/* *These* symbols are masked to allow static link. */
# define Perl_pregexec my_pregexec
# define Perl_reginitcolors my_reginitcolors
+
+# define PERL_NO_GET_CONTEXT
#endif
/*SUPPRESS 112*/
#define PERL_IN_REGEXEC_C
#include "perl.h"
+#ifdef PERL_IN_XSUB_RE
+# if defined(PERL_CAPI) || defined(PERL_OBJECT)
+# include "XSUB.h"
+# endif
+#endif
+
#include "regcomp.h"
#define RF_tainted 1 /* tainted information used? */
* Forwards.
*/
-#define REGINCLASS(p,c) (*(p) ? reginclass(p,c) : ANYOF_TEST(p,c))
+#define REGINCLASS(p,c) (ANYOF_FLAGS(p) ? reginclass(p,c) : ANYOF_BITMAP_TEST(p,c))
#define REGINCLASSUTF8(f,p) (ARG1(f) ? reginclassutf8(f,p) : swash_fetch((SV*)PL_regdata->data[ARG2(f)],p))
#define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
#define HOPc(pos,off) ((char*)HOP(pos,off))
#define HOPMAYBEc(pos,off) ((char*)HOPMAYBE(pos,off))
+static void restore_pos(pTHXo_ void *arg);
+
+
STATIC CHECKPOINT
S_regcppush(pTHX_ I32 parenfloor)
{
PL_reg_re = prog;
}
-STATIC void
-S_restore_pos(pTHX_ void *arg)
-{
- dTHR;
- if (PL_reg_eval_set) {
- if (PL_reg_oldsaved) {
- PL_reg_re->subbeg = PL_reg_oldsaved;
- PL_reg_re->sublen = PL_reg_oldsavedlen;
- RX_MATCH_COPIED_on(PL_reg_re);
- }
- PL_reg_magic->mg_len = PL_reg_oldpos;
- PL_reg_eval_set = 0;
- PL_curpm = PL_reg_oldcurpm;
- }
-}
-
/*
* Need to implement the following flags for reg_anch:
*
/* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
-/* If SCREAM, then sv should be compatible with strpos and strend.
+/* If SCREAM, then SvPVX(sv) should be compatible with strpos and strend.
Otherwise, only SvCUR(sv) is used to get strbeg. */
/* XXXX We assume that strpos is strbeg unless sv. */
+/* A failure to find a constant substring means that there is no need to make
+ an expensive call to REx engine, thus we celebrate a failure. Similarly,
+ finding a substring too deep into the string means that less calls to
+ regtry() should be needed.
+
+ REx compiler's optimizer found 4 possible hints:
+ a) Anchored substring;
+ b) Fixed substring;
+ c) Whether we are anchored (beginning-of-line or \G);
+ d) First node (of those at offset 0) which may distingush positions;
+ We use 'a', 'b', multiline-part of 'c', and try to find a position in the
+ string which does not contradict any of them.
+ */
+
char *
Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos,
char *strend, U32 flags, re_scream_pos_data *data)
{
- I32 start_shift;
+ register I32 start_shift;
/* Should be nonnegative! */
- I32 end_shift;
- char *s;
+ register I32 end_shift;
+ register char *s;
+ register SV *check;
char *t;
I32 ml_anch;
+ char *tmp;
+ register char *other_last = Nullch;
+#ifdef DEBUGGING
+ char *i_strpos = strpos;
+#endif
DEBUG_r( if (!PL_colorset) reginitcolors() );
DEBUG_r(PerlIO_printf(Perl_debug_log,
- "%sGuessing start of match:%s `%s%.60s%s%s' against `%s%.*s%s%s'\n",
+ "%sGuessing start of match, REx%s `%s%.60s%s%s' against `%s%.*s%s%s'...\n",
PL_colors[4],PL_colors[5],PL_colors[0],
prog->precomp,
PL_colors[1],
(strend - strpos > 60 ? "..." : ""))
);
- if (prog->minlen > strend - strpos)
+ if (prog->minlen > strend - strpos) {
+ DEBUG_r(PerlIO_printf(Perl_debug_log, "String too short...\n"));
goto fail;
-
- /* XXXX Move further down? */
- start_shift = prog->check_offset_min; /* okay to underestimate on CC */
- /* Should be nonnegative! */
- end_shift = prog->minlen - start_shift -
- CHR_SVLEN(prog->check_substr) + (SvTAIL(prog->check_substr) != 0);
-
- if (prog->reganch & ROPT_ANCH) {
+ }
+ if (prog->reganch & ROPT_ANCH) { /* Match at beg-of-str or after \n */
ml_anch = !( (prog->reganch & ROPT_ANCH_SINGLE)
|| ( (prog->reganch & ROPT_ANCH_BOL)
- && !PL_multiline ) );
+ && !PL_multiline ) ); /* Check after \n? */
if ((prog->check_offset_min == prog->check_offset_max) && !ml_anch) {
- /* Anchored... */
+ /* Substring at constant offset from beg-of-str... */
I32 slen;
if ( !(prog->reganch & ROPT_ANCH_GPOS) /* Checked by the caller */
- && (sv && (strpos + SvCUR(sv) != strend)) )
+ && (sv && (strpos + SvCUR(sv) != strend)) ) {
+ DEBUG_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
goto fail;
-
- s = (char*)HOP((U8*)strpos, prog->check_offset_min);
+ }
+ PL_regeol = strend; /* Used in HOP() */
+ s = HOPc(strpos, prog->check_offset_min);
if (SvTAIL(prog->check_substr)) {
slen = SvCUR(prog->check_substr); /* >= 1 */
- if ( strend - s > slen || strend - s < slen - 1 ) {
- s = Nullch;
- goto finish;
- }
- if ( strend - s == slen && strend[-1] != '\n') {
- s = Nullch;
- goto finish;
+ if ( strend - s > slen || strend - s < slen - 1
+ || (strend - s == slen && strend[-1] != '\n')) {
+ DEBUG_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
+ goto fail_finish;
}
/* Now should match s[0..slen-2] */
slen--;
if (slen && (*SvPVX(prog->check_substr) != *s
|| (slen > 1
- && memNE(SvPVX(prog->check_substr), s, slen))))
- s = Nullch;
+ && memNE(SvPVX(prog->check_substr), s, slen)))) {
+ report_neq:
+ DEBUG_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
+ goto fail_finish;
+ }
}
else if (*SvPVX(prog->check_substr) != *s
|| ((slen = SvCUR(prog->check_substr)) > 1
&& memNE(SvPVX(prog->check_substr), s, slen)))
- s = Nullch;
- else
- s = strpos;
- goto finish;
+ goto report_neq;
+ goto success_at_start;
}
+ /* Match is anchored, but substr is not anchored wrt beg-of-str. */
s = strpos;
- if (!ml_anch && (s + prog->check_offset_max < strend - prog->minlen))
- end_shift += strend - s - prog->minlen - prog->check_offset_max;
+ start_shift = prog->check_offset_min; /* okay to underestimate on CC */
+ /* Should be nonnegative! */
+ end_shift = prog->minlen - start_shift -
+ CHR_SVLEN(prog->check_substr) + (SvTAIL(prog->check_substr) != 0);
+ if (!ml_anch) {
+ I32 end = prog->check_offset_max + CHR_SVLEN(prog->check_substr)
+ - (SvTAIL(prog->check_substr) != 0);
+ I32 eshift = strend - s - end;
+
+ if (end_shift < eshift)
+ end_shift = eshift;
+ }
}
- else {
+ else { /* Can match at random position */
ml_anch = 0;
s = strpos;
+ start_shift = prog->check_offset_min; /* okay to underestimate on CC */
+ /* Should be nonnegative! */
+ end_shift = prog->minlen - start_shift -
+ CHR_SVLEN(prog->check_substr) + (SvTAIL(prog->check_substr) != 0);
}
+#ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
+ if (end_shift < 0)
+ Perl_croak(aTHX_ "panic: end_shift");
+#endif
+
+ check = prog->check_substr;
restart:
+ /* Find a possible match in the region s..strend by looking for
+ the "check" substring in the region corrected by start/end_shift. */
if (flags & REXEC_SCREAM) {
- SV *c = prog->check_substr;
char *strbeg = SvPVX(sv); /* XXXX Assume PV_force() on SCREAM! */
I32 p = -1; /* Internal iterator of scream. */
I32 *pp = data ? data->scream_pos : &p;
- if (PL_screamfirst[BmRARE(c)] >= 0
- || ( BmRARE(c) == '\n'
- && (BmPREVIOUS(c) == SvCUR(c) - 1)
- && SvTAIL(c) ))
- s = screaminstr(sv, prog->check_substr,
- start_shift + (strpos - strbeg), end_shift, pp, 0);
+ if (PL_screamfirst[BmRARE(check)] >= 0
+ || ( BmRARE(check) == '\n'
+ && (BmPREVIOUS(check) == SvCUR(check) - 1)
+ && SvTAIL(check) ))
+ s = screaminstr(sv, check,
+ start_shift + (s - strbeg), end_shift, pp, 0);
else
- s = Nullch;
+ goto fail_finish;
if (data)
*data->scream_olds = s;
}
else
s = fbm_instr((unsigned char*)s + start_shift,
(unsigned char*)strend - end_shift,
- prog->check_substr, PL_multiline ? FBMrf_MULTILINE : 0);
+ check, PL_multiline ? FBMrf_MULTILINE : 0);
/* Update the count-of-usability, remove useless subpatterns,
unshift s. */
- finish:
- if (!s) {
- ++BmUSEFUL(prog->check_substr); /* hooray */
- goto fail; /* not present */
+
+ DEBUG_r(PerlIO_printf(Perl_debug_log, "%s %s substr `%s%.*s%s'%s%s",
+ (s ? "Found" : "Did not find"),
+ ((check == prog->anchored_substr) ? "anchored" : "floating"),
+ PL_colors[0],
+ SvCUR(check) - (SvTAIL(check)!=0), SvPVX(check),
+ PL_colors[1], (SvTAIL(check) ? "$" : ""),
+ (s ? " at offset " : "...\n") ) );
+
+ if (!s)
+ goto fail_finish;
+
+ /* Finish the diagnostic message */
+ DEBUG_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
+
+ /* Got a candidate. Check MBOL anchoring, and the *other* substr.
+ Start with the other substr.
+ XXXX no SCREAM optimization yet - and a very coarse implementation
+ XXXX /ttx+/ results in anchored=`ttx', floating=`x'. floating will
+ *always* match. Probably should be marked during compile...
+ Probably it is right to do no SCREAM here...
+ */
+
+ if (prog->float_substr && prog->anchored_substr) {
+ /* Take into account the "other" substring. */
+ /* XXXX May be hopelessly wrong for UTF... */
+ if (!other_last)
+ other_last = strpos - 1;
+ if (check == prog->float_substr) {
+ do_other_anchored:
+ {
+ char *last = s - start_shift, *last1, *last2;
+ char *s1 = s;
+
+ tmp = PL_bostr;
+ t = s - prog->check_offset_max;
+ if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
+ && (!(prog->reganch & ROPT_UTF8)
+ || (PL_bostr = strpos, /* Used in regcopmaybe() */
+ (t = reghopmaybe_c(s, -(prog->check_offset_max)))
+ && t > strpos)))
+ /* EMPTY */;
+ else
+ t = strpos;
+ t += prog->anchored_offset;
+ if (t <= other_last)
+ t = other_last + 1;
+ PL_bostr = tmp;
+ last2 = last1 = strend - prog->minlen;
+ if (last < last1)
+ last1 = last;
+ /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
+ /* On end-of-str: see comment below. */
+ s = fbm_instr((unsigned char*)t,
+ (unsigned char*)last1 + prog->anchored_offset
+ + SvCUR(prog->anchored_substr)
+ - (SvTAIL(prog->anchored_substr)!=0),
+ prog->anchored_substr, PL_multiline ? FBMrf_MULTILINE : 0);
+ DEBUG_r(PerlIO_printf(Perl_debug_log, "%s anchored substr `%s%.*s%s'%s",
+ (s ? "Found" : "Contradicts"),
+ PL_colors[0],
+ SvCUR(prog->anchored_substr)
+ - (SvTAIL(prog->anchored_substr)!=0),
+ SvPVX(prog->anchored_substr),
+ PL_colors[1], (SvTAIL(prog->anchored_substr) ? "$" : "")));
+ if (!s) {
+ if (last1 >= last2) {
+ DEBUG_r(PerlIO_printf(Perl_debug_log,
+ ", giving up...\n"));
+ goto fail_finish;
+ }
+ DEBUG_r(PerlIO_printf(Perl_debug_log,
+ ", trying floating at offset %ld...\n",
+ (long)(s1 + 1 - i_strpos)));
+ PL_regeol = strend; /* Used in HOP() */
+ other_last = last1 + prog->anchored_offset;
+ s = HOPc(last, 1);
+ goto restart;
+ }
+ else {
+ DEBUG_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
+ (long)(s - i_strpos)));
+ t = s - prog->anchored_offset;
+ other_last = s - 1;
+ s = s1;
+ if (t == strpos)
+ goto try_at_start;
+ goto try_at_offset;
+ }
+ }
+ }
+ else { /* Take into account the floating substring. */
+ char *last, *last1;
+ char *s1 = s;
+
+ t = s - start_shift;
+ last1 = last = strend - prog->minlen + prog->float_min_offset;
+ if (last - t > prog->float_max_offset)
+ last = t + prog->float_max_offset;
+ s = t + prog->float_min_offset;
+ if (s <= other_last)
+ s = other_last + 1;
+ /* XXXX It is not documented what units *_offsets are in. Assume bytes. */
+ /* fbm_instr() takes into account exact value of end-of-str
+ if the check is SvTAIL(ed). Since false positives are OK,
+ and end-of-str is not later than strend we are OK. */
+ s = fbm_instr((unsigned char*)s,
+ (unsigned char*)last + SvCUR(prog->float_substr)
+ - (SvTAIL(prog->float_substr)!=0),
+ prog->float_substr, PL_multiline ? FBMrf_MULTILINE : 0);
+ DEBUG_r(PerlIO_printf(Perl_debug_log, "%s floating substr `%s%.*s%s'%s",
+ (s ? "Found" : "Contradicts"),
+ PL_colors[0],
+ SvCUR(prog->float_substr)
+ - (SvTAIL(prog->float_substr)!=0),
+ SvPVX(prog->float_substr),
+ PL_colors[1], (SvTAIL(prog->float_substr) ? "$" : "")));
+ if (!s) {
+ if (last1 == last) {
+ DEBUG_r(PerlIO_printf(Perl_debug_log,
+ ", giving up...\n"));
+ goto fail_finish;
+ }
+ DEBUG_r(PerlIO_printf(Perl_debug_log,
+ ", trying anchored starting at offset %ld...\n",
+ (long)(s1 + 1 - i_strpos)));
+ other_last = last;
+ PL_regeol = strend; /* Used in HOP() */
+ s = HOPc(t, 1);
+ goto restart;
+ }
+ else {
+ DEBUG_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
+ (long)(s - i_strpos)));
+ other_last = s - 1;
+ s = s1;
+ if (t == strpos)
+ goto try_at_start;
+ goto try_at_offset;
+ }
+ }
}
- else if (s - strpos > prog->check_offset_max &&
- ((prog->reganch & ROPT_UTF8)
- ? ((t = reghopmaybe_c(s, -(prog->check_offset_max)))
- && t >= strpos)
- : (t = s - prog->check_offset_max) != 0) ) {
+
+ t = s - prog->check_offset_max;
+ tmp = PL_bostr;
+ if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
+ && (!(prog->reganch & ROPT_UTF8)
+ || (PL_bostr = strpos, /* Used in regcopmaybe() */
+ ((t = reghopmaybe_c(s, -(prog->check_offset_max)))
+ && t > strpos)))) {
+ PL_bostr = tmp;
+ /* Fixed substring is found far enough so that the match
+ cannot start at strpos. */
+ try_at_offset:
if (ml_anch && t[-1] != '\n') {
+ /* Eventually fbm_*() should handle this, but often
+ anchored_offset is not 0, so this check will not be wasted. */
+ /* XXXX In the code below we prefer to look for "^" even in
+ presence of anchored substrings. And we search even
+ beyond the found float position. These pessimizations
+ are historical artefacts only. */
find_anchor:
- while (t < strend - end_shift - prog->minlen) {
+ while (t < strend - prog->minlen) {
if (*t == '\n') {
if (t < s - prog->check_offset_min) {
+ if (prog->anchored_substr) {
+ /* We definitely contradict the found anchored
+ substr. Due to the above check we do not
+ contradict "check" substr.
+ Thus we can arrive here only if check substr
+ is float. Redo checking for "other"=="fixed".
+ */
+ strpos = t + 1;
+ DEBUG_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
+ PL_colors[0],PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
+ goto do_other_anchored;
+ }
s = t + 1;
+ DEBUG_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
+ PL_colors[0],PL_colors[1], (long)(s - i_strpos)));
goto set_useful;
}
- s = t + 1;
+ DEBUG_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting at offset %ld...\n",
+ PL_colors[0],PL_colors[1], (long)(t + 1 - i_strpos)));
+ strpos = s = t + 1;
goto restart;
}
t++;
}
- s = Nullch;
- goto finish;
+ DEBUG_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
+ PL_colors[0],PL_colors[1]));
+ goto fail_finish;
}
s = t;
set_useful:
- ++BmUSEFUL(prog->check_substr); /* hooray/2 */
+ ++BmUSEFUL(prog->check_substr); /* hooray/5 */
}
else {
- if (ml_anch && sv
+ PL_bostr = tmp;
+ /* The found string does not prohibit matching at beg-of-str
+ - no optimization of calling REx engine can be performed,
+ unless it was an MBOL and we are not after MBOL. */
+ try_at_start:
+ /* Even in this situation we may use MBOL flag if strpos is offset
+ wrt the start of the string. */
+ if (ml_anch && sv
&& (strpos + SvCUR(sv) != strend) && strpos[-1] != '\n') {
t = strpos;
goto find_anchor;
}
- if (!(prog->reganch & ROPT_NAUGHTY)
+ DEBUG_r( if (ml_anch)
+ PerlIO_printf(Perl_debug_log, "Does not contradict /%s^%s/m...\n",
+ PL_colors[0],PL_colors[1]);
+ );
+ success_at_start:
+ if (!(prog->reganch & ROPT_NAUGHTY) /* XXXX If strpos moved? */
&& --BmUSEFUL(prog->check_substr) < 0
&& prog->check_substr == prog->float_substr) { /* boo */
/* If flags & SOMETHING - do not do it many times on the same match */
s = strpos;
}
- DEBUG_r(PerlIO_printf(Perl_debug_log, "%sFound%s at offset %ld\n",
- PL_colors[4],PL_colors[5], (long)(s - strpos)) );
+ DEBUG_r(PerlIO_printf(Perl_debug_log, "%sGuessed:%s match at offset %ld\n",
+ PL_colors[4], PL_colors[5], (long)(s - i_strpos)) );
return s;
+
+ fail_finish: /* Substring not found */
+ BmUSEFUL(prog->check_substr) += 5; /* hooray */
fail:
- DEBUG_r(PerlIO_printf(Perl_debug_log, "%sNot found...%s\n",
+ DEBUG_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
PL_colors[4],PL_colors[5]));
return Nullch;
}
register I32 tmp;
I32 minlen; /* must match at least this many chars */
I32 dontbother = 0; /* how many characters not to try at end */
- CURCUR cc;
I32 start_shift = 0; /* Offset of the start to find
constant substr. */ /* CC */
I32 end_shift = 0; /* Same for the end. */ /* CC */
char *scream_olds;
SV* oreplsv = GvSV(PL_replgv);
- cc.cur = 0;
- cc.oldcc = 0;
- PL_regcc = &cc;
+ PL_regcc = 0;
cache_re(prog);
#ifdef DEBUGGING
PL_reg_flags = 0;
PL_reg_eval_set = 0;
+ PL_reg_maxiter = 0;
if (prog->reganch & ROPT_UTF8)
PL_reg_flags |= RF_utf8;
DEBUG_r( if (!PL_colorset) reginitcolors() );
DEBUG_r(PerlIO_printf(Perl_debug_log,
- "%sMatching%s `%s%.60s%s%s' against `%s%.*s%s%s'\n",
+ "%sMatching REx%s `%s%.60s%s%s' against `%s%.*s%s%s'\n",
PL_colors[4],PL_colors[5],PL_colors[0],
prog->precomp,
PL_colors[1],
end = HOPc(strend, -dontbother) - 1;
/* for multiline we only have to try after newlines */
if (prog->check_substr) {
+ if (s == startpos)
+ goto after_try;
while (1) {
if (regtry(prog, s))
goto got_it;
+ after_try:
if (s >= end)
goto phooey;
s = re_intuit_start(prog, sv, s + 1, strend, flags, NULL);
/* We know what class it must start with. */
switch (OP(c)) {
case ANYOFUTF8:
- cc = (char *) OPERAND(c);
+ cc = MASK(c);
while (s < strend) {
if (REGINCLASSUTF8(c, (U8*)s)) {
if (tmp && regtry(prog, s))
}
break;
case ANYOF:
- cc = (char *) OPERAND(c);
+ cc = MASK(c);
while (s < strend) {
if (REGINCLASS(cc, *s)) {
if (tmp && regtry(prog, s))
s += UTF8SKIP(s);
}
break;
+ case DIGITL:
+ PL_reg_flags |= RF_tainted;
+ while (s < strend) {
+ if (isDIGIT_LC(*s)) {
+ if (tmp && regtry(prog, s))
+ goto got_it;
+ else
+ tmp = doevery;
+ }
+ else
+ tmp = 1;
+ s++;
+ }
+ break;
+ case DIGITLUTF8:
+ PL_reg_flags |= RF_tainted;
+ while (s < strend) {
+ if (isDIGIT_LC_utf8((U8*)s)) {
+ if (tmp && regtry(prog, s))
+ goto got_it;
+ else
+ tmp = doevery;
+ }
+ else
+ tmp = 1;
+ s += UTF8SKIP(s);
+ }
+ break;
case NDIGIT:
while (s < strend) {
if (!isDIGIT(*s)) {
s += UTF8SKIP(s);
}
break;
+ case NDIGITL:
+ PL_reg_flags |= RF_tainted;
+ while (s < strend) {
+ if (!isDIGIT_LC(*s)) {
+ if (tmp && regtry(prog, s))
+ goto got_it;
+ else
+ tmp = doevery;
+ }
+ else
+ tmp = 1;
+ s++;
+ }
+ break;
+ case NDIGITLUTF8:
+ PL_reg_flags |= RF_tainted;
+ while (s < strend) {
+ if (!isDIGIT_LC_utf8((U8*)s)) {
+ if (tmp && regtry(prog, s))
+ goto got_it;
+ else
+ tmp = doevery;
+ }
+ else
+ tmp = 1;
+ s += UTF8SKIP(s);
+ }
+ break;
}
}
else {
sv_setsv(oreplsv, GvSV(PL_replgv));/* So that when GvSV(replgv) is
restored, the value remains
the same. */
- restore_pos(0);
+ restore_pos(aTHXo_ 0);
}
/* make sure $`, $&, $', and $digit will work later */
phooey:
if (PL_reg_eval_set)
- restore_pos(0);
+ restore_pos(aTHXo_ 0);
return 0;
}
}
PL_reg_magic = mg;
PL_reg_oldpos = mg->mg_len;
- SAVEDESTRUCTOR(S_restore_pos, 0);
+ SAVEDESTRUCTOR(restore_pos, 0);
}
if (!PL_reg_curpm)
New(22,PL_reg_curpm, 1, PMOP);
nextchr = UCHARAT(++locinput);
break;
case EXACT:
- s = (char *) OPERAND(scan);
- ln = UCHARAT(s++);
+ s = STRING(scan);
+ ln = STR_LEN(scan);
/* Inline the first character, for speed. */
if (UCHARAT(s) != nextchr)
sayNO;
PL_reg_flags |= RF_tainted;
/* FALL THROUGH */
case EXACTF:
- s = (char *) OPERAND(scan);
- ln = UCHARAT(s++);
+ s = STRING(scan);
+ ln = STR_LEN(scan);
if (UTF) {
char *l = locinput;
nextchr = UCHARAT(locinput);
break;
case ANYOFUTF8:
- s = (char *) OPERAND(scan);
+ s = MASK(scan);
if (!REGINCLASSUTF8(scan, (U8*)locinput))
sayNO;
if (locinput >= PL_regeol)
nextchr = UCHARAT(locinput);
break;
case ANYOF:
- s = (char *) OPERAND(scan);
+ s = MASK(scan);
if (nextchr < 0)
nextchr = UCHARAT(locinput);
if (!REGINCLASS(s, nextchr))
sayNO;
nextchr = UCHARAT(++locinput);
break;
+ case DIGITL:
+ PL_reg_flags |= RF_tainted;
+ /* FALL THROUGH */
case DIGIT:
- if (!isDIGIT(nextchr))
+ if (!nextchr && locinput >= PL_regeol)
+ sayNO;
+ if (!(OP(scan) == DIGIT
+ ? isDIGIT(nextchr) : isDIGIT_LC(nextchr)))
sayNO;
nextchr = UCHARAT(++locinput);
break;
+ case DIGITLUTF8:
+ PL_reg_flags |= RF_tainted;
+ /* FALL THROUGH */
case DIGITUTF8:
+ if (!nextchr)
+ sayNO;
if (nextchr & 0x80) {
- if (!(swash_fetch(PL_utf8_digit,(U8*)locinput)))
+ if (OP(scan) == NDIGITUTF8
+ ? swash_fetch(PL_utf8_digit,(U8*)locinput)
+ : isDIGIT_LC_utf8((U8*)locinput))
+ {
sayNO;
+ }
locinput += PL_utf8skip[nextchr];
nextchr = UCHARAT(locinput);
break;
sayNO;
nextchr = UCHARAT(++locinput);
break;
+ case NDIGITL:
+ PL_reg_flags |= RF_tainted;
+ /* FALL THROUGH */
case NDIGIT:
- if (!nextchr && locinput >= PL_regeol)
+ if (!nextchr)
sayNO;
- if (isDIGIT(nextchr))
+ if (OP(scan) == DIGIT
+ ? isDIGIT(nextchr) : isDIGIT_LC(nextchr))
sayNO;
nextchr = UCHARAT(++locinput);
break;
+ case NDIGITLUTF8:
+ PL_reg_flags |= RF_tainted;
+ /* FALL THROUGH */
case NDIGITUTF8:
if (!nextchr && locinput >= PL_regeol)
sayNO;
case REFF:
n = ARG(scan); /* which paren pair */
ln = PL_regstartp[n];
+ PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
if (*PL_reglastparen < n || ln == -1)
sayNO; /* Do not match unless seen CLOSEn. */
if (ln == PL_regendp[n])
regexp *re;
MAGIC *mg = Null(MAGIC*);
re_cc_state state;
- CURCUR cctmp;
CHECKPOINT cp, lastcp;
if(SvROK(ret) || SvRMAGICAL(ret)) {
state.cc = PL_regcc;
state.re = PL_reg_re;
- cctmp.cur = 0;
- cctmp.oldcc = 0;
- PL_regcc = &cctmp;
+ PL_regcc = 0;
cp = regcppush(0); /* Save *all* the positions. */
REGCP_SET;
*PL_reglastparen = 0;
PL_reg_call_cc = &state;
PL_reginput = locinput;
+
+ /* XXXX This is too dramatic a measure... */
+ PL_reg_maxiter = 0;
+
if (regmatch(re->program + 1)) {
+ /* Even though we succeeded, we need to restore
+ global variables, since we may be wrapped inside
+ SUSPEND, thus the match may be not finished yet. */
+
+ /* XXXX Do this only if SUSPENDed? */
+ PL_reg_call_cc = state.prev;
+ PL_regcc = state.cc;
+ PL_reg_re = state.re;
+ cache_re(PL_reg_re);
+
+ /* XXXX This is too dramatic a measure... */
+ PL_reg_maxiter = 0;
+
+ /* These are needed even if not SUSPEND. */
ReREFCNT_dec(re);
regcpblow(cp);
sayYES;
PL_regcc = state.cc;
PL_reg_re = state.re;
cache_re(PL_reg_re);
+
+ /* XXXX This is too dramatic a measure... */
+ PL_reg_maxiter = 0;
+
sayNO;
}
sw = SvTRUE(ret);
sw = (*PL_reglastparen >= n && PL_regendp[n] != -1);
break;
case IFTHEN:
+ PL_reg_leftiter = PL_reg_maxiter; /* Void cache */
if (sw)
next = NEXTOPER(NEXTOPER(scan));
else {
case LOGICAL:
logical = scan->flags;
break;
+/*******************************************************************
+ PL_regcc contains infoblock about the innermost (...)* loop, and
+ a pointer to the next outer infoblock.
+
+ Here is how Y(A)*Z is processed (if it is compiled into CURLYX/WHILEM):
+
+ 1) After matching X, regnode for CURLYX is processed;
+
+ 2) This regnode creates infoblock on the stack, and calls
+ regmatch() recursively with the starting point at WHILEM node;
+
+ 3) Each hit of WHILEM node tries to match A and Z (in the order
+ depending on the current iteration, min/max of {min,max} and
+ greediness). The information about where are nodes for "A"
+ and "Z" is read from the infoblock, as is info on how many times "A"
+ was already matched, and greediness.
+
+ 4) After A matches, the same WHILEM node is hit again.
+
+ 5) Each time WHILEM is hit, PL_regcc is the infoblock created by CURLYX
+ of the same pair. Thus when WHILEM tries to match Z, it temporarily
+ resets PL_regcc, since this Y(A)*Z can be a part of some other loop:
+ as in (Y(A)*Z)*. If Z matches, the automaton will hit the WHILEM node
+ of the external loop.
+
+ Currently present infoblocks form a tree with a stem formed by PL_curcc
+ and whatever it mentions via ->next, and additional attached trees
+ corresponding to temporarily unset infoblocks as in "5" above.
+
+ In the following picture infoblocks for outer loop of
+ (Y(A)*?Z)*?T are denoted O, for inner I. NULL starting block
+ is denoted by x. The matched string is YAAZYAZT. Temporarily postponed
+ infoblocks are drawn below the "reset" infoblock.
+
+ In fact in the picture below we do not show failed matches for Z and T
+ by WHILEM blocks. [We illustrate minimal matches, since for them it is
+ more obvious *why* one needs to *temporary* unset infoblocks.]
+
+ Matched REx position InfoBlocks Comment
+ (Y(A)*?Z)*?T x
+ Y(A)*?Z)*?T x <- O
+ Y (A)*?Z)*?T x <- O
+ Y A)*?Z)*?T x <- O <- I
+ YA )*?Z)*?T x <- O <- I
+ YA A)*?Z)*?T x <- O <- I
+ YAA )*?Z)*?T x <- O <- I
+ YAA Z)*?T x <- O # Temporary unset I
+ I
+
+ YAAZ Y(A)*?Z)*?T x <- O
+ I
+
+ YAAZY (A)*?Z)*?T x <- O
+ I
+
+ YAAZY A)*?Z)*?T x <- O <- I
+ I
+
+ YAAZYA )*?Z)*?T x <- O <- I
+ I
+
+ YAAZYA Z)*?T x <- O # Temporary unset I
+ I,I
+
+ YAAZYAZ )*?T x <- O
+ I,I
+
+ YAAZYAZ T x # Temporary unset O
+ O
+ I,I
+
+ YAAZYAZT x
+ O
+ I,I
+ *******************************************************************/
case CURLYX: {
CURCUR cc;
CHECKPOINT cp = PL_savestack_ix;
/*
* This is really hard to understand, because after we match
* what we're trying to match, we must make sure the rest of
- * the RE is going to match for sure, and to do that we have
+ * the REx is going to match for sure, and to do that we have
* to go back UP the parse tree by recursing ever deeper. And
* if it fails, we have to reset our parent's current state
* that we can try again after backing off.
if (locinput == cc->lastloc && n >= cc->min) {
PL_regcc = cc->oldcc;
- ln = PL_regcc->cur;
+ if (PL_regcc)
+ ln = PL_regcc->cur;
DEBUG_r(
PerlIO_printf(Perl_debug_log,
"%*s empty match detected, try continuation...\n",
"%*s failed...\n",
REPORT_CODE_OFF+PL_regindent*2, "")
);
- PL_regcc->cur = ln;
+ if (PL_regcc)
+ PL_regcc->cur = ln;
PL_regcc = cc;
sayNO;
}
sayNO;
}
+ if (scan->flags) {
+ /* Check whether we already were at this position.
+ Postpone detection until we know the match is not
+ *that* much linear. */
+ if (!PL_reg_maxiter) {
+ PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4);
+ PL_reg_leftiter = PL_reg_maxiter;
+ }
+ if (PL_reg_leftiter-- == 0) {
+ I32 size = (PL_reg_maxiter + 7)/8;
+ if (PL_reg_poscache) {
+ if (PL_reg_poscache_size < size) {
+ Renew(PL_reg_poscache, size, char);
+ PL_reg_poscache_size = size;
+ }
+ Zero(PL_reg_poscache, size, char);
+ }
+ else {
+ PL_reg_poscache_size = size;
+ Newz(29, PL_reg_poscache, size, char);
+ }
+ DEBUG_r(
+ PerlIO_printf(Perl_debug_log,
+ "%sDetected a super-linear match, switching on caching%s...\n",
+ PL_colors[4], PL_colors[5])
+ );
+ }
+ if (PL_reg_leftiter < 0) {
+ I32 o = locinput - PL_bostr, b;
+
+ o = (scan->flags & 0xf) - 1 + o * (scan->flags>>4);
+ b = o % 8;
+ o /= 8;
+ if (PL_reg_poscache[o] & (1<<b)) {
+ DEBUG_r(
+ PerlIO_printf(Perl_debug_log,
+ "%*s already tried at this position...\n",
+ REPORT_CODE_OFF+PL_regindent*2, "")
+ );
+ sayNO;
+ }
+ PL_reg_poscache[o] |= (1<<b);
+ }
+ }
+
/* Prefer next over scan for minimal matching. */
if (cc->minmod) {
PL_regcc = cc->oldcc;
- ln = PL_regcc->cur;
+ if (PL_regcc)
+ ln = PL_regcc->cur;
cp = regcppush(cc->parenfloor);
REGCP_SET;
if (regmatch(cc->next)) {
}
REGCP_UNWIND;
regcppop();
- PL_regcc->cur = ln;
+ if (PL_regcc)
+ PL_regcc->cur = ln;
PL_regcc = cc;
if (n >= cc->max) { /* Maximum greed exceeded? */
/* Failed deeper matches of scan, so see if this one works. */
PL_regcc = cc->oldcc;
- ln = PL_regcc->cur;
+ if (PL_regcc)
+ ln = PL_regcc->cur;
if (regmatch(cc->next))
sayYES;
DEBUG_r(
PerlIO_printf(Perl_debug_log, "%*s failed...\n",
REPORT_CODE_OFF+PL_regindent*2, "")
);
- PL_regcc->cur = ln;
+ if (PL_regcc)
+ PL_regcc->cur = ln;
PL_regcc = cc;
cc->cur = n - 1;
cc->lastloc = lastloc;
ln = n;
locinput = PL_reginput;
if (PL_regkind[(U8)OP(next)] == EXACT) {
- c1 = UCHARAT(OPERAND(next) + 1);
+ c1 = (U8)*STRING(next);
if (OP(next) == EXACTF)
c2 = PL_fold[c1];
else if (OP(next) == EXACTFL)
);
if (n >= ln) {
if (PL_regkind[(U8)OP(next)] == EXACT) {
- c1 = UCHARAT(OPERAND(next) + 1);
+ c1 = (U8)*STRING(next);
if (OP(next) == EXACTF)
c2 = PL_fold[c1];
else if (OP(next) == EXACTFL)
* when we know what character comes next.
*/
if (PL_regkind[(U8)OP(next)] == EXACT) {
- c1 = UCHARAT(OPERAND(next) + 1);
+ c1 = (U8)*STRING(next);
if (OP(next) == EXACTF)
c2 = PL_fold[c1];
else if (OP(next) == EXACTFL)
scan = PL_reginput;
if (max != REG_INFTY && max < loceol - scan)
loceol = scan + max;
- opnd = (char *) OPERAND(p);
switch (OP(p)) {
case REG_ANY:
while (scan < loceol && *scan != '\n')
}
break;
case EXACT: /* length of string is 1 */
- c = UCHARAT(++opnd);
+ c = (U8)*STRING(p);
while (scan < loceol && UCHARAT(scan) == c)
scan++;
break;
case EXACTF: /* length of string is 1 */
- c = UCHARAT(++opnd);
+ c = (U8)*STRING(p);
while (scan < loceol &&
(UCHARAT(scan) == c || UCHARAT(scan) == PL_fold[c]))
scan++;
break;
case EXACTFL: /* length of string is 1 */
PL_reg_flags |= RF_tainted;
- c = UCHARAT(++opnd);
+ c = (U8)*STRING(p);
while (scan < loceol &&
(UCHARAT(scan) == c || UCHARAT(scan) == PL_fold_locale[c]))
scan++;
}
break;
case ANYOF:
+ opnd = MASK(p);
while (scan < loceol && REGINCLASS(opnd, *scan))
scan++;
break;
S_reginclass(pTHX_ register char *p, register I32 c)
{
dTHR;
- char flags = *p;
+ char flags = ANYOF_FLAGS(p);
bool match = FALSE;
c &= 0xFF;
- if (ANYOF_TEST(p, c))
+ if (ANYOF_BITMAP_TEST(p, c))
match = TRUE;
else if (flags & ANYOF_FOLD) {
I32 cf;
}
else
cf = PL_fold[c];
- if (ANYOF_TEST(p, cf))
+ if (ANYOF_BITMAP_TEST(p, cf))
match = TRUE;
}
- if (!match && (flags & ANYOF_ISA)) {
+ if (!match && (flags & ANYOF_CLASS)) {
PL_reg_flags |= RF_tainted;
-
- if (((flags & ANYOF_ALNUML) && isALNUM_LC(c)) ||
- ((flags & ANYOF_NALNUML) && !isALNUM_LC(c)) ||
- ((flags & ANYOF_SPACEL) && isSPACE_LC(c)) ||
- ((flags & ANYOF_NSPACEL) && !isSPACE_LC(c)))
+ if (
+ (ANYOF_CLASS_TEST(p, ANYOF_ALNUM) && isALNUM_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_NALNUM) && !isALNUM_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_SPACE) && isSPACE_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_NSPACE) && !isSPACE_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_DIGIT) && isDIGIT_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_NDIGIT) && !isDIGIT_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_ALNUMC) && isALNUMC_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_NALNUMC) && !isALNUMC_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_ALPHA) && isALPHA_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_NALPHA) && !isALPHA_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_ASCII) && isASCII(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_NASCII) && !isASCII(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_CNTRL) && isCNTRL_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_NCNTRL) && !isCNTRL_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_GRAPH) && isGRAPH_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_NGRAPH) && !isGRAPH_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_LOWER) && isLOWER_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_NLOWER) && !isLOWER_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_PRINT) && isPRINT_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_NPRINT) && !isPRINT_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_PUNCT) && isPUNCT_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_NPUNCT) && !isPUNCT_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_UPPER) && isUPPER_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_NUPPER) && !isUPPER_LC(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_XDIGIT) && isXDIGIT(c)) ||
+ (ANYOF_CLASS_TEST(p, ANYOF_NXDIGIT) && !isXDIGIT(c))
+ ) /* How's that for a conditional? */
{
match = TRUE;
}
match = TRUE;
}
- if (!match && (flags & ANYOF_ISA)) {
- PL_reg_flags |= RF_tainted;
-
- if (((flags & ANYOF_ALNUML) && isALNUM_LC_utf8(p)) ||
- ((flags & ANYOF_NALNUML) && !isALNUM_LC_utf8(p)) ||
- ((flags & ANYOF_SPACEL) && isSPACE_LC_utf8(p)) ||
- ((flags & ANYOF_NSPACEL) && !isSPACE_LC_utf8(p)))
- {
- match = TRUE;
- }
- }
+ /* UTF8 combined with ANYOF_CLASS is ill-defined. */
return (flags & ANYOF_INVERT) ? !match : match;
}
}
return s;
}
+
+#ifdef PERL_OBJECT
+#define NO_XSLOCKS
+#include "XSUB.h"
+#endif
+
+static void
+restore_pos(pTHXo_ void *arg)
+{
+ dTHR;
+ if (PL_reg_eval_set) {
+ if (PL_reg_oldsaved) {
+ PL_reg_re->subbeg = PL_reg_oldsaved;
+ PL_reg_re->sublen = PL_reg_oldsavedlen;
+ RX_MATCH_COPIED_on(PL_reg_re);
+ }
+ PL_reg_magic->mg_len = PL_reg_oldpos;
+ PL_reg_eval_set = 0;
+ PL_curpm = PL_reg_oldcurpm;
+ }
+}
+
projects / p5sagit/p5-mst-13.2.git / blobdiff