FliteEngine - Objective-C speech synthesizer

/* This is an altered version.  It was originally altered for use in
   the Edinburgh Speech Tools by Richard Caley and others.  It was
   later altered for use in Flite by Alan W Black.  It was then
   further altered by David Huggins-Daines.  Identifiers and
   structures have been changed, but the basic structure of the code
   is intact. */

/*
 * regcomp and regexec -- regsub and regerror are elsewhere
 *
 *	Copyright (c) 1986 by University of Toronto.
 *	Written by Henry Spencer.  Not derived from licensed software.
 *
 *	Permission is granted to anyone to use this software for any
 *	purpose on any computer system, and to redistribute it freely,
 *	subject to the following restrictions:
 *
 *	1. The author is not responsible for the consequences of use of
 *		this software, no matter how awful, even if they arise
 *		from defects in it.
 *
 *	2. The origin of this software must not be misrepresented, either
 *		by explicit claim or by omission.
 *
 *	3. Altered versions must be plainly marked as such, and must not
 *		be misrepresented as being the original software.
 *** THIS IS AN ALTERED VERSION.  It was altered by John Gilmore,
 *** hoptoad!gnu, on 27 Dec 1986, to add \n as an alternative to |
 *** to assist in implementing egrep.
 *** THIS IS AN ALTERED VERSION.  It was altered by John Gilmore,
 *** hoptoad!gnu, on 27 Dec 1986, to add \< and \> for word-matching
 *** as in BSD grep and ex.
 *** THIS IS AN ALTERED VERSION.  It was altered by John Gilmore,
 *** hoptoad!gnu, on 28 Dec 1986, to optimize characters quoted with \.
 *** THIS IS AN ALTERED VERSION.  It was altered by James A. Woods,
 *** ames!jaw, on 19 June 1987, to quash a regcomp() redundancy.
 *
 * Beware that some of this code is subtly aware of the way operator
 * precedence is structured in regular expressions.  Serious changes in
 * regular-expression syntax might require a total rethink.  */
#include <ctype.h>
#include "cst_alloc.h"
#include "cst_string.h"
#include "cst_file.h"
#include "cst_error.h"
#include "cst_regex.h"

/*
 * The "internal use only" fields in regexp.h are present to pass info from
 * compile to execute that permits the execute phase to run lots faster on
 * simple cases.  They are:
 *
 * regstart	char that must begin a match; '\0' if none obvious
 * reganch	is the match anchored (at beginning-of-line only)?
 * regmust	string (pointer into program) that match must include, or NULL
 * regmlen	length of regmust string
 *
 * Regstart and reganch permit very fast decisions on suitable starting points
 * for a match, cutting down the work a lot.  Regmust permits fast rejection
 * of lines that cannot possibly match.  The regmust tests are costly enough
 * that regcomp() supplies a regmust only if the r.e. contains something
 * potentially expensive (at present, the only such thing detected is * or +
 * at the start of the r.e., which can involve a lot of backup).  Regmlen is
 * supplied because the test in regexec() needs it and regcomp() is computing
 * it anyway.
 */

/*
 * Structure for regexp "program".  This is essentially a linear encoding
 * of a nondeterministic finite-state machine (aka syntax charts or
 * "railroad normal form" in parsing technology).  Each node is an opcode
 * plus a "next" pointer, possibly plus an operand.  "Next" pointers of
 * all nodes except BRANCH implement concatenation; a "next" pointer with
 * a BRANCH on both ends of it is connecting two alternatives.  (Here we
 * have one of the subtle syntax dependencies:  an individual BRANCH (as
 * opposed to a collection of them) is never concatenated with anything
 * because of operator precedence.)  The operand of some types of node is
 * a literal string; for others, it is a node leading into a sub-FSM.  In
 * particular, the operand of a BRANCH node is the first node of the branch.
 * (NB this is *not* a tree structure:  the tail of the branch connects
 * to the thing following the set of BRANCHes.)  The opcodes are:
 */

/* definition	number	opnd?	meaning */
#define	END	0	/* no	End of program. */
#define	BOL	1	/* no	Match "" at beginning of line. */
#define	EOL	2	/* no	Match "" at end of line. */
#define	ANY	3	/* no	Match any one character. */
#define	ANYOF	4	/* str	Match any character in this string. */
#define	ANYBUT	5	/* str	Match any character not in this string. */
#define	BRANCH	6	/* node	Match this alternative, or the next... */
#define	BACK	7	/* no	Match "", "next" ptr points backward. */
#define	EXACTLY	8	/* str	Match this string. */
#define	NOTHING	9	/* no	Match empty string. */
#define	STAR	10	/* node	Match this (simple) thing 0 or more times. */
#define	PLUS	11	/* node	Match this (simple) thing 1 or more times. */
#define	WORDA	12	/* no	Match "" at wordchar, where prev is nonword */
#define	WORDZ	13	/* no	Match "" at nonwordchar, where prev is word */
#define	OPEN	20	/* no	Mark this point in input as start of #n. */
			/*	OPEN+1 is number 1, etc. */
#define	CLOSE	30	/* no	Analogous to OPEN. */

/*
 * Opcode notes:
 *
 * BRANCH	The set of branches constituting a single choice are hooked
 *		together with their "next" pointers, since precedence prevents
 *		anything being concatenated to any individual branch.  The
 *		"next" pointer of the last BRANCH in a choice points to the
 *		thing following the whole choice.  This is also where the
 *		final "next" pointer of each individual branch points; each
 *		branch starts with the operand node of a BRANCH node.
 *
 * BACK		Normal "next" pointers all implicitly point forward; BACK
 *		exists to make loop structures possible.
 *
 * STAR,PLUS	'?', and complex '*' and '+', are implemented as circular
 *		BRANCH structures using BACK.  Simple cases (one character
 *		per match) are implemented with STAR and PLUS for speed
 *		and to minimize recursive plunges.
 *
 * OPEN,CLOSE	...are numbered at compile time.
 */

/*
 * A node is one char of opcode followed by two chars of "next" pointer.
 * "Next" pointers are stored as two 8-bit pieces, high order first.  The
 * value is a positive offset from the opcode of the node containing it.
 * An operand, if any, simply follows the node.  (Note that much of the
 * code generation knows about this implicit relationship.)
 *
 * Using two bytes for the "next" pointer is vast overkill for most things,
 * but allows patterns to get big without disasters.
 */
#define	OP(p)	(*(p))
#define	NEXT(p)	(((*((p)+1)&0377)<<8) + (*((p)+2)&0377))
#define	OPERAND(p)	((p) + 3)

/*
 * See regmagic.h for one further detail of program structure.
 */


/*
 * Utility definitions.
 */
#ifndef CHARBITS
#define	UCHARAT(p)	((int)*(unsigned char *)(p))
#else
#define	UCHARAT(p)	((int)*(p)&CHARBITS)
#endif

#define	FAIL(m)	{ cst_errmsg("regexp failure: %s\n", m); cst_error(); }
#define	ISMULT(c)	((c) == '*' || (c) == '+' || (c) == '?')

/*
 * Flags to be passed up and down.
 */
#define	HASWIDTH	01	/* Known never to match null string. */
#define	SIMPLE		02	/* Simple enough to be STAR/PLUS operand. */
#define	SPSTART		04	/* Starts with * or +. */
#define	WORST		0	/* Worst case. */

/*
 * Global work variables for regcomp().
 */
static const char *regparse;		/* Input-scan pointer. */
static int regnpar;		/* () count. */
static char regdummy;
static char *regcode;		/* Code-emit pointer; &regdummy = don't. */
static long regsize;		/* Code size. */

/*
 * Forward declarations for regcomp()'s friends.
 */
#ifndef STATIC
#define	STATIC	static
#endif
STATIC char *reg(int paren, int *flagp);
STATIC char *regbranch(int *flagp);
STATIC char *regpiece(int *flagp);
STATIC char *regatom(int *flagp);
STATIC char *regnode(char op);
STATIC char *regnext(register char *p);
STATIC void regc(char b);
STATIC void reginsert(char op, char *opnd);
STATIC void regtail(char *p, char *val);
STATIC void regoptail(char *p, char *val);
#ifdef STRCSPN
STATIC int strcspn();
#endif

/*
 - regcomp - compile a regular expression into internal code
 *
 * We can't allocate space until we know how big the compiled form will be,
 * but we can't compile it (and thus know how big it is) until we've got a
 * place to put the code.  So we cheat:  we compile it twice, once with code
 * generation turned off and size counting turned on, and once "for real".
 * This also means that we don't allocate space until we are sure that the
 * thing really will compile successfully, and we never have to move the
 * code and thus invalidate pointers into it.  (Note that it has to be in
 * one piece because free() must be able to free it all.)
 *
 * Beware that the optimization-preparation code in here knows about some
 * of the structure of the compiled regexp.
 */
cst_regex *
hs_regcomp(const char *exp)
{
        cst_regex *r;
	char *scan;
	char *longest;
	unsigned int len;
	int flags;
	if (exp == NULL)
		FAIL("NULL argument");

	/* First pass: determine size, legality. */
#ifdef notdef
	if (exp[0] == '.' && exp[1] == '*') exp += 2;  /* aid grep */
#endif
	regparse = exp;
	regnpar = 1;
	regsize = 0L;
	regcode = &regdummy;
	regc(CST_REGMAGIC);
	if (reg(0, &flags) == NULL)
		return(NULL);

	/* Small enough for pointer-storage convention? */
	if (regsize >= 32767L)		/* Probably could be 65535L. */
		FAIL("regexp too big");

	/* Allocate space. */
	r = cst_alloc(cst_regex,1);
	r->regsize = regsize;
	r->program = cst_alloc(char,regsize);
	if (r == NULL)
		FAIL("out of space");

	/* Second pass: emit code. */
	regparse = exp;
	regnpar = 1;
	regcode = r->program;
	regc(CST_REGMAGIC);
	if (reg(0, &flags) == NULL)
		return(NULL);

	/* Dig out information for optimizations. */
	r->regstart = '\0';	/* Worst-case defaults. */
	r->reganch = 0;
	r->regmust = NULL;
	r->regmlen = 0;
	scan = r->program+1;			/* First BRANCH. */

	if (OP(regnext(scan)) == END) {		/* Only one top-level choice. */
		scan = OPERAND(scan);

		/* Starting-point info. */
		if (OP(scan) == EXACTLY)
			r->regstart = *OPERAND(scan);
		else if (OP(scan) == BOL)
			r->reganch++;

		/*
		 * If there's something expensive in the r.e., find the
		 * longest literal string that must appear and make it the
		 * regmust.  Resolve ties in favor of later strings, since
		 * the regstart check works with the beginning of the r.e.
		 * and avoiding duplication strengthens checking.  Not a
		 * strong reason, but sufficient in the absence of others.
		 */
		if (flags&SPSTART) {
			longest = NULL;
			len = 0;
			for (; scan != NULL; scan = regnext(scan))
				if ((OP(scan) == EXACTLY) && 
				    (strlen(OPERAND(scan)) >= len)) {
					longest = OPERAND(scan);
					len = strlen(OPERAND(scan));
				}
			r->regmust = longest;
			r->regmlen = len;
		}
	}
	return(r);

}

void
hs_regdelete(cst_regex *r)
{
	cst_free(r->program);
	cst_free(r);
}

/*
 - reg - regular expression, i.e. main body or parenthesized thing
 *
 * Caller must absorb opening parenthesis.
 *
 * Combining parenthesis handling with the base level of regular expression
 * is a trifle forced, but the need to tie the tails of the branches to what
 * follows makes it hard to avoid.
 */
static char *
reg(int paren, int *flagp)
          			/* Parenthesized? */
           
{
	char *ret;
	char *br;
	char *ender;
	int parno=0;
	int flags;

	*flagp = HASWIDTH;	/* Tentatively. */

	/* Make an OPEN node, if parenthesized. */
	if (paren) {
		if (regnpar >= CST_NSUBEXP)
			FAIL("too many ()");
		parno = regnpar;
		regnpar++;
		ret = regnode(OPEN+parno);
	} else
		ret = NULL;

	/* Pick up the branches, linking them together. */
	br = regbranch(&flags);
	if (br == NULL)
		return(NULL);
	if (ret != NULL)
		regtail(ret, br);	/* OPEN -> first. */
	else
		ret = br;
	if (!(flags&HASWIDTH))
		*flagp &= ~HASWIDTH;
	*flagp |= flags&SPSTART;
	while (*regparse == '|' || *regparse == '\n') {
		regparse++;
		br = regbranch(&flags);
		if (br == NULL)
			return(NULL);
		regtail(ret, br);	/* BRANCH -> BRANCH. */
		if (!(flags&HASWIDTH))
			*flagp &= ~HASWIDTH;
		*flagp |= flags&SPSTART;
	}

	/* Make a closing node, and hook it on the end. */
	ender = regnode((paren) ? CLOSE+parno : END);
	regtail(ret, ender);

	/* Hook the tails of the branches to the closing node. */
	for (br = ret; br != NULL; br = regnext(br))
		regoptail(br, ender);

	/* Check for proper termination. */
	if (paren && *regparse++ != ')') {
		FAIL("unmatched ()");
	} else if (!paren && *regparse != '\0') {
		if (*regparse == ')') {
			FAIL("unmatched ()");
		} else
			FAIL("junk on end");	/* "Can't happen". */
		/* NOTREACHED */
	}

	return(ret);
}

/*
 - regbranch - one alternative of an | operator
 *
 * Implements the concatenation operator.
 */
static char *
regbranch(int *flagp)
{
	char *ret;
	char *chain;
	char *latest;
	int flags;

	*flagp = WORST;		/* Tentatively. */

	ret = regnode(BRANCH);
	chain = NULL;
	while (*regparse != '\0' && *regparse != ')' &&
	       *regparse != '\n' && *regparse != '|') {
		latest = regpiece(&flags);
		if (latest == NULL)
			return(NULL);
		*flagp |= flags&HASWIDTH;
		if (chain == NULL)	/* First piece. */
			*flagp |= flags&SPSTART;
		else
			regtail(chain, latest);
		chain = latest;
	}
	if (chain == NULL)	/* Loop ran zero times. */
		(void) regnode(NOTHING);

	return(ret);
}

/*
 - regpiece - something followed by possible [*+?]
 *
 * Note that the branching code sequences used for ? and the general cases
 * of * and + are somewhat optimized:  they use the same NOTHING node as
 * both the endmarker for their branch list and the body of the last branch.
 * It might seem that this node could be dispensed with entirely, but the
 * endmarker role is not redundant.
 */
static char *
regpiece(int *flagp)
{
	char *ret;
	char op;
	char *next;
	int flags;

	ret = regatom(&flags);
	if (ret == NULL)
		return(NULL);

	op = *regparse;
	if (!ISMULT(op)) {
		*flagp = flags;
		return(ret);
	}

	if (!(flags&HASWIDTH) && op != '?')
		FAIL("*+ operand could be empty");
	*flagp = (op != '+') ? (WORST|SPSTART) : (WORST|HASWIDTH);

	if (op == '*' && (flags&SIMPLE))
		reginsert(STAR, ret);
	else if (op == '*') {
		/* Emit x* as (x&|), where & means "self". */
		reginsert(BRANCH, ret);			/* Either x */
		regoptail(ret, regnode(BACK));		/* and loop */
		regoptail(ret, ret);			/* back */
		regtail(ret, regnode(BRANCH));		/* or */
		regtail(ret, regnode(NOTHING));		/* null. */
	} else if (op == '+' && (flags&SIMPLE))
		reginsert(PLUS, ret);
	else if (op == '+') {
		/* Emit x+ as x(&|), where & means "self". */
		next = regnode(BRANCH);			/* Either */
		regtail(ret, next);
		regtail(regnode(BACK), ret);		/* loop back */
		regtail(next, regnode(BRANCH));		/* or */
		regtail(ret, regnode(NOTHING));		/* null. */
	} else if (op == '?') {
		/* Emit x? as (x|) */
		reginsert(BRANCH, ret);			/* Either x */
		regtail(ret, regnode(BRANCH));		/* or */
		next = regnode(NOTHING);		/* null. */
		regtail(ret, next);
		regoptail(ret, next);
	}
	regparse++;
	if (ISMULT(*regparse))
		FAIL("nested *?+");

	return(ret);
}

/*
 - regatom - the lowest level
 *
 * Optimization:  gobbles an entire sequence of ordinary characters so that
 * it can turn them into a single node, which is smaller to store and
 * faster to run.  Backslashed characters are exceptions, each becoming a
 * separate node; the code is simpler that way and it's not worth fixing.
 */
static char *
regatom(int *flagp)
{
	char *ret = NULL;
	int flags;

	*flagp = WORST;		/* Tentatively. */

	switch (*regparse++) {
	/* FIXME: these chars only have meaning at beg/end of pat? */
	case '^':
		ret = regnode(BOL);
		break;
	case '$':
		ret = regnode(EOL);
		break;
	case '.':
		ret = regnode(ANY);
		*flagp |= HASWIDTH|SIMPLE;
		break;
	case '[': {
			int class1;
			int classend;

			if (*regparse == '^') {	/* Complement of range. */
				ret = regnode(ANYBUT);
				regparse++;
			} else
				ret = regnode(ANYOF);
			if (*regparse == ']' || *regparse == '-')
				regc(*regparse++);
			while (*regparse != '\0' && *regparse != ']') {
				if (*regparse == '-') {
					regparse++;
					if (*regparse == ']' || *regparse == '\0')
						regc('-');
					else {
						class1 = UCHARAT(regparse-2)+1;
						classend = UCHARAT(regparse);
						if (class1 > classend+1)
							FAIL("invalid [] range");
						for (; class1 <= classend; class1++)
							regc(class1);
						regparse++;
					}
				} else
					regc(*regparse++);
			}
			regc('\0');
			if (*regparse != ']')
				FAIL("unmatched []");
			regparse++;
			*flagp |= HASWIDTH|SIMPLE;
		}
		break;
	case '(':
		ret = reg(1, &flags);
		if (ret == NULL)
			return(NULL);
		*flagp |= flags&(HASWIDTH|SPSTART);
		break;
	case '\0':
	case '|':
	case '\n':
	case ')':
		FAIL("internal urp");	/* Supposed to be caught earlier. */
		break;
	case '?':
	case '+':
	case '*':
		FAIL("?+* follows nothing");
		break;
	case '\\':
		switch (*regparse++) {
		case '\0':
			FAIL("trailing \\");
			break;
		case '<':
			ret = regnode(WORDA);
			break;
		case '>':
			ret = regnode(WORDZ);
			break;
		/* FIXME: Someday handle \1, \2, ... */
		default:
			/* Handle general quoted chars in exact-match routine */
			goto de_fault;
		}
		break;
	de_fault:
	default:
		/*
		 * Encode a string of characters to be matched exactly.
		 *
		 * This is a bit tricky due to quoted chars and due to
		 * '*', '+', and '?' taking the SINGLE char previous
		 * as their operand.
		 *
		 * On entry, the char at regparse[-1] is going to go
		 * into the string, no matter what it is.  (It could be
		 * following a \ if we are entered from the '\' case.)
		 *
		 * Basic idea is to pick up a good char in  ch  and
		 * examine the next char.  If it's *+? then we twiddle.
		 * If it's \ then we frozzle.  If it's other magic char
		 * we push  ch  and terminate the string.  If none of the
		 * above, we push  ch  on the string and go around again.
		 *
		 *  regprev  is used to remember where "the current char"
		 * starts in the string, if due to a *+? we need to back
		 * up and put the current char in a separate, 1-char, string.
		 * When  regprev  is NULL,  ch  is the only char in the
		 * string; this is used in *+? handling, and in setting
		 * flags |= SIMPLE at the end.
		 */
		{
			const char *regprev;
			char ch = 0;

			regparse--;			/* Look at cur char */
			ret = regnode(EXACTLY);
			for ( regprev = 0 ; ; ) {
				ch = *regparse++;	/* Get current char */
				switch (*regparse) {	/* look at next one */

				default:
					regc(ch);	/* Add cur to string */
					break;

				case '.': case '[': case '(':
				case ')': case '|': case '\n':
				case '$': case '^':
				case '\0':
				/* FIXME, $ and ^ should not always be magic */
				magic:
					regc(ch);	/* dump cur char */
					goto done;	/* and we are done */

				case '?': case '+': case '*':
					if (!regprev) 	/* If just ch in str, */
						goto magic;	/* use it */
					/* End mult-char string one early */
					regparse = regprev; /* Back up parse */
					goto done;

				case '\\':
					regc(ch);	/* Cur char OK */
					switch (regparse[1]){ /* Look after \ */
					case '\0':
					case '<':
					case '>':
					/* FIXME: Someday handle \1, \2, ... */
						goto done; /* Not quoted */
					default:
						/* Backup point is \, scan							 * point is after it. */
						regprev = regparse;
						regparse++;
						continue;	/* NOT break; */
					}
				}
				regprev = regparse;	/* Set backup point */
			}
		done:
			regc('\0');
			*flagp |= HASWIDTH;
			if (!regprev)		/* One char? */
				*flagp |= SIMPLE;
		}
		break;
	}

	return(ret);
}

/*
 - regnode - emit a node
 */
static char *			/* Location. */
regnode(char op)
{
	char *ret;
	char *ptr;

	ret = regcode;
	if (ret == &regdummy) {
		regsize += 3;
		return(ret);
	}

	ptr = ret;
	*ptr++ = op;
	*ptr++ = '\0';		/* Null "next" pointer. */
	*ptr++ = '\0';
	regcode = ptr;

	return(ret);
}

/*
 - regc - emit (if appropriate) a byte of code
 */
static void
regc(char b)
{
	if (regcode != &regdummy)
		*regcode++ = b;
	else
		regsize++;
}

/*
 - reginsert - insert an operator in front of already-emitted operand
 *
 * Means relocating the operand.
 */
static void
reginsert(char op, char *opnd)
{
	char *src;
	char *dst;
	char *place;

	if (regcode == &regdummy) {
		regsize += 3;
		return;
	}

	src = regcode;
	regcode += 3;
	dst = regcode;
	while (src > opnd)
		*--dst = *--src;

	place = opnd;		/* Op node, where operand used to be. */
	*place++ = op;
	*place++ = '\0';
	*place++ = '\0';
}

/*
 - regtail - set the next-pointer at the end of a node chain
 */
static void
regtail(char *p, char *val)
{
	char *scan;
	char *temp;
	int offset;

	if (p == &regdummy)
		return;

	/* Find last node. */
	scan = p;
	for (;;) {
		temp = regnext(scan);
		if (temp == NULL)
			break;
		scan = temp;
	}

	if (OP(scan) == BACK)
		offset = scan - val;
	else
		offset = val - scan;
	*(scan+1) = (offset>>8)&0377;
	*(scan+2) = offset&0377;
}

/*
 - regoptail - regtail on operand of first argument; nop if operandless
 */
static void
regoptail(char *p, char *val)
{
	/* "Operandless" and "op != BRANCH" are synonymous in practice. */
	if (p == NULL || p == &regdummy || OP(p) != BRANCH)
		return;
	regtail(OPERAND(p), val);
}

/*
 * regexec and friends
 */

/* dhd@cepstral.com changed all this stuff to use a state structure
   (and thus, be re-entrant) 2001-10-18 */

/*
 * Forwards.
 */
STATIC int regtry(cst_regstate *state, const char *string, char *prog);
STATIC int regmatch(cst_regstate *state, char *scan);
STATIC int regrepeat(cst_regstate *state, char *p);

#ifdef UNDER_CE
#undef DEBUG /* Just does not work on WinCE... */
#endif

#ifdef DEBUG
#define regnarrate stdout
void regdump();
STATIC char *regprop(char *scan);
#endif

/*
 - regexec - match a regexp against a string
 */
cst_regstate *
hs_regexec(const cst_regex *prog, const char *string)
{
	cst_regstate *state;
	char *s;

	/* Be paranoid... */
	if (prog == NULL || string == NULL) {
		FAIL("NULL parameter");
		return(0);
	}

	/* Check validity of program. */
	if (UCHARAT(prog->program) != CST_REGMAGIC) {
		FAIL("corrupted program");
		return(0);
	}

	/* If there is a "must appear" string, look for it. */
	if (prog->regmust != NULL) {
		s = (char *)string;
		while ((s = strchr(s, prog->regmust[0])) != NULL) {
			if (strncmp(s, prog->regmust, prog->regmlen) == 0)
				break;	/* Found it. */
			s++;
		}
		if (s == NULL)	/* Not present. */
			return(0);
	}

	state = cst_alloc(cst_regstate, 1);
	/* Mark beginning of line for ^ . */
	state->bol = string;

	/* Simplest case:  anchored match need be tried only once. */
	if (prog->reganch) {
		if (regtry(state, string, prog->program+1))
			return state;
		else {
			cst_free(state);
			return NULL;
		}
	}

	/* Messy cases:  unanchored match. */
	s = (char *)string;
	if (prog->regstart != '\0')
		/* We know what char it must start with. */
		while ((s = strchr(s, prog->regstart)) != NULL) {
			if (regtry(state, s, prog->program+1))
				return state;
			s++;
		}
	else
		/* We don't -- general case. */
		do {
			if (regtry(state, s, prog->program+1))
				return state;
		} while (*s++ != '\0');

	cst_free(state);
	return NULL;
}

/*
 - regtry - try match at specific point
 */
static int			/* 0 failure, 1 success */
regtry(cst_regstate *state, const char *string, char *prog)
{
	int i;
	const char **sp;
	const char **ep;

	state->input = string;

	sp = state->startp;
	ep = state->endp;
	for (i = CST_NSUBEXP; i > 0; i--) {
		*sp++ = NULL;
		*ep++ = NULL;
	}
	if (regmatch(state, prog)) {
	  state->startp[0] = (char *)string;
	  state->endp[0] = (char *)state->input;
		return(1);
	} else
		return(0);
}

/*
 - regmatch - main matching routine
 *
 * Conceptually the strategy is simple:  check to see whether the current
 * node matches, call self recursively to see whether the rest matches,
 * and then act accordingly.  In practice we make some effort to avoid
 * recursion, in particular by going through "ordinary" nodes (that don't
 * need to know whether the rest of the match failed) by a loop instead of
 * by recursion.
 */
static int			/* 0 failure, 1 success */
regmatch(cst_regstate *state, char *scan)
{
	char *next;		/* Next node. */

#ifdef DEBUG
	if (scan != NULL && regnarrate)
		fprintf(regnarrate, "%s(\n", regprop(scan));
#endif
	while (scan != NULL) {
#ifdef DEBUG
		if (regnarrate)
			fprintf(regnarrate, "%s...\n", regprop(scan));
#endif
		next = regnext(scan);

		switch (OP(scan)) {
		case BOL:
			if (state->input != state->bol)
				return(0);
			break;
		case EOL:
			if (*state->input != '\0')
				return(0);
			break;
		case WORDA:
			/* Must be looking at a letter, digit, or _ */
			if ((!isalnum((int)*state->input)) && *state->input != '_')
				return(0);
			/* Prev must be BOL or nonword */
			if (state->input > state->bol &&
			    (isalnum((int)state->input[-1]) || state->input[-1] == '_'))
				return(0);
			break;
		case WORDZ:
			/* Must be looking at non letter, digit, or _ */
			if (isalnum((int)*state->input) || *state->input == '_')
				return(0);
			/* We don't care what the previous char was */
			break;
		case ANY:
			if (*state->input == '\0')
				return(0);
			state->input++;
			break;
		case EXACTLY: {
				int len;
				char *opnd;

				opnd = OPERAND(scan);
				/* Inline the first character, for speed. */
				if (*opnd != *state->input)
					return(0);
				len = strlen(opnd);
				if (len > 1 && strncmp(opnd, state->input, len) != 0)
					return(0);
				state->input += len;
			}
			break;
		case ANYOF:
 			if (*state->input == '\0' || strchr(OPERAND(scan), *state->input) == NULL)
				return(0);
			state->input++;
			break;
		case ANYBUT:
 			if (*state->input == '\0' || strchr(OPERAND(scan), *state->input) != NULL)
				return(0);
			state->input++;
			break;
		case NOTHING:
			break;
		case BACK:
			break;
		case OPEN+1:
		case OPEN+2:
		case OPEN+3:
		case OPEN+4:
		case OPEN+5:
		case OPEN+6:
		case OPEN+7:
		case OPEN+8:
		case OPEN+9: {
				int no;
				const char *save;

				no = OP(scan) - OPEN;
				save = state->input;

				if (regmatch(state, next)) {
					/*
					 * Don't set startp if some later
					 * invocation of the same parentheses
					 * already has.
					 */
					if (state->startp[no] == NULL)
					  state->startp[no] = save;
					return(1);
				} else
					return(0);
			}
			break;
		case CLOSE+1:
		case CLOSE+2:
		case CLOSE+3:
		case CLOSE+4:
		case CLOSE+5:
		case CLOSE+6:
		case CLOSE+7:
		case CLOSE+8:
		case CLOSE+9: {
				int no;
				const char *save;

				no = OP(scan) - CLOSE;
				save = state->input;

				if (regmatch(state, next)) {
					/*
					 * Don't set endp if some later
					 * invocation of the same parentheses
					 * already has.
					 */
					if (state->endp[no] == NULL)
					  state->endp[no] = save;
					return(1);
				} else
					return(0);
			}
			break;
		case BRANCH: {
				const char *save;

				if (OP(next) != BRANCH)		/* No choice. */
					next = OPERAND(scan);	/* Avoid recursion. */
				else {
					do {
						save = state->input;
						if (regmatch(state, OPERAND(scan)))
							return(1);
						state->input = save;
						scan = regnext(scan);
					} while (scan != NULL && OP(scan) == BRANCH);
					return(0);
					/* NOTREACHED */
				}
			}
			break;
		case STAR:
		case PLUS: {
				char nextch;
				int no;
				const char *save;
				int min;

				/*
				 * Lookahead to avoid useless match attempts
				 * when we know what character comes next.
				 */
				nextch = '\0';
				if (OP(next) == EXACTLY)
					nextch = *OPERAND(next);
				min = (OP(scan) == STAR) ? 0 : 1;
				save = state->input;

				no = regrepeat(state, OPERAND(scan));
				while (no >= min) {
					/* If it could work, try it. */
					if (nextch == '\0' || *state->input == nextch)
					  {
					    if (regmatch(state, next))
					      return(1);
					  }
					/* Couldn't or didn't -- back up. */
					no--;
					state->input = save + no;
				}
				return(0);
			}
			break;
		case END:
			return(1);	/* Success! */
			break;
		default:
			FAIL("memory corruption");
			return(0);
			break;
		}

		scan = next;
	}

	/*
	 * We get here only if there's trouble -- normally "case END" is
	 * the terminating point.
	 */
	FAIL("corrupted pointers");
	return(0);
}

/*
 - regrepeat - repeatedly match something simple, report how many
 */
static int
regrepeat(cst_regstate *state, char *p)
{
	int count = 0;
	const char *scan;
	char *opnd;

	scan = state->input;
	opnd = OPERAND(p);
	switch (OP(p)) {
	case ANY:
		count = strlen(scan);
		scan += count;
		break;
	case EXACTLY:
		while (*opnd == *scan) {
			count++;
			scan++;
		}
		break;
	case ANYOF:
		while (*scan != '\0' && strchr(opnd, *scan) != NULL) {
			count++;
			scan++;
		}
		break;
	case ANYBUT:
		while (*scan != '\0' && strchr(opnd, *scan) == NULL) {
			count++;
			scan++;
		}
		break;
	default:		/* Oh dear.  Called inappropriately. */
		FAIL("internal foulup");
		count = 0;	/* Best compromise. */
		break;
	}
	state->input = scan;

	return(count);
}

/*
 - regnext - dig the "next" pointer out of a node
 */
static char *
regnext(char *p)
{
	int offset;

	if (p == &regdummy)
		return(NULL);

	offset = NEXT(p);
	if (offset == 0)
		return(NULL);

	if (OP(p) == BACK)
		return(p-offset);
	else
		return(p+offset);
}

#ifdef DEBUG

STATIC char *regprop(char *scan);

/*
 - regdump - dump a regexp onto stdout in vaguely comprehensible form
 */
void
regdump(cst_regex *r)
{
	char *s;
	char op = EXACTLY;	/* Arbitrary non-END op. */
	char *next;


	s = r->program + 1;
	while (op != END) {	/* While that wasn't END last time... */
		op = OP(s);
		printf("%2d%s", s-r->program, regprop(s));	/* Where, what. */
		next = regnext(s);
		if (next == NULL)		/* Next ptr. */
			printf("(0)");
		else
			printf("(%d)", (s-r->program)+(next-s));
		s += 3;
		if (op == ANYOF || op == ANYBUT || op == EXACTLY) {
			/* Literal string, where present. */
			while (*s != '\0') {
				putchar(*s);
				s++;
			}
			s++;
		}
		putchar('\n');
	}

	/* Header fields of interest. */
	if (r->regstart != '\0')
		printf("start `%c' ", r->regstart);
	if (r->reganch)
		printf("anchored ");
	if (r->regmust != NULL)
		printf("must have \"%s\"", r->regmust);
	printf("\n");
}

/*
 - regprop - printable representation of opcode
 */
static char *
regprop(char *op)
{
	char *p=NULL;
	static char buf[50];

	(void) strcpy(buf, ":");

	switch (OP(op)) {
	case BOL:
		p = "BOL";
		break;
	case EOL:
		p = "EOL";
		break;
	case ANY:
		p = "ANY";
		break;
	case ANYOF:
		p = "ANYOF";
		break;
	case ANYBUT:
		p = "ANYBUT";
		break;
	case BRANCH:
		p = "BRANCH";
		break;
	case EXACTLY:
		p = "EXACTLY";
		break;
	case NOTHING:
		p = "NOTHING";
		break;
	case BACK:
		p = "BACK";
		break;
	case END:
		p = "END";
		break;
	case OPEN+1:
	case OPEN+2:
	case OPEN+3:
	case OPEN+4:
	case OPEN+5:
	case OPEN+6:
	case OPEN+7:
	case OPEN+8:
	case OPEN+9:
	    cst_sprintf(buf+strlen(buf), "OPEN%d", OP(op)-OPEN);
		p = NULL;
		break;
	case CLOSE+1:
	case CLOSE+2:
	case CLOSE+3:
	case CLOSE+4:
	case CLOSE+5:
	case CLOSE+6:
	case CLOSE+7:
	case CLOSE+8:
	case CLOSE+9:
	    cst_sprintf(buf+strlen(buf), "CLOSE%d", OP(op)-CLOSE);
		p = NULL;
		break;
	case STAR:
		p = "STAR";
		break;
	case PLUS:
		p = "PLUS";
		break;
	case WORDA:
		p = "WORDA";
		break;
	case WORDZ:
		p = "WORDZ";
		break;
	default:
		FAIL("corrupted opcode");
		break;
	}
	if (p != NULL)
		(void) strcat(buf, p);
	return(buf);
}
#endif

/*
 * The following is provided for those people who do not have strcspn() in
 * their C libraries.  They should get off their butts and do something
 * about it; at least one public-domain implementation of those (highly
 * useful) string routines has been published on Usenet.
 */
#ifdef STRCSPN
/*
 * strcspn - find length of initial segment of s1 consisting entirely
 * of characters not from s2
 */

static int
strcspn(s1, s2)
char *s1;
char *s2;
{
	char *scan1;
	char *scan2;
	int count;

	count = 0;
	for (scan1 = s1; *scan1 != '\0'; scan1++) {
		for (scan2 = s2; *scan2 != '\0';)	/* ++ moved down. */
			if (*scan1 == *scan2++)
				return(count);
		count++;
	}
	return(count);
}
#endif