5 * "One Ring to rule them all, One Ring to find them..."
8 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
9 * confused with the original package (see point 3 below). Thanks, Henry!
12 /* Additional note: this code is very heavily munged from Henry's version
13 * in places. In some spots I've traded clarity for efficiency, so don't
14 * blame Henry for some of the lack of readability.
17 /* The names of the functions have been changed from regcomp and
18 * regexec to pregcomp and pregexec in order to avoid conflicts
19 * with the POSIX routines of the same names.
24 * pregcomp and pregexec -- regsub and regerror are not used in perl
26 * Copyright (c) 1986 by University of Toronto.
27 * Written by Henry Spencer. Not derived from licensed software.
29 * Permission is granted to anyone to use this software for any
30 * purpose on any computer system, and to redistribute it freely,
31 * subject to the following restrictions:
33 * 1. The author is not responsible for the consequences of use of
34 * this software, no matter how awful, even if they arise
37 * 2. The origin of this software must not be misrepresented, either
38 * by explicit claim or by omission.
40 * 3. Altered versions must be plainly marked as such, and must not
41 * be misrepresented as being the original software.
43 **** Alterations to Henry's code are...
45 **** Copyright (c) 1991-1997, Larry Wall
47 **** You may distribute under the terms of either the GNU General Public
48 **** License or the Artistic License, as specified in the README file.
50 * Beware that some of this code is subtly aware of the way operator
51 * precedence is structured in regular expressions. Serious changes in
52 * regular-expression syntax might require a total rethink.
63 static I32 regnarrate = 0;
64 static char* regprogram = 0;
67 /* Current curly descriptor */
68 typedef struct curcur CURCUR;
70 int parenfloor; /* how far back to strip paren data */
71 int cur; /* how many instances of scan we've matched */
72 int min; /* the minimal number of scans to match */
73 int max; /* the maximal number of scans to match */
74 int minmod; /* whether to work our way up or down */
75 char * scan; /* the thing to match */
76 char * next; /* what has to match after it */
77 char * lastloc; /* where we started matching this scan */
78 CURCUR * oldcc; /* current curly before we started this one */
83 typedef I32 CHECKPOINT;
85 static CHECKPOINT regcppush _((I32 parenfloor));
86 static char * regcppop _((void));
93 int retval = savestack_ix;
94 int i = (regsize - parenfloor) * 3;
98 for (p = regsize; p > parenfloor; p--) {
99 SSPUSHPTR(regendp[p]);
100 SSPUSHPTR(regstartp[p]);
104 SSPUSHINT(*reglastparen);
107 SSPUSHINT(SAVEt_REGCONTEXT);
119 assert(i == SAVEt_REGCONTEXT);
121 input = (char *) SSPOPPTR;
122 *reglastparen = SSPOPINT;
124 for (i -= 3; i > 0; i -= 3) {
125 paren = (U32)SSPOPINT;
126 regstartp[paren] = (char *) SSPOPPTR;
127 tmps = (char*)SSPOPPTR;
128 if (paren <= *reglastparen)
129 regendp[paren] = tmps;
131 for (paren = *reglastparen + 1; paren <= regnpar; paren++) {
133 regstartp[paren] = Nullch;
134 regendp[paren] = Nullch;
139 /* After a successful match in WHILEM, we want to restore paren matches
140 * that have been overwritten by a failed match attempt in the process
141 * of reaching this success. We do this by restoring regstartp[i]
142 * wherever regendp[i] has not changed; if OPEN is changed to modify
143 * regendp[], the '== endp' test below should be changed to match.
144 * This corrects the error of:
145 * 0 > length [ "foobar" =~ / ( (foo) | (bar) )* /x ]->[1]
156 assert(i == SAVEt_REGCONTEXT);
158 /* input, lastparen, size */
159 SSPOPPTR; SSPOPINT; SSPOPINT;
160 for (i -= 3; i > 0; i -= 3) {
161 paren = (U32)SSPOPINT;
162 startp = (char *) SSPOPPTR;
163 endp = (char *) SSPOPPTR;
164 if (paren <= *reglastparen && regendp[paren] == endp)
165 regstartp[paren] = startp;
167 assert(savestack_ix == base);
170 #define regcpblow(cp) leave_scope(cp)
173 * pregexec and friends
180 static I32 regmatch _((char *prog));
181 static I32 regrepeat _((char *p, I32 max));
182 static I32 regtry _((regexp *prog, char *startpos));
183 static bool reginclass _((char *p, I32 c));
185 static bool regtainted; /* tainted information used? */
188 - pregexec - match a regexp against a string
191 pregexec(prog, stringarg, strend, strbeg, minend, screamer, safebase)
192 register regexp *prog;
194 register char *strend; /* pointer to null at end of string */
195 char *strbeg; /* real beginning of string */
196 I32 minend; /* end of match must be at least minend after stringarg */
198 I32 safebase; /* no need to remember string in subbase */
202 register char *startpos = stringarg;
204 I32 minlen = 0; /* must match at least this many chars */
205 I32 dontbother = 0; /* how many characters not to try at end */
213 regnarrate = debug & 512;
214 regprogram = prog->program;
218 if (prog == NULL || startpos == NULL) {
219 croak("NULL regexp parameter");
223 if (startpos == strbeg) /* is ^ valid at stringarg? */
226 regprev = stringarg[-1];
227 if (!multiline && regprev == '\n')
228 regprev = '\0'; /* force ^ to NOT match */
231 regprecomp = prog->precomp;
232 /* Check validity of program. */
233 if (UCHARAT(prog->program) != MAGIC) {
234 FAIL("corrupted regexp program");
237 regnpar = prog->nparens;
240 /* If there is a "must appear" string, look for it. */
242 if (prog->regmust != Nullsv &&
243 !(prog->reganch & ROPT_ANCH_GPOS) &&
244 (!(prog->reganch & ROPT_ANCH_BOL)
245 || (multiline && prog->regback >= 0)) )
247 if (stringarg == strbeg && screamer) {
248 if (screamfirst[BmRARE(prog->regmust)] >= 0)
249 s = screaminstr(screamer,prog->regmust);
254 s = fbm_instr((unsigned char*)s, (unsigned char*)strend,
257 ++BmUSEFUL(prog->regmust); /* hooray */
258 goto phooey; /* not present */
260 else if (prog->regback >= 0) {
264 minlen = prog->regback + SvCUR(prog->regmust);
266 else if (!prog->naughty && --BmUSEFUL(prog->regmust) < 0) { /* boo */
267 SvREFCNT_dec(prog->regmust);
268 prog->regmust = Nullsv; /* disable regmust */
273 minlen = SvCUR(prog->regmust);
277 /* Mark beginning of line for ^ . */
280 /* Mark end of line for $ (and such) */
283 /* see how far we have to get to not match where we matched before */
284 regtill = startpos+minend;
286 /* Simplest case: anchored match need be tried only once. */
287 /* [unless only anchor is BOL and multiline is set] */
288 if (prog->reganch & ROPT_ANCH) {
289 if (regtry(prog, startpos))
291 else if (!(prog->reganch & ROPT_ANCH_GPOS) &&
292 (multiline || (prog->reganch & ROPT_IMPLICIT)))
295 dontbother = minlen - 1;
296 strend -= dontbother;
297 /* for multiline we only have to try after newlines */
302 if (s < strend && regtry(prog, s))
310 /* Messy cases: unanchored match. */
311 if (prog->regstart) {
312 if (prog->reganch & ROPT_SKIP) { /* we have /x+whatever/ */
313 /* it must be a one character string */
314 char ch = SvPVX(prog->regstart)[0];
320 while (s < strend && *s == ch)
326 else if (SvTYPE(prog->regstart) == SVt_PVBM) {
327 /* We know what string it must start with. */
328 while ((s = fbm_instr((unsigned char*)s,
329 (unsigned char*)strend, prog->regstart)) != NULL)
336 else { /* Optimized fbm_instr: */
337 c = SvPVX(prog->regstart);
338 while ((s = ninstr(s, strend, c, c + SvCUR(prog->regstart))) != NULL)
348 if (c = prog->regstclass) {
349 I32 doevery = (prog->reganch & ROPT_SKIP) == 0;
352 dontbother = minlen - 1;
353 strend -= dontbother; /* don't bother with what can't match */
355 /* We know what class it must start with. */
360 if (reginclass(c, *s)) {
361 if (tmp && regtry(prog, s))
376 dontbother++,strend--;
377 tmp = (s != startpos) ? UCHARAT(s - 1) : regprev;
378 tmp = ((OP(c) == BOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
380 if (tmp == !(OP(c) == BOUND ? isALNUM(*s) : isALNUM_LC(*s))) {
387 if ((minlen || tmp) && regtry(prog,s))
395 dontbother++,strend--;
396 tmp = (s != startpos) ? UCHARAT(s - 1) : regprev;
397 tmp = ((OP(c) == NBOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0);
399 if (tmp == !(OP(c) == NBOUND ? isALNUM(*s) : isALNUM_LC(*s)))
401 else if (regtry(prog, s))
405 if ((minlen || !tmp) && regtry(prog,s))
411 if (tmp && regtry(prog, s))
424 if (isALNUM_LC(*s)) {
425 if (tmp && regtry(prog, s))
438 if (tmp && regtry(prog, s))
451 if (!isALNUM_LC(*s)) {
452 if (tmp && regtry(prog, s))
465 if (tmp && regtry(prog, s))
478 if (isSPACE_LC(*s)) {
479 if (tmp && regtry(prog, s))
492 if (tmp && regtry(prog, s))
505 if (!isSPACE_LC(*s)) {
506 if (tmp && regtry(prog, s))
519 if (tmp && regtry(prog, s))
532 if (tmp && regtry(prog, s))
546 dontbother = minlen - 1;
547 strend -= dontbother;
548 /* We don't know much -- general case. */
552 } while (s++ < strend);
559 strend += dontbother; /* uncheat */
560 prog->subbeg = strbeg;
561 prog->subend = strend;
562 prog->exec_tainted = regtainted;
564 /* make sure $`, $&, $', and $digit will work later */
565 if (strbeg != prog->subbase) {
568 Safefree(prog->subbase);
569 prog->subbase = Nullch;
573 I32 i = strend - startpos + (stringarg - strbeg);
574 s = savepvn(strbeg, i);
575 Safefree(prog->subbase);
577 prog->subbeg = prog->subbase;
578 prog->subend = prog->subbase + i;
579 s = prog->subbase + (stringarg - strbeg);
580 for (i = 0; i <= prog->nparens; i++) {
582 prog->startp[i] = s + (prog->startp[i] - startpos);
583 prog->endp[i] = s + (prog->endp[i] - startpos);
595 - regtry - try match at specific point
597 static I32 /* 0 failure, 1 success */
598 regtry(prog, startpos)
607 regstartp = prog->startp;
608 regendp = prog->endp;
609 reglastparen = &prog->lastparen;
616 for (i = prog->nparens; i >= 0; i--) {
621 if (regmatch(prog->program + 1) && reginput >= regtill) {
622 prog->startp[0] = startpos;
623 prog->endp[0] = reginput;
631 - regmatch - main matching routine
633 * Conceptually the strategy is simple: check to see whether the current
634 * node matches, call self recursively to see whether the rest matches,
635 * and then act accordingly. In practice we make some effort to avoid
636 * recursion, in particular by going through "ordinary" nodes (that don't
637 * need to know whether the rest of the match failed) by a loop instead of
640 /* [lwall] I've hoisted the register declarations to the outer block in order to
641 * maybe save a little bit of pushing and popping on the stack. It also takes
642 * advantage of machines that use a register save mask on subroutine entry.
644 static I32 /* 0 failure, 1 success */
648 register char *scan; /* Current node. */
649 char *next; /* Next node. */
650 register I32 nextchar;
651 register I32 n; /* no or next */
652 register I32 ln; /* len or last */
653 register char *s; /* operand or save */
654 register char *locinput = reginput;
655 register I32 c1, c2; /* case fold search */
658 static int regindent = 0;
662 nextchar = UCHARAT(locinput);
664 while (scan != NULL) {
666 #define sayYES goto yes
667 #define sayNO goto no
668 #define saySAME(x) if (x) goto yes; else goto no
670 SV *prop = sv_newmortal();
672 PerlIO_printf(Perl_debug_log, "%*s%2ld%-8.8s\t<%.10s>\n",
673 regindent*2, "", (long)(scan - regprogram),
674 SvPVX(prop), locinput);
677 #define sayYES return 1
678 #define sayNO return 0
679 #define saySAME(x) return x
683 next = scan + NEXT(scan);
687 next = regnext(scan);
692 if (locinput == regbol
694 : ((nextchar || locinput < regeol) && locinput[-1] == '\n') )
696 /* regtill = regbol; */
701 if (locinput == regbol
703 : ((nextchar || locinput < regeol) && locinput[-1] == '\n') )
709 if (locinput == regbol && regprev == '\n')
713 if (locinput == regbol)
723 if ((nextchar || locinput < regeol) && nextchar != '\n')
728 if ((nextchar || locinput < regeol) && nextchar != '\n')
730 if (regeol - locinput > 1)
734 if (!nextchar && locinput >= regeol)
736 nextchar = UCHARAT(++locinput);
739 if (!nextchar && locinput >= regeol || nextchar == '\n')
741 nextchar = UCHARAT(++locinput);
746 /* Inline the first character, for speed. */
747 if (UCHARAT(s) != nextchar)
749 if (regeol - locinput < ln)
751 if (ln > 1 && memNE(s, locinput, ln))
754 nextchar = UCHARAT(locinput);
762 /* Inline the first character, for speed. */
763 if (UCHARAT(s) != nextchar &&
764 UCHARAT(s) != ((OP(scan) == EXACTF)
765 ? fold : fold_locale)[nextchar])
767 if (regeol - locinput < ln)
769 if (ln > 1 && (OP(scan) == EXACTF
770 ? ibcmp(s, locinput, ln)
771 : ibcmp_locale(s, locinput, ln)))
774 nextchar = UCHARAT(locinput);
779 nextchar = UCHARAT(locinput);
780 if (!reginclass(s, nextchar))
782 if (!nextchar && locinput >= regeol)
784 nextchar = UCHARAT(++locinput);
792 if (!(OP(scan) == ALNUM
793 ? isALNUM(nextchar) : isALNUM_LC(nextchar)))
795 nextchar = UCHARAT(++locinput);
801 if (!nextchar && locinput >= regeol)
803 if (OP(scan) == NALNUM
804 ? isALNUM(nextchar) : isALNUM_LC(nextchar))
806 nextchar = UCHARAT(++locinput);
814 /* was last char in word? */
815 ln = (locinput != regbol) ? UCHARAT(locinput - 1) : regprev;
816 if (OP(scan) == BOUND || OP(scan) == NBOUND) {
818 n = isALNUM(nextchar);
822 n = isALNUM_LC(nextchar);
824 if (((!ln) == (!n)) == (OP(scan) == BOUND || OP(scan) == BOUNDL))
831 if (!nextchar && locinput >= regeol)
833 if (!(OP(scan) == SPACE
834 ? isSPACE(nextchar) : isSPACE_LC(nextchar)))
836 nextchar = UCHARAT(++locinput);
844 if (OP(scan) == SPACE
845 ? isSPACE(nextchar) : isSPACE_LC(nextchar))
847 nextchar = UCHARAT(++locinput);
850 if (!isDIGIT(nextchar))
852 nextchar = UCHARAT(++locinput);
855 if (!nextchar && locinput >= regeol)
857 if (isDIGIT(nextchar))
859 nextchar = UCHARAT(++locinput);
866 n = ARG1(scan); /* which paren pair */
874 /* Inline the first character, for speed. */
875 if (UCHARAT(s) != nextchar &&
877 (UCHARAT(s) != ((OP(scan) == REFF
878 ? fold : fold_locale)[nextchar]))))
881 if (locinput + ln > regeol)
883 if (ln > 1 && (OP(scan) == REF
884 ? memNE(s, locinput, ln)
886 ? ibcmp(s, locinput, ln)
887 : ibcmp_locale(s, locinput, ln))))
890 nextchar = UCHARAT(locinput);
898 n = ARG1(scan); /* which paren pair */
899 regstartp[n] = locinput;
904 n = ARG1(scan); /* which paren pair */
905 regendp[n] = locinput;
906 if (n > *reglastparen)
912 CHECKPOINT cp = savestack_ix;
915 cc.parenfloor = *reglastparen;
919 cc.scan = NEXTOPER(scan) + 4;
924 n = regmatch(PREVOPER(next)); /* start on the WHILEM */
932 * This is really hard to understand, because after we match
933 * what we're trying to match, we must make sure the rest of
934 * the RE is going to match for sure, and to do that we have
935 * to go back UP the parse tree by recursing ever deeper. And
936 * if it fails, we have to reset our parent's current state
937 * that we can try again after backing off.
942 n = cc->cur + 1; /* how many we know we matched */
947 PerlIO_printf(Perl_debug_log, "%*s %ld %lx\n", regindent*2, "",
951 /* If degenerate scan matches "", assume scan done. */
953 if (locinput == cc->lastloc && n >= cc->min) {
956 if (regmatch(cc->next))
963 /* First just match a string of min scans. */
967 cc->lastloc = locinput;
968 if (regmatch(cc->scan))
974 /* Prefer next over scan for minimal matching. */
979 cp = regcppush(cc->parenfloor);
980 if (regmatch(cc->next)) {
982 sayYES; /* All done. */
988 if (n >= cc->max) /* Maximum greed exceeded? */
991 /* Try scanning more and see if it helps. */
994 cc->lastloc = locinput;
995 cp = regcppush(cc->parenfloor);
996 if (regmatch(cc->scan)) {
1005 /* Prefer scan over next for maximal matching. */
1007 if (n < cc->max) { /* More greed allowed? */
1008 cp = regcppush(cc->parenfloor);
1010 cc->lastloc = locinput;
1011 if (regmatch(cc->scan)) {
1015 regcppop(); /* Restore some previous $<digit>s? */
1016 reginput = locinput;
1019 /* Failed deeper matches of scan, so see if this one works. */
1022 if (regmatch(cc->next))
1031 if (OP(next) != BRANCH) /* No choice. */
1032 next = NEXTOPER(scan);/* Avoid recursion. */
1034 int lastparen = *reglastparen;
1036 reginput = locinput;
1037 if (regmatch(NEXTOPER(scan)))
1039 for (n = *reglastparen; n > lastparen; n--)
1050 scan = regnext(scan);
1052 } while (scan != NULL && OP(scan) == BRANCH);
1062 ln = ARG1(scan); /* min to match */
1063 n = ARG2(scan); /* max to match */
1064 scan = NEXTOPER(scan) + 4;
1069 scan = NEXTOPER(scan);
1073 * Lookahead to avoid useless match attempts
1074 * when we know what character comes next.
1078 scan = NEXTOPER(scan);
1080 if (regkind[(U8)OP(next)] == EXACT) {
1081 c1 = UCHARAT(OPERAND(next) + 1);
1082 if (OP(next) == EXACTF)
1084 else if (OP(next) == EXACTFL)
1085 c2 = fold_locale[c1];
1091 reginput = locinput;
1094 if (ln && regrepeat(scan, ln) < ln)
1096 while (n >= ln || (n == 32767 && ln > 0)) { /* ln overflow ? */
1097 /* If it could work, try it. */
1099 UCHARAT(reginput) == c1 ||
1100 UCHARAT(reginput) == c2)
1105 /* Couldn't or didn't -- back up. */
1106 reginput = locinput + ln;
1107 if (regrepeat(scan, 1)) {
1109 reginput = locinput + ln;
1116 n = regrepeat(scan, n);
1117 if (ln < n && regkind[(U8)OP(next)] == EOL &&
1118 (!multiline || OP(next) == SEOL))
1119 ln = n; /* why back off? */
1121 /* If it could work, try it. */
1123 UCHARAT(reginput) == c1 ||
1124 UCHARAT(reginput) == c2)
1129 /* Couldn't or didn't -- back up. */
1131 reginput = locinput + n;
1137 reginput = locinput; /* put where regtry can find it */
1138 sayYES; /* Success! */
1140 reginput = locinput;
1141 scan = NEXTOPER(scan);
1142 if (!regmatch(scan))
1146 reginput = locinput;
1147 scan = NEXTOPER(scan);
1152 PerlIO_printf(PerlIO_stderr(), "%lx %d\n",
1153 (unsigned long)scan, scan[1]);
1154 FAIL("regexp memory corruption");
1160 * We get here only if there's trouble -- normally "case END" is
1161 * the terminating point.
1163 FAIL("corrupted regexp pointers");
1181 - regrepeat - repeatedly match something simple, report how many
1184 * [This routine now assumes that it will only match on things of length 1.
1185 * That was true before, but now we assume scan - reginput is the count,
1186 * rather than incrementing count on every character.]
1193 register char *scan;
1194 register char *opnd;
1196 register char *loceol = regeol;
1199 if (max != 32767 && max < loceol - scan)
1200 loceol = scan + max;
1204 while (scan < loceol && *scan != '\n')
1210 case EXACT: /* length of string is 1 */
1211 c = UCHARAT(++opnd);
1212 while (scan < loceol && UCHARAT(scan) == c)
1215 case EXACTF: /* length of string is 1 */
1216 c = UCHARAT(++opnd);
1217 while (scan < loceol &&
1218 (UCHARAT(scan) == c || UCHARAT(scan) == fold[c]))
1221 case EXACTFL: /* length of string is 1 */
1223 c = UCHARAT(++opnd);
1224 while (scan < loceol &&
1225 (UCHARAT(scan) == c || UCHARAT(scan) == fold_locale[c]))
1229 while (scan < loceol && reginclass(opnd, *scan))
1233 while (scan < loceol && isALNUM(*scan))
1238 while (scan < loceol && isALNUM_LC(*scan))
1242 while (scan < loceol && !isALNUM(*scan))
1247 while (scan < loceol && !isALNUM_LC(*scan))
1251 while (scan < loceol && isSPACE(*scan))
1256 while (scan < loceol && isSPACE_LC(*scan))
1260 while (scan < loceol && !isSPACE(*scan))
1265 while (scan < loceol && !isSPACE_LC(*scan))
1269 while (scan < loceol && isDIGIT(*scan))
1273 while (scan < loceol && !isDIGIT(*scan))
1276 default: /* Called on something of 0 width. */
1277 break; /* So match right here or not at all. */
1280 c = scan - reginput;
1287 - regclass - determine if a character falls into a character class
1299 if (p[1 + (c >> 3)] & (1 << (c & 7)))
1301 else if (flags & ANYOF_FOLD) {
1303 if (flags & ANYOF_LOCALE) {
1305 cf = fold_locale[c];
1309 if (p[1 + (cf >> 3)] & (1 << (cf & 7)))
1313 if (!match && (flags & ANYOF_ISA)) {
1316 if (((flags & ANYOF_ALNUML) && isALNUM_LC(c)) ||
1317 ((flags & ANYOF_NALNUML) && !isALNUM_LC(c)) ||
1318 ((flags & ANYOF_SPACEL) && isSPACE_LC(c)) ||
1319 ((flags & ANYOF_NSPACEL) && !isSPACE_LC(c)))
1325 return match ^ ((flags & ANYOF_INVERT) != 0);
1329 - regnext - dig the "next" pointer out of a node
1331 * [Note, when REGALIGN is defined there are two places in regmatch()
1332 * that bypass this code for speed.]
1338 register I32 offset;