1 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
2 * confused with the original package (see point 3 below). Thanks, Henry!
5 /* Additional note: this code is very heavily munged from Henry's version
6 * in places. In some spots I've traded clarity for efficiency, so don't
7 * blame Henry for some of the lack of readability.
10 /* $Header: regcomp.c,v 3.0.1.4 90/08/09 05:05:33 lwall Locked $
13 * Revision 3.0.1.4 90/08/09 05:05:33 lwall
14 * patch19: sped up /x+y/ patterns greatly by not retrying on every x
15 * patch19: inhibited backoff on patterns anchored to the end like /\s+$/
16 * patch19: sped up {m,n} on simple items
17 * patch19: optimized /.*whatever/ to /^.*whatever/
18 * patch19: fixed character classes to allow backslashing hyphen
20 * Revision 3.0.1.3 90/03/12 16:59:22 lwall
21 * patch13: pattern matches can now use \0 to mean \000
23 * Revision 3.0.1.2 90/02/28 18:08:35 lwall
24 * patch9: /[\200-\377]/ didn't work on machines with signed chars
26 * Revision 3.0.1.1 89/11/11 04:51:04 lwall
27 * patch2: /[\000]/ didn't work
29 * Revision 3.0 89/10/18 15:22:29 lwall
35 * regcomp and regexec -- regsub and regerror are not used in perl
37 * Copyright (c) 1986 by University of Toronto.
38 * Written by Henry Spencer. Not derived from licensed software.
40 * Permission is granted to anyone to use this software for any
41 * purpose on any computer system, and to redistribute it freely,
42 * subject to the following restrictions:
44 * 1. The author is not responsible for the consequences of use of
45 * this software, no matter how awful, even if they arise
48 * 2. The origin of this software must not be misrepresented, either
49 * by explicit claim or by omission.
51 * 3. Altered versions must be plainly marked as such, and must not
52 * be misrepresented as being the original software.
55 **** Alterations to Henry's code are...
57 **** Copyright (c) 1989, Larry Wall
59 **** You may distribute under the terms of the GNU General Public License
60 **** as specified in the README file that comes with the perl 3.0 kit.
62 * Beware that some of this code is subtly aware of the way operator
63 * precedence is structured in regular expressions. Serious changes in
64 * regular-expression syntax might require a total rethink.
75 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
76 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
77 ((*s) == '{' && regcurly(s)))
78 #define META "^$.[()|?+*\\"
81 * Flags to be passed up and down.
83 #define HASWIDTH 01 /* Known never to match null string. */
84 #define SIMPLE 02 /* Simple enough to be STAR/PLUS operand. */
85 #define SPSTART 04 /* Starts with * or +. */
86 #define WORST 0 /* Worst case. */
89 * Global work variables for regcomp().
91 static char *regprecomp; /* uncompiled string. */
92 static char *regparse; /* Input-scan pointer. */
93 static char *regxend; /* End of input for compile */
94 static int regnpar; /* () count. */
95 static char *regcode; /* Code-emit pointer; ®dummy = don't. */
96 static long regsize; /* Code size. */
98 static int regsawbracket; /* Did we do {d,d} trick? */
101 * Forward declarations for regcomp()'s friends.
103 STATIC int regcurly();
105 STATIC char *regbranch();
106 STATIC char *regpiece();
107 STATIC char *regatom();
108 STATIC char *regclass();
109 STATIC char *regnode();
111 STATIC void reginsert();
112 STATIC void regtail();
113 STATIC void regoptail();
116 - regcomp - compile a regular expression into internal code
118 * We can't allocate space until we know how big the compiled form will be,
119 * but we can't compile it (and thus know how big it is) until we've got a
120 * place to put the code. So we cheat: we compile it twice, once with code
121 * generation turned off and size counting turned on, and once "for real".
122 * This also means that we don't allocate space until we are sure that the
123 * thing really will compile successfully, and we never have to move the
124 * code and thus invalidate pointers into it. (Note that it has to be in
125 * one piece because free() must be able to free it all.) [NB: not true in perl]
127 * Beware that the optimization-preparation code in here knows about some
128 * of the structure of the compiled regexp. [I'll say.]
131 regcomp(exp,xend,fold)
138 register STR *longest;
140 register char *first;
144 extern char *safemalloc();
145 extern char *savestr();
149 fatal("NULL regexp argument");
151 /* First pass: determine size, legality. */
155 regprecomp = nsavestr(exp,xend-exp);
161 if (reg(0, &flags) == NULL) {
162 Safefree(regprecomp);
166 /* Small enough for pointer-storage convention? */
167 if (regsize >= 32767L) /* Probably could be 65535L. */
168 FAIL("regexp too big");
170 /* Allocate space. */
171 Newc(1001, r, sizeof(regexp) + (unsigned)regsize, char, regexp);
173 FAIL("regexp out of space");
175 /* Second pass: emit code. */
177 bcopy(regprecomp,exp,xend-exp);
178 r->precomp = regprecomp;
182 regcode = r->program;
184 if (reg(0, &flags) == NULL)
187 /* Dig out information for optimizations. */
188 r->regstart = Nullstr; /* Worst-case defaults. */
190 r->regmust = Nullstr;
192 r->regstclass = Nullch;
193 scan = r->program+1; /* First BRANCH. */
194 if (OP(regnext(scan)) == END) {/* Only one top-level choice. */
195 scan = NEXTOPER(scan);
198 while ((OP(first) > OPEN && OP(first) < CLOSE) ||
199 (OP(first) == BRANCH && OP(regnext(first)) != BRANCH) ||
200 (OP(first) == PLUS) ||
201 (OP(first) == CURLY && ARG1(first) > 0) ) {
202 if (OP(first) == CURLY)
204 else if (OP(first) == PLUS)
206 first = NEXTOPER(first);
209 /* Starting-point info. */
210 if (OP(first) == EXACTLY) {
212 str_make(OPERAND(first)+1,*OPERAND(first));
213 if (r->regstart->str_cur > !(sawstudy|fold))
214 fbmcompile(r->regstart,fold);
216 else if ((exp = index(simple,OP(first))) && exp > simple)
217 r->regstclass = first;
218 else if (OP(first) == BOUND || OP(first) == NBOUND)
219 r->regstclass = first;
220 else if (OP(first) == BOL ||
221 (OP(first) == STAR && OP(NEXTOPER(first)) == ANY) )
222 r->reganch = 1; /* kinda turn .* into ^.* */
223 r->reganch |= sawplus;
227 fprintf(stderr,"first %d next %d offset %d\n",
228 OP(first), OP(NEXTOPER(first)), first - scan);
231 * If there's something expensive in the r.e., find the
232 * longest literal string that must appear and make it the
233 * regmust. Resolve ties in favor of later strings, since
234 * the regstart check works with the beginning of the r.e.
235 * and avoiding duplication strengthens checking. Not a
236 * strong reason, but sufficient in the absence of others.
237 * [Now we resolve ties in favor of the earlier string if
238 * it happens that curback has been invalidated, since the
239 * earlier string may buy us something the later one won't.]
241 longest = str_make("",0);
245 while (scan != NULL) {
246 if (OP(scan) == BRANCH) {
247 if (OP(regnext(scan)) == BRANCH) {
249 while (OP(scan) == BRANCH)
250 scan = regnext(scan);
252 else /* single branch is ok */
253 scan = NEXTOPER(scan);
255 if (OP(scan) == EXACTLY) {
257 while (OP(regnext(scan)) >= CLOSE)
258 scan = regnext(scan);
259 if (curback - back == len) {
260 str_ncat(longest, OPERAND(first)+1,
262 len += *OPERAND(first);
263 curback += *OPERAND(first);
264 first = regnext(scan);
266 else if (*OPERAND(first) >= len + (curback >= 0)) {
267 len = *OPERAND(first);
268 str_nset(longest, OPERAND(first)+1,len);
271 first = regnext(scan);
274 curback += *OPERAND(first);
276 else if (index(varies,OP(scan)))
278 else if (index(simple,OP(scan)))
280 scan = regnext(scan);
283 r->regmust = longest;
287 if (len > !(sawstudy||fold||OP(first)==EOL))
288 fbmcompile(r->regmust,fold);
289 r->regmust->str_u.str_useful = 100;
290 if (OP(first) == EOL) /* is match anchored to EOL? */
291 r->regmust->str_pok |= SP_TAIL;
297 r->do_folding = fold;
298 r->nparens = regnpar - 1;
307 - reg - regular expression, i.e. main body or parenthesized thing
309 * Caller must absorb opening parenthesis.
311 * Combining parenthesis handling with the base level of regular expression
312 * is a trifle forced, but the need to tie the tails of the branches to what
313 * follows makes it hard to avoid.
317 int paren; /* Parenthesized? */
322 register char *ender;
326 *flagp = HASWIDTH; /* Tentatively. */
328 /* Make an OPEN node, if parenthesized. */
330 if (regnpar >= NSUBEXP)
331 FAIL("too many () in regexp");
334 ret = regnode(OPEN+parno);
338 /* Pick up the branches, linking them together. */
339 br = regbranch(&flags);
343 regtail(ret, br); /* OPEN -> first. */
346 if (!(flags&HASWIDTH))
348 *flagp |= flags&SPSTART;
349 while (*regparse == '|') {
351 br = regbranch(&flags);
354 regtail(ret, br); /* BRANCH -> BRANCH. */
355 if (!(flags&HASWIDTH))
357 *flagp |= flags&SPSTART;
360 /* Make a closing node, and hook it on the end. */
361 ender = regnode((paren) ? CLOSE+parno : END);
364 /* Hook the tails of the branches to the closing node. */
365 for (br = ret; br != NULL; br = regnext(br))
366 regoptail(br, ender);
368 /* Check for proper termination. */
369 if (paren && *regparse++ != ')') {
370 FAIL("unmatched () in regexp");
371 } else if (!paren && regparse < regxend) {
372 if (*regparse == ')') {
373 FAIL("unmatched () in regexp");
375 FAIL("junk on end of regexp"); /* "Can't happen". */
383 - regbranch - one alternative of an | operator
385 * Implements the concatenation operator.
392 register char *chain;
393 register char *latest;
396 *flagp = WORST; /* Tentatively. */
398 ret = regnode(BRANCH);
400 while (regparse < regxend && *regparse != '|' && *regparse != ')') {
401 latest = regpiece(&flags);
404 *flagp |= flags&HASWIDTH;
405 if (chain == NULL) /* First piece. */
406 *flagp |= flags&SPSTART;
408 regtail(chain, latest);
411 if (chain == NULL) /* Loop ran zero times. */
412 (void) regnode(NOTHING);
418 - regpiece - something followed by possible [*+?]
420 * Note that the branching code sequences used for ? and the general cases
421 * of * and + are somewhat optimized: they use the same NOTHING node as
422 * both the endmarker for their branch list and the body of the last branch.
423 * It might seem that this node could be dispensed with entirely, but the
424 * endmarker role is not redundant.
434 char *origparse = regparse;
435 int orignpar = regnpar;
440 ret = regatom(&flags);
446 /* Here's a total kludge: if after the atom there's a {\d+,?\d*}
447 * then we decrement the first number by one and reset our
448 * parsing back to the beginning of the same atom. If the first number
449 * is down to 0, decrement the second number instead and fake up
450 * a ? after it. Given the way this compiler doesn't keep track
451 * of offsets on the first pass, this is the only way to replicate
452 * a piece of code. Sigh.
454 if (op == '{' && regcurly(regparse)) {
457 while (isdigit(*next) || *next == ',') {
466 if (*next == '}') { /* got one */
470 iter = atoi(regparse);
471 if (flags&SIMPLE) { /* we can do it right after all */
474 reginsert(CURLY, ret);
478 if (tmp && tmp < iter)
479 fatal("Can't do {n,m} with n > m");
480 if (regcode != ®dummy) {
482 *(unsigned short *)(ret+3) = iter;
483 *(unsigned short *)(ret+5) = tmp;
485 ret[3] = iter >> 8; ret[4] = iter & 0377;
486 ret[5] = tmp >> 8; ret[6] = tmp & 0377;
492 regsawbracket++; /* remember we clobbered exp */
495 sprintf(regparse,"%.*d", max-regparse, iter - 1);
497 if (*max == ',' && max[1] != '}') {
498 if (atoi(max+1) <= 0)
499 fatal("Can't do {n,m} with n > m");
501 sprintf(max+1,"%.*d", next-(max+1), atoi(max+1) - 1);
504 if (iter != 1 || *max == ',') {
505 regparse = origparse; /* back up input pointer */
506 regnpar = orignpar; /* don't make more parens */
518 if (max == next) { /* any number more? */
520 op = '*'; /* fake up one with a star */
523 op = '?'; /* fake up optional atom */
525 sprintf(max,"%.*d", next-max, iter - 1);
530 regparse = origparse - 1; /* offset ++ below */
535 fatal("Can't do {n,0}");
538 fatal("Can't do {0}");
547 if (!(flags&HASWIDTH) && op != '?')
548 FAIL("regexp *+ operand could be empty");
549 *flagp = (op != '+') ? (WORST|SPSTART) : (WORST|HASWIDTH);
551 if (op == '*' && (flags&SIMPLE))
552 reginsert(STAR, ret);
553 else if (op == '*') {
554 /* Emit x* as (x&|), where & means "self". */
555 reginsert(BRANCH, ret); /* Either x */
556 regoptail(ret, regnode(BACK)); /* and loop */
557 regoptail(ret, ret); /* back */
558 regtail(ret, regnode(BRANCH)); /* or */
559 regtail(ret, regnode(NOTHING)); /* null. */
560 } else if (op == '+' && (flags&SIMPLE))
561 reginsert(PLUS, ret);
562 else if (op == '+') {
563 /* Emit x+ as x(&|), where & means "self". */
564 next = regnode(BRANCH); /* Either */
566 regtail(regnode(BACK), ret); /* loop back */
567 regtail(next, regnode(BRANCH)); /* or */
568 regtail(ret, regnode(NOTHING)); /* null. */
569 } else if (op == '?') {
570 /* Emit x? as (x|) */
571 reginsert(BRANCH, ret); /* Either x */
572 regtail(ret, regnode(BRANCH)); /* or */
573 next = regnode(NOTHING); /* null. */
575 regoptail(ret, next);
579 if (ISMULT2(regparse))
580 FAIL("nested *?+ in regexp");
586 - regatom - the lowest level
588 * Optimization: gobbles an entire sequence of ordinary characters so that
589 * it can turn them into a single node, which is smaller to store and
590 * faster to run. Backslashed characters are exceptions, each becoming a
591 * separate node; the code is simpler that way and it's not worth fixing.
593 * [Yes, it is worth fixing, some scripts can run twice the speed.]
602 *flagp = WORST; /* Tentatively. */
604 switch (*regparse++) {
613 *flagp |= HASWIDTH|SIMPLE;
617 *flagp |= HASWIDTH|SIMPLE;
620 ret = reg(1, &flags);
623 *flagp |= flags&(HASWIDTH|SPSTART);
627 FAIL("internal urp in regexp"); /* Supposed to be caught earlier. */
632 FAIL("?+* follows nothing in regexp");
637 ret = regnode(ALNUM);
638 *flagp |= HASWIDTH|SIMPLE;
642 ret = regnode(NALNUM);
643 *flagp |= HASWIDTH|SIMPLE;
647 ret = regnode(BOUND);
652 ret = regnode(NBOUND);
657 ret = regnode(SPACE);
658 *flagp |= HASWIDTH|SIMPLE;
662 ret = regnode(NSPACE);
663 *flagp |= HASWIDTH|SIMPLE;
667 ret = regnode(DIGIT);
668 *flagp |= HASWIDTH|SIMPLE;
672 ret = regnode(NDIGIT);
673 *flagp |= HASWIDTH|SIMPLE;
681 case '0': case '1': case '2': case '3': case '4':
682 case '5': case '6': case '7': case '8': case '9':
683 if (isdigit(regparse[1]) || *regparse == '0')
686 ret = regnode(REF + *regparse++ - '0');
691 if (regparse >= regxend)
692 FAIL("trailing \\ in regexp");
706 ret = regnode(EXACTLY);
707 regc(0); /* save spot for len */
708 for (len=0, p=regparse-1;
709 len < 127 && p < regxend;
750 case '0': case '1': case '2': case '3':case '4':
751 case '5': case '6': case '7': case '8':case '9':
752 if (isdigit(p[1]) || *p == '0') {
755 foo = (foo<<3) + *++p - '0';
757 foo = (foo<<3) + *++p - '0';
768 FAIL("trailing \\ in regexp");
779 if (regfold && isupper(ender))
780 ender = tolower(ender);
781 if (ISMULT2(p)) { /* Back off on ?+*. */
795 FAIL("internal disaster in regexp");
799 if (regcode != ®dummy)
815 if (regcode == ®dummy)
819 bits[c >> 3] &= ~(1 << (c & 7));
821 bits[c >> 3] |= (1 << (c & 7));
829 register int lastclass;
830 register int range = 0;
834 ret = regnode(ANYOF);
835 if (*regparse == '^') { /* Complement of range. */
842 for (class = 0; class < 32; class++)
844 if (*regparse == ']' || *regparse == '-')
845 goto skipcond; /* allow 1st char to be ] or - */
846 while (regparse < regxend && *regparse != ']') {
848 class = UCHARAT(regparse++);
850 class = UCHARAT(regparse++);
853 for (class = 'a'; class <= 'z'; class++)
854 regset(bits,def,class);
855 for (class = 'A'; class <= 'Z'; class++)
856 regset(bits,def,class);
857 for (class = '0'; class <= '9'; class++)
858 regset(bits,def,class);
859 regset(bits,def,'_');
863 regset(bits,def,' ');
864 regset(bits,def,'\t');
865 regset(bits,def,'\r');
866 regset(bits,def,'\f');
867 regset(bits,def,'\n');
871 for (class = '0'; class <= '9'; class++)
872 regset(bits,def,class);
890 case '0': case '1': case '2': case '3': case '4':
891 case '5': case '6': case '7': case '8': case '9':
893 if (isdigit(*regparse)) {
895 class += *regparse++ - '0';
897 if (isdigit(*regparse)) {
899 class += *regparse++ - '0';
905 if (lastclass > class)
906 FAIL("invalid [] range in regexp");
911 if (*regparse == '-' && regparse+1 < regxend &&
912 regparse[1] != ']') {
915 continue; /* do it next time */
918 for ( ; lastclass <= class; lastclass++) {
919 regset(bits,def,lastclass);
920 if (regfold && isupper(lastclass))
921 regset(bits,def,tolower(lastclass));
925 if (*regparse != ']')
926 FAIL("unmatched [] in regexp");
932 - regnode - emit a node
934 static char * /* Location. */
942 if (ret == ®dummy) {
953 if (!((long)ret & 1))
959 *ptr++ = '\0'; /* Null "next" pointer. */
967 - regc - emit (if appropriate) a byte of code
973 if (regcode != ®dummy)
980 - reginsert - insert an operator in front of already-emitted operand
982 * Means relocating the operand.
991 register char *place;
992 register offset = (op == CURLY ? 4 : 0);
994 if (regcode == ®dummy) {
996 regsize += 4 + offset;
998 regsize += 3 + offset;
1005 regcode += 4 + offset;
1007 regcode += 3 + offset;
1013 place = opnd; /* Op node, where operand used to be. */
1017 while (offset-- > 0)
1022 - regtail - set the next-pointer at the end of a node chain
1029 register char *scan;
1030 register char *temp;
1031 register int offset;
1036 /* Find last node. */
1039 temp = regnext(scan);
1046 offset = val - scan;
1048 *(short*)(scan+1) = offset;
1053 if (OP(scan) == BACK)
1054 offset = scan - val;
1056 offset = val - scan;
1057 *(scan+1) = (offset>>8)&0377;
1058 *(scan+2) = offset&0377;
1063 - regoptail - regtail on operand of first argument; nop if operandless
1070 /* "Operandless" and "op != BRANCH" are synonymous in practice. */
1071 if (p == NULL || p == ®dummy || OP(p) != BRANCH)
1073 regtail(NEXTOPER(p), val);
1077 - regcurly - a little FSA that accepts {\d+,?\d*}
1101 - regdump - dump a regexp onto stderr in vaguely comprehensible form
1108 register char op = EXACTLY; /* Arbitrary non-END op. */
1109 register char *next;
1110 extern char *index();
1114 while (op != END) { /* While that wasn't END last time... */
1120 fprintf(stderr,"%2d%s", s-r->program, regprop(s)); /* Where, what. */
1122 if (next == NULL) /* Next ptr. */
1123 fprintf(stderr,"(0)");
1125 fprintf(stderr,"(%d)", (s-r->program)+(next-s));
1130 if (op == EXACTLY) {
1131 /* Literal string, where present. */
1133 while (*s != '\0') {
1139 (void)putchar('\n');
1142 /* Header fields of interest. */
1144 fprintf(stderr,"start `%s' ", r->regstart->str_ptr);
1146 fprintf(stderr,"stclass `%s' ", regprop(r->regstclass));
1148 fprintf(stderr,"anchored ");
1150 fprintf(stderr,"plus ");
1151 if (r->regmust != NULL)
1152 fprintf(stderr,"must have \"%s\" back %d ", r->regmust->str_ptr,
1154 fprintf(stderr,"\n");
1158 - regprop - printable representation of opcode
1166 (void) strcpy(buf, ":");
1221 (void)sprintf(buf+strlen(buf), "CURLY {%d,%d}",
1235 (void)sprintf(buf+strlen(buf), "REF%d", OP(op)-REF);
1247 (void)sprintf(buf+strlen(buf), "OPEN%d", OP(op)-OPEN);
1259 (void)sprintf(buf+strlen(buf), "CLOSE%d", OP(op)-CLOSE);
1269 FAIL("corrupted regexp opcode");
1272 (void) strcat(buf, p);
1275 #endif /* DEBUGGING */
1281 Safefree(r->precomp);
1283 Safefree(r->subbase);
1285 str_free(r->regmust);
1287 str_free(r->regstart);