*/
/*
- * "One Ring to rule them all, One Ring to find them..."
+ * One Ring to rule them all, One Ring to find them
+ &
+ * [p.v of _The Lord of the Rings_, opening poem]
+ * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
+ * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
*/
/* This file contains functions for executing a regular expression. See
**** Alterations to Henry's code are...
****
**** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
- **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007 by Larry Wall and others
+ **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
+ **** by Larry Wall and others
****
**** You may distribute under the terms of either the GNU General Public
**** License or the Artistic License, as specified in the README file.
dVAR;
U32 i;
char *input;
-
GET_RE_DEBUG_FLAGS_DECL;
+ PERL_ARGS_ASSERT_REGCPPOP;
+
/* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
i = SSPOPINT;
assert(i == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
* requiring null fields (pat.t#187 and split.t#{13,14}
* (as of patchlevel 7877) will fail. Then again,
* this code seems to be necessary or otherwise
- * building DynaLoader will fail:
- * "Error: '*' not in typemap in DynaLoader.xs, line 164"
- * --jhi */
+ * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
+ * --jhi updated by dapm */
for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
if (i > PL_regsize)
PL_regoffs[i].start = -1;
/* minend: end of match must be >=minend after stringarg. */
/* nosave: For optimizations. */
{
+ PERL_ARGS_ASSERT_PREGEXEC;
+
return
regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
nosave ? 0 : REXEC_COPY_STR);
deleted from the finite automaton. */
char *
-Perl_re_intuit_start(pTHX_ REGEXP * const prog, SV *sv, char *strpos,
+Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos,
char *strend, const U32 flags, re_scream_pos_data *data)
{
dVAR;
+ struct regexp *const prog = (struct regexp *)SvANY(rx);
register I32 start_shift = 0;
/* Should be nonnegative! */
register I32 end_shift = 0;
#ifdef DEBUGGING
const char * const i_strpos = strpos;
#endif
-
GET_RE_DEBUG_FLAGS_DECL;
- RX_MATCH_UTF8_set(prog,do_utf8);
+ PERL_ARGS_ASSERT_RE_INTUIT_START;
- if (prog->extflags & RXf_UTF8) {
+ RX_MATCH_UTF8_set(rx,do_utf8);
+
+ if (RX_UTF8(rx)) {
PL_reg_flags |= RF_utf8;
}
DEBUG_EXECUTE_r(
- debug_start_match(prog, do_utf8, strpos, strend,
+ debug_start_match(rx, do_utf8, strpos, strend,
sv ? "Guessing start of match in sv for"
: "Guessing start of match in string for");
);
#ifdef QDEBUGGING /* 7/99: reports of failure (with the older version) */
if (end_shift < 0)
Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
- (IV)end_shift, prog->precomp);
+ (IV)end_shift, RX_PRECOMP(prog));
#endif
restart:
else
goto fail_finish;
/* we may be pointing at the wrong string */
- if (s && RX_MATCH_COPIED(prog))
+ if (s && RXp_MATCH_COPIED(prog))
s = strbeg + (s - SvPVX_const(sv));
if (data)
*data->scream_olds = s;
return NULL;
}
-
+#define DECL_TRIE_TYPE(scan) \
+ const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
+ trie_type = (scan->flags != EXACT) \
+ ? (do_utf8 ? trie_utf8_fold : (UTF ? trie_latin_utf8_fold : trie_plain)) \
+ : (do_utf8 ? trie_utf8 : trie_plain)
#define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
+ UV uvc_unfolded = 0; \
switch (trie_type) { \
case trie_utf8_fold: \
if ( foldlen>0 ) { \
- uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
+ uvc_unfolded = uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
foldlen -= len; \
uscan += len; \
len=0; \
} else { \
- uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
+ uvc_unfolded = uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
foldlen -= UNISKIP( uvc ); \
uscan = foldbuf + UNISKIP( uvc ); \
} \
break; \
+ case trie_latin_utf8_fold: \
+ if ( foldlen>0 ) { \
+ uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
+ foldlen -= len; \
+ uscan += len; \
+ len=0; \
+ } else { \
+ len = 1; \
+ uvc = to_uni_fold( *(U8*)uc, foldbuf, &foldlen ); \
+ foldlen -= UNISKIP( uvc ); \
+ uscan = foldbuf + UNISKIP( uvc ); \
+ } \
+ break; \
case trie_utf8: \
uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
break; \
charid = (U16)SvIV(*svpp); \
} \
} \
+ if (!charid && trie_type == trie_utf8_fold && !UTF) { \
+ charid = trie->charmap[uvc_unfolded]; \
+ } \
} STMT_END
-#define REXEC_FBC_EXACTISH_CHECK(CoNd) \
+#define REXEC_FBC_EXACTISH_CHECK(CoNd) \
+{ \
+ char *my_strend= (char *)strend; \
if ( (CoNd) \
&& (ln == len || \
- ibcmp_utf8(s, NULL, 0, do_utf8, \
+ !ibcmp_utf8(s, &my_strend, 0, do_utf8, \
m, NULL, ln, (bool)UTF)) \
- && (!reginfo || regtry(reginfo, &s)) ) \
+ && (!reginfo || regtry(reginfo, &s)) ) \
goto got_it; \
else { \
U8 foldbuf[UTF8_MAXBYTES_CASE+1]; \
f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); \
if ( f != c \
&& (f == c1 || f == c2) \
- && (ln == foldlen || \
- !ibcmp_utf8((char *) foldbuf, \
- NULL, foldlen, do_utf8, \
- m, \
- NULL, ln, (bool)UTF)) \
- && (!reginfo || regtry(reginfo, &s)) ) \
+ && (ln == len || \
+ !ibcmp_utf8(s, &my_strend, 0, do_utf8,\
+ m, NULL, ln, (bool)UTF)) \
+ && (!reginfo || regtry(reginfo, &s)) ) \
goto got_it; \
} \
- s += len
+} \
+s += len
#define REXEC_FBC_EXACTISH_SCAN(CoNd) \
STMT_START { \
register I32 tmp = 1; /* Scratch variable? */
register const bool do_utf8 = PL_reg_match_utf8;
RXi_GET_DECL(prog,progi);
+
+ PERL_ARGS_ASSERT_FIND_BYCLASS;
/* We know what class it must start with. */
switch (OP(c)) {
U8 *sm = (U8 *) m;
U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
- const U32 uniflags = UTF8_ALLOW_DEFAULT;
-
- to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
- to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
-
+ /* used by commented-out code below */
+ /*const U32 uniflags = UTF8_ALLOW_DEFAULT;*/
+
+ /* XXX: Since the node will be case folded at compile
+ time this logic is a little odd, although im not
+ sure that its actually wrong. --dmq */
+
+ c1 = to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
+ c2 = to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
+
+ /* XXX: This is kinda strange. to_utf8_XYZ returns the
+ codepoint of the first character in the converted
+ form, yet originally we did the extra step.
+ No tests fail by commenting this code out however
+ so Ive left it out. -- dmq.
+
c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
0, uniflags);
c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
0, uniflags);
+ */
+
lnc = 0;
while (sm < ((U8 *) m + ln)) {
lnc++;
* matching (called "loose matching" in Unicode).
* ibcmp_utf8() will do just that. */
- if (do_utf8) {
+ if (do_utf8 || UTF) {
UV c, f;
U8 tmpbuf [UTF8_MAXBYTES+1];
- STRLEN len, foldlen;
+ STRLEN len = 1;
+ STRLEN foldlen;
const U32 uniflags = UTF8_ALLOW_DEFAULT;
if (c1 == c2) {
/* Upper and lower of 1st char are equal -
* probably not a "letter". */
while (s <= e) {
- c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
+ if (do_utf8) {
+ c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
uniflags);
+ } else {
+ c = *((U8*)s);
+ }
REXEC_FBC_EXACTISH_CHECK(c == c1);
}
}
else {
while (s <= e) {
- c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
+ if (do_utf8) {
+ c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
uniflags);
+ } else {
+ c = *((U8*)s);
+ }
/* Handle some of the three Greek sigmas cases.
* Note that not all the possible combinations
}
}
else {
+ /* Neither pattern nor string are UTF8 */
if (c1 == c2)
REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
else
case AHOCORASICKC:
case AHOCORASICK:
{
- const enum { trie_plain, trie_utf8, trie_utf8_fold }
- trie_type = do_utf8 ?
- (c->flags == EXACT ? trie_utf8 : trie_utf8_fold)
- : trie_plain;
+ DECL_TRIE_TYPE(c);
/* what trie are we using right now */
reg_ac_data *aho
= (reg_ac_data*)progi->data->data[ ARG( c ) ];
reg_trie_data *trie
= (reg_trie_data*)progi->data->data[ aho->trie ];
- HV *widecharmap = (HV*) progi->data->data[ aho->trie + 1 ];
+ HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
const char *last_start = strend - trie->minlen;
#ifdef DEBUGGING
U8 **points; /* map of where we were in the input string
when reading a given char. For ASCII this
is unnecessary overhead as the relationship
- is always 1:1, but for unicode, especially
- case folded unicode this is not true. */
+ is always 1:1, but for Unicode, especially
+ case folded Unicode this is not true. */
U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
U8 *bitmap=NULL;
return s;
}
-static void
-S_swap_match_buff (pTHX_ regexp *prog) {
- regexp_paren_pair *t;
-
- if (!prog->swap) {
- /* We have to be careful. If the previous successful match
- was from this regex we don't want a subsequent paritally
- successful match to clobber the old results.
- So when we detect this possibility we add a swap buffer
- to the re, and switch the buffer each match. If we fail
- we switch it back, otherwise we leave it swapped.
- */
- Newxz(prog->swap, (prog->nparens + 1), regexp_paren_pair);
- }
- t = prog->swap;
- prog->swap = prog->offs;
- prog->offs = t;
-}
-
/*
- regexec_flags - match a regexp against a string
*/
I32
-Perl_regexec_flags(pTHX_ REGEXP * const prog, char *stringarg, register char *strend,
+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 */
/* nosave: For optimizations. */
{
dVAR;
+ struct regexp *const prog = (struct regexp *)SvANY(rx);
/*register*/ char *s;
register regnode *c;
/*register*/ char *startpos = stringarg;
I32 multiline;
RXi_GET_DECL(prog,progi);
regmatch_info reginfo; /* create some info to pass to regtry etc */
- bool swap_on_fail = 0;
-
+ regexp_paren_pair *swap = NULL;
GET_RE_DEBUG_FLAGS_DECL;
+ PERL_ARGS_ASSERT_REGEXEC_FLAGS;
PERL_UNUSED_ARG(data);
/* Be paranoid... */
}
multiline = prog->extflags & RXf_PMf_MULTILINE;
- reginfo.prog = prog;
+ reginfo.prog = rx; /* Yes, sorry that this is confusing. */
- RX_MATCH_UTF8_set(prog, do_utf8);
+ RX_MATCH_UTF8_set(rx, do_utf8);
DEBUG_EXECUTE_r(
- debug_start_match(prog, do_utf8, startpos, strend,
+ debug_start_match(rx, do_utf8, startpos, strend,
"Matching");
);
PL_reg_eval_set = 0;
PL_reg_maxiter = 0;
- if (prog->extflags & RXf_UTF8)
+ if (RX_UTF8(rx))
PL_reg_flags |= RF_utf8;
/* Mark beginning of line for ^ and lookbehind. */
if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
MAGIC *mg;
-
- if (flags & REXEC_IGNOREPOS) /* Means: check only at start */
+ if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
reginfo.ganch = startpos + prog->gofs;
- else if (sv && SvTYPE(sv) >= SVt_PVMG
+ DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
+ "GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",prog->gofs));
+ } else if (sv && SvTYPE(sv) >= SVt_PVMG
&& SvMAGIC(sv)
&& (mg = mg_find(sv, PERL_MAGIC_regex_global))
&& mg->mg_len >= 0) {
reginfo.ganch = strbeg + mg->mg_len; /* Defined pos() */
+ DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
+ "GPOS MAGIC: reginfo.ganch = strbeg + %"IVdf"\n",mg->mg_len));
+
if (prog->extflags & RXf_ANCH_GPOS) {
if (s > reginfo.ganch)
goto phooey;
s = reginfo.ganch - prog->gofs;
+ DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
+ "GPOS ANCH_GPOS: s = ganch - %"UVxf"\n",prog->gofs));
+ if (s < strbeg)
+ goto phooey;
}
}
else if (data) {
reginfo.ganch = strbeg + PTR2UV(data);
- } else /* pos() not defined */
+ DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
+ "GPOS DATA: reginfo.ganch= strbeg + %"UVxf"\n",PTR2UV(data)));
+
+ } else { /* pos() not defined */
reginfo.ganch = strbeg;
+ DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
+ "GPOS: reginfo.ganch = strbeg\n"));
+ }
}
- if (PL_curpm && (PM_GETRE(PL_curpm) == prog)) {
- swap_on_fail = 1;
- swap_match_buff(prog); /* do we need a save destructor here for
- eval dies? */
+ if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
+ /* We have to be careful. If the previous successful match
+ was from this regex we don't want a subsequent partially
+ successful match to clobber the old results.
+ So when we detect this possibility we add a swap buffer
+ to the re, and switch the buffer each match. If we fail
+ we switch it back, otherwise we leave it swapped.
+ */
+ swap = prog->offs;
+ /* do we need a save destructor here for eval dies? */
+ Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
}
if (!(flags & REXEC_CHECKED) && (prog->check_substr != NULL || prog->check_utf8 != NULL)) {
re_scream_pos_data d;
d.scream_olds = &scream_olds;
d.scream_pos = &scream_pos;
- s = re_intuit_start(prog, sv, s, strend, flags, &d);
+ s = re_intuit_start(rx, sv, s, strend, flags, &d);
if (!s) {
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not present...\n"));
goto phooey; /* not present */
if (regtry(®info, &s))
goto got_it;
after_try:
- if (s >= end)
+ if (s > end)
goto phooey;
if (prog->extflags & RXf_USE_INTUIT) {
- s = re_intuit_start(prog, sv, s + 1, strend, flags, NULL);
+ s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
if (!s)
goto phooey;
}
is bogus -- we set it above, when prog->extflags & RXf_GPOS_SEEN
and we only enter this block when the same bit is set. */
char *tmp_s = reginfo.ganch - prog->gofs;
- if (regtry(®info, &tmp_s))
+
+ if (tmp_s >= strbeg && regtry(®info, &tmp_s))
goto got_it;
goto phooey;
}
(unsigned char*)strend, must,
multiline ? FBMrf_MULTILINE : 0))) ) {
/* we may be pointing at the wrong string */
- if ((flags & REXEC_SCREAM) && RX_MATCH_COPIED(prog))
+ if ((flags & REXEC_SCREAM) && RXp_MATCH_COPIED(prog))
s = strbeg + (s - SvPVX_const(sv));
DEBUG_EXECUTE_r( did_match = 1 );
if (HOPc(s, -back_max) > last1) {
if (!last)
last = scream_olds; /* Only one occurrence. */
/* we may be pointing at the wrong string */
- else if (RX_MATCH_COPIED(prog))
+ else if (RXp_MATCH_COPIED(prog))
s = strbeg + (s - SvPVX_const(sv));
}
else {
goto phooey;
got_it:
- RX_MATCH_TAINTED_set(prog, PL_reg_flags & RF_tainted);
+ Safefree(swap);
+ RX_MATCH_TAINTED_set(rx, PL_reg_flags & RF_tainted);
if (PL_reg_eval_set)
restore_pos(aTHX_ prog);
- if (prog->paren_names)
- (void)hv_iterinit(prog->paren_names);
+ if (RXp_PAREN_NAMES(prog))
+ (void)hv_iterinit(RXp_PAREN_NAMES(prog));
/* make sure $`, $&, $', and $digit will work later */
if ( !(flags & REXEC_NOT_FIRST) ) {
- RX_MATCH_COPY_FREE(prog);
+ RX_MATCH_COPY_FREE(rx);
if (flags & REXEC_COPY_STR) {
const I32 i = PL_regeol - startpos + (stringarg - strbeg);
#ifdef PERL_OLD_COPY_ON_WRITE
} else
#endif
{
- RX_MATCH_COPIED_on(prog);
+ RX_MATCH_COPIED_on(rx);
s = savepvn(strbeg, i);
prog->subbeg = s;
}
PL_colors[4], PL_colors[5]));
if (PL_reg_eval_set)
restore_pos(aTHX_ prog);
- if (swap_on_fail)
+ if (swap) {
/* we failed :-( roll it back */
- swap_match_buff(prog);
-
+ Safefree(prog->offs);
+ prog->offs = swap;
+ }
+
return 0;
}
{
dVAR;
CHECKPOINT lastcp;
- regexp *prog = reginfo->prog;
+ REGEXP *const rx = reginfo->prog;
+ regexp *const prog = (struct regexp *)SvANY(rx);
RXi_GET_DECL(prog,progi);
GET_RE_DEBUG_FLAGS_DECL;
+
+ PERL_ARGS_ASSERT_REGTRY;
+
reginfo->cutpoint=NULL;
if ((prog->extflags & RXf_EVAL_SEEN) && !PL_reg_eval_set) {
/* Make $_ available to executed code. */
if (reginfo->sv != DEFSV) {
SAVE_DEFSV;
- DEFSV = reginfo->sv;
+ DEFSV_set(reginfo->sv);
}
if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
Newxz(PL_reg_curpm, 1, PMOP);
#ifdef USE_ITHREADS
{
- SV* const repointer = newSViv(0);
- /* so we know which PL_regex_padav element is PL_reg_curpm */
- SvFLAGS(repointer) |= SVf_BREAK;
- av_push(PL_regex_padav,repointer);
+ SV* const repointer = &PL_sv_undef;
+ /* this regexp is also owned by the new PL_reg_curpm, which
+ will try to free it. */
+ av_push(PL_regex_padav, repointer);
PL_reg_curpm->op_pmoffset = av_len(PL_regex_padav);
PL_regex_pad = AvARRAY(PL_regex_padav);
}
#endif
}
- PM_SETRE(PL_reg_curpm, prog);
+#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. */
+ ReREFCNT_inc(rx);
+#endif
+ PM_SETRE(PL_reg_curpm, rx);
PL_reg_oldcurpm = PL_curpm;
PL_curpm = PL_reg_curpm;
- if (RX_MATCH_COPIED(prog)) {
+ if (RXp_MATCH_COPIED(prog)) {
/* Here is a serious problem: we cannot rewrite subbeg,
since it may be needed if this match fails. Thus
$` inside (?{}) could fail... */
#ifdef PERL_OLD_COPY_ON_WRITE
PL_nrs = prog->saved_copy;
#endif
- RX_MATCH_COPIED_off(prog);
+ RXp_MATCH_COPIED_off(prog);
}
else
PL_reg_oldsaved = NULL;
/* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
* Actually, the code in regcppop() (which Ilya may be meaning by
* PL_reglastparen), is not needed at all by the test suite
- * (op/regexp, op/pat, op/split), but that code is needed, oddly
- * enough, for building DynaLoader, or otherwise this
- * "Error: '*' not in typemap in DynaLoader.xs, line 164"
- * will happen. Meanwhile, this code *is* needed for the
+ * (op/regexp, op/pat, op/split), but that code is needed otherwise
+ * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
+ * Meanwhile, this code *is* needed for the
* above-mentioned test suite tests to succeed. The common theme
* on those tests seems to be returning null fields from matches.
- * --jhi */
+ * --jhi updated by dapm */
#if 1
if (prog->nparens) {
regexp_paren_pair *pp = PL_regoffs;
#ifdef DEBUGGING
STATIC void
-S_debug_start_match(pTHX_ const regexp *prog, const bool do_utf8,
+S_debug_start_match(pTHX_ const REGEXP *prog, const bool do_utf8,
const char *start, const char *end, const char *blurb)
{
- const bool utf8_pat= prog->extflags & RXf_UTF8 ? 1 : 0;
+ const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
+
+ PERL_ARGS_ASSERT_DEBUG_START_MATCH;
+
if (!PL_colorset)
reginitcolors();
{
RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
- prog->precomp, prog->prelen, 60);
+ RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
RE_PV_QUOTED_DECL(s1, do_utf8, PERL_DEBUG_PAD_ZERO(1),
start, end - start, 60);
? (5 + taill) - l : locinput - loc_bostr;
int pref0_len;
+ PERL_ARGS_ASSERT_DUMP_EXEC_POS;
+
while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
pref_len++;
pref0_len = pref_len - (locinput - loc_reg_starttry);
* or 0 if non of the buffers matched.
*/
STATIC I32
-S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan) {
+S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
+{
I32 n;
RXi_GET_DECL(rex,rexi);
- SV *sv_dat=(SV*)rexi->data->data[ ARG( scan ) ];
+ SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
I32 *nums=(I32*)SvPVX(sv_dat);
+
+ PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
+
for ( n=0; n<SvIVX(sv_dat); n++ ) {
if ((I32)*PL_reglastparen >= nums[n] &&
PL_regoffs[nums[n]].end != -1)
dVAR;
register const bool do_utf8 = PL_reg_match_utf8;
const U32 uniflags = UTF8_ALLOW_DEFAULT;
-
- regexp *rex = reginfo->prog;
+ 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;
-
/* cache heavy used fields of st in registers */
register regnode *scan;
register regnode *next;
const U32 max_nochange_depth =
(3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
-
regmatch_state *yes_state = NULL; /* state to pop to on success of
subpattern */
/* mark_state piggy backs on the yes_state logic so that when we unwind
the stack on success we can update the mark_state as we go */
regmatch_state *mark_state = NULL; /* last mark state we have seen */
-
regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
U32 state_num;
during a successfull match */
U32 lastopen = 0; /* last open we saw */
bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
-
SV* const oreplsv = GvSV(PL_replgv);
-
-
/* these three flags are set by various ops to signal information to
* the very next op. They have a useful lifetime of exactly one loop
* iteration, and are not preserved or restored by state pushes/pops
false: plain (?=foo)
true: used as a condition: (?(?=foo))
*/
-
#ifdef DEBUGGING
GET_RE_DEBUG_FLAGS_DECL;
#endif
+ PERL_ARGS_ASSERT_REGMATCH;
+
DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
PerlIO_printf(Perl_debug_log,"regmatch start\n");
}));
state_num = OP(scan);
reenter_switch:
+
+ assert(PL_reglastparen == &rex->lastparen);
+ assert(PL_reglastcloseparen == &rex->lastcloseparen);
+ assert(PL_regoffs == rex->offs);
+
switch (state_num) {
case BOL:
if (locinput == PL_bostr)
case TRIE:
{
/* what type of TRIE am I? (utf8 makes this contextual) */
- const enum { trie_plain, trie_utf8, trie_utf8_fold }
- trie_type = do_utf8 ?
- (scan->flags == EXACT ? trie_utf8 : trie_utf8_fold)
- : trie_plain;
+ DECL_TRIE_TYPE(scan);
/* what trie are we using right now */
reg_trie_data * const trie
= (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
- HV * widecharmap = (HV *)rexi->data->data[ ARG( scan ) + 1 ];
+ HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
U32 state = trie->startstate;
if (trie->bitmap && trie_type != trie_utf8_fold &&
if ( got_wordnum ) {
if ( ! ST.accepted ) {
ENTER;
- SAVETMPS;
+ SAVETMPS; /* XXX is this necessary? dmq */
bufflen = TRIE_INITAL_ACCEPT_BUFFLEN;
sv_accept_buff=newSV(bufflen *
sizeof(reg_trie_accepted) - 1);
/* only one choice left - just continue */
DEBUG_EXECUTE_r({
AV *const trie_words
- = (AV *) rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET];
+ = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
SV ** const tmp = av_fetch( trie_words,
ST.accept_buff[ 0 ].wordnum-1, 0 );
SV *sv= tmp ? sv_newmortal() : NULL;
DEBUG_EXECUTE_r({
AV *const trie_words
- = (AV *) rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET];
+ = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
SV ** const tmp = av_fetch( trie_words,
ST.accept_buff[ best ].wordnum - 1, 0 );
regnode *nextop=(!ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) ?
PL_reginput = (char *)ST.accept_buff[ best ].endpos;
if ( !ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) {
scan = ST.B;
- /* NOTREACHED */
} else {
scan = ST.me + ST.jump[ST.accept_buff[best].wordnum];
- /* NOTREACHED */
- }
- if (has_cutgroup) {
- PUSH_YES_STATE_GOTO(TRIE_next, scan);
- /* NOTREACHED */
- } else {
- PUSH_STATE_GOTO(TRIE_next, scan);
- /* NOTREACHED */
}
+ PUSH_YES_STATE_GOTO(TRIE_next, scan);
/* NOTREACHED */
}
/* NOTREACHED */
case TRIE_next:
+ /* we dont want to throw this away, see bug 57042*/
+ if (oreplsv != GvSV(PL_replgv))
+ sv_setsv(oreplsv, GvSV(PL_replgv));
FREETMPS;
LEAVE;
sayYES;
#define ST st->u.eval
{
SV *ret;
+ REGEXP *re_sv;
regexp *re;
regexp_internal *rei;
regnode *startpoint;
} else {
nochange_depth = 0;
}
+ re_sv = rex_sv;
re = rex;
rei = rexi;
- (void)ReREFCNT_inc(rex);
+ (void)ReREFCNT_inc(rex_sv);
if (OP(scan)==GOSUB) {
startpoint = scan + ARG2L(scan);
ST.close_paren = ARG(scan);
OP_4tree * const oop = PL_op;
COP * const ocurcop = PL_curcop;
PAD *old_comppad;
+ char *saved_regeol = PL_regeol;
n = ARG(scan);
PL_op = (OP_4tree*)rexi->data->data[n];
PL_op = oop;
PAD_RESTORE_LOCAL(old_comppad);
PL_curcop = ocurcop;
+ PL_regeol = saved_regeol;
if (!logical) {
/* /(?{...})/ */
sv_setsv(save_scalar(PL_replgv), ret);
{
/* extract RE object from returned value; compiling if
* necessary */
-
MAGIC *mg = NULL;
- const SV *sv;
- if(SvROK(ret) && SvSMAGICAL(sv = SvRV(ret)))
- mg = mg_find(sv, PERL_MAGIC_qr);
- else if (SvSMAGICAL(ret)) {
- if (SvGMAGICAL(ret))
+ 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
+ }
+ 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) {
- re = reg_temp_copy((regexp *)mg->mg_obj); /*XXX:dmq*/
+ rx = (REGEXP *) mg->mg_obj; /*XXX:dmq*/
+ assert(rx);
+ }
+ if (rx) {
+ rx = reg_temp_copy(rx);
}
else {
U32 pm_flags = 0;
const I32 osize = PL_regsize;
- if (DO_UTF8(ret)) pm_flags |= RXf_UTF8;
- re = CALLREGCOMP(ret, pm_flags);
+ 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. */
+ STRLEN len;
+ const char *const p = SvPV(ret, len);
+ ret = newSVpvn_flags(p, len, SVs_TEMP);
+ }
+ rx = CALLREGCOMP(ret, pm_flags);
if (!(SvFLAGS(ret)
& (SVs_TEMP | SVs_PADTMP | SVf_READONLY
- | SVs_GMG)))
- sv_magic(ret,(SV*)ReREFCNT_inc(re),
- PERL_MAGIC_qr,0,0);
+ | 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);
+ }
PL_regsize = osize;
}
+ re_sv = rx;
+ re = (struct regexp *)SvANY(rx);
}
- RX_MATCH_COPIED_off(re);
+ RXp_MATCH_COPIED_off(re);
re->subbeg = rex->subbeg;
re->sublen = rex->sublen;
rei = RXi_GET(re);
DEBUG_EXECUTE_r(
- debug_start_match(re, do_utf8, locinput, PL_regeol,
+ debug_start_match(re_sv, do_utf8, locinput, PL_regeol,
"Matching embedded");
);
startpoint = rei->program + 1;
PL_regoffs = re->offs; /* essentially NOOP on GOSUB */
- *PL_reglastparen = 0;
- *PL_reglastcloseparen = 0;
+ /* see regtry, specifically PL_reglast(?:close)?paren is a pointer! (i dont know why) :dmq */
+ PL_reglastparen = &re->lastparen;
+ PL_reglastcloseparen = &re->lastcloseparen;
+ re->lastparen = 0;
+ re->lastcloseparen = 0;
+
PL_reginput = locinput;
PL_regsize = 0;
PL_reg_maxiter = 0;
ST.toggle_reg_flags = PL_reg_flags;
- if (re->extflags & RXf_UTF8)
+ if (RX_UTF8(re_sv))
PL_reg_flags |= RF_utf8;
else
PL_reg_flags &= ~RF_utf8;
ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */
- ST.prev_rex = rex;
+ ST.prev_rex = rex_sv;
ST.prev_curlyx = cur_curlyx;
- SETREX(rex,re);
+ SETREX(rex_sv,re_sv);
+ rex = re;
rexi = rei;
cur_curlyx = NULL;
ST.B = next;
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);
- SETREX(rex,ST.prev_rex);
+ ReREFCNT_dec(rex_sv);
+ SETREX(rex_sv,ST.prev_rex);
+ rex = (struct regexp *)SvANY(rex_sv);
rexi = RXi_GET(rex);
regcpblow(ST.cp);
cur_eval = ST.prev_eval;
cur_curlyx = ST.prev_curlyx;
+
+ /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
+ PL_reglastparen = &rex->lastparen;
+ PL_reglastcloseparen = &rex->lastcloseparen;
+ /* also update PL_regoffs */
+ PL_regoffs = rex->offs;
+
/* XXXX This is too dramatic a measure... */
PL_reg_maxiter = 0;
if ( nochange_depth )
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);
- SETREX(rex,ST.prev_rex);
+ ReREFCNT_dec(rex_sv);
+ SETREX(rex_sv,ST.prev_rex);
+ rex = (struct regexp *)SvANY(rex_sv);
rexi = RXi_GET(rex);
+ /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
+ PL_reglastparen = &rex->lastparen;
+ PL_reglastcloseparen = &rex->lastcloseparen;
+
PL_reginput = locinput;
REGCP_UNWIND(ST.lastcp);
regcppop(rex);
case BRANCH: /* /(...|A|...)/ */
scan = NEXTOPER(scan); /* scan now points to inner node */
- if ((!next || (OP(next) != BRANCH && OP(next) != BRANCHJ))
- && !has_cutgroup)
- {
- /* last branch; skip state push and jump direct to node */
- continue;
- }
ST.lastparen = *PL_reglastparen;
ST.next_branch = next;
REGCP_SET(ST.cp);
case CUTGROUP:
PL_reginput = locinput;
sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
- (SV*)rexi->data->data[ ARG( scan ) ];
+ MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
PUSH_STATE_GOTO(CUTGROUP_next,next);
/* NOTREACHED */
case CUTGROUP_next_fail:
case CURLYM: /* /A{m,n}B/ where A is fixed-length */
/* This is an optimisation of CURLYX that enables us to push
- * only a single backtracking state, no matter now many matches
+ * only a single backtracking state, no matter how many matches
* there are in {m,n}. It relies on the pattern being constant
* length, with no parens to influence future backrefs
*/
cur_eval->u.eval.close_paren == (U32)ST.me->flags)
goto fake_end;
- if ( ST.count < (ST.minmod ? ARG1(ST.me) : ARG2(ST.me)) )
- goto curlym_do_A; /* try to match another A */
+ {
+ I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
+ if ( max == REG_INFTY || ST.count < max )
+ goto curlym_do_A; /* try to match another A */
+ }
goto curlym_do_B; /* try to match B */
case CURLYM_A_fail: /* just failed to match an A */
case CURLYM_B_fail: /* just failed to match a B */
REGCP_UNWIND(ST.cp);
if (ST.minmod) {
- if (ST.count == ARG2(ST.me) /* max */)
+ I32 max = ARG2(ST.me);
+ if (max != REG_INFTY && ST.count == max)
sayNO;
goto curlym_do_A; /* try to match a further A */
}
= cur_eval->u.eval.toggle_reg_flags;
PL_reg_flags ^= st->u.eval.toggle_reg_flags;
- st->u.eval.prev_rex = rex; /* inner */
- SETREX(rex,cur_eval->u.eval.prev_rex);
+ st->u.eval.prev_rex = rex_sv; /* inner */
+ SETREX(rex_sv,cur_eval->u.eval.prev_rex);
+ rex = (struct regexp *)SvANY(rex_sv);
rexi = RXi_GET(rex);
cur_curlyx = cur_eval->u.eval.prev_curlyx;
- ReREFCNT_inc(rex);
+ ReREFCNT_inc(rex_sv);
st->u.eval.cp = regcppush(0); /* Save *all* the positions. */
+
+ /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */
+ PL_reglastparen = &rex->lastparen;
+ PL_reglastcloseparen = &rex->lastcloseparen;
+
REGCP_SET(st->u.eval.lastcp);
PL_reginput = locinput;
do_ifmatch:
ST.me = scan;
ST.logical = logical;
+ 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 */
case PRUNE:
PL_reginput = locinput;
if (!scan->flags)
- sv_yes_mark = sv_commit = (SV*)rexi->data->data[ ARG( scan ) ];
+ sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
PUSH_STATE_GOTO(COMMIT_next,next);
/* NOTREACHED */
case COMMIT_next_fail:
case MARKPOINT:
ST.prev_mark = mark_state;
ST.mark_name = sv_commit = sv_yes_mark
- = (SV*)rexi->data->data[ ARG( scan ) ];
+ = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
mark_state = st;
ST.mark_loc = PL_reginput = locinput;
PUSH_YES_STATE_GOTO(MARKPOINT_next,next);
otherwise do nothing. Meaning we need to scan
*/
regmatch_state *cur = mark_state;
- SV *find = (SV*)rexi->data->data[ ARG( scan ) ];
+ SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
while (cur) {
if ( sv_eq( cur->u.mark.mark_name,
#undef ST
case FOLDCHAR:
n = ARG(scan);
- if ( n == what_len_TRICKYFOLD(locinput,do_utf8,ln) ) {
+ if ( n == (U32)what_len_TRICKYFOLD(locinput,do_utf8,ln) ) {
locinput += ln;
} else if ( 0xDF == n && !do_utf8 && !UTF ) {
sayNO;
PERL_UNUSED_ARG(depth);
#endif
+ PERL_ARGS_ASSERT_REGREPEAT;
+
scan = PL_reginput;
if (max == REG_INFTY)
max = I32_MAX;
} else {
while (scan < loceol && !isSPACE(*scan))
scan++;
- break;
}
+ break;
case NSPACEL:
PL_reg_flags |= RF_tainted;
if (do_utf8) {
RXi_GET_DECL(prog,progi);
const struct reg_data * const data = prog ? progi->data : NULL;
+ PERL_ARGS_ASSERT_REGCLASS_SWASH;
+
if (data && data->count) {
const U32 n = ARG(node);
if (data->what[n] == 's') {
- SV * const rv = (SV*)data->data[n];
- AV * const av = (AV*)SvRV((SV*)rv);
+ SV * const rv = MUTABLE_SV(data->data[n]);
+ AV * const av = MUTABLE_AV(SvRV(rv));
SV **const ary = AvARRAY(av);
SV **a, **b;
* documentation of these array elements. */
si = *ary;
- a = SvROK(ary[1]) ? &ary[1] : 0;
- b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : 0;
+ a = SvROK(ary[1]) ? &ary[1] : NULL;
+ b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : NULL;
if (a)
sw = *a;
STRLEN len = 0;
STRLEN plen;
+ PERL_ARGS_ASSERT_REGINCLASS;
+
if (do_utf8 && !UTF8_IS_INVARIANT(c)) {
c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &len,
(UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV) | UTF8_CHECK_ONLY);
SV * const sw = regclass_swash(prog, n, TRUE, 0, (SV**)&av);
if (sw) {
- if (swash_fetch(sw, p, do_utf8))
+ U8 * utf8_p;
+ if (do_utf8) {
+ utf8_p = (U8 *) p;
+ } else {
+ STRLEN len = 1;
+ utf8_p = bytes_to_utf8(p, &len);
+ }
+ if (swash_fetch(sw, utf8_p, 1))
match = TRUE;
else if (flags & ANYOF_FOLD) {
if (!match && lenp && av) {
SV* const sv = *av_fetch(av, i, FALSE);
STRLEN len;
const char * const s = SvPV_const(sv, len);
-
- if (len <= plen && memEQ(s, (char*)p, len)) {
+ if (len <= plen && memEQ(s, (char*)utf8_p, len)) {
*lenp = len;
match = TRUE;
break;
}
if (!match) {
U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
- STRLEN tmplen;
- to_utf8_fold(p, tmpbuf, &tmplen);
- if (swash_fetch(sw, tmpbuf, do_utf8))
+ STRLEN tmplen;
+ to_utf8_fold(utf8_p, tmpbuf, &tmplen);
+ if (swash_fetch(sw, tmpbuf, 1))
match = TRUE;
}
}
+
+ /* If we allocated a string above, free it */
+ if (! do_utf8) Safefree(utf8_p);
}
}
if (match && lenp && *lenp == 0)
S_reghop3(U8 *s, I32 off, const U8* lim)
{
dVAR;
+
+ PERL_ARGS_ASSERT_REGHOP3;
+
if (off >= 0) {
while (off-- && s < lim) {
/* XXX could check well-formedness here */
S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim)
{
dVAR;
+
+ PERL_ARGS_ASSERT_REGHOP4;
+
if (off >= 0) {
while (off-- && s < rlim) {
/* XXX could check well-formedness here */
S_reghopmaybe3(U8* s, I32 off, const U8* lim)
{
dVAR;
+
+ PERL_ARGS_ASSERT_REGHOPMAYBE3;
+
if (off >= 0) {
while (off-- && s < lim) {
/* XXX could check well-formedness here */
#ifdef PERL_OLD_COPY_ON_WRITE
rex->saved_copy = PL_nrs;
#endif
- RX_MATCH_COPIED_on(rex);
+ RXp_MATCH_COPIED_on(rex);
}
PL_reg_magic->mg_len = PL_reg_oldpos;
PL_reg_eval_set = 0;
S_to_utf8_substr(pTHX_ register regexp *prog)
{
int i = 1;
+
+ PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
+
do {
if (prog->substrs->data[i].substr
&& !prog->substrs->data[i].utf8_substr) {
{
dVAR;
int i = 1;
+
+ PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
+
do {
if (prog->substrs->data[i].utf8_substr
&& !prog->substrs->data[i].substr) {