# include "regcomp.h"
#endif
+#include "inline_invlist.c"
+#include "unicode_constants.h"
+
#define RF_tainted 1 /* tainted information used? e.g. locale */
#define RF_warned 2 /* warned about big count? */
#define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
/* these are unrolled below in the CCC_TRY_XXX defined */
-#ifdef EBCDIC
- /* Often 'str' is a hard-coded utf8 string instead of utfebcdic. so just
- * skip the check on EBCDIC platforms */
-# define LOAD_UTF8_CHARCLASS(class,str) LOAD_UTF8_CHARCLASS_NO_CHECK(class)
-#else
-# define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
+#define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \
if (!CAT2(PL_utf8_,class)) { \
bool ok; \
ENTER; save_re_context(); \
ok=CAT2(is_utf8_,class)((const U8*)str); \
+ PERL_UNUSED_VAR(ok); \
assert(ok); assert(CAT2(PL_utf8_,class)); LEAVE; } } STMT_END
-#endif
-
/* Doesn't do an assert to verify that is correct */
#define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
if (!CAT2(PL_utf8_,class)) { \
#define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
#define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
- LOAD_UTF8_CHARCLASS(X_begin, " "); \
- LOAD_UTF8_CHARCLASS(X_non_hangul, "A"); \
- /* These are utf8 constants, and not utf-ebcdic constants, so the \
- * assert should likely and hopefully fail on an EBCDIC machine */ \
- LOAD_UTF8_CHARCLASS(X_extend, "\xcc\x80"); /* U+0300 */ \
- \
- /* No asserts are done for these, in case called on an early \
- * Unicode version in which they map to nothing */ \
- LOAD_UTF8_CHARCLASS_NO_CHECK(X_prepend);/* U+0E40 "\xe0\xb9\x80" */ \
- LOAD_UTF8_CHARCLASS_NO_CHECK(X_L); /* U+1100 "\xe1\x84\x80" */ \
- LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV); /* U+AC00 "\xea\xb0\x80" */ \
- LOAD_UTF8_CHARCLASS_NO_CHECK(X_LVT); /* U+AC01 "\xea\xb0\x81" */ \
- LOAD_UTF8_CHARCLASS_NO_CHECK(X_LV_LVT_V);/* U+AC01 "\xea\xb0\x81" */\
- LOAD_UTF8_CHARCLASS_NO_CHECK(X_T); /* U+11A8 "\xe1\x86\xa8" */ \
- LOAD_UTF8_CHARCLASS_NO_CHECK(X_V) /* U+1160 "\xe1\x85\xa0" */
+ /* No asserts are done for some of these, in case called on a */ \
+ /* Unicode version in which they map to nothing */ \
+ LOAD_UTF8_CHARCLASS(X_regular_begin, HYPHEN_UTF8); \
+ LOAD_UTF8_CHARCLASS(X_extend, COMBINING_GRAVE_ACCENT_UTF8); \
#define PLACEHOLDER /* Something for the preprocessor to grab onto */
if (POS_OR_NEG (UTF8_TEST)) { \
sayNO; \
} \
- locinput += PL_utf8skip[nextchr]; \
- nextchr = UCHARAT(locinput); \
- break; \
} \
- if (POS_OR_NEG (FUNC(nextchr))) { \
- sayNO; \
+ else if (POS_OR_NEG (FUNC(nextchr))) { \
+ sayNO; \
} \
- nextchr = UCHARAT(++locinput); \
- break;
+ goto increment_locinput;
/* Handle the non-locale cases for a character class and its complement. It
* calls _CCC_TRY_CODE with a ! to complement the test for the character class.
sayNO; \
} \
/* Matched a utf8-invariant, so don't have to worry about utf8 */ \
- nextchr = UCHARAT(++locinput); \
+ locinput++; \
break; \
case NNAMEA: \
if (locinput >= PL_regeol || FUNCA(nextchr)) { \
sayNO; \
} \
- if (utf8_target) { \
- locinput += PL_utf8skip[nextchr]; \
- nextchr = UCHARAT(locinput); \
- } \
- else { \
- nextchr = UCHARAT(++locinput); \
- } \
- break; \
+ goto increment_locinput; \
/* Generate the non-locale cases */ \
_CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR)
static void restore_pos(pTHX_ void *arg);
-#define REGCP_PAREN_ELEMS 4
-#define REGCP_OTHER_ELEMS 4
+#define REGCP_PAREN_ELEMS 3
+#define REGCP_OTHER_ELEMS 3
#define REGCP_FRAME_ELEMS 1
/* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
* are needed for the regexp context stack bookkeeping. */
const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS;
const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
- int p;
+ I32 p;
GET_RE_DEBUG_FLAGS_DECL;
PERL_ARGS_ASSERT_REGCPPUSH;
SSGROW(total_elems + REGCP_FRAME_ELEMS);
- for (p = PL_regsize; p > parenfloor; p--) {
+ DEBUG_BUFFERS_r(
+ if ((int)PL_regsize > (int)parenfloor)
+ PerlIO_printf(Perl_debug_log,
+ "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
+ PTR2UV(rex),
+ PTR2UV(rex->offs)
+ );
+ );
+ for (p = parenfloor+1; p <= (I32)PL_regsize; p++) {
/* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
SSPUSHINT(rex->offs[p].end);
SSPUSHINT(rex->offs[p].start);
- SSPUSHPTR(PL_reg_start_tmp[p]);
- SSPUSHINT(p);
+ SSPUSHINT(rex->offs[p].start_tmp);
DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
- " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n",
- (UV)p, (IV)rex->offs[p].start,
- (IV)(PL_reg_start_tmp[p] - PL_bostr),
- (IV)rex->offs[p].end
+ " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
+ (UV)p,
+ (IV)rex->offs[p].start,
+ (IV)rex->offs[p].start_tmp,
+ (IV)rex->offs[p].end
));
}
/* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
SSPUSHINT(PL_regsize);
SSPUSHINT(rex->lastparen);
SSPUSHINT(rex->lastcloseparen);
- SSPUSHPTR(PL_reginput);
SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
return retval;
(IV)(cp), (IV)PL_savestack_ix)); \
regcpblow(cp)
-STATIC char *
+#define UNWIND_PAREN(lp, lcp) \
+ for (n = rex->lastparen; n > lp; n--) \
+ rex->offs[n].end = -1; \
+ rex->lastparen = n; \
+ rex->lastcloseparen = lcp;
+
+
+STATIC void
S_regcppop(pTHX_ regexp *rex)
{
dVAR;
UV i;
- char *input;
+ U32 paren;
GET_RE_DEBUG_FLAGS_DECL;
PERL_ARGS_ASSERT_REGCPPOP;
i = SSPOPUV;
assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
- input = (char *) SSPOPPTR;
rex->lastcloseparen = SSPOPINT;
rex->lastparen = SSPOPINT;
PL_regsize = SSPOPINT;
i -= REGCP_OTHER_ELEMS;
/* Now restore the parentheses context. */
+ DEBUG_BUFFERS_r(
+ if (i || rex->lastparen + 1 <= rex->nparens)
+ PerlIO_printf(Perl_debug_log,
+ "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
+ PTR2UV(rex),
+ PTR2UV(rex->offs)
+ );
+ );
+ paren = PL_regsize;
for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
I32 tmps;
- U32 paren = (U32)SSPOPINT;
- PL_reg_start_tmp[paren] = (char *) SSPOPPTR;
+ rex->offs[paren].start_tmp = SSPOPINT;
rex->offs[paren].start = SSPOPINT;
tmps = SSPOPINT;
if (paren <= rex->lastparen)
rex->offs[paren].end = tmps;
- DEBUG_BUFFERS_r(
- PerlIO_printf(Perl_debug_log,
- " restoring \\%"UVuf" to %"IVdf"(%"IVdf")..%"IVdf"%s\n",
- (UV)paren, (IV)rex->offs[paren].start,
- (IV)(PL_reg_start_tmp[paren] - PL_bostr),
- (IV)rex->offs[paren].end,
- (paren > rex->lastparen ? "(no)" : ""));
+ DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
+ " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
+ (UV)paren,
+ (IV)rex->offs[paren].start,
+ (IV)rex->offs[paren].start_tmp,
+ (IV)rex->offs[paren].end,
+ (paren > rex->lastparen ? "(skipped)" : ""));
);
+ paren--;
}
- DEBUG_BUFFERS_r(
- if (rex->lastparen + 1 <= rex->nparens) {
- PerlIO_printf(Perl_debug_log,
- " restoring \\%"IVdf"..\\%"IVdf" to undef\n",
- (IV)(rex->lastparen + 1), (IV)rex->nparens);
- }
- );
#if 1
/* It would seem that the similar code in regtry()
* already takes care of this, and in fact it is in
if (i > PL_regsize)
rex->offs[i].start = -1;
rex->offs[i].end = -1;
+ DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
+ " \\%"UVuf": %s ..-1 undeffing\n",
+ (UV)i,
+ (i > PL_regsize) ? "-1" : " "
+ ));
}
#endif
- return input;
+}
+
+/* restore the parens and associated vars at savestack position ix,
+ * but without popping the stack */
+
+STATIC void
+S_regcp_restore(pTHX_ regexp *rex, I32 ix)
+{
+ I32 tmpix = PL_savestack_ix;
+ PL_savestack_ix = ix;
+ regcppop(rex);
+ PL_savestack_ix = tmpix;
}
#define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
I32
Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
char *strbeg, I32 minend, SV *screamer, U32 nosave)
-/* strend: pointer to null at end of string */
-/* strbeg: real beginning of string */
-/* minend: end of match must be >=minend after stringarg. */
-/* nosave: For optimizations. */
+/* stringarg: the point in the string at which to begin matching */
+/* strend: pointer to null at end of string */
+/* strbeg: real beginning of string */
+/* minend: end of match must be >= minend bytes after stringarg. */
+/* screamer: SV being matched: only used for utf8 flag, pos() etc; string
+ * itself is accessed via the pointers above */
+/* nosave: For optimizations. */
{
PERL_ARGS_ASSERT_PREGEXEC;
{
dVAR;
struct regexp *const prog = (struct regexp *)SvANY(rx);
- register I32 start_shift = 0;
+ I32 start_shift = 0;
/* Should be nonnegative! */
- register I32 end_shift = 0;
- register char *s;
- register SV *check;
+ I32 end_shift = 0;
+ char *s;
+ SV *check;
char *strbeg;
char *t;
const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
I32 ml_anch;
- register char *other_last = NULL; /* other substr checked before this */
+ char *other_last = NULL; /* other substr checked before this */
char *check_at = NULL; /* check substr found at this pos */
char *checked_upto = NULL; /* how far into the string we have already checked using find_byclass*/
const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
const U8 *fold_array; /* array for folding ords < 256 */
STRLEN ln;
STRLEN lnc;
- register STRLEN uskip;
+ STRLEN uskip;
U8 c1;
U8 c2;
char *e;
- register I32 tmp = 1; /* Scratch variable? */
- register const bool utf8_target = PL_reg_match_utf8;
+ I32 tmp = 1; /* Scratch variable? */
+ const bool utf8_target = PL_reg_match_utf8;
UV utf8_fold_flags = 0;
RXi_GET_DECL(prog,progi);
!is_HORIZWS_latin1(s)
);
break;
+ case POSIXA:
+ /* Don't need to worry about utf8, as it can match only a single
+ * byte invariant character. The flag in this node type is the
+ * class number to pass to _generic_isCC() to build a mask for
+ * searching in PL_charclass[] */
+ REXEC_FBC_CLASS_SCAN( _generic_isCC_A(*s, FLAGS(c)));
+ break;
+ case NPOSIXA:
+ REXEC_FBC_CSCAN(
+ !_generic_isCC_A(*s, FLAGS(c)),
+ !_generic_isCC_A(*s, FLAGS(c))
+ );
+ break;
+
case AHOCORASICKC:
case AHOCORASICK:
{
I32
Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
-/* strend: pointer to null at end of string */
-/* strbeg: real beginning of string */
-/* minend: end of match must be >=minend after stringarg. */
-/* data: May be used for some additional optimizations.
- Currently its only used, with a U32 cast, for transmitting
- the ganch offset when doing a /g match. This will change */
-/* nosave: For optimizations. */
+/* stringarg: the point in the string at which to begin matching */
+/* strend: pointer to null at end of string */
+/* strbeg: real beginning of string */
+/* minend: end of match must be >= minend bytes after stringarg. */
+/* sv: SV being matched: only used for utf8 flag, pos() etc; string
+ * itself is accessed via the pointers above */
+/* data: May be used for some additional optimizations.
+ Currently its only used, with a U32 cast, for transmitting
+ the ganch offset when doing a /g match. This will change */
+/* nosave: For optimizations. */
+
{
dVAR;
struct regexp *const prog = (struct regexp *)SvANY(rx);
/*register*/ char *s;
- register regnode *c;
+ regnode *c;
/*register*/ char *startpos = stringarg;
I32 minlen; /* must match at least this many chars */
I32 dontbother = 0; /* how many characters not to try at end */
swap = prog->offs;
/* do we need a save destructor here for eval dies? */
Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
+ DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
+ "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
+ PTR2UV(prog),
+ PTR2UV(swap),
+ PTR2UV(prog->offs)
+ ));
}
if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
re_scream_pos_data d;
goto phooey;
got_it:
+ DEBUG_BUFFERS_r(
+ if (swap)
+ PerlIO_printf(Perl_debug_log,
+ "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
+ PTR2UV(prog),
+ PTR2UV(swap)
+ );
+ );
Safefree(swap);
RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
/* make sure $`, $&, $', and $digit will work later */
if ( !(flags & REXEC_NOT_FIRST) ) {
- RX_MATCH_COPY_FREE(rx);
if (flags & REXEC_COPY_STR) {
- const I32 i = PL_regeol - startpos + (stringarg - strbeg);
#ifdef PERL_OLD_COPY_ON_WRITE
if ((SvIsCOW(sv)
|| (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
"Copy on write: regexp capture, type %d\n",
(int) SvTYPE(sv));
}
+ RX_MATCH_COPY_FREE(rx);
prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
assert (SvPOKp(prog->saved_copy));
+ prog->sublen = PL_regeol - strbeg;
+ prog->suboffset = 0;
+ prog->subcoffset = 0;
} else
#endif
{
- RX_MATCH_COPIED_on(rx);
- s = savepvn(strbeg, i);
- prog->subbeg = s;
- }
- prog->sublen = i;
+ I32 min = 0;
+ I32 max = PL_regeol - strbeg;
+ I32 sublen;
+
+ if ( (flags & REXEC_COPY_SKIP_POST)
+ && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
+ && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
+ ) { /* don't copy $' part of string */
+ U32 n = 0;
+ max = -1;
+ /* calculate the right-most part of the string covered
+ * by a capture. Due to look-ahead, this may be to
+ * the right of $&, so we have to scan all captures */
+ while (n <= prog->lastparen) {
+ if (prog->offs[n].end > max)
+ max = prog->offs[n].end;
+ n++;
+ }
+ if (max == -1)
+ max = (PL_sawampersand & SAWAMPERSAND_LEFT)
+ ? prog->offs[0].start
+ : 0;
+ assert(max >= 0 && max <= PL_regeol - strbeg);
+ }
+
+ if ( (flags & REXEC_COPY_SKIP_PRE)
+ && !(RX_EXTFLAGS(rx) & RXf_PMf_KEEPCOPY) /* //p */
+ && !(PL_sawampersand & SAWAMPERSAND_LEFT)
+ ) { /* don't copy $` part of string */
+ U32 n = 0;
+ min = max;
+ /* calculate the left-most part of the string covered
+ * by a capture. Due to look-behind, this may be to
+ * the left of $&, so we have to scan all captures */
+ while (min && n <= prog->lastparen) {
+ if ( prog->offs[n].start != -1
+ && prog->offs[n].start < min)
+ {
+ min = prog->offs[n].start;
+ }
+ n++;
+ }
+ if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
+ && min > prog->offs[0].end
+ )
+ min = prog->offs[0].end;
+
+ }
+
+ assert(min >= 0 && min <= max && min <= PL_regeol - strbeg);
+ sublen = max - min;
+
+ if (RX_MATCH_COPIED(rx)) {
+ if (sublen > prog->sublen)
+ prog->subbeg =
+ (char*)saferealloc(prog->subbeg, sublen+1);
+ }
+ else
+ prog->subbeg = (char*)safemalloc(sublen+1);
+ Copy(strbeg + min, prog->subbeg, sublen, char);
+ prog->subbeg[sublen] = '\0';
+ prog->suboffset = min;
+ prog->sublen = sublen;
+ RX_MATCH_COPIED_on(rx);
+ }
+ prog->subcoffset = prog->suboffset;
+ if (prog->suboffset && utf8_target) {
+ /* Convert byte offset to chars.
+ * XXX ideally should only compute this if @-/@+
+ * has been seen, a la PL_sawampersand ??? */
+
+ /* If there's a direct correspondence between the
+ * string which we're matching and the original SV,
+ * then we can use the utf8 len cache associated with
+ * the SV. In particular, it means that under //g,
+ * sv_pos_b2u() will use the previously cached
+ * position to speed up working out the new length of
+ * subcoffset, rather than counting from the start of
+ * the string each time. This stops
+ * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
+ * from going quadratic */
+ if (SvPOKp(sv) && SvPVX(sv) == strbeg)
+ sv_pos_b2u(sv, &(prog->subcoffset));
+ else
+ prog->subcoffset = utf8_length((U8*)strbeg,
+ (U8*)(strbeg+prog->suboffset));
+ }
}
else {
+ RX_MATCH_COPY_FREE(rx);
prog->subbeg = strbeg;
+ prog->suboffset = 0;
+ prog->subcoffset = 0;
prog->sublen = PL_regeol - strbeg; /* strend may have been modified */
}
}
restore_pos(aTHX_ prog);
if (swap) {
/* we failed :-( roll it back */
+ DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
+ "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
+ PTR2UV(prog),
+ PTR2UV(prog->offs),
+ PTR2UV(swap)
+ ));
Safefree(prog->offs);
prog->offs = swap;
}
}
+/* Set which rex is pointed to by PL_reg_state, handling ref counting.
+ * Do inc before dec, in case old and new rex are the same */
+#define SET_reg_curpm(Re2) \
+ if (PL_reg_state.re_state_eval_setup_done) { \
+ (void)ReREFCNT_inc(Re2); \
+ ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
+ PM_SETRE((PL_reg_curpm), (Re2)); \
+ }
+
+
/*
- regtry - try match at specific point
*/
STATIC I32 /* 0 failure, 1 success */
-S_regtry(pTHX_ regmatch_info *reginfo, char **startpos)
+S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
{
dVAR;
CHECKPOINT lastcp;
REGEXP *const rx = reginfo->prog;
regexp *const prog = (struct regexp *)SvANY(rx);
+ I32 result;
RXi_GET_DECL(prog,progi);
GET_RE_DEBUG_FLAGS_DECL;
}
#endif
}
-#ifdef USE_ITHREADS
- /* It seems that non-ithreads works both with and without this code.
- So for efficiency reasons it seems best not to have the code
- compiled when it is not needed. */
- /* This is safe against NULLs: */
- ReREFCNT_dec(PM_GETRE(PL_reg_curpm));
- /* PM_reg_curpm owns a reference to this regexp. */
- (void)ReREFCNT_inc(rx);
-#endif
- PM_SETRE(PL_reg_curpm, rx);
+ SET_reg_curpm(rx);
PL_reg_oldcurpm = PL_curpm;
PL_curpm = PL_reg_curpm;
if (RXp_MATCH_COPIED(prog)) {
$` inside (?{}) could fail... */
PL_reg_oldsaved = prog->subbeg;
PL_reg_oldsavedlen = prog->sublen;
+ PL_reg_oldsavedoffset = prog->suboffset;
+ PL_reg_oldsavedcoffset = prog->suboffset;
#ifdef PERL_OLD_COPY_ON_WRITE
PL_nrs = prog->saved_copy;
#endif
else
PL_reg_oldsaved = NULL;
prog->subbeg = PL_bostr;
+ prog->suboffset = 0;
+ prog->subcoffset = 0;
prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */
}
- DEBUG_EXECUTE_r(PL_reg_starttry = *startpos);
- prog->offs[0].start = *startpos - PL_bostr;
- PL_reginput = *startpos;
+#ifdef DEBUGGING
+ PL_reg_starttry = *startposp;
+#endif
+ prog->offs[0].start = *startposp - PL_bostr;
prog->lastparen = 0;
prog->lastcloseparen = 0;
PL_regsize = 0;
- if (PL_reg_start_tmpl <= prog->nparens) {
- PL_reg_start_tmpl = prog->nparens*3/2 + 3;
- if(PL_reg_start_tmp)
- Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
- else
- Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
- }
/* XXXX What this code is doing here?!!! There should be no need
to do this again and again, prog->lastparen should take care of
#if 1
if (prog->nparens) {
regexp_paren_pair *pp = prog->offs;
- register I32 i;
+ I32 i;
for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
++pp;
pp->start = -1;
}
#endif
REGCP_SET(lastcp);
- if (regmatch(reginfo, progi->program + 1)) {
- prog->offs[0].end = PL_reginput - PL_bostr;
+ result = regmatch(reginfo, *startposp, progi->program + 1);
+ if (result != -1) {
+ prog->offs[0].end = result;
return 1;
}
if (reginfo->cutpoint)
- *startpos= reginfo->cutpoint;
+ *startposp= reginfo->cutpoint;
REGCP_UNWIND(lastcp);
return 0;
}
/* push a new state then goto it */
-#define PUSH_STATE_GOTO(state, node) \
+#define PUSH_STATE_GOTO(state, node, input) \
+ pushinput = input; \
scan = node; \
st->resume_state = state; \
goto push_state;
/* push a new state with success backtracking, then goto it */
-#define PUSH_YES_STATE_GOTO(state, node) \
+#define PUSH_YES_STATE_GOTO(state, node, input) \
+ pushinput = input; \
scan = node; \
st->resume_state = state; \
goto push_yes_state;
+
/*
regmatch() - main matching routine
// push a yes backtrack state with a resume value of
// IFMATCH_A/IFMATCH_A_fail, then continue execution at the
// first node of A:
- PUSH_YES_STATE_GOTO(IFMATCH_A, A);
+ PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
// NOTREACHED
case IFMATCH_A: // we have successfully executed A; now continue with B
want to claim it, populate any ST.foo fields in it with values you wish to
save, then do one of
- PUSH_STATE_GOTO(resume_state, node);
- PUSH_YES_STATE_GOTO(resume_state, node);
+ PUSH_STATE_GOTO(resume_state, node, newinput);
+ PUSH_YES_STATE_GOTO(resume_state, node, newinput);
which sets that backtrack state's resume value to 'resume_state', pushes a
new free entry to the top of the backtrack stack, then goes to 'node'.
}
-#define SETREX(Re1,Re2) \
- if (PL_reg_state.re_state_eval_setup_done) \
- PM_SETRE((PL_reg_curpm), (Re2)); \
- Re1 = (Re2)
-
-STATIC I32 /* 0 failure, 1 success */
-S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog)
+/* returns -1 on failure, $+[0] on success */
+STATIC I32
+S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
{
#if PERL_VERSION < 9 && !defined(PERL_CORE)
dMY_CXT;
#endif
dVAR;
- register const bool utf8_target = PL_reg_match_utf8;
+ const bool utf8_target = PL_reg_match_utf8;
const U32 uniflags = UTF8_ALLOW_DEFAULT;
REGEXP *rex_sv = reginfo->prog;
regexp *rex = (struct regexp *)SvANY(rex_sv);
RXi_GET_DECL(rex,rexi);
I32 oldsave;
/* the current state. This is a cached copy of PL_regmatch_state */
- register regmatch_state *st;
+ regmatch_state *st;
/* cache heavy used fields of st in registers */
- register regnode *scan;
- register regnode *next;
- register U32 n = 0; /* general value; init to avoid compiler warning */
- register I32 ln = 0; /* len or last; init to avoid compiler warning */
- register char *locinput = PL_reginput;
- register I32 nextchr; /* is always set to UCHARAT(locinput) */
+ regnode *scan;
+ regnode *next;
+ U32 n = 0; /* general value; init to avoid compiler warning */
+ I32 ln = 0; /* len or last; init to avoid compiler warning */
+ char *locinput = startpos;
+ char *pushinput; /* where to continue after a PUSH */
+ I32 nextchr; /* is always set to UCHARAT(locinput) */
bool result = 0; /* return value of S_regmatch */
int depth = 0; /* depth of backtrack stack */
U32 state_num;
bool no_final = 0; /* prevent failure from backtracking? */
bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
- char *startpoint = PL_reginput;
+ char *startpoint = locinput;
SV *popmark = NULL; /* are we looking for a mark? */
SV *sv_commit = NULL; /* last mark name seen in failure */
SV *sv_yes_mark = NULL; /* last mark name we have seen
I32 gimme = G_SCALAR;
CV *caller_cv = NULL; /* who called us */
CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
+ CHECKPOINT runops_cp; /* savestack position before executing EVAL */
#ifdef DEBUGGING
GET_RE_DEBUG_FLAGS_DECL;
multicall_oldcatch = 0;
multicall_cv = NULL;
cx = NULL;
+ PERL_UNUSED_VAR(multicall_cop);
+ PERL_UNUSED_VAR(newsp);
PERL_ARGS_ASSERT_REGMATCH;
reenter_switch:
+ nextchr = UCHARAT(locinput);
+ assert(nextchr >= 0);
+
switch (state_num) {
case BOL:
if (locinput == PL_bostr)
case KEEPS:
/* update the startpoint */
st->u.keeper.val = rex->offs[0].start;
- PL_reginput = locinput;
rex->offs[0].start = locinput - PL_bostr;
- PUSH_STATE_GOTO(KEEPS_next, next);
+ PUSH_STATE_GOTO(KEEPS_next, next, locinput);
/*NOT-REACHED*/
case KEEPS_next_fail:
/* rollback the start point change */
case SANY:
if (!nextchr && locinput >= PL_regeol)
sayNO;
- if (utf8_target) {
- locinput += PL_utf8skip[nextchr];
- if (locinput > PL_regeol)
- sayNO;
- nextchr = UCHARAT(locinput);
- }
- else
- nextchr = UCHARAT(++locinput);
- break;
+ goto increment_locinput;
case CANY:
if (!nextchr && locinput >= PL_regeol)
sayNO;
- nextchr = UCHARAT(++locinput);
+ locinput++;
break;
case REG_ANY:
if ((!nextchr && locinput >= PL_regeol) || nextchr == '\n')
sayNO;
- if (utf8_target) {
- locinput += PL_utf8skip[nextchr];
- if (locinput > PL_regeol)
- sayNO;
- nextchr = UCHARAT(locinput);
- }
- else
- nextchr = UCHARAT(++locinput);
- break;
+ goto increment_locinput;
+
#undef ST
#define ST st->u.trie
/* In this case the charclass data is available inline so
we can fail fast without a lot of extra overhead.
*/
- if(!ANYOF_BITMAP_TEST(scan, *locinput)) {
+ if(!ANYOF_BITMAP_TEST(scan, nextchr)) {
DEBUG_EXECUTE_r(
PerlIO_printf(Perl_debug_log,
"%*s %sfailed to match trie start class...%s\n",
REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
);
sayNO_SILENT;
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
}
/* FALL THROUGH */
case TRIE:
HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
U32 state = trie->startstate;
- if (trie->bitmap && !TRIE_BITMAP_TEST(trie,*locinput) ) {
+ if (trie->bitmap && !TRIE_BITMAP_TEST(trie, nextchr) ) {
if (trie->states[ state ].wordnum) {
DEBUG_EXECUTE_r(
PerlIO_printf(Perl_debug_log,
U32 charcount = 0; /* how many input chars we have matched */
U32 accepted = 0; /* have we seen any accepting states? */
- ST.B = next;
ST.jump = trie->jump;
ST.me = scan;
ST.firstpos = NULL;
);
goto trie_first_try; /* jump into the fail handler */
}}
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
case TRIE_next_fail: /* we failed - try next alternative */
+ {
+ U8 *uc;
if ( ST.jump) {
REGCP_UNWIND(ST.cp);
- for (n = rex->lastparen; n > ST.lastparen; n--)
- rex->offs[n].end = -1;
- rex->lastparen = n;
+ UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
}
if (!--ST.accepted) {
DEBUG_EXECUTE_r({
{
/* Find next-highest word to process. Note that this code
* is O(N^2) per trie run (O(N) per branch), so keep tight */
- register U16 min = 0;
- register U16 word;
- register U16 const nextword = ST.nextword;
- register reg_trie_wordinfo * const wordinfo
+ U16 min = 0;
+ U16 word;
+ U16 const nextword = ST.nextword;
+ reg_trie_wordinfo * const wordinfo
= ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
for (word=ST.topword; word; word=wordinfo[word].prev) {
if (word > nextword && (!min || word < min))
if ( ST.jump) {
ST.lastparen = rex->lastparen;
+ ST.lastcloseparen = rex->lastcloseparen;
REGCP_SET(ST.cp);
}
/* find start char of end of current word */
{
U32 chars; /* how many chars to skip */
- U8 *uc = ST.firstpos;
reg_trie_data * const trie
= (reg_trie_data*)rexi->data->data[ARG(ST.me)];
>= ST.firstchars);
chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
- ST.firstchars;
+ uc = ST.firstpos;
if (ST.longfold) {
/* the hard option - fold each char in turn and find
else
uc += chars;
}
- PL_reginput = (char *)uc;
}
- scan = (ST.jump && ST.jump[ST.nextword])
- ? ST.me + ST.jump[ST.nextword]
- : ST.B;
+ scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
+ ? ST.jump[ST.nextword]
+ : NEXT_OFF(ST.me));
DEBUG_EXECUTE_r({
PerlIO_printf( Perl_debug_log,
});
if (ST.accepted > 1 || has_cutgroup) {
- PUSH_STATE_GOTO(TRIE_next, scan);
- /* NOTREACHED */
+ PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
+ assert(0); /* NOTREACHED */
}
/* only one choice left - just continue */
DEBUG_EXECUTE_r({
PL_colors[5] );
});
- locinput = PL_reginput;
- nextchr = UCHARAT(locinput);
+ locinput = (char*)uc;
continue; /* execute rest of RE */
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
+ }
#undef ST
case EXACT: {
const char * const e = s + ln;
if (utf8_target) {
- /* The target is utf8, the pattern is not utf8. */
+ /* The target is utf8, the pattern is not utf8.
+ * Above-Latin1 code points can't match the pattern;
+ * invariants match exactly, and the other Latin1 ones need
+ * to be downgraded to a single byte in order to do the
+ * comparison. (If we could be confident that the target
+ * is not malformed, this could be refactored to have fewer
+ * tests by just assuming that if the first bytes match, it
+ * is an invariant, but there are tests in the test suite
+ * dealing with (??{...}) which violate this) */
while (s < e) {
- STRLEN ulen;
if (l >= PL_regeol)
sayNO;
- if (NATIVE_TO_UNI(*(U8*)s) !=
- utf8n_to_uvuni((U8*)l, UTF8_MAXBYTES, &ulen,
- uniflags))
- sayNO;
- l += ulen;
- s ++;
+ if (UTF8_IS_ABOVE_LATIN1(* (U8*) l)) {
+ sayNO;
+ }
+ if (UTF8_IS_INVARIANT(*(U8*)l)) {
+ if (*l != *s) {
+ sayNO;
+ }
+ l++;
+ }
+ else {
+ if (TWO_BYTE_UTF8_TO_UNI(*l, *(l+1)) != * (U8*) s) {
+ sayNO;
+ }
+ l += 2;
+ }
+ s++;
}
}
else {
/* The target is not utf8, the pattern is utf8. */
while (s < e) {
- STRLEN ulen;
- if (l >= PL_regeol)
- sayNO;
- if (NATIVE_TO_UNI(*((U8*)l)) !=
- utf8n_to_uvuni((U8*)s, UTF8_MAXBYTES, &ulen,
- uniflags))
- sayNO;
- s += ulen;
- l ++;
+ if (l >= PL_regeol || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
+ {
+ sayNO;
+ }
+ if (UTF8_IS_INVARIANT(*(U8*)s)) {
+ if (*s != *l) {
+ sayNO;
+ }
+ s++;
+ }
+ else {
+ if (TWO_BYTE_UTF8_TO_UNI(*s, *(s+1)) != * (U8*) l) {
+ sayNO;
+ }
+ s += 2;
+ }
+ l++;
}
}
locinput = l;
- nextchr = UCHARAT(locinput);
break;
}
/* The target and the pattern have the same utf8ness. */
if (ln > 1 && memNE(s, locinput, ln))
sayNO;
locinput += ln;
- nextchr = UCHARAT(locinput);
break;
}
case EXACTFL: {
sayNO;
}
locinput = e;
- nextchr = UCHARAT(locinput);
break;
}
if (ln > 1 && ! folder(s, locinput, ln))
sayNO;
locinput += ln;
- nextchr = UCHARAT(locinput);
break;
}
if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, utf8_target))
sayNO;
locinput += inclasslen;
- nextchr = UCHARAT(locinput);
break;
}
else {
- if (nextchr < 0)
- nextchr = UCHARAT(locinput);
if (!nextchr && locinput >= PL_regeol)
sayNO;
if (!REGINCLASS(rex, scan, (U8*)locinput))
sayNO;
- nextchr = UCHARAT(++locinput);
+ locinput++;
break;
}
break;
DIGITA, NDIGITA, isDIGIT_A,
digit, "0");
+ case POSIXA:
+ if (locinput >= PL_regeol || ! _generic_isCC_A(nextchr, FLAGS(scan))) {
+ sayNO;
+ }
+ /* Matched a utf8-invariant, so don't have to worry about utf8 */
+ locinput++;
+ break;
+ case NPOSIXA:
+ if (locinput >= PL_regeol || _generic_isCC_A(nextchr, FLAGS(scan))) {
+ sayNO;
+ }
+ goto increment_locinput;
+
case CLUMP: /* Match \X: logical Unicode character. This is defined as
a Unicode extended Grapheme Cluster */
/* From http://www.unicode.org/reports/tr29 (5.2 version). An
| Prepend* Begin Extend*
| .
- Begin is (Hangul-syllable | ! Control)
- Extend is (Grapheme_Extend | Spacing_Mark)
- Control is [ GCB_Control CR LF ]
-
- The discussion below shows how the code for CLUMP is derived
- from this regex. Note that most of these concepts are from
- property values of the Grapheme Cluster Boundary (GCB) property.
- No code point can have multiple property values for a given
- property. Thus a code point in Prepend can't be in Control, but
- it must be in !Control. This is why Control above includes
- GCB_Control plus CR plus LF. The latter two are used in the GCB
- property separately, and so can't be in GCB_Control, even though
- they logically are controls. Control is not the same as gc=cc,
- but includes format and other characters as well.
-
- The Unicode definition of Hangul-syllable is:
- L+
- | (L* ( ( V | LV ) V* | LVT ) T*)
- | T+
- )
- Each of these is a value for the GCB property, and hence must be
- disjoint, so the order they are tested is immaterial, so the
- above can safely be changed to
- T+
- | L+
- | (L* ( LVT | ( V | LV ) V*) T*)
-
- The last two terms can be combined like this:
- L* ( L
- | (( LVT | ( V | LV ) V*) T*))
-
- And refactored into this:
- L* (L | LVT T* | V V* T* | LV V* T*)
-
- That means that if we have seen any L's at all we can quit
- there, but if the next character is an LVT, a V, or an LV we
- should keep going.
+ Begin is: ( Special_Begin | ! Control )
+ Special_Begin is: ( Regional-Indicator+ | Hangul-syllable )
+ Extend is: ( Grapheme_Extend | Spacing_Mark )
+ Control is: [ GCB_Control CR LF ]
+ Hangul-syllable is: ( T+ | ( L* ( L | ( LVT | ( V | LV ) V* ) T* ) ))
+
+ If we create a 'Regular_Begin' = Begin - Special_Begin, then
+ we can rewrite
+
+ Begin is ( Regular_Begin + Special Begin )
+
+ It turns out that 98.4% of all Unicode code points match
+ Regular_Begin. Doing it this way eliminates a table match in
+ the previous implementation for almost all Unicode code points.
There is a subtlety with Prepend* which showed up in testing.
Note that the Begin, and only the Begin is required in:
locinput += 2;
}
else {
+ STRLEN len;
+
/* In case have to backtrack to beginning, then match '.' */
char *starting = locinput;
LOAD_UTF8_CHARCLASS_GCB();
- /* Match (prepend)* */
- while (locinput < PL_regeol
- && swash_fetch(PL_utf8_X_prepend,
- (U8*)locinput, utf8_target))
- {
- previous_prepend = locinput;
- locinput += UTF8SKIP(locinput);
- }
+ /* Match (prepend)* */
+ while (locinput < PL_regeol
+ && (len = is_GCB_Prepend_utf8(locinput)))
+ {
+ previous_prepend = locinput;
+ locinput += len;
+ }
/* As noted above, if we matched a prepend character, but
* the next thing won't match, back off the last prepend we
* matched, as it is guaranteed to match the begin */
if (previous_prepend
&& (locinput >= PL_regeol
- || ! swash_fetch(PL_utf8_X_begin,
- (U8*)locinput, utf8_target)))
+ || (! swash_fetch(PL_utf8_X_regular_begin,
+ (U8*)locinput, utf8_target)
+ && ! is_GCB_SPECIAL_BEGIN_utf8(locinput)))
+ )
{
locinput = previous_prepend;
}
* moved locinput forward, we tested the result just above
* and it either passed, or we backed off so that it will
* now pass */
- if (! swash_fetch(PL_utf8_X_begin, (U8*)locinput, utf8_target)) {
+ if (swash_fetch(PL_utf8_X_regular_begin,
+ (U8*)locinput, utf8_target)) {
+ locinput += UTF8SKIP(locinput);
+ }
+ else if (! is_GCB_SPECIAL_BEGIN_utf8(locinput)) {
/* Here did not match the required 'Begin' in the
* second term. So just match the very first
* character, the '.' of the final term of the regex */
locinput = starting + UTF8SKIP(starting);
+ goto exit_utf8;
} else {
- /* Here is the beginning of a character that can have
- * an extender. It is either a hangul syllable, or a
- * non-control */
- if (swash_fetch(PL_utf8_X_non_hangul,
- (U8*)locinput, utf8_target))
- {
-
- /* Here not a Hangul syllable, must be a
- * ('! * Control') */
- locinput += UTF8SKIP(locinput);
- } else {
-
- /* Here is a Hangul syllable. It can be composed
- * of several individual characters. One
- * possibility is T+ */
- if (swash_fetch(PL_utf8_X_T,
- (U8*)locinput, utf8_target))
- {
- while (locinput < PL_regeol
- && swash_fetch(PL_utf8_X_T,
- (U8*)locinput, utf8_target))
- {
- locinput += UTF8SKIP(locinput);
- }
- } else {
-
- /* Here, not T+, but is a Hangul. That means
- * it is one of the others: L, LV, LVT or V,
- * and matches:
- * L* (L | LVT T* | V V* T* | LV V* T*) */
-
- /* Match L* */
- while (locinput < PL_regeol
- && swash_fetch(PL_utf8_X_L,
- (U8*)locinput, utf8_target))
- {
- locinput += UTF8SKIP(locinput);
- }
+ /* Here is a special begin. It can be composed of
+ * several individual characters. One possibility is
+ * RI+ */
+ if ((len = is_GCB_RI_utf8(locinput))) {
+ locinput += len;
+ while (locinput < PL_regeol
+ && (len = is_GCB_RI_utf8(locinput)))
+ {
+ locinput += len;
+ }
+ } else if ((len = is_GCB_T_utf8(locinput))) {
+ /* Another possibility is T+ */
+ locinput += len;
+ while (locinput < PL_regeol
+ && (len = is_GCB_T_utf8(locinput)))
+ {
+ locinput += len;
+ }
+ } else {
+
+ /* Here, neither RI+ nor T+; must be some other
+ * Hangul. That means it is one of the others: L,
+ * LV, LVT or V, and matches:
+ * L* (L | LVT T* | V * V* T* | LV V* T*) */
+
+ /* Match L* */
+ while (locinput < PL_regeol
+ && (len = is_GCB_L_utf8(locinput)))
+ {
+ locinput += len;
+ }
- /* Here, have exhausted L*. If the next
- * character is not an LV, LVT nor V, it means
- * we had to have at least one L, so matches L+
- * in the original equation, we have a complete
- * hangul syllable. Are done. */
-
- if (locinput < PL_regeol
- && swash_fetch(PL_utf8_X_LV_LVT_V,
- (U8*)locinput, utf8_target))
- {
-
- /* Otherwise keep going. Must be LV, LVT
- * or V. See if LVT */
- if (swash_fetch(PL_utf8_X_LVT,
- (U8*)locinput, utf8_target))
- {
- locinput += UTF8SKIP(locinput);
- } else {
-
- /* Must be V or LV. Take it, then
- * match V* */
- locinput += UTF8SKIP(locinput);
- while (locinput < PL_regeol
- && swash_fetch(PL_utf8_X_V,
- (U8*)locinput, utf8_target))
- {
- locinput += UTF8SKIP(locinput);
- }
- }
+ /* Here, have exhausted L*. If the next character
+ * is not an LV, LVT nor V, it means we had to have
+ * at least one L, so matches L+ in the original
+ * equation, we have a complete hangul syllable.
+ * Are done. */
+
+ if (locinput < PL_regeol
+ && is_GCB_LV_LVT_V_utf8(locinput))
+ {
+
+ /* Otherwise keep going. Must be LV, LVT or V.
+ * See if LVT */
+ if (is_utf8_X_LVT((U8*)locinput)) {
+ locinput += UTF8SKIP(locinput);
+ } else {
+
+ /* Must be V or LV. Take it, then match
+ * V* */
+ locinput += UTF8SKIP(locinput);
+ while (locinput < PL_regeol
+ && (len = is_GCB_V_utf8(locinput)))
+ {
+ locinput += len;
+ }
+ }
- /* And any of LV, LVT, or V can be followed
- * by T* */
- while (locinput < PL_regeol
- && swash_fetch(PL_utf8_X_T,
- (U8*)locinput,
- utf8_target))
- {
- locinput += UTF8SKIP(locinput);
- }
- }
- }
- }
+ /* And any of LV, LVT, or V can be followed
+ * by T* */
+ while (locinput < PL_regeol
+ && (len = is_GCB_T_utf8(locinput)))
+ {
+ locinput += len;
+ }
+ }
+ }
+ }
- /* Match any extender */
- while (locinput < PL_regeol
- && swash_fetch(PL_utf8_X_extend,
- (U8*)locinput, utf8_target))
- {
- locinput += UTF8SKIP(locinput);
- }
- }
+ /* Match any extender */
+ while (locinput < PL_regeol
+ && swash_fetch(PL_utf8_X_extend,
+ (U8*)locinput, utf8_target))
+ {
+ locinput += UTF8SKIP(locinput);
+ }
}
+ exit_utf8:
if (locinput > PL_regeol) sayNO;
}
- nextchr = UCHARAT(locinput);
break;
case NREFFL:
sayNO;
}
locinput = limit;
- nextchr = UCHARAT(locinput);
break;
}
: ! folder(s, locinput, ln)))
sayNO;
locinput += ln;
- nextchr = UCHARAT(locinput);
break;
}
case NOTHING:
re_sv = rex_sv;
re = rex;
rei = rexi;
- (void)ReREFCNT_inc(rex_sv);
if (OP(scan)==GOSUB) {
startpoint = scan + ARG2L(scan);
ST.close_paren = ARG(scan);
ST.close_paren = 0;
}
goto eval_recurse_doit;
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
if (cur_eval && cur_eval->locinput==locinput) {
if ( ++nochange_depth > max_nochange_depth )
struct re_save_state saved_state;
CV *newcv;
+ /* save *all* paren positions */
+ regcppush(rex, 0);
+ REGCP_SET(runops_cp);
+
/* To not corrupt the existing regex state while executing the
* eval we would normally put it on the save stack, like with
* save_re_context. However, re-evals have a weird scoping so we
nop = (OP*)rexi->data->data[n];
}
+ /* normally if we're about to execute code from the same
+ * CV that we used previously, we just use the existing
+ * CX stack entry. However, its possible that in the
+ * meantime we may have backtracked, popped from the save
+ * stack, and undone the SAVECOMPPAD(s) associated with
+ * PUSH_MULTICALL; in which case PL_comppad no longer
+ * points to newcv's pad. */
+ if (newcv != last_pushed_cv || PL_comppad != last_pad)
+ {
+ I32 depth = (newcv == caller_cv) ? 0 : 1;
+ if (last_pushed_cv) {
+ CHANGE_MULTICALL_WITHDEPTH(newcv, depth);
+ }
+ else {
+ PUSH_MULTICALL_WITHDEPTH(newcv, depth);
+ }
+ last_pushed_cv = newcv;
+ }
+ last_pad = PL_comppad;
+
/* the initial nextstate you would normally execute
* at the start of an eval (which would cause error
* messages to come from the eval), may be optimised
DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
" re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) );
- /* normally if we're about to execute code from the same
- * CV that we used previously, we just use the existing
- * CX stack entry. However, its possible that in the
- * meantime we may have backtracked, popped from the save
- * stack, and undone the SAVECOMPPAD(s) associated with
- * PUSH_MULTICALL; in which case PL_comppad no longer
- * points to newcv's pad. */
- if (newcv != last_pushed_cv || PL_comppad != last_pad)
- {
- I32 depth = (newcv == caller_cv) ? 0 : 1;
- if (last_pushed_cv) {
- CHANGE_MULTICALL_WITHDEPTH(newcv, depth);
- }
- else {
- PUSH_MULTICALL_WITHDEPTH(newcv, depth);
- }
- last_pushed_cv = newcv;
- }
- last_pad = PL_comppad;
-
rex->offs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr;
if (sv_yes_mark) {
PUTBACK;
}
+ /* before restoring everything, evaluate the returned
+ * value, so that 'uninit' warnings don't use the wrong
+ * PL_op or pad. Also need to process any magic vars
+ * (e.g. $1) *before* parentheses are restored */
+
+ PL_op = NULL;
+
+ re_sv = NULL;
+ if (logical == 0) /* (?{})/ */
+ sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
+ else if (logical == 1) { /* /(?(?{...})X|Y)/ */
+ sw = cBOOL(SvTRUE(ret));
+ logical = 0;
+ }
+ else { /* /(??{}) */
+ /* if its overloaded, let the regex compiler handle
+ * it; otherwise extract regex, or stringify */
+ if (!SvAMAGIC(ret)) {
+ SV *sv = ret;
+ if (SvROK(sv))
+ sv = SvRV(sv);
+ if (SvTYPE(sv) == SVt_REGEXP)
+ re_sv = (REGEXP*) sv;
+ else if (SvSMAGICAL(sv)) {
+ MAGIC *mg = mg_find(sv, PERL_MAGIC_qr);
+ if (mg)
+ re_sv = (REGEXP *) mg->mg_obj;
+ }
+
+ /* force any magic, undef warnings here */
+ if (!re_sv) {
+ ret = sv_mortalcopy(ret);
+ (void) SvPV_force_nolen(ret);
+ }
+ }
+
+ }
+
Copy(&saved_state, &PL_reg_state, 1, struct re_save_state);
/* *** Note that at this point we don't restore
PL_op = oop;
PL_curcop = ocurcop;
PL_regeol = saved_regeol;
- if (!logical) {
- /* /(?{...})/ */
- sv_setsv(save_scalar(PL_replgv), ret);
+ S_regcp_restore(aTHX_ rex, runops_cp);
+
+ if (logical != 2)
break;
- }
}
- if (logical == 2) { /* Postponed subexpression: /(??{...})/ */
+
+ /* only /(??{})/ from now on */
logical = 0;
{
/* extract RE object from returned value; compiling if
* necessary */
- MAGIC *mg = NULL;
- REGEXP *rx = NULL;
-
- if (SvROK(ret)) {
- SV *const sv = SvRV(ret);
-
- if (SvTYPE(sv) == SVt_REGEXP) {
- rx = (REGEXP*) sv;
- } else if (SvSMAGICAL(sv)) {
- mg = mg_find(sv, PERL_MAGIC_qr);
- assert(mg);
- }
- } else if (SvTYPE(ret) == SVt_REGEXP) {
- rx = (REGEXP*) ret;
- } else if (SvSMAGICAL(ret)) {
- if (SvGMAGICAL(ret)) {
- /* I don't believe that there is ever qr magic
- here. */
- assert(!mg_find(ret, PERL_MAGIC_qr));
- sv_unmagic(ret, PERL_MAGIC_qr);
- }
- else {
- mg = mg_find(ret, PERL_MAGIC_qr);
- /* testing suggests mg only ends up non-NULL for
- scalars who were upgraded and compiled in the
- else block below. In turn, this is only
- triggered in the "postponed utf8 string" tests
- in t/op/pat.t */
- }
- }
- if (mg) {
- rx = (REGEXP *) mg->mg_obj; /*XXX:dmq*/
- assert(rx);
- }
- if (rx) {
- rx = reg_temp_copy(NULL, rx);
+ if (re_sv) {
+ re_sv = reg_temp_copy(NULL, re_sv);
}
else {
U32 pm_flags = 0;
const I32 osize = PL_regsize;
- if (DO_UTF8(ret)) {
- assert (SvUTF8(ret));
- } else if (SvUTF8(ret)) {
- /* Not doing UTF-8, despite what the SV says. Is
- this only if we're trapped in use 'bytes'? */
- /* Make a copy of the octet sequence, but without
- the flag on, as the compiler now honours the
- SvUTF8 flag on ret. */
+ if (SvUTF8(ret) && IN_BYTES) {
+ /* In use 'bytes': make a copy of the octet
+ * sequence, but without the flag on */
STRLEN len;
const char *const p = SvPV(ret, len);
ret = newSVpvn_flags(p, len, SVs_TEMP);
}
- rx = CALLREGCOMP(ret, pm_flags);
+ if (rex->intflags & PREGf_USE_RE_EVAL)
+ pm_flags |= PMf_USE_RE_EVAL;
+
+ /* if we got here, it should be an engine which
+ * supports compiling code blocks and stuff */
+ assert(rex->engine && rex->engine->op_comp);
+ assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
+ re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
+ rex->engine, NULL, NULL,
+ /* copy /msix etc to inner pattern */
+ scan->flags,
+ pm_flags);
+
if (!(SvFLAGS(ret)
& (SVs_TEMP | SVs_PADTMP | SVf_READONLY
| SVs_GMG))) {
/* This isn't a first class regexp. Instead, it's
caching a regexp onto an existing, Perl visible
scalar. */
- sv_magic(ret, MUTABLE_SV(rx), PERL_MAGIC_qr, 0, 0);
+ sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
}
PL_regsize = osize;
+ /* safe to do now that any $1 etc has been
+ * interpolated into the new pattern string and
+ * compiled */
+ S_regcp_restore(aTHX_ rex, runops_cp);
}
- re_sv = rx;
- re = (struct regexp *)SvANY(rx);
+ re = (struct regexp *)SvANY(re_sv);
}
RXp_MATCH_COPIED_off(re);
re->subbeg = rex->subbeg;
re->sublen = rex->sublen;
+ re->suboffset = rex->suboffset;
+ re->subcoffset = rex->subcoffset;
rei = RXi_GET(re);
DEBUG_EXECUTE_r(
debug_start_match(re_sv, utf8_target, locinput, PL_regeol,
);
startpoint = rei->program + 1;
ST.close_paren = 0; /* only used for GOSUB */
- /* borrowed from regtry */
- if (PL_reg_start_tmpl <= re->nparens) {
- PL_reg_start_tmpl = re->nparens*3/2 + 3;
- if(PL_reg_start_tmp)
- Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
- else
- Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*);
- }
eval_recurse_doit: /* Share code with GOSUB below this line */
/* run the pattern returned from (??{...}) */
ST.cp = regcppush(rex, 0); /* Save *all* the positions. */
REGCP_SET(ST.lastcp);
- /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */
re->lastparen = 0;
re->lastcloseparen = 0;
- PL_reginput = locinput;
PL_regsize = 0;
/* XXXX This is too dramatic a measure... */
ST.prev_rex = rex_sv;
ST.prev_curlyx = cur_curlyx;
- SETREX(rex_sv,re_sv);
+ rex_sv = re_sv;
+ SET_reg_curpm(rex_sv);
rex = re;
rexi = rei;
cur_curlyx = NULL;
ST.prev_eval = cur_eval;
cur_eval = st;
/* now continue from first node in postoned RE */
- PUSH_YES_STATE_GOTO(EVAL_AB, startpoint);
- /* NOTREACHED */
- }
- /* logical is 1, /(?(?{...})X|Y)/ */
- sw = cBOOL(SvTRUE(ret));
- logical = 0;
- break;
+ PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput);
+ assert(0); /* NOTREACHED */
}
case EVAL_AB: /* cleanup after a successful (??{A})B */
/* note: this is called twice; first after popping B, then A */
PL_reg_flags ^= ST.toggle_reg_flags;
- ReREFCNT_dec(rex_sv);
- SETREX(rex_sv,ST.prev_rex);
+ rex_sv = ST.prev_rex;
+ SET_reg_curpm(rex_sv);
rex = (struct regexp *)SvANY(rex_sv);
rexi = RXi_GET(rex);
regcpblow(ST.cp);
case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
/* note: this is called twice; first after popping B, then A */
PL_reg_flags ^= ST.toggle_reg_flags;
- ReREFCNT_dec(rex_sv);
- SETREX(rex_sv,ST.prev_rex);
+ rex_sv = ST.prev_rex;
+ SET_reg_curpm(rex_sv);
rex = (struct regexp *)SvANY(rex_sv);
rexi = RXi_GET(rex);
- PL_reginput = locinput;
REGCP_UNWIND(ST.lastcp);
regcppop(rex);
cur_eval = ST.prev_eval;
case OPEN:
n = ARG(scan); /* which paren pair */
- PL_reg_start_tmp[n] = locinput;
+ rex->offs[n].start_tmp = locinput - PL_bostr;
if (n > PL_regsize)
PL_regsize = n;
+ DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
+ "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; regsize=%"UVuf"\n",
+ PTR2UV(rex),
+ PTR2UV(rex->offs),
+ (UV)n,
+ (IV)rex->offs[n].start_tmp,
+ (UV)PL_regsize
+ ));
lastopen = n;
break;
+
+/* XXX really need to log other places start/end are set too */
+#define CLOSE_CAPTURE \
+ rex->offs[n].start = rex->offs[n].start_tmp; \
+ rex->offs[n].end = locinput - PL_bostr; \
+ DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \
+ "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \
+ PTR2UV(rex), \
+ PTR2UV(rex->offs), \
+ (UV)n, \
+ (IV)rex->offs[n].start, \
+ (IV)rex->offs[n].end \
+ ))
+
case CLOSE:
n = ARG(scan); /* which paren pair */
- rex->offs[n].start = PL_reg_start_tmp[n] - PL_bostr;
- rex->offs[n].end = locinput - PL_bostr;
+ CLOSE_CAPTURE;
/*if (n > PL_regsize)
PL_regsize = n;*/
if (n > rex->lastparen)
if ( OP(cursor)==CLOSE ){
n = ARG(cursor);
if ( n <= lastopen ) {
- rex->offs[n].start
- = PL_reg_start_tmp[n] - PL_bostr;
- rex->offs[n].end = locinput - PL_bostr;
+ CLOSE_CAPTURE;
/*if (n > PL_regsize)
PL_regsize = n;*/
if (n > rex->lastparen)
ST.count = -1; /* this will be updated by WHILEM */
ST.lastloc = NULL; /* this will be updated by WHILEM */
- PL_reginput = locinput;
- PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next));
- /* NOTREACHED */
+ PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput);
+ assert(0); /* NOTREACHED */
}
case CURLYX_end: /* just finished matching all of A*B */
cur_curlyx = ST.prev_curlyx;
sayYES;
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
case CURLYX_end_fail: /* just failed to match all of A*B */
regcpblow(ST.cp);
cur_curlyx = ST.prev_curlyx;
sayNO;
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
#undef ST
ST.cache_offset = 0;
ST.cache_mask = 0;
- PL_reginput = locinput;
DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
"%*s whilem: matched %ld out of %d..%d\n",
cur_curlyx->u.curlyx.lastloc = locinput;
REGCP_SET(ST.lastcp);
- PUSH_STATE_GOTO(WHILEM_A_pre, A);
- /* NOTREACHED */
+ PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput);
+ assert(0); /* NOTREACHED */
}
/* If degenerate A matches "", assume A done. */
cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor);
REGCP_SET(ST.lastcp);
- PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B);
- /* NOTREACHED */
+ PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
+ locinput);
+ assert(0); /* NOTREACHED */
}
/* Prefer A over B for maximal matching. */
ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor);
cur_curlyx->u.curlyx.lastloc = locinput;
REGCP_SET(ST.lastcp);
- PUSH_STATE_GOTO(WHILEM_A_max, A);
- /* NOTREACHED */
+ PUSH_STATE_GOTO(WHILEM_A_max, A, locinput);
+ assert(0); /* NOTREACHED */
}
goto do_whilem_B_max;
}
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
case WHILEM_B_min: /* just matched B in a minimal match */
case WHILEM_B_max: /* just matched B in a maximal match */
cur_curlyx = ST.save_curlyx;
sayYES;
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
cur_curlyx = ST.save_curlyx;
cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
cur_curlyx->u.curlyx.count--;
CACHEsayNO;
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
/* FALL THROUGH */
cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
cur_curlyx->u.curlyx.count--;
CACHEsayNO;
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
REGCP_UNWIND(ST.lastcp);
regcppop(rex); /* Restore some previous $<digit>s? */
- PL_reginput = locinput;
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
"%*s whilem: failed, trying continuation...\n",
REPORT_CODE_OFF+depth*2, "")
/* now try B */
ST.save_curlyx = cur_curlyx;
cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
- PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B);
- /* NOTREACHED */
+ PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
+ locinput);
+ assert(0); /* NOTREACHED */
case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
cur_curlyx = ST.save_curlyx;
"%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
);
/* Try grabbing another A and see if it helps. */
- PL_reginput = locinput;
cur_curlyx->u.curlyx.lastloc = locinput;
ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor);
REGCP_SET(ST.lastcp);
PUSH_STATE_GOTO(WHILEM_A_min,
- /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS);
- /* NOTREACHED */
+ /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
+ locinput);
+ assert(0); /* NOTREACHED */
#undef ST
#define ST st->u.branch
case BRANCH: /* /(...|A|...)/ */
scan = NEXTOPER(scan); /* scan now points to inner node */
ST.lastparen = rex->lastparen;
+ ST.lastcloseparen = rex->lastcloseparen;
ST.next_branch = next;
REGCP_SET(ST.cp);
- PL_reginput = locinput;
/* Now go into the branch */
if (has_cutgroup) {
- PUSH_YES_STATE_GOTO(BRANCH_next, scan);
+ PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput);
} else {
- PUSH_STATE_GOTO(BRANCH_next, scan);
+ PUSH_STATE_GOTO(BRANCH_next, scan, locinput);
}
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
case CUTGROUP:
- PL_reginput = locinput;
sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
- PUSH_STATE_GOTO(CUTGROUP_next,next);
- /* NOTREACHED */
+ PUSH_STATE_GOTO(CUTGROUP_next, next, locinput);
+ assert(0); /* NOTREACHED */
case CUTGROUP_next_fail:
do_cutgroup = 1;
no_final = 1;
if (st->u.mark.mark_name)
sv_commit = st->u.mark.mark_name;
sayNO;
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
case BRANCH_next:
sayYES;
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
case BRANCH_next_fail: /* that branch failed; try the next, if any */
if (do_cutgroup) {
do_cutgroup = 0;
no_final = 0;
}
REGCP_UNWIND(ST.cp);
- for (n = rex->lastparen; n > ST.lastparen; n--)
- rex->offs[n].end = -1;
- rex->lastparen = n;
- /*dmq: rex->lastcloseparen = n; */
+ UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
scan = ST.next_branch;
/* no more branches? */
if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
sayNO_SILENT;
}
continue; /* execute next BRANCH[J] op */
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
case MINMOD:
minmod = 1;
ST.me = scan;
scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
+ ST.lastparen = rex->lastparen;
+ ST.lastcloseparen = rex->lastcloseparen;
+
/* if paren positive, emulate an OPEN/CLOSE around A */
if (ST.me->flags) {
U32 paren = ST.me->flags;
if (paren > PL_regsize)
PL_regsize = paren;
- if (paren > rex->lastparen)
- rex->lastparen = paren;
scan += NEXT_OFF(scan); /* Skip former OPEN. */
}
ST.A = scan;
goto curlym_do_B;
curlym_do_A: /* execute the A in /A{m,n}B/ */
- PL_reginput = locinput;
- PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */
- /* NOTREACHED */
+ PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */
+ assert(0); /* NOTREACHED */
case CURLYM_A: /* we've just matched an A */
- locinput = st->locinput;
- nextchr = UCHARAT(locinput);
-
ST.count++;
/* after first match, determine A's length: u.curlym.alen */
if (ST.count == 1) {
if (PL_reg_match_utf8) {
- char *s = locinput;
- while (s < PL_reginput) {
+ char *s = st->locinput;
+ while (s < locinput) {
ST.alen++;
s += UTF8SKIP(s);
}
}
else {
- ST.alen = PL_reginput - locinput;
+ ST.alen = locinput - st->locinput;
}
if (ST.alen == 0)
ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
(IV) ST.count, (IV)ST.alen)
);
- locinput = PL_reginput;
-
if (cur_eval && cur_eval->u.eval.close_paren &&
cur_eval->u.eval.close_paren == (U32)ST.me->flags)
goto fake_end;
sayNO;
curlym_do_B: /* execute the B in /A{m,n}B/ */
- PL_reginput = locinput;
if (ST.c1 == CHRTEST_UNINIT) {
/* calculate c1 and c2 for possible match of 1st char
* following curly */
"", (IV)ST.count)
);
if (ST.c1 != CHRTEST_VOID
- && UCHARAT(PL_reginput) != ST.c1
- && UCHARAT(PL_reginput) != ST.c2)
+ && nextchr != ST.c1
+ && nextchr != ST.c2)
{
/* simulate B failing */
DEBUG_OPTIMISE_r(
}
if (ST.me->flags) {
- /* mark current A as captured */
+ /* emulate CLOSE: mark current A as captured */
I32 paren = ST.me->flags;
if (ST.count) {
rex->offs[paren].start
- = HOPc(PL_reginput, -ST.alen) - PL_bostr;
- rex->offs[paren].end = PL_reginput - PL_bostr;
- /*dmq: rex->lastcloseparen = paren; */
+ = HOPc(locinput, -ST.alen) - PL_bostr;
+ rex->offs[paren].end = locinput - PL_bostr;
+ if ((U32)paren > rex->lastparen)
+ rex->lastparen = paren;
+ rex->lastcloseparen = paren;
}
else
rex->offs[paren].end = -1;
}
}
- PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */
- /* NOTREACHED */
+ PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */
+ assert(0); /* NOTREACHED */
case CURLYM_B_fail: /* just failed to match a B */
REGCP_UNWIND(ST.cp);
+ UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
if (ST.minmod) {
I32 max = ARG2(ST.me);
if (max != REG_INFTY && ST.count == max)
sayNO;
ST.count--;
locinput = HOPc(locinput, -ST.alen);
+ nextchr = UCHARAT(locinput);
goto curlym_do_B; /* try to match B */
#undef ST
if (success) { \
rex->offs[paren].start = HOPc(locinput, -1) - PL_bostr; \
rex->offs[paren].end = locinput - PL_bostr; \
+ if (paren > rex->lastparen) \
+ rex->lastparen = paren; \
rex->lastcloseparen = paren; \
} \
- else \
+ else { \
rex->offs[paren].end = -1; \
+ rex->lastparen = ST.lastparen; \
+ rex->lastcloseparen = ST.lastcloseparen; \
+ } \
}
case STAR: /* /A*B/ where A is width 1 */
goto repeat;
case CURLYN: /* /(A){m,n}B/ where A is width 1 */
ST.paren = scan->flags; /* Which paren to set */
+ ST.lastparen = rex->lastparen;
+ ST.lastcloseparen = rex->lastcloseparen;
if (ST.paren > PL_regsize)
PL_regsize = ST.paren;
- if (ST.paren > rex->lastparen)
- rex->lastparen = ST.paren;
ST.min = ARG1(scan); /* min to match */
ST.max = ARG2(scan); /* max to match */
if (cur_eval && cur_eval->u.eval.close_paren &&
ST.A = scan;
ST.B = next;
- PL_reginput = locinput;
if (minmod) {
+ char *li = locinput;
minmod = 0;
- if (ST.min && regrepeat(rex, ST.A, ST.min, depth) < ST.min)
+ if (ST.min && regrepeat(rex, &li, ST.A, ST.min, depth) < ST.min)
sayNO;
+ locinput = li;
+ nextchr = UCHARAT(locinput);
ST.count = ST.min;
- locinput = PL_reginput;
REGCP_SET(ST.cp);
if (ST.c1 == CHRTEST_VOID)
goto curly_try_B_min;
}
else {
- ST.count = regrepeat(rex, ST.A, ST.max, depth);
- locinput = PL_reginput;
+ /* avoid taking address of locinput, so it can remain
+ * a register var */
+ char *li = locinput;
+ ST.count = regrepeat(rex, &li, ST.A, ST.max, depth);
if (ST.count < ST.min)
sayNO;
+ locinput = li;
+ nextchr = UCHARAT(locinput);
if ((ST.count > ST.min)
&& (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
{
/* ...except that $ and \Z can match before *and* after
newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
We may back off by one in this case. */
- if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS)
+ if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
ST.min--;
}
REGCP_SET(ST.cp);
goto curly_try_B_max;
}
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
case CURLY_B_min_known_fail:
/* failed to find B in a non-greedy match where c1,c2 valid */
- if (ST.paren && ST.count)
- rex->offs[ST.paren].end = -1;
- PL_reginput = locinput; /* Could be reset... */
REGCP_UNWIND(ST.cp);
+ if (ST.paren) {
+ UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
+ }
/* Couldn't or didn't -- move forward. */
ST.oldloc = locinput;
if (utf8_target)
}
if (locinput > ST.maxpos)
sayNO;
- /* PL_reginput == oldloc now */
if (n) {
+ /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
+ * at b; check that everything between oldloc and
+ * locinput matches */
+ char *li = ST.oldloc;
ST.count += n;
- if (regrepeat(rex, ST.A, n, depth) < n)
+ if (regrepeat(rex, &li, ST.A, n, depth) < n)
sayNO;
+ assert(n == REG_INFTY || locinput == li);
}
- PL_reginput = locinput;
CURLY_SETPAREN(ST.paren, ST.count);
if (cur_eval && cur_eval->u.eval.close_paren &&
cur_eval->u.eval.close_paren == (U32)ST.paren) {
goto fake_end;
}
- PUSH_STATE_GOTO(CURLY_B_min_known, ST.B);
+ PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput);
}
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
case CURLY_B_min_fail:
/* failed to find B in a non-greedy match where c1,c2 invalid */
- if (ST.paren && ST.count)
- rex->offs[ST.paren].end = -1;
REGCP_UNWIND(ST.cp);
+ if (ST.paren) {
+ UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
+ }
/* failed -- move forward one */
- PL_reginput = locinput;
- if (regrepeat(rex, ST.A, 1, depth)) {
+ {
+ char *li = locinput;
+ if (!regrepeat(rex, &li, ST.A, 1, depth)) {
+ sayNO;
+ }
+ locinput = li;
+ }
+ {
ST.count++;
- locinput = PL_reginput;
if (ST.count <= ST.max || (ST.max == REG_INFTY &&
ST.count > 0)) /* count overflow ? */
{
cur_eval->u.eval.close_paren == (U32)ST.paren) {
goto fake_end;
}
- PUSH_STATE_GOTO(CURLY_B_min, ST.B);
+ PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput);
}
}
- sayNO;
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
curly_try_B_max:
{
UV c = 0;
if (ST.c1 != CHRTEST_VOID)
- c = utf8_target ? utf8n_to_uvchr((U8*)PL_reginput,
+ c = utf8_target ? utf8n_to_uvchr((U8*)locinput,
UTF8_MAXBYTES, 0, uniflags)
- : (UV) UCHARAT(PL_reginput);
+ : (UV) UCHARAT(locinput);
/* If it could work, try it. */
if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) {
CURLY_SETPAREN(ST.paren, ST.count);
- PUSH_STATE_GOTO(CURLY_B_max, ST.B);
- /* NOTREACHED */
+ PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput);
+ assert(0); /* NOTREACHED */
}
}
/* FALL THROUGH */
case CURLY_B_max_fail:
/* failed to find B in a greedy match */
- if (ST.paren && ST.count)
- rex->offs[ST.paren].end = -1;
REGCP_UNWIND(ST.cp);
+ if (ST.paren) {
+ UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
+ }
/* back up. */
if (--ST.count < ST.min)
sayNO;
- PL_reginput = locinput = HOPc(locinput, -1);
+ locinput = HOPc(locinput, -1);
goto curly_try_B_max;
#undef ST
fake_end:
if (cur_eval) {
/* we've just finished A in /(??{A})B/; now continue with B */
- I32 tmpix;
st->u.eval.toggle_reg_flags
= cur_eval->u.eval.toggle_reg_flags;
PL_reg_flags ^= st->u.eval.toggle_reg_flags;
st->u.eval.prev_rex = rex_sv; /* inner */
st->u.eval.cp = regcppush(rex, 0); /* Save *all* the positions. */
- SETREX(rex_sv,cur_eval->u.eval.prev_rex);
+ rex_sv = cur_eval->u.eval.prev_rex;
+ SET_reg_curpm(rex_sv);
rex = (struct regexp *)SvANY(rex_sv);
rexi = RXi_GET(rex);
cur_curlyx = cur_eval->u.eval.prev_curlyx;
- (void)ReREFCNT_inc(rex_sv);
REGCP_SET(st->u.eval.lastcp);
- PL_reginput = locinput;
/* Restore parens of the outer rex without popping the
* savestack */
- tmpix = PL_savestack_ix;
- PL_savestack_ix = cur_eval->u.eval.lastcp;
- regcppop(rex);
- PL_savestack_ix = tmpix;
+ S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp);
st->u.eval.prev_eval = cur_eval;
cur_eval = cur_eval->u.eval.prev_eval;
if ( nochange_depth )
nochange_depth--;
- PUSH_YES_STATE_GOTO(EVAL_AB,
- st->u.eval.prev_eval->u.eval.B); /* match B */
+ PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B,
+ locinput); /* match B */
}
if (locinput < reginfo->till) {
sayNO_SILENT; /* Cannot match: too short. */
}
- PL_reginput = locinput; /* put where regtry can find it */
sayYES; /* Success! */
case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
PerlIO_printf(Perl_debug_log,
"%*s %ssubpattern success...%s\n",
REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
- PL_reginput = locinput; /* put where regtry can find it */
sayYES; /* Success! */
#undef ST
#define ST st->u.ifmatch
+ {
+ char *newstart;
+
case SUSPEND: /* (?>A) */
ST.wanted = 1;
- PL_reginput = locinput;
+ newstart = locinput;
goto do_ifmatch;
case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
next = NULL;
break;
}
- PL_reginput = s;
+ newstart = s;
}
else
- PL_reginput = locinput;
+ newstart = locinput;
do_ifmatch:
ST.me = scan;
logical = 0; /* XXX: reset state of logical once it has been saved into ST */
/* execute body of (?...A) */
- PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)));
- /* NOTREACHED */
+ PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart);
+ assert(0); /* NOTREACHED */
+ }
case IFMATCH_A_fail: /* body of (?...A) failed */
ST.wanted = !ST.wanted;
else if (!ST.wanted)
sayNO;
- if (OP(ST.me) == SUSPEND)
- locinput = PL_reginput;
- else {
- locinput = PL_reginput = st->locinput;
- nextchr = UCHARAT(locinput);
+ if (OP(ST.me) != SUSPEND) {
+ /* restore old position except for (?>...) */
+ locinput = st->locinput;
}
scan = ST.me + ARG(ST.me);
if (scan == ST.me)
reginfo->cutpoint = PL_regeol;
/* FALLTHROUGH */
case PRUNE:
- PL_reginput = locinput;
if (!scan->flags)
sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
- PUSH_STATE_GOTO(COMMIT_next,next);
- /* NOTREACHED */
+ PUSH_STATE_GOTO(COMMIT_next, next, locinput);
+ assert(0); /* NOTREACHED */
case COMMIT_next_fail:
no_final = 1;
/* FALLTHROUGH */
case OPFAIL:
sayNO;
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
#define ST st->u.mark
case MARKPOINT:
ST.mark_name = sv_commit = sv_yes_mark
= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
mark_state = st;
- ST.mark_loc = PL_reginput = locinput;
- PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
- /* NOTREACHED */
+ ST.mark_loc = locinput;
+ PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput);
+ assert(0); /* NOTREACHED */
case MARKPOINT_next:
mark_state = ST.prev_mark;
sayYES;
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
case MARKPOINT_next_fail:
if (popmark && sv_eq(ST.mark_name,popmark))
{
sv_yes_mark = mark_state ?
mark_state->u.mark.mark_name : NULL;
sayNO;
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
case SKIP:
- PL_reginput = locinput;
if (scan->flags) {
/* (*SKIP) : if we fail we cut here*/
ST.mark_name = NULL;
ST.mark_loc = locinput;
- PUSH_STATE_GOTO(SKIP_next,next);
+ PUSH_STATE_GOTO(SKIP_next,next, locinput);
} else {
/* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
otherwise do nothing. Meaning we need to scan
find ) )
{
ST.mark_name = find;
- PUSH_STATE_GOTO( SKIP_next, next );
+ PUSH_STATE_GOTO( SKIP_next, next, locinput);
}
cur = cur->u.mark.prev_mark;
}
}
no_final = 1;
sayNO;
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
#undef ST
case LNBREAK:
if ((n=is_LNBREAK(locinput,utf8_target))) {
locinput += n;
- nextchr = UCHARAT(locinput);
} else
sayNO;
break;
sayNO; \
if ((n=is_##nAmE(locinput,utf8_target))) { \
locinput += n; \
- nextchr = UCHARAT(locinput); \
} else \
sayNO; \
break; \
sayNO; \
} else { \
locinput += UTF8SKIP(locinput); \
- nextchr = UCHARAT(locinput); \
} \
break
PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
PTR2UV(scan), OP(scan));
Perl_croak(aTHX_ "regexp memory corruption");
+
+ /* this is a point to jump to in order to increment
+ * locinput by one character */
+ increment_locinput:
+ if (utf8_target) {
+ locinput += PL_utf8skip[nextchr];
+ if (locinput > PL_regeol)
+ sayNO;
+ }
+ else
+ locinput++;
+ break;
} /* end switch */
/* switch break jumps here */
scan = next; /* prepare to execute the next op and ... */
continue; /* ... jump back to the top, reusing st */
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
push_yes_state:
/* push a state that backtracks on success */
newst = S_push_slab(aTHX);
PL_regmatch_state = newst;
- locinput = PL_reginput;
- nextchr = UCHARAT(locinput);
+ locinput = pushinput;
st = newst;
continue;
- /* NOTREACHED */
+ assert(0); /* NOTREACHED */
}
}
yes_state = st->u.yes.prev_yes_state;
PL_regmatch_state = st;
- if (no_final) {
+ if (no_final)
locinput= st->locinput;
- nextchr = UCHARAT(locinput);
- }
state_num = st->resume_state + no_final;
goto reenter_switch;
}
}
PL_regmatch_state = st;
locinput= st->locinput;
- nextchr = UCHARAT(locinput);
DEBUG_STATE_pp("pop");
depth--;
if (last_pushed_cv) {
dSP;
POP_MULTICALL;
+ PERL_UNUSED_VAR(SP);
}
/* clean up; in particular, free all slabs above current one */
LEAVE_SCOPE(oldsave);
- return result;
+ assert(!result || locinput - PL_bostr >= 0);
+ return result ? locinput - PL_bostr : -1;
}
/*
- regrepeat - repeatedly match something simple, report how many
- */
-/*
- * [This routine now assumes that it will only match on things of length 1.
- * That was true before, but now we assume scan - reginput is the count,
- * rather than incrementing count on every character. [Er, except utf8.]]
+ *
+ * startposp - pointer a pointer to the start position. This is updated
+ * to point to the byte following the highest successful
+ * match.
+ * p - the regnode to be repeatedly matched against.
+ * max - maximum number of characters to match.
+ * depth - (for debugging) backtracking depth.
*/
STATIC I32
-S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth)
+S_regrepeat(pTHX_ const regexp *prog, char **startposp, const regnode *p, I32 max, int depth)
{
dVAR;
- register char *scan;
- register I32 c;
- register char *loceol = PL_regeol;
- register I32 hardcount = 0;
- register bool utf8_target = PL_reg_match_utf8;
+ char *scan;
+ I32 c;
+ char *loceol = PL_regeol;
+ I32 hardcount = 0;
+ bool utf8_target = PL_reg_match_utf8;
UV utf8_flags;
#ifndef DEBUGGING
PERL_UNUSED_ARG(depth);
PERL_ARGS_ASSERT_REGREPEAT;
- scan = PL_reginput;
+ scan = *startposp;
if (max == REG_INFTY)
max = I32_MAX;
else if (max < loceol - scan)
scan++;
}
break;
+
+ case POSIXA:
+ while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
+ scan++;
+ }
+ break;
+ case NPOSIXA:
+ if (utf8_target) {
+ while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
+ scan += UTF8SKIP(scan);
+ }
+ }
+ else {
+ while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
+ scan++;
+ }
+ }
+ break;
case NALNUMA:
if (utf8_target) {
while (scan < loceol && ! isWORDCHAR_A((U8) *scan)) {
if (hardcount)
c = hardcount;
else
- c = scan - PL_reginput;
- PL_reginput = scan;
+ c = scan - *startposp;
+ *startposp = scan;
DEBUG_r({
GET_RE_DEBUG_FLAGS_DECL;
SV * const rv = MUTABLE_SV(data->data[n]);
AV * const av = MUTABLE_AV(SvRV(rv));
SV **const ary = AvARRAY(av);
- bool invlist_has_user_defined_property;
+ U8 swash_init_flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
si = *ary; /* ary[0] = the string to initialize the swash with */
* that inversion list has any user-defined properties in it. */
if (av_len(av) >= 3) {
invlist = ary[3];
- invlist_has_user_defined_property = cBOOL(SvUV(ary[4]));
+ if (SvUV(ary[4])) {
+ swash_init_flags |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
+ }
}
else {
invlist = NULL;
- invlist_has_user_defined_property = FALSE;
}
/* Element [1] is reserved for the set-up swash. If already there,
si,
1, /* binary */
0, /* not from tr/// */
- FALSE, /* is error if can't find
- property */
invlist,
- invlist_has_user_defined_property);
+ &swash_init_flags);
(void)av_store(av, 1, sw);
}
if (listsvp) {
SV* matches_string = newSVpvn("", 0);
- SV** invlistsvp;
/* Use the swash, if any, which has to have incorporated into it all
* possibilities */
- if ( sw
- && SvROK(sw)
- && SvTYPE(SvRV(sw)) == SVt_PVHV
- && (invlistsvp = hv_fetchs(MUTABLE_HV(SvRV(sw)), "INVLIST", FALSE)))
- {
- invlist = *invlistsvp;
- }
- else if (si && si != &PL_sv_undef) {
+ if ((! sw || (invlist = _get_swash_invlist(sw)) == NULL)
+ && (si && si != &PL_sv_undef))
+ {
- /* If no swash, use the input nitialization string, if available */
+ /* If no swash, use the input initialization string, if available */
sv_catsv(matches_string, si);
}
if (! utf8_target) Safefree(utf8_p);
}
}
+
+ if (UNICODE_IS_SUPER(c)
+ && (flags & ANYOF_WARN_SUPER)
+ && ckWARN_d(WARN_NON_UNICODE))
+ {
+ Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
+ "Code point 0x%04"UVXf" is not Unicode, all \\p{} matches fail; all \\P{} matches succeed", c);
+ }
}
- return (flags & ANYOF_INVERT) ? !match : match;
+ /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
+ return cBOOL(flags & ANYOF_INVERT) ^ match;
}
STATIC U8 *
if (PL_reg_oldsaved) {
rex->subbeg = PL_reg_oldsaved;
rex->sublen = PL_reg_oldsavedlen;
+ rex->suboffset = PL_reg_oldsavedoffset;
+ rex->subcoffset = PL_reg_oldsavedcoffset;
#ifdef PERL_OLD_COPY_ON_WRITE
rex->saved_copy = PL_nrs;
#endif
} while (i--);
}
+/* These constants are for finding GCB=LV and GCB=LVT. These are for the
+ * pre-composed Hangul syllables, which are all in a contiguous block and
+ * arranged there in such a way so as to facilitate alorithmic determination of
+ * their characteristics. As such, they don't need a swash, but can be
+ * determined by simple arithmetic. Almost all are GCB=LVT, but every 28th one
+ * is a GCB=LV */
+#define SBASE 0xAC00 /* Start of block */
+#define SCount 11172 /* Length of block */
+#define TCount 28
+
+#if 0 /* This routine is not currently used */
+PERL_STATIC_INLINE bool
+S_is_utf8_X_LV(pTHX_ const U8 *p)
+{
+ /* Unlike most other similarly named routines here, this does not create a
+ * swash, so swash_fetch() cannot be used on PL_utf8_X_LV. */
+
+ dVAR;
+
+ UV cp = valid_utf8_to_uvchr(p, NULL);
+
+ PERL_ARGS_ASSERT_IS_UTF8_X_LV;
+
+ /* The earliest Unicode releases did not have these precomposed Hangul
+ * syllables. Set to point to undef in that case, so will return false on
+ * every call */
+ if (! PL_utf8_X_LV) { /* Set up if this is the first time called */
+ PL_utf8_X_LV = swash_init("utf8", "_X_GCB_LV", &PL_sv_undef, 1, 0);
+ if (_invlist_len(_get_swash_invlist(PL_utf8_X_LV)) == 0) {
+ SvREFCNT_dec(PL_utf8_X_LV);
+ PL_utf8_X_LV = &PL_sv_undef;
+ }
+ }
+
+ return (PL_utf8_X_LV != &PL_sv_undef
+ && cp >= SBASE && cp < SBASE + SCount
+ && (cp - SBASE) % TCount == 0); /* Only every TCount one is LV */
+}
+#endif
+
+PERL_STATIC_INLINE bool
+S_is_utf8_X_LVT(pTHX_ const U8 *p)
+{
+ /* Unlike most other similarly named routines here, this does not create a
+ * swash, so swash_fetch() cannot be used on PL_utf8_X_LVT. */
+
+ dVAR;
+
+ UV cp = valid_utf8_to_uvchr(p, NULL);
+
+ PERL_ARGS_ASSERT_IS_UTF8_X_LVT;
+
+ /* The earliest Unicode releases did not have these precomposed Hangul
+ * syllables. Set to point to undef in that case, so will return false on
+ * every call */
+ if (! PL_utf8_X_LVT) { /* Set up if this is the first time called */
+ PL_utf8_X_LVT = swash_init("utf8", "_X_GCB_LVT", &PL_sv_undef, 1, 0);
+ if (_invlist_len(_get_swash_invlist(PL_utf8_X_LVT)) == 0) {
+ SvREFCNT_dec(PL_utf8_X_LVT);
+ PL_utf8_X_LVT = &PL_sv_undef;
+ }
+ }
+
+ return (PL_utf8_X_LVT != &PL_sv_undef
+ && cp >= SBASE && cp < SBASE + SCount
+ && (cp - SBASE) % TCount != 0); /* All but every TCount one is LV */
+}
+
/*
* Local variables:
* c-indentation-style: bsd