/*
- regbranch - one alternative of an | operator
*
* Implements the concatenation operator.
*/
static char *
regbranch(int *flagp)
{
register char *ret;
register char *chain;
register char *latest;
int flags;
*flagp = WORST; /* Tentatively. */
ret = regnode(BRANCH);
chain = NULL;
while (*regparse != '\0' && *regparse != '|' && *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);
}
/*
- 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 == (char *) NULL || p == ®dummy || OP(p) != BRANCH)
return;
regtail(OPERAND(p), val);
}
/*
- regbranch - one alternative of an | operator
*
* Implements the concatenation operator.
*/
static int regbranch(regex_t *preg, int *flagp )
{
int ret;
int chain;
int latest;
int flags;
*flagp = WORST; /* Tentatively. */
ret = regnode(preg, BRANCH);
chain = 0;
while (*preg->regparse != '\0' && *preg->regparse != ')' &&
*preg->regparse != '|') {
latest = regpiece(preg, &flags);
if (latest == 0)
return 0;
*flagp |= flags&HASWIDTH;
if (chain == 0) {/* First piece. */
*flagp |= flags&SPSTART;
}
else {
regtail(preg, chain, latest);
}
chain = latest;
}
if (chain == 0) /* Loop ran zero times. */
(void) regnode(preg, NOTHING);
return(ret);
}
TCHAR *CRegExp::regbranch(int *flagp)
{
TCHAR *ret;
TCHAR *chain;
TCHAR *latest;
int flags;
int c;
*flagp = WORST; // Tentatively.
ret = regnode(BRANCH);
chain = NULL;
while ((c = *regparse) != _T('\0') && c != _T('|') && c != _T(')')) {
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);
}
void CRegExp::regoptail(TCHAR *p, TCHAR *val)
{
// "Operandless" and "op != BRANCH" are synonymous in practice.
if (!bEmitCode || OP(p) != BRANCH)
return;
regtail(OPERAND(p), val);
}
static void regoptail(regex_t *preg, int p, int val )
{
/* "Operandless" and "op != BRANCH" are synonymous in practice. */
if (p != 0 && OP(preg, p) == BRANCH) {
regtail(preg, OPERAND(p), val);
}
}
/*
- regbranch - one alternative of an | operator
*
* Implements the concatenation operator.
*/
static char *
regbranch(int *flagp)
{
register char *ret;
register char *chain;
register char *latest;
int flags;
int len = 0;
*flagp = WORST; /* Tentatively. */
ret = regnode(BRANCH);
chain = NULL;
/* Solaris 2.6 motif diff bug 1236359 1 line */
while ((len = CHARLEN(regparse)) > 0 &&
(len != 1 || (*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);
}
/*
- regbranch - one alternative of an | operator
*
* Implements the concatenation operator.
*/
char* ossimRegExp::regbranch (int *flagp) {
char* ret;
char* chain;
char* latest;
int flags;
*flagp = WORST; // Tentatively.
ret = regnode(BRANCH);
chain = NULL;
while (*regparse != '\0' && *regparse != '|' && *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.
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)
{
register char *ret;
register char *op;
register char *next;
int flags;
int len = 0;
ret = regatom(&flags);
if (ret == NULL)
return(NULL);
op = regparse;
if (!ISMULT(op)) {
*flagp = flags;
return(ret);
}
len = CHARLEN(op);
if (!(flags & HASWIDTH) && ((len != 1) || (*op != '?')) )
FAIL("*+ operand could be empty");
*flagp = ((len != 1) || (*op != '+')) ? (WORST | SPSTART) : (WORST | HASWIDTH);
if ((len == 1) && (*op == '*') && (flags & SIMPLE))
reginsert(STAR, ret);
else if ((len == 1) &&
(*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 ((len == 1) && (*op == '+') && (flags & SIMPLE))
reginsert(PLUS, ret);
else if ((len == 1) && (*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 ((len == 1) && (*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 += INCRLEN(len);
if (ISMULT(regparse))
FAIL("nested *?+");
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(i4 *flagp)
{
register char *ret;
register char op;
register char *next;
i4 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);
}
CMnext( regparse );
if (ISMULT(*regparse))
_FAIL("nested *?+");
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, op, *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( ERR_RE_EMPTY_OPERAND );
}
*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( ERR_RE_NESTED_OPERAND );
}
return( ret );
}
/*
- regpiece - something followed by possible [*+?]
*
* Note that the branching code sequence 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)
{
register char *ret;
register short op;
register char *nxt;
int flags;
ret = regatom(&flags);
if (ret == (char *) NULL)
return ((char *) NULL);
op = *regparse;
if (!ISMULT(op)) {
*flagp = flags;
return (ret);
}
if (!(flags & HASWIDTH) && op != QMARK)
FAIL("*+ operand could be empty\n");
*flagp = (op != PLUSS) ? (WORST | SPSTART) : (WORST | HASWIDTH);
if (op == ASTERIX && (flags & SIMPLE))
reginsert(STAR, ret);
else if (op == ASTERIX) {
/* 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 == PLUSS && (flags & SIMPLE))
reginsert(PLUS, ret);
else if (op == PLUSS) {
/* Emit x+ as x(&|), where & means "self". */
nxt = regnode(BRANCH); /* Either */
regtail(ret, nxt);
regtail(regnode(BACK), ret); /* loop back */
regtail(nxt, regnode(BRANCH)); /* or */
regtail(ret, regnode(NOTHING)); /* null. */
} else if (op == QMARK) {
/* Emit x? as (x|) */
reginsert(BRANCH, ret); /* Either x */
regtail(ret, regnode(BRANCH)); /* or */
nxt = regnode(NOTHING); /* null. */
regtail(ret, nxt);
regoptail(ret, nxt);
}
regparse++;
if (ISMULT(*regparse))
FAIL("nested *?+\n");
return (ret);
}
/*
- 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.
*/
char* ossimRegExp::reg (int paren, int *flagp) {
char* ret;
char* br;
char* ender;
int parno =0;
int flags;
*flagp = HASWIDTH; // Tentatively.
// Make an OPEN node, if parenthesized.
if (paren) {
if (regnpar >= NSUBEXP) {
//RAISE Error, SYM(ossimRegExp), SYM(Too_Many_Parens),
printf ("ossimRegExp::compile(): Too many parentheses.\n");
return 0;
}
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++;
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++ != ')') {
//RAISE Error, SYM(ossimRegExp), SYM(Unmatched_Parens),
printf ("ossimRegExp::compile(): Unmatched parentheses.\n");
return 0;
}
else if (!paren && *regparse != '\0') {
if (*regparse == ')') {
//RAISE Error, SYM(ossimRegExp), SYM(Unmatched_Parens),
printf ("ossimRegExp::compile(): Unmatched parentheses.\n");
return 0;
}
else {
//RAISE Error, SYM(ossimRegExp), SYM(Internal_Error),
printf ("ossimRegExp::compile(): Internal error.\n");
return 0;
}
// NOTREACHED
}
return (ret);
}
/*
* 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 )
{
char *ret, *br, *ender;
int flags;
char parno = 0;
*flagp = HASWIDTH; /* Tentatively. */
/* Make an OPEN node, if parenthesized. */
if( paren ) {
if( regnpar >= NSUBEXP ) {
FAIL( ERR_RE_TOO_MANY_ROUND_BRACKETS );
}
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++;
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( ERR_RE_UNMATCHED_ROUND_BRACKETS );
} else if( !paren && *regparse != '\0' ) {
if( *regparse == ')' ) {
FAIL( ERR_RE_UNMATCHED_ROUND_BRACKETS );
} else {
FAIL( ERR_RE_INTERNAL_FOULUP ); /* "Can't happen". */
}
}
return( ret );
}
/*
- 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 int reg(regex_t *preg, int paren /* Parenthesized? */, int *flagp )
{
int ret;
int br;
int ender;
int parno = 0;
int flags;
*flagp = HASWIDTH; /* Tentatively. */
/* Make an OPEN node, if parenthesized. */
if (paren) {
if (preg->regparse[0] == '?' && preg->regparse[1] == ':') {
/* non-capturing paren */
preg->regparse += 2;
parno = -1;
}
else {
parno = ++preg->re_nsub;
}
ret = regnode(preg, OPEN+parno);
} else
ret = 0;
/* Pick up the branches, linking them together. */
br = regbranch(preg, &flags);
if (br == 0)
return 0;
if (ret != 0)
regtail(preg, ret, br); /* OPEN -> first. */
else
ret = br;
if (!(flags&HASWIDTH))
*flagp &= ~HASWIDTH;
*flagp |= flags&SPSTART;
while (*preg->regparse == '|') {
preg->regparse++;
br = regbranch(preg, &flags);
if (br == 0)
return 0;
regtail(preg, ret, br); /* BRANCH -> BRANCH. */
if (!(flags&HASWIDTH))
*flagp &= ~HASWIDTH;
*flagp |= flags&SPSTART;
}
/* Make a closing node, and hook it on the end. */
ender = regnode(preg, (paren) ? CLOSE+parno : END);
regtail(preg, ret, ender);
/* Hook the tails of the branches to the closing node. */
for (br = ret; br != 0; br = regnext(preg, br))
regoptail(preg, br, ender);
/* Check for proper termination. */
if (paren && *preg->regparse++ != ')') {
preg->err = REG_ERR_UNMATCHED_PAREN;
return 0;
} else if (!paren && *preg->regparse != '\0') {
if (*preg->regparse == ')') {
preg->err = REG_ERR_UNMATCHED_PAREN;
return 0;
} else {
preg->err = REG_ERR_JUNK_ON_END;
return 0;
}
}
return(ret);
}
TCHAR *CRegExp::regpiece(int *flagp)
{
TCHAR *ret;
TCHAR op;
TCHAR *next;
int flags;
ret = regatom(&flags);
if (ret == NULL)
return(NULL);
op = *regparse;
if (!ISREPN(op)) {
*flagp = flags;
return(ret);
}
if (!(flags&HASWIDTH) && op != _T('?'))
{
TRACE0("*+ operand could be empty\n");
return NULL;
}
switch (op) {
case _T('*'): *flagp = WORST|SPSTART; break;
case _T('+'): *flagp = WORST|SPSTART|HASWIDTH; break;
case _T('?'): *flagp = WORST; break;
}
if (op == _T('*') && (flags&SIMPLE))
reginsert(STAR, ret);
else if (op == _T('*')) {
// 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 == _T('+') && (flags&SIMPLE))
reginsert(PLUS, ret);
else if (op == _T('+')) {
// 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 == _T('?')) {
// 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 (ISREPN(*regparse))
{
TRACE0("nested *?+\n");
return NULL;
}
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 int regpiece(regex_t *preg, int *flagp)
{
int ret;
char op;
int next;
int flags;
int min;
int max;
ret = regatom(preg, &flags);
if (ret == 0)
return 0;
op = *preg->regparse;
if (!ISMULT(op)) {
*flagp = flags;
return(ret);
}
if (!(flags&HASWIDTH) && op != '?') {
preg->err = REG_ERR_OPERAND_COULD_BE_EMPTY;
return 0;
}
/* Handle braces (counted repetition) by expansion */
if (op == '{') {
char *end;
min = strtoul(preg->regparse + 1, &end, 10);
if (end == preg->regparse + 1) {
preg->err = REG_ERR_BAD_COUNT;
return 0;
}
if (*end == '}') {
max = min;
}
else {
preg->regparse = end;
max = strtoul(preg->regparse + 1, &end, 10);
if (*end != '}') {
preg->err = REG_ERR_UNMATCHED_BRACES;
return 0;
}
}
if (end == preg->regparse + 1) {
max = MAX_REP_COUNT;
}
else if (max < min || max >= 100) {
preg->err = REG_ERR_BAD_COUNT;
return 0;
}
if (min >= 100) {
preg->err = REG_ERR_BAD_COUNT;
return 0;
}
preg->regparse = strchr(preg->regparse, '}');
}
else {
min = (op == '+');
max = (op == '?' ? 1 : MAX_REP_COUNT);
}
if (preg->regparse[1] == '?') {
preg->regparse++;
next = reginsert(preg, flags & SIMPLE ? REPMIN : REPXMIN, 5, ret);
}
else {
next = reginsert(preg, flags & SIMPLE ? REP: REPX, 5, ret);
}
preg->program[ret + 2] = max;
preg->program[ret + 3] = min;
preg->program[ret + 4] = 0;
*flagp = (min) ? (WORST|HASWIDTH) : (WORST|SPSTART);
if (!(flags & SIMPLE)) {
int back = regnode(preg, BACK);
regtail(preg, back, ret);
regtail(preg, next, back);
}
preg->regparse++;
if (ISMULT(*preg->regparse)) {
preg->err = REG_ERR_NESTED_COUNT;
return 0;
}
return ret;
}
TCHAR *CRegExp::reg(int paren, int *flagp)
{
char *ret;
char *br;
char *ender;
int parno;
int flags;
*flagp = HASWIDTH; // Tentatively.
if (paren)
{
// Make an OPEN node.
if (regnpar >= NSUBEXP)
{
TRACE1("Too many (). NSUBEXP is set to %d\n", NSUBEXP );
return NULL;
}
parno = regnpar;
regnpar++;
ret = regnode(OPEN+parno);
}
// Pick up the branches, linking them together.
br = regbranch(&flags);
if (br == NULL)
return(NULL);
if (paren)
regtail(ret, br); // OPEN -> first.
else
ret = br;
*flagp &= ~(~flags&HASWIDTH); // Clear bit if bit 0.
*flagp |= flags&SPSTART;
while (*regparse == _T('|')) {
regparse++;
br = regbranch(&flags);
if (br == NULL)
return(NULL);
regtail(ret, br); // BRANCH -> BRANCH.
*flagp &= ~(~flags&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++ != _T(')'))
{
TRACE0("unterminated ()\n");
return NULL;
}
else if (!paren && *regparse != _T('\0'))
{
if (*regparse == _T(')'))
{
TRACE0("unmatched ()\n");
return NULL;
}
else
{
TRACE0("internal error: junk on end\n");
return NULL;
}
// NOTREACHED
}
return(ret);
}
/*
- regoptail - regtail on operand of first argument; nop if operandless
*/
void ossimRegExp::regoptail (char* p, const char* val) {
// "Operandless" and "op != BRANCH" are synonymous in practice.
if (p == NULL || p == ®dummy || OP(p) != BRANCH)
return;
regtail(OPERAND(p), val);
}
/*
- 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, /* Parenthesized? */
int *flagp )
{
register char *ret;
register char *br;
register char *ender;
register int parno;
int flags;
*flagp = HASWIDTH; /* Tentatively. */
/* Make an OPEN node, if parenthesized. */
if (paren) {
if (regnpar >= 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);
}
/*
- 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)
{
register char *next;
register char *ret;
register char op;
unsigned char max;
unsigned char min;
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 != '+' && 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 == '{') {
for (min = 0, regparse++ ; *regparse && isdigit(*regparse) ; regparse++) {
min = min * 10 + (*regparse - '0');
}
for (max = 0, regparse++ ; *regparse && isdigit(*regparse) ; regparse++) {
max = max * 10 + (*regparse - '0');
}
reginsert(max, ret);
next = OPERAND(ret);
reginsert(min, ret);
next = OPERAND(next);
reginsert(MINMAX, ret);
regtail(ret, OPERAND(next)); /* MINMAX->next = x */
} 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);
}
