regexp.c

Boost provides free peer-reviewed portable C++ source libraries. We emphasize libraries that work

		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 )
{
	register char *ret;
	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 '[': {
			register int classr;
			register 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 {
						classr = UCHARAT(regparse-2)+1;
						classend = UCHARAT(regparse);
						if (classr > classend+1)
							FAIL("invalid [] range");
						for (; classr <= classend; classr++)
							regc(classr);
						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.
		 */
		{
			char *regprev;
			register char ch;

			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( int op )
{
	register char *ret;
	register 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( int 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 )
{
	register char *src;
	register char *dst;
	register 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 )
{
	register char *scan;
	register char *temp;
	register 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
 */

/*
 * Global work variables for regexec().
 */
static char *reginput;		/* String-input pointer. */
static char *regbol;		/* Beginning of input, for ^ check. */
static char **regstartp;	/* Pointer to startp array. */
static char **regendp;		/* Ditto for endp. */

/*
 * Forwards.
 */
STATIC int regtry( regexp *prog, char *string );
STATIC int regmatch( char *prog );
STATIC int regrepeat( char *p );

#ifdef DEBUG
int regnarrate = 0;
void regdump();
STATIC char *regprop();
#endif

/*
 - regexec - match a regexp against a string
 */
int
regexec(
	register regexp *prog,
	register char *string )
{
	register char *s;

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

	/* Check validity of program. */
	if (UCHARAT(prog->program) != MAGIC) {
		regerror("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);
	}

	/* Mark beginning of line for ^ . */
	regbol = (char *)string;

	/* Simplest case:  anchored match need be tried only once. */
	if (prog->reganch)
		return(regtry(prog, string));

	/* 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(prog, s))
				return(1);
			s++;
		}
	else
		/* We don't -- general case. */
		do {
			if (regtry(prog, s))
				return(1);
		} while (*s++ != '\0');

	/* Failure. */
	return(0);
}

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

	reginput = string;
	regstartp = prog->startp;
	regendp = prog->endp;

	sp = prog->startp;
	ep = prog->endp;
	for (i = NSUBEXP; i > 0; i--) {
		*sp++ = NULL;
		*ep++ = NULL;
	}
	if (regmatch(prog->program + 1)) {
		prog->startp[0] = string;
		prog->endp[0] = reginput;
		return(1);
	} else
		return(0);
}

/*