@0: 		ARCV('^', 1);	/* BOL */
sl@0: 		ARCV('^', 0);	/* or BOS */
sl@0: 		NEXT();
sl@0: 		return;
sl@0: 		break;
sl@0: 	case SEND:
sl@0: 		ARCV('$', 1);	/* EOL */
sl@0: 		ARCV('$', 0);	/* or EOS */
sl@0: 		NEXT();
sl@0: 		return;
sl@0: 		break;
sl@0: 	case '<':
sl@0: 		wordchrs(v);	/* does NEXT() */
sl@0: 		s = newstate(v->nfa);
sl@0: 		NOERR();
sl@0: 		nonword(v, BEHIND, lp, s);
sl@0: 		word(v, AHEAD, s, rp);
sl@0: 		return;
sl@0: 		break;
sl@0: 	case '>':
sl@0: 		wordchrs(v);	/* does NEXT() */
sl@0: 		s = newstate(v->nfa);
sl@0: 		NOERR();
sl@0: 		word(v, BEHIND, lp, s);
sl@0: 		nonword(v, AHEAD, s, rp);
sl@0: 		return;
sl@0: 		break;
sl@0: 	case WBDRY:
sl@0: 		wordchrs(v);	/* does NEXT() */
sl@0: 		s = newstate(v->nfa);
sl@0: 		NOERR();
sl@0: 		nonword(v, BEHIND, lp, s);
sl@0: 		word(v, AHEAD, s, rp);
sl@0: 		s = newstate(v->nfa);
sl@0: 		NOERR();
sl@0: 		word(v, BEHIND, lp, s);
sl@0: 		nonword(v, AHEAD, s, rp);
sl@0: 		return;
sl@0: 		break;
sl@0: 	case NWBDRY:
sl@0: 		wordchrs(v);	/* does NEXT() */
sl@0: 		s = newstate(v->nfa);
sl@0: 		NOERR();
sl@0: 		word(v, BEHIND, lp, s);
sl@0: 		word(v, AHEAD, s, rp);
sl@0: 		s = newstate(v->nfa);
sl@0: 		NOERR();
sl@0: 		nonword(v, BEHIND, lp, s);
sl@0: 		nonword(v, AHEAD, s, rp);
sl@0: 		return;
sl@0: 		break;
sl@0: 	case LACON:	/* lookahead constraint */
sl@0: 		pos = v->nextvalue;
sl@0: 		NEXT();
sl@0: 		s = newstate(v->nfa);
sl@0: 		s2 = newstate(v->nfa);
sl@0: 		NOERR();
sl@0: 		t = parse(v, ')', LACON, s, s2);
sl@0: 		freesubre(v, t);	/* internal structure irrelevant */
sl@0: 		assert(SEE(')') || ISERR());
sl@0: 		NEXT();
sl@0: 		n = newlacon(v, s, s2, pos);
sl@0: 		NOERR();
sl@0: 		ARCV(LACON, n);
sl@0: 		return;
sl@0: 		break;
sl@0: 	/* then errors, to get them out of the way */
sl@0: 	case '*':
sl@0: 	case '+':
sl@0: 	case '?':
sl@0: 	case '{':
sl@0: 		ERR(REG_BADRPT);
sl@0: 		return;
sl@0: 		break;
sl@0: 	default:
sl@0: 		ERR(REG_ASSERT);
sl@0: 		return;
sl@0: 		break;
sl@0: 	/* then plain characters, and minor variants on that theme */
sl@0: 	case ')':		/* unbalanced paren */
sl@0: 		if ((v->cflags&REG_ADVANCED) != REG_EXTENDED) {
sl@0: 			ERR(REG_EPAREN);
sl@0: 			return;
sl@0: 		}
sl@0: 		/* legal in EREs due to specification botch */
sl@0: 		NOTE(REG_UPBOTCH);
sl@0: 		/* fallthrough into case PLAIN */
sl@0: 	case PLAIN:
sl@0: 		onechr(v, v->nextvalue, lp, rp);
sl@0: 		okcolors(v->nfa, v->cm);
sl@0: 		NOERR();
sl@0: 		NEXT();
sl@0: 		break;
sl@0: 	case '[':
sl@0: 		if (v->nextvalue == 1)
sl@0: 			bracket(v, lp, rp);
sl@0: 		else
sl@0: 			cbracket(v, lp, rp);
sl@0: 		assert(SEE(']') || ISERR());
sl@0: 		NEXT();
sl@0: 		break;
sl@0: 	case '.':
sl@0: 		rainbow(v->nfa, v->cm, PLAIN,
sl@0: 				(v->cflags&REG_NLSTOP) ? v->nlcolor : COLORLESS,
sl@0: 				lp, rp);
sl@0: 		NEXT();
sl@0: 		break;
sl@0: 	/* and finally the ugly stuff */
sl@0: 	case '(':	/* value flags as capturing or non */
sl@0: 		cap = (type == LACON) ? 0 : v->nextvalue;
sl@0: 		if (cap) {
sl@0: 			v->nsubexp++;
sl@0: 			subno = v->nsubexp;
sl@0: 			if ((size_t)subno >= v->nsubs)
sl@0: 				moresubs(v, subno);
sl@0: 			assert((size_t)subno < v->nsubs);
sl@0: 		} else
sl@0: 			atomtype = PLAIN;	/* something that's not '(' */
sl@0: 		NEXT();
sl@0: 		/* need new endpoints because tree will contain pointers */
sl@0: 		s = newstate(v->nfa);
sl@0: 		s2 = newstate(v->nfa);
sl@0: 		NOERR();
sl@0: 		EMPTYARC(lp, s);
sl@0: 		EMPTYARC(s2, rp);
sl@0: 		NOERR();
sl@0: 		atom = parse(v, ')', PLAIN, s, s2);
sl@0: 		assert(SEE(')') || ISERR());
sl@0: 		NEXT();
sl@0: 		NOERR();
sl@0: 		if (cap) {
sl@0: 			v->subs[subno] = atom;
sl@0: 			t = subre(v, '(', atom->flags|CAP, lp, rp);
sl@0: 			NOERR();
sl@0: 			t->subno = subno;
sl@0: 			t->left = atom;
sl@0: 			atom = t;
sl@0: 		}
sl@0: 		/* postpone everything else pending possible {0} */
sl@0: 		break;
sl@0: 	case BACKREF:	/* the Feature From The Black Lagoon */
sl@0: 		INSIST(type != LACON, REG_ESUBREG);
sl@0: 		INSIST(v->nextvalue < v->nsubs, REG_ESUBREG);
sl@0: 		INSIST(v->subs[v->nextvalue] != NULL, REG_ESUBREG);
sl@0: 		NOERR();
sl@0: 		assert(v->nextvalue > 0);
sl@0: 		atom = subre(v, 'b', BACKR, lp, rp);
sl@0: 		subno = v->nextvalue;
sl@0: 		atom->subno = subno;
sl@0: 		EMPTYARC(lp, rp);	/* temporarily, so there's something */
sl@0: 		NEXT();
sl@0: 		break;
sl@0: 	}
sl@0: 
sl@0: 	/* ...and an atom may be followed by a quantifier */
sl@0: 	switch (v->nexttype) {
sl@0: 	case '*':
sl@0: 		m = 0;
sl@0: 		n = INFINITY;
sl@0: 		qprefer = (v->nextvalue) ? LONGER : SHORTER;
sl@0: 		NEXT();
sl@0: 		break;
sl@0: 	case '+':
sl@0: 		m = 1;
sl@0: 		n = INFINITY;
sl@0: 		qprefer = (v->nextvalue) ? LONGER : SHORTER;
sl@0: 		NEXT();
sl@0: 		break;
sl@0: 	case '?':
sl@0: 		m = 0;
sl@0: 		n = 1;
sl@0: 		qprefer = (v->nextvalue) ? LONGER : SHORTER;
sl@0: 		NEXT();
sl@0: 		break;
sl@0: 	case '{':
sl@0: 		NEXT();
sl@0: 		m = scannum(v);
sl@0: 		if (EAT(',')) {
sl@0: 			if (SEE(DIGIT))
sl@0: 				n = scannum(v);
sl@0: 			else
sl@0: 				n = INFINITY;
sl@0: 			if (m > n) {
sl@0: 				ERR(REG_BADBR);
sl@0: 				return;
sl@0: 			}
sl@0: 			/* {m,n} exercises preference, even if it's {m,m} */
sl@0: 			qprefer = (v->nextvalue) ? LONGER : SHORTER;
sl@0: 		} else {
sl@0: 			n = m;
sl@0: 			/* {m} passes operand's preference through */
sl@0: 			qprefer = 0;
sl@0: 		}
sl@0: 		if (!SEE('}')) {	/* catches errors too */
sl@0: 			ERR(REG_BADBR);
sl@0: 			return;
sl@0: 		}
sl@0: 		NEXT();
sl@0: 		break;
sl@0: 	default:		/* no quantifier */
sl@0: 		m = n = 1;
sl@0: 		qprefer = 0;
sl@0: 		break;
sl@0: 	}
sl@0: 
sl@0: 	/* annoying special case:  {0} or {0,0} cancels everything */
sl@0: 	if (m == 0 && n == 0) {
sl@0: 		if (atom != NULL)
sl@0: 			freesubre(v, atom);
sl@0: 		if (atomtype == '(')
sl@0: 			v->subs[subno] = NULL;
sl@0: 		delsub(v->nfa, lp, rp);
sl@0: 		EMPTYARC(lp, rp);
sl@0: 		return;
sl@0: 	}
sl@0: 
sl@0: 	/* if not a messy case, avoid hard part */
sl@0: 	assert(!MESSY(top->flags));
sl@0: 	f = top->flags | qprefer | ((atom != NULL) ? atom->flags : 0);
sl@0: 	if (atomtype != '(' && atomtype != BACKREF && !MESSY(UP(f))) {
sl@0: 		if (!(m == 1 && n == 1))
sl@0: 			repeat(v, lp, rp, m, n);
sl@0: 		if (atom != NULL)
sl@0: 			freesubre(v, atom);
sl@0: 		top->flags = f;
sl@0: 		return;
sl@0: 	}
sl@0: 
sl@0: 	/*
sl@0: 	 * hard part:  something messy
sl@0: 	 * That is, capturing parens, back reference, short/long clash, or
sl@0: 	 * an atom with substructure containing one of those.
sl@0: 	 */
sl@0: 
sl@0: 	/* now we'll need a subre for the contents even if they're boring */
sl@0: 	if (atom == NULL) {
sl@0: 		atom = subre(v, '=', 0, lp, rp);
sl@0: 		NOERR();
sl@0: 	}
sl@0: 
sl@0: 	/*
sl@0: 	 * prepare a general-purpose state skeleton
sl@0: 	 *
sl@0: 	 *    ---> [s] ---prefix---> [begin] ---atom---> [end] ----rest---> [rp]
sl@0: 	 *   /                                            /
sl@0: 	 * [lp] ----> [s2] ----bypass---------------------
sl@0: 	 *
sl@0: 	 * where bypass is an empty, and prefix is some repetitions of atom
sl@0: 	 */
sl@0: 	s = newstate(v->nfa);		/* first, new endpoints for the atom */
sl@0: 	s2 = newstate(v->nfa);
sl@0: 	NOERR();
sl@0: 	moveouts(v->nfa, lp, s);
sl@0: 	moveins(v->nfa, rp, s2);
sl@0: 	NOERR();
sl@0: 	atom->begin = s;
sl@0: 	atom->end = s2;
sl@0: 	s = newstate(v->nfa);		/* and spots for prefix and bypass */
sl@0: 	s2 = newstate(v->nfa);
sl@0: 	NOERR();
sl@0: 	EMPTYARC(lp, s);
sl@0: 	EMPTYARC(lp, s2);
sl@0: 	NOERR();
sl@0: 
sl@0: 	/* break remaining subRE into x{...} and what follows */
sl@0: 	t = subre(v, '.', COMBINE(qprefer, atom->flags), lp, rp);
sl@0: 	t->left = atom;
sl@0: 	atomp = &t->left;
sl@0: 	/* here we should recurse... but we must postpone that to the end */
sl@0: 
sl@0: 	/* split top into prefix and remaining */
sl@0: 	assert(top->op == '=' && top->left == NULL && top->right == NULL);
sl@0: 	top->left = subre(v, '=', top->flags, top->begin, lp);
sl@0: 	top->op = '.';
sl@0: 	top->right = t;
sl@0: 
sl@0: 	/* if it's a backref, now is the time to replicate the subNFA */
sl@0: 	if (atomtype == BACKREF) {
sl@0: 		assert(atom->begin->nouts == 1);	/* just the EMPTY */
sl@0: 		delsub(v->nfa, atom->begin, atom->end);
sl@0: 		assert(v->subs[subno] != NULL);
sl@0: 		/* and here's why the recursion got postponed:  it must */
sl@0: 		/* wait until the skeleton is filled in, because it may */
sl@0: 		/* hit a backref that wants to copy the filled-in skeleton */
sl@0: 		dupnfa(v->nfa, v->subs[subno]->begin, v->subs[subno]->end,
sl@0: 						atom->begin, atom->end);
sl@0: 		NOERR();
sl@0: 	}
sl@0: 
sl@0: 	/* it's quantifier time; first, turn x{0,...} into x{1,...}|empty */
sl@0: 	if (m == 0) {
sl@0: 		EMPTYARC(s2, atom->end);		/* the bypass */
sl@0: 		assert(PREF(qprefer) != 0);
sl@0: 		f = COMBINE(qprefer, atom->flags);
sl@0: 		t = subre(v, '|', f, lp, atom->end);
sl@0: 		NOERR();
sl@0: 		t->left = atom;
sl@0: 		t->right = subre(v, '|', PREF(f), s2, atom->end);
sl@0: 		NOERR();
sl@0: 		t->right->left = subre(v, '=', 0, s2, atom->end);
sl@0: 		NOERR();
sl@0: 		*atomp = t;
sl@0: 		atomp = &t->left;
sl@0: 		m = 1;
sl@0: 	}
sl@0: 
sl@0: 	/* deal with the rest of the quantifier */
sl@0: 	if (atomtype == BACKREF) {
sl@0: 		/* special case:  backrefs have internal quantifiers */
sl@0: 		EMPTYARC(s, atom->begin);	/* empty prefix */
sl@0: 		/* just stuff everything into atom */
sl@0: 		repeat(v, atom->begin, atom->end, m, n);
sl@0: 		atom->min = (short)m;
sl@0: 		atom->max = (short)n;
sl@0: 		atom->flags |= COMBINE(qprefer, atom->flags);
sl@0: 	} else if (m == 1 && n == 1) {
sl@0: 		/* no/vacuous quantifier:  done */
sl@0: 		EMPTYARC(s, atom->begin);	/* empty prefix */
sl@0: 	} else {
sl@0: 		/* turn x{m,n} into x{m-1,n-1}x, with capturing */
sl@0: 		/*  parens in only second x */
sl@0: 		dupnfa(v->nfa, atom->begin, atom->end, s, atom->begin);
sl@0: 		assert(m >= 1 && m != INFINITY && n >= 1);
sl@0: 		repeat(v, s, atom->begin, m-1, (n == INFINITY) ? n : n-1);
sl@0: 		f = COMBINE(qprefer, atom->flags);
sl@0: 		t = subre(v, '.', f, s, atom->end);	/* prefix and atom */
sl@0: 		NOERR();
sl@0: 		t->left = subre(v, '=', PREF(f), s, atom->begin);
sl@0: 		NOERR();
sl@0: 		t->right = atom;
sl@0: 		*atomp = t;
sl@0: 	}
sl@0: 
sl@0: 	/* and finally, look after that postponed recursion */
sl@0: 	t = top->right;
sl@0: 	if (!(SEE('|') || SEE(stopper) || SEE(EOS)))
sl@0: 		t->right = parsebranch(v, stopper, type, atom->end, rp, 1);
sl@0: 	else {
sl@0: 		EMPTYARC(atom->end, rp);
sl@0: 		t->right = subre(v, '=', 0, atom->end, rp);
sl@0: 	}
sl@0: 	assert(SEE('|') || SEE(stopper) || SEE(EOS));
sl@0: 	t->flags |= COMBINE(t->flags, t->right->flags);
sl@0: 	top->flags |= COMBINE(top->flags, t->flags);
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - nonword - generate arcs for non-word-character ahead or behind
sl@0:  ^ static VOID nonword(struct vars *, int, struct state *, struct state *);
sl@0:  */
sl@0: static VOID
sl@0: nonword(v, dir, lp, rp)
sl@0: struct vars *v;
sl@0: int dir;			/* AHEAD or BEHIND */
sl@0: struct state *lp;
sl@0: struct state *rp;
sl@0: {
sl@0: 	int anchor = (dir == AHEAD) ? '$' : '^';
sl@0: 
sl@0: 	assert(dir == AHEAD || dir == BEHIND);
sl@0: 	newarc(v->nfa, anchor, 1, lp, rp);
sl@0: 	newarc(v->nfa, anchor, 0, lp, rp);
sl@0: 	colorcomplement(v->nfa, v->cm, dir, v->wordchrs, lp, rp);
sl@0: 	/* (no need for special attention to \n) */
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - word - generate arcs for word character ahead or behind
sl@0:  ^ static VOID word(struct vars *, int, struct state *, struct state *);
sl@0:  */
sl@0: static VOID
sl@0: word(v, dir, lp, rp)
sl@0: struct vars *v;
sl@0: int dir;			/* AHEAD or BEHIND */
sl@0: struct state *lp;
sl@0: struct state *rp;
sl@0: {
sl@0: 	assert(dir == AHEAD || dir == BEHIND);
sl@0: 	cloneouts(v->nfa, v->wordchrs, lp, rp, dir);
sl@0: 	/* (no need for special attention to \n) */
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - scannum - scan a number
sl@0:  ^ static int scannum(struct vars *);
sl@0:  */
sl@0: static int			/* value, <= DUPMAX */
sl@0: scannum(v)
sl@0: struct vars *v;
sl@0: {
sl@0: 	int n = 0;
sl@0: 
sl@0: 	while (SEE(DIGIT) && n < DUPMAX) {
sl@0: 		n = n*10 + v->nextvalue;
sl@0: 		NEXT();
sl@0: 	}
sl@0: 	if (SEE(DIGIT) || n > DUPMAX) {
sl@0: 		ERR(REG_BADBR);
sl@0: 		return 0;
sl@0: 	}
sl@0: 	return n;
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - repeat - replicate subNFA for quantifiers
sl@0:  * The duplication sequences used here are chosen carefully so that any
sl@0:  * pointers starting out pointing into the subexpression end up pointing into
sl@0:  * the last occurrence.  (Note that it may not be strung between the same
sl@0:  * left and right end states, however!)  This used to be important for the
sl@0:  * subRE tree, although the important bits are now handled by the in-line
sl@0:  * code in parse(), and when this is called, it doesn't matter any more.
sl@0:  ^ static VOID repeat(struct vars *, struct state *, struct state *, int, int);
sl@0:  */
sl@0: static VOID
sl@0: repeat(v, lp, rp, m, n)
sl@0: struct vars *v;
sl@0: struct state *lp;
sl@0: struct state *rp;
sl@0: int m;
sl@0: int n;
sl@0: {
sl@0: #	define	SOME	2
sl@0: #	define	INF	3
sl@0: #	define	PAIR(x, y)	((x)*4 + (y))
sl@0: #	define	REDUCE(x)	( ((x) == INFINITY) ? INF : (((x) > 1) ? SOME : (x)) )
sl@0: 	CONST int rm = REDUCE(m);
sl@0: 	CONST int rn = REDUCE(n);
sl@0: 	struct state *s;
sl@0: 	struct state *s2;
sl@0: 
sl@0: 	switch (PAIR(rm, rn)) {
sl@0: 	case PAIR(0, 0):		/* empty string */
sl@0: 		delsub(v->nfa, lp, rp);
sl@0: 		EMPTYARC(lp, rp);
sl@0: 		break;
sl@0: 	case PAIR(0, 1):		/* do as x| */
sl@0: 		EMPTYARC(lp, rp);
sl@0: 		break;
sl@0: 	case PAIR(0, SOME):		/* do as x{1,n}| */
sl@0: 		repeat(v, lp, rp, 1, n);
sl@0: 		NOERR();
sl@0: 		EMPTYARC(lp, rp);
sl@0: 		break;
sl@0: 	case PAIR(0, INF):		/* loop x around */
sl@0: 		s = newstate(v->nfa);
sl@0: 		NOERR();
sl@0: 		moveouts(v->nfa, lp, s);
sl@0: 		moveins(v->nfa, rp, s);
sl@0: 		EMPTYARC(lp, s);
sl@0: 		EMPTYARC(s, rp);
sl@0: 		break;
sl@0: 	case PAIR(1, 1):		/* no action required */
sl@0: 		break;
sl@0: 	case PAIR(1, SOME):		/* do as x{0,n-1}x = (x{1,n-1}|)x */
sl@0: 		s = newstate(v->nfa);
sl@0: 		NOERR();
sl@0: 		moveouts(v->nfa, lp, s);
sl@0: 		dupnfa(v->nfa, s, rp, lp, s);
sl@0: 		NOERR();
sl@0: 		repeat(v, lp, s, 1, n-1);
sl@0: 		NOERR();
sl@0: 		EMPTYARC(lp, s);
sl@0: 		break;
sl@0: 	case PAIR(1, INF):		/* add loopback arc */
sl@0: 		s = newstate(v->nfa);
sl@0: 		s2 = newstate(v->nfa);
sl@0: 		NOERR();
sl@0: 		moveouts(v->nfa, lp, s);
sl@0: 		moveins(v->nfa, rp, s2);
sl@0: 		EMPTYARC(lp, s);
sl@0: 		EMPTYARC(s2, rp);
sl@0: 		EMPTYARC(s2, s);
sl@0: 		break;
sl@0: 	case PAIR(SOME, SOME):		/* do as x{m-1,n-1}x */
sl@0: 		s = newstate(v->nfa);
sl@0: 		NOERR();
sl@0: 		moveouts(v->nfa, lp, s);
sl@0: 		dupnfa(v->nfa, s, rp, lp, s);
sl@0: 		NOERR();
sl@0: 		repeat(v, lp, s, m-1, n-1);
sl@0: 		break;
sl@0: 	case PAIR(SOME, INF):		/* do as x{m-1,}x */
sl@0: 		s = newstate(v->nfa);
sl@0: 		NOERR();
sl@0: 		moveouts(v->nfa, lp, s);
sl@0: 		dupnfa(v->nfa, s, rp, lp, s);
sl@0: 		NOERR();
sl@0: 		repeat(v, lp, s, m-1, n);
sl@0: 		break;
sl@0: 	default:
sl@0: 		ERR(REG_ASSERT);
sl@0: 		break;
sl@0: 	}
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - bracket - handle non-complemented bracket expression
sl@0:  * Also called from cbracket for complemented bracket expressions.
sl@0:  ^ static VOID bracket(struct vars *, struct state *, struct state *);
sl@0:  */
sl@0: static VOID
sl@0: bracket(v, lp, rp)
sl@0: struct vars *v;
sl@0: struct state *lp;
sl@0: struct state *rp;
sl@0: {
sl@0: 	assert(SEE('['));
sl@0: 	NEXT();
sl@0: 	while (!SEE(']') && !SEE(EOS))
sl@0: 		brackpart(v, lp, rp);
sl@0: 	assert(SEE(']') || ISERR());
sl@0: 	okcolors(v->nfa, v->cm);
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - cbracket - handle complemented bracket expression
sl@0:  * We do it by calling bracket() with dummy endpoints, and then complementing
sl@0:  * the result.  The alternative would be to invoke rainbow(), and then delete
sl@0:  * arcs as the b.e. is seen... but that gets messy.
sl@0:  ^ static VOID cbracket(struct vars *, struct state *, struct state *);
sl@0:  */
sl@0: static VOID
sl@0: cbracket(v, lp, rp)
sl@0: struct vars *v;
sl@0: struct state *lp;
sl@0: struct state *rp;
sl@0: {
sl@0: 	struct state *left = newstate(v->nfa);
sl@0: 	struct state *right = newstate(v->nfa);
sl@0: 	struct state *s;
sl@0: 	struct arc *a;			/* arc from lp */
sl@0: 	struct arc *ba;			/* arc from left, from bracket() */
sl@0: 	struct arc *pa;			/* MCCE-prototype arc */
sl@0: 	color co;
sl@0: 	chr *p;
sl@0: 	int i;
sl@0: 
sl@0: 	NOERR();
sl@0: 	bracket(v, left, right);
sl@0: 	if (v->cflags&REG_NLSTOP)
sl@0: 		newarc(v->nfa, PLAIN, v->nlcolor, left, right);
sl@0: 	NOERR();
sl@0: 
sl@0: 	assert(lp->nouts == 0);		/* all outarcs will be ours */
sl@0: 
sl@0: 	/* easy part of complementing */
sl@0: 	colorcomplement(v->nfa, v->cm, PLAIN, left, lp, rp);
sl@0: 	NOERR();
sl@0: 	if (v->mcces == NULL) {		/* no MCCEs -- we're done */
sl@0: 		dropstate(v->nfa, left);
sl@0: 		assert(right->nins == 0);
sl@0: 		freestate(v->nfa, right);
sl@0: 		return;
sl@0: 	}
sl@0: 
sl@0: 	/* but complementing gets messy in the presence of MCCEs... */
sl@0: 	NOTE(REG_ULOCALE);
sl@0: 	for (p = v->mcces->chrs, i = v->mcces->nchrs; i > 0; p++, i--) {
sl@0: 		co = GETCOLOR(v->cm, *p);
sl@0: 		a = findarc(lp, PLAIN, co);
sl@0: 		ba = findarc(left, PLAIN, co);
sl@0: 		if (ba == NULL) {
sl@0: 			assert(a != NULL);
sl@0: 			freearc(v->nfa, a);
sl@0: 		} else {
sl@0: 			assert(a == NULL);
sl@0: 		}
sl@0: 		s = newstate(v->nfa);
sl@0: 		NOERR();
sl@0: 		newarc(v->nfa, PLAIN, co, lp, s);
sl@0: 		NOERR();
sl@0: 		pa = findarc(v->mccepbegin, PLAIN, co);
sl@0: 		assert(pa != NULL);
sl@0: 		if (ba == NULL) {	/* easy case, need all of them */
sl@0: 			cloneouts(v->nfa, pa->to, s, rp, PLAIN);
sl@0: 			newarc(v->nfa, '$', 1, s, rp);
sl@0: 			newarc(v->nfa, '$', 0, s, rp);
sl@0: 			colorcomplement(v->nfa, v->cm, AHEAD, pa->to, s, rp);
sl@0: 		} else {		/* must be selective */
sl@0: 			if (findarc(ba->to, '$', 1) == NULL) {
sl@0: 				newarc(v->nfa, '$', 1, s, rp);
sl@0: 				newarc(v->nfa, '$', 0, s, rp);
sl@0: 				colorcomplement(v->nfa, v->cm, AHEAD, pa->to,
sl@0: 									 s, rp);
sl@0: 			}
sl@0: 			for (pa = pa->to->outs; pa != NULL; pa = pa->outchain)
sl@0: 				if (findarc(ba->to, PLAIN, pa->co) == NULL)
sl@0: 					newarc(v->nfa, PLAIN, pa->co, s, rp);
sl@0: 			if (s->nouts == 0)	/* limit of selectivity: none */
sl@0: 				dropstate(v->nfa, s);	/* frees arc too */
sl@0: 		}
sl@0: 		NOERR();
sl@0: 	}
sl@0: 
sl@0: 	delsub(v->nfa, left, right);
sl@0: 	assert(left->nouts == 0);
sl@0: 	freestate(v->nfa, left);
sl@0: 	assert(right->nins == 0);
sl@0: 	freestate(v->nfa, right);
sl@0: }
sl@0: 			
sl@0: /*
sl@0:  - brackpart - handle one item (or range) within a bracket expression
sl@0:  ^ static VOID brackpart(struct vars *, struct state *, struct state *);
sl@0:  */
sl@0: static VOID
sl@0: brackpart(v, lp, rp)
sl@0: struct vars *v;
sl@0: struct state *lp;
sl@0: struct state *rp;
sl@0: {
sl@0: 	celt startc;
sl@0: 	celt endc;
sl@0: 	struct cvec *cv;
sl@0: 	chr *startp;
sl@0: 	chr *endp;
sl@0: 	chr c[1];
sl@0: 
sl@0: 	/* parse something, get rid of special cases, take shortcuts */
sl@0: 	switch (v->nexttype) {
sl@0: 	case RANGE:			/* a-b-c or other botch */
sl@0: 		ERR(REG_ERANGE);
sl@0: 		return;
sl@0: 		break;
sl@0: 	case PLAIN:
sl@0: 		c[0] = v->nextvalue;
sl@0: 		NEXT();
sl@0: 		/* shortcut for ordinary chr (not range, not MCCE leader) */
sl@0: 		if (!SEE(RANGE) && !ISCELEADER(v, c[0])) {
sl@0: 			onechr(v, c[0], lp, rp);
sl@0: 			return;
sl@0: 		}
sl@0: 		startc = element(v, c, c+1);
sl@0: 		NOERR();
sl@0: 		break;
sl@0: 	case COLLEL:
sl@0: 		startp = v->now;
sl@0: 		endp = scanplain(v);
sl@0: 		INSIST(startp < endp, REG_ECOLLATE);
sl@0: 		NOERR();
sl@0: 		startc = element(v, startp, endp);
sl@0: 		NOERR();
sl@0: 		break;
sl@0: 	case ECLASS:
sl@0: 		startp = v->now;
sl@0: 		endp = scanplain(v);
sl@0: 		INSIST(startp < endp, REG_ECOLLATE);
sl@0: 		NOERR();
sl@0: 		startc = element(v, startp, endp);
sl@0: 		NOERR();
sl@0: 		cv = eclass(v, startc, (v->cflags&REG_ICASE));
sl@0: 		NOERR();
sl@0: 		dovec(v, cv, lp, rp);
sl@0: 		return;
sl@0: 		break;
sl@0: 	case CCLASS:
sl@0: 		startp = v->now;
sl@0: 		endp = scanplain(v);
sl@0: 		INSIST(startp < endp, REG_ECTYPE);
sl@0: 		NOERR();
sl@0: 		cv = cclass(v, startp, endp, (v->cflags&REG_ICASE));
sl@0: 		NOERR();
sl@0: 		dovec(v, cv, lp, rp);
sl@0: 		return;
sl@0: 		break;
sl@0: 	default:
sl@0: 		ERR(REG_ASSERT);
sl@0: 		return;
sl@0: 		break;
sl@0: 	}
sl@0: 
sl@0: 	if (SEE(RANGE)) {
sl@0: 		NEXT();
sl@0: 		switch (v->nexttype) {
sl@0: 		case PLAIN:
sl@0: 		case RANGE:
sl@0: 			c[0] = v->nextvalue;
sl@0: 			NEXT();
sl@0: 			endc = element(v, c, c+1);
sl@0: 			NOERR();
sl@0: 			break;
sl@0: 		case COLLEL:
sl@0: 			startp = v->now;
sl@0: 			endp = scanplain(v);
sl@0: 			INSIST(startp < endp, REG_ECOLLATE);
sl@0: 			NOERR();
sl@0: 			endc = element(v, startp, endp);
sl@0: 			NOERR();
sl@0: 			break;
sl@0: 		default:
sl@0: 			ERR(REG_ERANGE);
sl@0: 			return;
sl@0: 			break;
sl@0: 		}
sl@0: 	} else
sl@0: 		endc = startc;
sl@0: 
sl@0: 	/*
sl@0: 	 * Ranges are unportable.  Actually, standard C does
sl@0: 	 * guarantee that digits are contiguous, but making
sl@0: 	 * that an exception is just too complicated.
sl@0: 	 */
sl@0: 	if (startc != endc)
sl@0: 		NOTE(REG_UUNPORT);
sl@0: 	cv = range(v, startc, endc, (v->cflags&REG_ICASE));
sl@0: 	NOERR();
sl@0: 	dovec(v, cv, lp, rp);
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - scanplain - scan PLAIN contents of [. etc.
sl@0:  * Certain bits of trickery in lex.c know that this code does not try
sl@0:  * to look past the final bracket of the [. etc.
sl@0:  ^ static chr *scanplain(struct vars *);
sl@0:  */
sl@0: static chr *			/* just after end of sequence */
sl@0: scanplain(v)
sl@0: struct vars *v;
sl@0: {
sl@0: 	chr *endp;
sl@0: 
sl@0: 	assert(SEE(COLLEL) || SEE(ECLASS) || SEE(CCLASS));
sl@0: 	NEXT();
sl@0: 
sl@0: 	endp = v->now;
sl@0: 	while (SEE(PLAIN)) {
sl@0: 		endp = v->now;
sl@0: 		NEXT();
sl@0: 	}
sl@0: 
sl@0: 	assert(SEE(END) || ISERR());
sl@0: 	NEXT();
sl@0: 
sl@0: 	return endp;
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - leaders - process a cvec of collating elements to also include leaders
sl@0:  * Also gives all characters involved their own colors, which is almost
sl@0:  * certainly necessary, and sets up little disconnected subNFA.
sl@0:  ^ static VOID leaders(struct vars *, struct cvec *);
sl@0:  */
sl@0: static VOID
sl@0: leaders(v, cv)
sl@0: struct vars *v;
sl@0: struct cvec *cv;
sl@0: {
sl@0: 	int mcce;
sl@0: 	chr *p;
sl@0: 	chr leader;
sl@0: 	struct state *s;
sl@0: 	struct arc *a;
sl@0: 
sl@0: 	v->mccepbegin = newstate(v->nfa);
sl@0: 	v->mccepend = newstate(v->nfa);
sl@0: 	NOERR();
sl@0: 
sl@0: 	for (mcce = 0; mcce < cv->nmcces; mcce++) {
sl@0: 		p = cv->mcces[mcce];
sl@0: 		leader = *p;
sl@0: 		if (!haschr(cv, leader)) {
sl@0: 			addchr(cv, leader);
sl@0: 			s = newstate(v->nfa);
sl@0: 			newarc(v->nfa, PLAIN, subcolor(v->cm, leader),
sl@0: 							v->mccepbegin, s);
sl@0: 			okcolors(v->nfa, v->cm);
sl@0: 		} else {
sl@0: 			a = findarc(v->mccepbegin, PLAIN,
sl@0: 						GETCOLOR(v->cm, leader));
sl@0: 			assert(a != NULL);
sl@0: 			s = a->to;
sl@0: 			assert(s != v->mccepend);
sl@0: 		}
sl@0: 		p++;
sl@0: 		assert(*p != 0 && *(p+1) == 0);	/* only 2-char MCCEs for now */
sl@0: 		newarc(v->nfa, PLAIN, subcolor(v->cm, *p), s, v->mccepend);
sl@0: 		okcolors(v->nfa, v->cm);
sl@0: 	}
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - onechr - fill in arcs for a plain character, and possible case complements
sl@0:  * This is mostly a shortcut for efficient handling of the common case.
sl@0:  ^ static VOID onechr(struct vars *, pchr, struct state *, struct state *);
sl@0:  */
sl@0: static VOID
sl@0: onechr(v, c, lp, rp)
sl@0: struct vars *v;
sl@0: pchr c;
sl@0: struct state *lp;
sl@0: struct state *rp;
sl@0: {
sl@0: 	if (!(v->cflags&REG_ICASE)) {
sl@0: 		newarc(v->nfa, PLAIN, subcolor(v->cm, c), lp, rp);
sl@0: 		return;
sl@0: 	}
sl@0: 
sl@0: 	/* rats, need general case anyway... */
sl@0: 	dovec(v, allcases(v, c), lp, rp);
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - dovec - fill in arcs for each element of a cvec
sl@0:  * This one has to handle the messy cases, like MCCEs and MCCE leaders.
sl@0:  ^ static VOID dovec(struct vars *, struct cvec *, struct state *,
sl@0:  ^ 	struct state *);
sl@0:  */
sl@0: static VOID
sl@0: dovec(v, cv, lp, rp)
sl@0: struct vars *v;
sl@0: struct cvec *cv;
sl@0: struct state *lp;
sl@0: struct state *rp;
sl@0: {
sl@0: 	chr ch, from, to;
sl@0: 	celt ce;
sl@0: 	chr *p;
sl@0: 	int i;
sl@0: 	color co;
sl@0: 	struct cvec *leads;
sl@0: 	struct arc *a;
sl@0: 	struct arc *pa;		/* arc in prototype */
sl@0: 	struct state *s;
sl@0: 	struct state *ps;	/* state in prototype */
sl@0: 
sl@0: 	/* need a place to store leaders, if any */
sl@0: 	if (nmcces(v) > 0) {
sl@0: 		assert(v->mcces != NULL);
sl@0: 		if (v->cv2 == NULL || v->cv2->nchrs < v->mcces->nchrs) {
sl@0: 			if (v->cv2 != NULL)
sl@0: 				free(v->cv2);
sl@0: 			v->cv2 = newcvec(v->mcces->nchrs, 0, v->mcces->nmcces);
sl@0: 			NOERR();
sl@0: 			leads = v->cv2;
sl@0: 		} else
sl@0: 			leads = clearcvec(v->cv2);
sl@0: 	} else
sl@0: 		leads = NULL;
sl@0: 
sl@0: 	/* first, get the ordinary characters out of the way */
sl@0: 	for (p = cv->chrs, i = cv->nchrs; i > 0; p++, i--) {
sl@0: 		ch = *p;
sl@0: 		if (!ISCELEADER(v, ch))
sl@0: 			newarc(v->nfa, PLAIN, subcolor(v->cm, ch), lp, rp);
sl@0: 		else {
sl@0: 			assert(singleton(v->cm, ch));
sl@0: 			assert(leads != NULL);
sl@0: 			if (!haschr(leads, ch))
sl@0: 				addchr(leads, ch);
sl@0: 		}
sl@0: 	}
sl@0: 
sl@0: 	/* and the ranges */
sl@0: 	for (p = cv->ranges, i = cv->nranges; i > 0; p += 2, i--) {
sl@0: 		from = *p;
sl@0: 		to = *(p+1);
sl@0: 		while (from <= to && (ce = nextleader(v, from, to)) != NOCELT) {
sl@0: 			if (from < ce)
sl@0: 				subrange(v, from, ce - 1, lp, rp);
sl@0: 			assert(singleton(v->cm, ce));
sl@0: 			assert(leads != NULL);
sl@0: 			if (!haschr(leads, ce))
sl@0: 				addchr(leads, ce);
sl@0: 			from = ce + 1;
sl@0: 		}
sl@0: 		if (from <= to)
sl@0: 			subrange(v, from, to, lp, rp);
sl@0: 	}
sl@0: 
sl@0: 	if ((leads == NULL || leads->nchrs == 0) && cv->nmcces == 0)
sl@0: 		return;
sl@0: 
sl@0: 	/* deal with the MCCE leaders */
sl@0: 	NOTE(REG_ULOCALE);
sl@0: 	for (p = leads->chrs, i = leads->nchrs; i > 0; p++, i--) {
sl@0: 		co = GETCOLOR(v->cm, *p);
sl@0: 		a = findarc(lp, PLAIN, co);
sl@0: 		if (a != NULL)
sl@0: 			s = a->to;
sl@0: 		else {
sl@0: 			s = newstate(v->nfa);
sl@0: 			NOERR();
sl@0: 			newarc(v->nfa, PLAIN, co, lp, s);
sl@0: 			NOERR();
sl@0: 		}
sl@0: 		pa = findarc(v->mccepbegin, PLAIN, co);
sl@0: 		assert(pa != NULL);
sl@0: 		ps = pa->to;
sl@0: 		newarc(v->nfa, '$', 1, s, rp);
sl@0: 		newarc(v->nfa, '$', 0, s, rp);
sl@0: 		colorcomplement(v->nfa, v->cm, AHEAD, ps, s, rp);
sl@0: 		NOERR();
sl@0: 	}
sl@0: 
sl@0: 	/* and the MCCEs */
sl@0: 	for (i = 0; i < cv->nmcces; i++) {
sl@0: 		p = cv->mcces[i];
sl@0: 		assert(singleton(v->cm, *p));
sl@0: 		if (!singleton(v->cm, *p)) {
sl@0: 			ERR(REG_ASSERT);
sl@0: 			return;
sl@0: 		}
sl@0: 		ch = *p++;
sl@0: 		co = GETCOLOR(v->cm, ch);
sl@0: 		a = findarc(lp, PLAIN, co);
sl@0: 		if (a != NULL)
sl@0: 			s = a->to;
sl@0: 		else {
sl@0: 			s = newstate(v->nfa);
sl@0: 			NOERR();
sl@0: 			newarc(v->nfa, PLAIN, co, lp, s);
sl@0: 			NOERR();
sl@0: 		}
sl@0: 		assert(*p != 0);	/* at least two chars */
sl@0: 		assert(singleton(v->cm, *p));
sl@0: 		ch = *p++;
sl@0: 		co = GETCOLOR(v->cm, ch);
sl@0: 		assert(*p == 0);	/* and only two, for now */
sl@0: 		newarc(v->nfa, PLAIN, co, s, rp);
sl@0: 		NOERR();
sl@0: 	}
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - nextleader - find next MCCE leader within range
sl@0:  ^ static celt nextleader(struct vars *, pchr, pchr);
sl@0:  */
sl@0: static celt			/* NOCELT means none */
sl@0: nextleader(v, from, to)
sl@0: struct vars *v;
sl@0: pchr from;
sl@0: pchr to;
sl@0: {
sl@0: 	int i;
sl@0: 	chr *p;
sl@0: 	chr ch;
sl@0: 	celt it = NOCELT;
sl@0: 
sl@0: 	if (v->mcces == NULL)
sl@0: 		return it;
sl@0: 
sl@0: 	for (i = v->mcces->nchrs, p = v->mcces->chrs; i > 0; i--, p++) {
sl@0: 		ch = *p;
sl@0: 		if (from <= ch && ch <= to)
sl@0: 			if (it == NOCELT || ch < it)
sl@0: 				it = ch;
sl@0: 	}
sl@0: 	return it;
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - wordchrs - set up word-chr list for word-boundary stuff, if needed
sl@0:  * The list is kept as a bunch of arcs between two dummy states; it's
sl@0:  * disposed of by the unreachable-states sweep in NFA optimization.
sl@0:  * Does NEXT().  Must not be called from any unusual lexical context.
sl@0:  * This should be reconciled with the \w etc. handling in lex.c, and
sl@0:  * should be cleaned up to reduce dependencies on input scanning.
sl@0:  ^ static VOID wordchrs(struct vars *);
sl@0:  */
sl@0: static VOID
sl@0: wordchrs(v)
sl@0: struct vars *v;
sl@0: {
sl@0: 	struct state *left;
sl@0: 	struct state *right;
sl@0: 
sl@0: 	if (v->wordchrs != NULL) {
sl@0: 		NEXT();		/* for consistency */
sl@0: 		return;
sl@0: 	}
sl@0: 
sl@0: 	left = newstate(v->nfa);
sl@0: 	right = newstate(v->nfa);
sl@0: 	NOERR();
sl@0: 	/* fine point:  implemented with [::], and lexer will set REG_ULOCALE */
sl@0: 	lexword(v);
sl@0: 	NEXT();
sl@0: 	assert(v->savenow != NULL && SEE('['));
sl@0: 	bracket(v, left, right);
sl@0: 	assert((v->savenow != NULL && SEE(']')) || ISERR());
sl@0: 	NEXT();
sl@0: 	NOERR();
sl@0: 	v->wordchrs = left;
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - subre - allocate a subre
sl@0:  ^ static struct subre *subre(struct vars *, int, int, struct state *,
sl@0:  ^	struct state *);
sl@0:  */
sl@0: static struct subre *
sl@0: subre(v, op, flags, begin, end)
sl@0: struct vars *v;
sl@0: int op;
sl@0: int flags;
sl@0: struct state *begin;
sl@0: struct state *end;
sl@0: {
sl@0: 	struct subre *ret;
sl@0: 
sl@0: 	ret = v->treefree;
sl@0: 	if (ret != NULL)
sl@0: 		v->treefree = ret->left;
sl@0: 	else {
sl@0: 		ret = (struct subre *)MALLOC(sizeof(struct subre));
sl@0: 		if (ret == NULL) {
sl@0: 			ERR(REG_ESPACE);
sl@0: 			return NULL;
sl@0: 		}
sl@0: 		ret->chain = v->treechain;
sl@0: 		v->treechain = ret;
sl@0: 	}
sl@0: 
sl@0: 	assert(strchr("|.b(=", op) != NULL);
sl@0: 
sl@0: 	ret->op = op;
sl@0: 	ret->flags = flags;
sl@0: 	ret->retry = 0;
sl@0: 	ret->subno = 0;
sl@0: 	ret->min = ret->max = 1;
sl@0: 	ret->left = NULL;
sl@0: 	ret->right = NULL;
sl@0: 	ret->begin = begin;
sl@0: 	ret->end = end;
sl@0: 	ZAPCNFA(ret->cnfa);
sl@0: 
sl@0: 	return ret;
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - freesubre - free a subRE subtree
sl@0:  ^ static VOID freesubre(struct vars *, struct subre *);
sl@0:  */
sl@0: static VOID
sl@0: freesubre(v, sr)
sl@0: struct vars *v;			/* might be NULL */
sl@0: struct subre *sr;
sl@0: {
sl@0: 	if (sr == NULL)
sl@0: 		return;
sl@0: 
sl@0: 	if (sr->left != NULL)
sl@0: 		freesubre(v, sr->left);
sl@0: 	if (sr->right != NULL)
sl@0: 		freesubre(v, sr->right);
sl@0: 
sl@0: 	freesrnode(v, sr);
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - freesrnode - free one node in a subRE subtree
sl@0:  ^ static VOID freesrnode(struct vars *, struct subre *);
sl@0:  */
sl@0: static VOID
sl@0: freesrnode(v, sr)
sl@0: struct vars *v;			/* might be NULL */
sl@0: struct subre *sr;
sl@0: {
sl@0: 	if (sr == NULL)
sl@0: 		return;
sl@0: 
sl@0: 	if (!NULLCNFA(sr->cnfa))
sl@0: 		freecnfa(&sr->cnfa);
sl@0: 	sr->flags = 0;
sl@0: 
sl@0: 	if (v != NULL) {
sl@0: 		sr->left = v->treefree;
sl@0: 		v->treefree = sr;
sl@0: 	} else
sl@0: 		FREE(sr);
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - optst - optimize a subRE subtree
sl@0:  ^ static VOID optst(struct vars *, struct subre *);
sl@0:  */
sl@0: static VOID
sl@0: optst(v, t)
sl@0: struct vars *v;
sl@0: struct subre *t;
sl@0: {
sl@0: 	if (t == NULL)
sl@0: 		return;
sl@0: 
sl@0: 	/* recurse through children */
sl@0: 	if (t->left != NULL)
sl@0: 		optst(v, t->left);
sl@0: 	if (t->right != NULL)
sl@0: 		optst(v, t->right);
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - numst - number tree nodes (assigning retry indexes)
sl@0:  ^ static int numst(struct subre *, int);
sl@0:  */
sl@0: static int			/* next number */
sl@0: numst(t, start)
sl@0: struct subre *t;
sl@0: int start;			/* starting point for subtree numbers */
sl@0: {
sl@0: 	int i;
sl@0: 
sl@0: 	assert(t != NULL);
sl@0: 
sl@0: 	i = start;
sl@0: 	t->retry = (short)i++;
sl@0: 	if (t->left != NULL)
sl@0: 		i = numst(t->left, i);
sl@0: 	if (t->right != NULL)
sl@0: 		i = numst(t->right, i);
sl@0: 	return i;
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - markst - mark tree nodes as INUSE
sl@0:  ^ static VOID markst(struct subre *);
sl@0:  */
sl@0: static VOID
sl@0: markst(t)
sl@0: struct subre *t;
sl@0: {
sl@0: 	assert(t != NULL);
sl@0: 
sl@0: 	t->flags |= INUSE;
sl@0: 	if (t->left != NULL)
sl@0: 		markst(t->left);
sl@0: 	if (t->right != NULL)
sl@0: 		markst(t->right);
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - cleanst - free any tree nodes not marked INUSE
sl@0:  ^ static VOID cleanst(struct vars *);
sl@0:  */
sl@0: static VOID
sl@0: cleanst(v)
sl@0: struct vars *v;
sl@0: {
sl@0: 	struct subre *t;
sl@0: 	struct subre *next;
sl@0: 
sl@0: 	for (t = v->treechain; t != NULL; t = next) {
sl@0: 		next = t->chain;
sl@0: 		if (!(t->flags&INUSE))
sl@0: 			FREE(t);
sl@0: 	}
sl@0: 	v->treechain = NULL;
sl@0: 	v->treefree = NULL;		/* just on general principles */
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - nfatree - turn a subRE subtree into a tree of compacted NFAs
sl@0:  ^ static long nfatree(struct vars *, struct subre *, FILE *);
sl@0:  */
sl@0: static long			/* optimize results from top node */
sl@0: nfatree(v, t, f)
sl@0: struct vars *v;
sl@0: struct subre *t;
sl@0: FILE *f;			/* for debug output */
sl@0: {
sl@0: 	assert(t != NULL && t->begin != NULL);
sl@0: 
sl@0: 	if (t->left != NULL)
sl@0: 		(DISCARD)nfatree(v, t->left, f);
sl@0: 	if (t->right != NULL)
sl@0: 		(DISCARD)nfatree(v, t->right, f);
sl@0: 
sl@0: 	return nfanode(v, t, f);
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - nfanode - do one NFA for nfatree
sl@0:  ^ static long nfanode(struct vars *, struct subre *, FILE *);
sl@0:  */
sl@0: static long			/* optimize results */
sl@0: nfanode(v, t, f)
sl@0: struct vars *v;
sl@0: struct subre *t;
sl@0: FILE *f;			/* for debug output */
sl@0: {
sl@0: 	struct nfa *nfa;
sl@0: 	long ret = 0;
sl@0: 	char idbuf[50];
sl@0: 
sl@0: 	assert(t->begin != NULL);
sl@0: 
sl@0: 	if (f != NULL)
sl@0: 		fprintf(f, "\n\n\n========= TREE NODE %s ==========\n",
sl@0: 						stid(t, idbuf, sizeof(idbuf)));
sl@0: 	nfa = newnfa(v, v->cm, v->nfa);
sl@0: 	NOERRZ();
sl@0: 	dupnfa(nfa, t->begin, t->end, nfa->init, nfa->final);
sl@0: 	if (!ISERR()) {
sl@0: 		specialcolors(nfa);
sl@0: 		ret = optimize(nfa, f);
sl@0: 	}
sl@0: 	if (!ISERR())
sl@0: 		compact(nfa, &t->cnfa);
sl@0: 
sl@0: 	freenfa(nfa);
sl@0: 	return ret;
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - newlacon - allocate a lookahead-constraint subRE
sl@0:  ^ static int newlacon(struct vars *, struct state *, struct state *, int);
sl@0:  */
sl@0: static int			/* lacon number */
sl@0: newlacon(v, begin, end, pos)
sl@0: struct vars *v;
sl@0: struct state *begin;
sl@0: struct state *end;
sl@0: int pos;
sl@0: {
sl@0: 	int n;
sl@0: 	struct subre *sub;
sl@0: 
sl@0: 	if (v->nlacons == 0) {
sl@0: 		v->lacons = (struct subre *)MALLOC(2 * sizeof(struct subre));
sl@0: 		n = 1;		/* skip 0th */
sl@0: 		v->nlacons = 2;
sl@0: 	} else {
sl@0: 		v->lacons = (struct subre *)REALLOC(v->lacons,
sl@0: 					(v->nlacons+1)*sizeof(struct subre));
sl@0: 		n = v->nlacons++;
sl@0: 	}
sl@0: 	if (v->lacons == NULL) {
sl@0: 		ERR(REG_ESPACE);
sl@0: 		return 0;
sl@0: 	}
sl@0: 	sub = &v->lacons[n];
sl@0: 	sub->begin = begin;
sl@0: 	sub->end = end;
sl@0: 	sub->subno = pos;
sl@0: 	ZAPCNFA(sub->cnfa);
sl@0: 	return n;
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - freelacons - free lookahead-constraint subRE vector
sl@0:  ^ static VOID freelacons(struct subre *, int);
sl@0:  */
sl@0: static VOID
sl@0: freelacons(subs, n)
sl@0: struct subre *subs;
sl@0: int n;
sl@0: {
sl@0: 	struct subre *sub;
sl@0: 	int i;
sl@0: 
sl@0: 	assert(n > 0);
sl@0: 	for (sub = subs + 1, i = n - 1; i > 0; sub++, i--)	/* no 0th */
sl@0: 		if (!NULLCNFA(sub->cnfa))
sl@0: 			freecnfa(&sub->cnfa);
sl@0: 	FREE(subs);
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - rfree - free a whole RE (insides of regfree)
sl@0:  ^ static VOID rfree(regex_t *);
sl@0:  */
sl@0: static VOID
sl@0: rfree(re)
sl@0: regex_t *re;
sl@0: {
sl@0: 	struct guts *g;
sl@0: 
sl@0: 	if (re == NULL || re->re_magic != REMAGIC)
sl@0: 		return;
sl@0: 
sl@0: 	re->re_magic = 0;	/* invalidate RE */
sl@0: 	g = (struct guts *)re->re_guts;
sl@0: 	re->re_guts = NULL;
sl@0: 	re->re_fns = NULL;
sl@0: 	g->magic = 0;
sl@0: 	freecm(&g->cmap);
sl@0: 	if (g->tree != NULL)
sl@0: 		freesubre((struct vars *)NULL, g->tree);
sl@0: 	if (g->lacons != NULL)
sl@0: 		freelacons(g->lacons, g->nlacons);
sl@0: 	if (!NULLCNFA(g->search))
sl@0: 		freecnfa(&g->search);
sl@0: 	FREE(g);
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - dump - dump an RE in human-readable form
sl@0:  ^ static VOID dump(regex_t *, FILE *);
sl@0:  */
sl@0: static VOID
sl@0: dump(re, f)
sl@0: regex_t *re;
sl@0: FILE *f;
sl@0: {
sl@0: #ifdef REG_DEBUG
sl@0: 	struct guts *g;
sl@0: 	int i;
sl@0: 
sl@0: 	if (re->re_magic != REMAGIC)
sl@0: 		fprintf(f, "bad magic number (0x%x not 0x%x)\n", re->re_magic,
sl@0: 								REMAGIC);
sl@0: 	if (re->re_guts == NULL) {
sl@0: 		fprintf(f, "NULL guts!!!\n");
sl@0: 		return;
sl@0: 	}
sl@0: 	g = (struct guts *)re->re_guts;
sl@0: 	if (g->magic != GUTSMAGIC)
sl@0: 		fprintf(f, "bad guts magic number (0x%x not 0x%x)\n", g->magic,
sl@0: 								GUTSMAGIC);
sl@0: 
sl@0: 	fprintf(f, "\n\n\n========= DUMP ==========\n");
sl@0: 	fprintf(f, "nsub %d, info 0%lo, csize %d, ntree %d\n", 
sl@0: 		re->re_nsub, re->re_info, re->re_csize, g->ntree);
sl@0: 
sl@0: 	dumpcolors(&g->cmap, f);
sl@0: 	if (!NULLCNFA(g->search)) {
sl@0: 		printf("\nsearch:\n");
sl@0: 		dumpcnfa(&g->search, f);
sl@0: 	}
sl@0: 	for (i = 1; i < g->nlacons; i++) {
sl@0: 		fprintf(f, "\nla%d (%s):\n", i,
sl@0: 				(g->lacons[i].subno) ? "positive" : "negative");
sl@0: 		dumpcnfa(&g->lacons[i].cnfa, f);
sl@0: 	}
sl@0: 	fprintf(f, "\n");
sl@0: 	dumpst(g->tree, f, 0);
sl@0: #endif
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - dumpst - dump a subRE tree
sl@0:  ^ static VOID dumpst(struct subre *, FILE *, int);
sl@0:  */
sl@0: static VOID
sl@0: dumpst(t, f, nfapresent)
sl@0: struct subre *t;
sl@0: FILE *f;
sl@0: int nfapresent;			/* is the original NFA still around? */
sl@0: {
sl@0: 	if (t == NULL)
sl@0: 		fprintf(f, "null tree\n");
sl@0: 	else
sl@0: 		stdump(t, f, nfapresent);
sl@0: 	fflush(f);
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - stdump - recursive guts of dumpst
sl@0:  ^ static VOID stdump(struct subre *, FILE *, int);
sl@0:  */
sl@0: static VOID
sl@0: stdump(t, f, nfapresent)
sl@0: struct subre *t;
sl@0: FILE *f;
sl@0: int nfapresent;			/* is the original NFA still around? */
sl@0: {
sl@0: 	char idbuf[50];
sl@0: 
sl@0: 	fprintf(f, "%s. `%c'", stid(t, idbuf, sizeof(idbuf)), t->op);
sl@0: 	if (t->flags&LONGER)
sl@0: 		fprintf(f, " longest");
sl@0: 	if (t->flags&SHORTER)
sl@0: 		fprintf(f, " shortest");
sl@0: 	if (t->flags&MIXED)
sl@0: 		fprintf(f, " hasmixed");
sl@0: 	if (t->flags&CAP)
sl@0: 		fprintf(f, " hascapture");
sl@0: 	if (t->flags&BACKR)
sl@0: 		fprintf(f, " hasbackref");
sl@0: 	if (!(t->flags&INUSE))
sl@0: 		fprintf(f, " UNUSED");
sl@0: 	if (t->subno != 0)
sl@0: 		fprintf(f, " (#%d)", t->subno);
sl@0: 	if (t->min != 1 || t->max != 1) {
sl@0: 		fprintf(f, " {%d,", t->min);
sl@0: 		if (t->max != INFINITY)
sl@0: 			fprintf(f, "%d", t->max);
sl@0: 		fprintf(f, "}");
sl@0: 	}
sl@0: 	if (nfapresent)
sl@0: 		fprintf(f, " %ld-%ld", (long)t->begin->no, (long)t->end->no);
sl@0: 	if (t->left != NULL)
sl@0: 		fprintf(f, " L:%s", stid(t->left, idbuf, sizeof(idbuf)));
sl@0: 	if (t->right != NULL)
sl@0: 		fprintf(f, " R:%s", stid(t->right, idbuf, sizeof(idbuf)));
sl@0: 	if (!NULLCNFA(t->cnfa)) {
sl@0: 		fprintf(f, "\n");
sl@0: 		dumpcnfa(&t->cnfa, f);
sl@0: 		fprintf(f, "\n");
sl@0: 	}
sl@0: 	if (t->left != NULL)
sl@0: 		stdump(t->left, f, nfapresent);
sl@0: 	if (t->right != NULL)
sl@0: 		stdump(t->right, f, nfapresent);
sl@0: }
sl@0: 
sl@0: /*
sl@0:  - stid - identify a subtree node for dumping
sl@0:  ^ static char *stid(struct subre *, char *, size_t);
sl@0:  */
sl@0: static char *			/* points to buf or constant string */
sl@0: stid(t, buf, bufsize)
sl@0: struct subre *t;
sl@0: char *buf;
sl@0: size_t bufsize;
sl@0: {
sl@0: 	/* big enough for hex int or decimal t->retry? */
sl@0: 	if (bufsize < sizeof(void*)*2 + 3 || bufsize < sizeof(t->retry)*3 + 1)
sl@0: 		return "unable";
sl@0: 	if (t->retry != 0)
sl@0: 		sprintf(buf, "%d", t->retry);
sl@0: 	else
sl@0: 		sprintf(buf, "%p", t);
sl@0: 	return buf;
sl@0: }
sl@0: 
sl@0: #include "regc_lex.c"
sl@0: #include "regc_color.c"
sl@0: #include "regc_nfa.c"
sl@0: #include "regc_cvec.c"
sl@0: #include "regc_locale.c"