Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with
or
.
Download ZIP
tree: b71c54b89a
Fetching contributors…

Cannot retrieve contributors at this time

7404 lines (6690 sloc) 246.504 kb
/* regexec.c
*/
/*
* 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
* also regcomp.c which funnily enough, contains functions for compiling
* a regular expression.
*
* This file is also copied at build time to ext/re/re_exec.c, where
* it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
* This causes the main functions to be compiled under new names and with
* debugging support added, which makes "use re 'debug'" work.
*/
/* NOTE: this is derived from Henry Spencer's regexp code, and should not
* confused with the original package (see point 3 below). Thanks, Henry!
*/
/* Additional note: this code is very heavily munged from Henry's version
* in places. In some spots I've traded clarity for efficiency, so don't
* blame Henry for some of the lack of readability.
*/
/* The names of the functions have been changed from regcomp and
* regexec to pregcomp and pregexec in order to avoid conflicts
* with the POSIX routines of the same names.
*/
#ifdef PERL_EXT_RE_BUILD
#include "re_top.h"
#endif
/*
* pregcomp and pregexec -- regsub and regerror are not used in perl
*
* Copyright (c) 1986 by University of Toronto.
* Written by Henry Spencer. Not derived from licensed software.
*
* Permission is granted to anyone to use this software for any
* purpose on any computer system, and to redistribute it freely,
* subject to the following restrictions:
*
* 1. The author is not responsible for the consequences of use of
* this software, no matter how awful, even if they arise
* from defects in it.
*
* 2. The origin of this software must not be misrepresented, either
* by explicit claim or by omission.
*
* 3. Altered versions must be plainly marked as such, and must not
* be misrepresented as being the original software.
*
**** 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, 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.
*
* Beware that some of this code is subtly aware of the way operator
* precedence is structured in regular expressions. Serious changes in
* regular-expression syntax might require a total rethink.
*/
#include "EXTERN.h"
#define PERL_IN_REGEXEC_C
#include "perl.h"
#ifdef PERL_IN_XSUB_RE
# include "re_comp.h"
#else
# 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 RF_utf8 8 /* Pattern contains multibyte chars? */
#define UTF_PATTERN ((PL_reg_flags & RF_utf8) != 0)
#ifndef STATIC
#define STATIC static
#endif
/* Valid for non-utf8 strings, non-ANYOFV nodes only: avoids the reginclass
* call if there are no complications: i.e., if everything matchable is
* straight forward in the bitmap */
#define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,0,0) \
: ANYOF_BITMAP_TEST(p,*(c)))
/*
* Forwards.
*/
#define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
#define CHR_DIST(a,b) (PL_reg_match_utf8 ? utf8_distance(a,b) : a - b)
#define HOPc(pos,off) \
(char *)(PL_reg_match_utf8 \
? reghop3((U8*)pos, off, (U8*)(off >= 0 ? PL_regeol : PL_bostr)) \
: (U8*)(pos + off))
#define HOPBACKc(pos, off) \
(char*)(PL_reg_match_utf8\
? reghopmaybe3((U8*)pos, -off, (U8*)PL_bostr) \
: (pos - off >= PL_bostr) \
? (U8*)pos - off \
: NULL)
#define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
#define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
/* these are unrolled below in the CCC_TRY_XXX defined */
#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
/* Doesn't do an assert to verify that is correct */
#define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \
if (!CAT2(PL_utf8_,class)) { \
bool throw_away PERL_UNUSED_DECL; \
ENTER; save_re_context(); \
throw_away = CAT2(is_utf8_,class)((const U8*)" "); \
LEAVE; } } STMT_END
#define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a")
#define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0")
#define LOAD_UTF8_CHARCLASS_SPACE() LOAD_UTF8_CHARCLASS(space," ")
#define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
/* 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 */
/* The actual code for CCC_TRY, which uses several variables from the routine
* it's callable from. It is designed to be the bulk of a case statement.
* FUNC is the macro or function to call on non-utf8 targets that indicate if
* nextchr matches the class.
* UTF8_TEST is the whole test string to use for utf8 targets
* LOAD is what to use to test, and if not present to load in the swash for the
* class
* POS_OR_NEG is either empty or ! to complement the results of FUNC or
* UTF8_TEST test.
* The logic is: Fail if we're at the end-of-string; otherwise if the target is
* utf8 and a variant, load the swash if necessary and test using the utf8
* test. Advance to the next character if test is ok, otherwise fail; If not
* utf8 or an invariant under utf8, use the non-utf8 test, and fail if it
* fails, or advance to the next character */
#define _CCC_TRY_CODE(POS_OR_NEG, FUNC, UTF8_TEST, CLASS, STR) \
if (locinput >= PL_regeol) { \
sayNO; \
} \
if (utf8_target && UTF8_IS_CONTINUED(nextchr)) { \
LOAD_UTF8_CHARCLASS(CLASS, STR); \
if (POS_OR_NEG (UTF8_TEST)) { \
sayNO; \
} \
locinput += PL_utf8skip[nextchr]; \
nextchr = UCHARAT(locinput); \
break; \
} \
if (POS_OR_NEG (FUNC(nextchr))) { \
sayNO; \
} \
nextchr = UCHARAT(++locinput); \
break;
/* 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.
* This is because that code fails when the test succeeds, so we want to have
* the test fail so that the code succeeds. The swash is stored in a
* predictable PL_ place */
#define _CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, \
CLASS, STR) \
case NAME: \
_CCC_TRY_CODE( !, FUNC, \
cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
(U8*)locinput, TRUE)), \
CLASS, STR) \
case NNAME: \
_CCC_TRY_CODE( PLACEHOLDER , FUNC, \
cBOOL(swash_fetch(CAT2(PL_utf8_,CLASS), \
(U8*)locinput, TRUE)), \
CLASS, STR) \
/* Generate the case statements for both locale and non-locale character
* classes in regmatch for classes that don't have special unicode semantics.
* Locales don't use an immediate swash, but an intermediary special locale
* function that is called on the pointer to the current place in the input
* string. That function will resolve to needing the same swash. One might
* think that because we don't know what the locale will match, we shouldn't
* check with the swash loading function that it loaded properly; ie, that we
* should use LOAD_UTF8_CHARCLASS_NO_CHECK for those, but what is passed to the
* regular LOAD_UTF8_CHARCLASS is in non-locale terms, and so locale is
* irrelevant here */
#define CCC_TRY(NAME, NNAME, FUNC, \
NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
NAMEA, NNAMEA, FUNCA, \
CLASS, STR) \
case NAMEL: \
PL_reg_flags |= RF_tainted; \
_CCC_TRY_CODE( !, LCFUNC, LCFUNC_utf8((U8*)locinput), CLASS, STR) \
case NNAMEL: \
PL_reg_flags |= RF_tainted; \
_CCC_TRY_CODE( PLACEHOLDER, LCFUNC, LCFUNC_utf8((U8*)locinput), \
CLASS, STR) \
case NAMEA: \
if (locinput >= PL_regeol || ! FUNCA(nextchr)) { \
sayNO; \
} \
/* Matched a utf8-invariant, so don't have to worry about utf8 */ \
nextchr = UCHARAT(++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; \
/* Generate the non-locale cases */ \
_CCC_TRY_NONLOCALE(NAME, NNAME, FUNC, CLASS, STR)
/* This is like CCC_TRY, but has an extra set of parameters for generating case
* statements to handle separate Unicode semantics nodes */
#define CCC_TRY_U(NAME, NNAME, FUNC, \
NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
NAMEU, NNAMEU, FUNCU, \
NAMEA, NNAMEA, FUNCA, \
CLASS, STR) \
CCC_TRY(NAME, NNAME, FUNC, \
NAMEL, NNAMEL, LCFUNC, LCFUNC_utf8, \
NAMEA, NNAMEA, FUNCA, \
CLASS, STR) \
_CCC_TRY_NONLOCALE(NAMEU, NNAMEU, FUNCU, CLASS, STR)
/* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
/* for use after a quantifier and before an EXACT-like node -- japhy */
/* it would be nice to rework regcomp.sym to generate this stuff. sigh
*
* NOTE that *nothing* that affects backtracking should be in here, specifically
* VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
* node that is in between two EXACT like nodes when ascertaining what the required
* "follow" character is. This should probably be moved to regex compile time
* although it may be done at run time beause of the REF possibility - more
* investigation required. -- demerphq
*/
#define JUMPABLE(rn) ( \
OP(rn) == OPEN || \
(OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
OP(rn) == EVAL || \
OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
OP(rn) == PLUS || OP(rn) == MINMOD || \
OP(rn) == KEEPS || \
(PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
)
#define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
#define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
#if 0
/* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
we don't need this definition. */
#define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
#define IS_TEXTF(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn)==EXACTFA || OP(rn)==EXACTF || OP(rn)==REFF || OP(rn)==NREFF )
#define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
#else
/* ... so we use this as its faster. */
#define IS_TEXT(rn) ( OP(rn)==EXACT )
#define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn)==EXACTFU_TRICKYFOLD || OP(rn) == EXACTFA)
#define IS_TEXTF(rn) ( OP(rn)==EXACTF )
#define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
#endif
/*
Search for mandatory following text node; for lookahead, the text must
follow but for lookbehind (rn->flags != 0) we skip to the next step.
*/
#define FIND_NEXT_IMPT(rn) STMT_START { \
while (JUMPABLE(rn)) { \
const OPCODE type = OP(rn); \
if (type == SUSPEND || PL_regkind[type] == CURLY) \
rn = NEXTOPER(NEXTOPER(rn)); \
else if (type == PLUS) \
rn = NEXTOPER(rn); \
else if (type == IFMATCH) \
rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
else rn += NEXT_OFF(rn); \
} \
} STMT_END
static void restore_pos(pTHX_ void *arg);
#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. */
STATIC CHECKPOINT
S_regcppush(pTHX_ const regexp *rex, I32 parenfloor)
{
dVAR;
const int retval = PL_savestack_ix;
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;
I32 p;
GET_RE_DEBUG_FLAGS_DECL;
PERL_ARGS_ASSERT_REGCPPUSH;
if (paren_elems_to_push < 0)
Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0",
paren_elems_to_push);
if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
" out of range (%lu-%ld)",
total_elems, (unsigned long)PL_regsize, (long)parenfloor);
SSGROW(total_elems + REGCP_FRAME_ELEMS);
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);
SSPUSHINT(rex->offs[p].start_tmp);
DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
" \\%"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);
SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
return retval;
}
/* These are needed since we do not localize EVAL nodes: */
#define REGCP_SET(cp) \
DEBUG_STATE_r( \
PerlIO_printf(Perl_debug_log, \
" Setting an EVAL scope, savestack=%"IVdf"\n", \
(IV)PL_savestack_ix)); \
cp = PL_savestack_ix
#define REGCP_UNWIND(cp) \
DEBUG_STATE_r( \
if (cp != PL_savestack_ix) \
PerlIO_printf(Perl_debug_log, \
" Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
(IV)(cp), (IV)PL_savestack_ix)); \
regcpblow(cp)
#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;
U32 paren;
GET_RE_DEBUG_FLAGS_DECL;
PERL_ARGS_ASSERT_REGCPPOP;
/* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
i = SSPOPUV;
assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
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;
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,
" \\%"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--;
}
#if 1
/* It would seem that the similar code in regtry()
* already takes care of this, and in fact it is in
* a better location to since this code can #if 0-ed out
* but the code in regtry() is needed or otherwise tests
* 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
* this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
* --jhi updated by dapm */
for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
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
}
/* 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. */
/*
* pregexec and friends
*/
#ifndef PERL_IN_XSUB_RE
/*
- pregexec - match a regexp against a string
*/
I32
Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, register char *strend,
char *strbeg, I32 minend, SV *screamer, U32 nosave)
/* 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;
return
regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
nosave ? 0 : REXEC_COPY_STR);
}
#endif
/*
* Need to implement the following flags for reg_anch:
*
* USE_INTUIT_NOML - Useful to call re_intuit_start() first
* USE_INTUIT_ML
* INTUIT_AUTORITATIVE_NOML - Can trust a positive answer
* INTUIT_AUTORITATIVE_ML
* INTUIT_ONCE_NOML - Intuit can match in one location only.
* INTUIT_ONCE_ML
*
* Another flag for this function: SECOND_TIME (so that float substrs
* with giant delta may be not rechecked).
*/
/* Assumptions: if ANCH_GPOS, then strpos is anchored. XXXX Check GPOS logic */
/* If SCREAM, then SvPVX_const(sv) should be compatible with strpos and strend.
Otherwise, only SvCUR(sv) is used to get strbeg. */
/* XXXX We assume that strpos is strbeg unless sv. */
/* XXXX Some places assume that there is a fixed substring.
An update may be needed if optimizer marks as "INTUITable"
RExen without fixed substrings. Similarly, it is assumed that
lengths of all the strings are no more than minlen, thus they
cannot come from lookahead.
(Or minlen should take into account lookahead.)
NOTE: Some of this comment is not correct. minlen does now take account
of lookahead/behind. Further research is required. -- demerphq
*/
/* A failure to find a constant substring means that there is no need to make
an expensive call to REx engine, thus we celebrate a failure. Similarly,
finding a substring too deep into the string means that less calls to
regtry() should be needed.
REx compiler's optimizer found 4 possible hints:
a) Anchored substring;
b) Fixed substring;
c) Whether we are anchored (beginning-of-line or \G);
d) First node (of those at offset 0) which may distinguish positions;
We use a)b)d) and multiline-part of c), and try to find a position in the
string which does not contradict any of them.
*/
/* Most of decisions we do here should have been done at compile time.
The nodes of the REx which we used for the search should have been
deleted from the finite automaton. */
char *
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);
I32 start_shift = 0;
/* Should be nonnegative! */
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;
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;
RXi_GET_DECL(prog,progi);
#ifdef DEBUGGING
const char * const i_strpos = strpos;
#endif
GET_RE_DEBUG_FLAGS_DECL;
PERL_ARGS_ASSERT_RE_INTUIT_START;
PERL_UNUSED_ARG(flags);
PERL_UNUSED_ARG(data);
RX_MATCH_UTF8_set(rx,utf8_target);
if (RX_UTF8(rx)) {
PL_reg_flags |= RF_utf8;
}
DEBUG_EXECUTE_r(
debug_start_match(rx, utf8_target, strpos, strend,
sv ? "Guessing start of match in sv for"
: "Guessing start of match in string for");
);
/* CHR_DIST() would be more correct here but it makes things slow. */
if (prog->minlen > strend - strpos) {
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
"String too short... [re_intuit_start]\n"));
goto fail;
}
strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos;
PL_regeol = strend;
if (utf8_target) {
if (!prog->check_utf8 && prog->check_substr)
to_utf8_substr(prog);
check = prog->check_utf8;
} else {
if (!prog->check_substr && prog->check_utf8)
to_byte_substr(prog);
check = prog->check_substr;
}
if (check == &PL_sv_undef) {
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
"Non-utf8 string cannot match utf8 check string\n"));
goto fail;
}
if (prog->extflags & RXf_ANCH) { /* Match at beg-of-str or after \n */
ml_anch = !( (prog->extflags & RXf_ANCH_SINGLE)
|| ( (prog->extflags & RXf_ANCH_BOL)
&& !multiline ) ); /* Check after \n? */
if (!ml_anch) {
if ( !(prog->extflags & RXf_ANCH_GPOS) /* Checked by the caller */
&& !(prog->intflags & PREGf_IMPLICIT) /* not a real BOL */
/* SvCUR is not set on references: SvRV and SvPVX_const overlap */
&& sv && !SvROK(sv)
&& (strpos != strbeg)) {
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Not at start...\n"));
goto fail;
}
if (prog->check_offset_min == prog->check_offset_max &&
!(prog->extflags & RXf_CANY_SEEN)) {
/* Substring at constant offset from beg-of-str... */
I32 slen;
s = HOP3c(strpos, prog->check_offset_min, strend);
if (SvTAIL(check)) {
slen = SvCUR(check); /* >= 1 */
if ( strend - s > slen || strend - s < slen - 1
|| (strend - s == slen && strend[-1] != '\n')) {
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too long...\n"));
goto fail_finish;
}
/* Now should match s[0..slen-2] */
slen--;
if (slen && (*SvPVX_const(check) != *s
|| (slen > 1
&& memNE(SvPVX_const(check), s, slen)))) {
report_neq:
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String not equal...\n"));
goto fail_finish;
}
}
else if (*SvPVX_const(check) != *s
|| ((slen = SvCUR(check)) > 1
&& memNE(SvPVX_const(check), s, slen)))
goto report_neq;
check_at = s;
goto success_at_start;
}
}
/* Match is anchored, but substr is not anchored wrt beg-of-str. */
s = strpos;
start_shift = prog->check_offset_min; /* okay to underestimate on CC */
end_shift = prog->check_end_shift;
if (!ml_anch) {
const I32 end = prog->check_offset_max + CHR_SVLEN(check)
- (SvTAIL(check) != 0);
const I32 eshift = CHR_DIST((U8*)strend, (U8*)s) - end;
if (end_shift < eshift)
end_shift = eshift;
}
}
else { /* Can match at random position */
ml_anch = 0;
s = strpos;
start_shift = prog->check_offset_min; /* okay to underestimate on CC */
end_shift = prog->check_end_shift;
/* end shift should be non negative here */
}
#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, RX_PRECOMP(prog));
#endif
restart:
/* Find a possible match in the region s..strend by looking for
the "check" substring in the region corrected by start/end_shift. */
{
I32 srch_start_shift = start_shift;
I32 srch_end_shift = end_shift;
U8* start_point;
U8* end_point;
if (srch_start_shift < 0 && strbeg - s > srch_start_shift) {
srch_end_shift -= ((strbeg - s) - srch_start_shift);
srch_start_shift = strbeg - s;
}
DEBUG_OPTIMISE_MORE_r({
PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n",
(IV)prog->check_offset_min,
(IV)srch_start_shift,
(IV)srch_end_shift,
(IV)prog->check_end_shift);
});
if (prog->extflags & RXf_CANY_SEEN) {
start_point= (U8*)(s + srch_start_shift);
end_point= (U8*)(strend - srch_end_shift);
} else {
start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend);
end_point= HOP3(strend, -srch_end_shift, strbeg);
}
DEBUG_OPTIMISE_MORE_r({
PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n",
(int)(end_point - start_point),
(int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
start_point);
});
s = fbm_instr( start_point, end_point,
check, multiline ? FBMrf_MULTILINE : 0);
}
/* Update the count-of-usability, remove useless subpatterns,
unshift s. */
DEBUG_EXECUTE_r({
RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s",
(s ? "Found" : "Did not find"),
(check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
? "anchored" : "floating"),
quoted,
RE_SV_TAIL(check),
(s ? " at offset " : "...\n") );
});
if (!s)
goto fail_finish;
/* Finish the diagnostic message */
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) );
/* XXX dmq: first branch is for positive lookbehind...
Our check string is offset from the beginning of the pattern.
So we need to do any stclass tests offset forward from that
point. I think. :-(
*/
check_at=s;
/* Got a candidate. Check MBOL anchoring, and the *other* substr.
Start with the other substr.
XXXX no SCREAM optimization yet - and a very coarse implementation
XXXX /ttx+/ results in anchored="ttx", floating="x". floating will
*always* match. Probably should be marked during compile...
Probably it is right to do no SCREAM here...
*/
if (utf8_target ? (prog->float_utf8 && prog->anchored_utf8)
: (prog->float_substr && prog->anchored_substr))
{
/* Take into account the "other" substring. */
/* XXXX May be hopelessly wrong for UTF... */
if (!other_last)
other_last = strpos;
if (check == (utf8_target ? prog->float_utf8 : prog->float_substr)) {
do_other_anchored:
{
char * const last = HOP3c(s, -start_shift, strbeg);
char *last1, *last2;
char * const saved_s = s;
SV* must;
t = s - prog->check_offset_max;
if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
&& (!utf8_target
|| ((t = (char*)reghopmaybe3((U8*)s, -(prog->check_offset_max), (U8*)strpos))
&& t > strpos)))
NOOP;
else
t = strpos;
t = HOP3c(t, prog->anchored_offset, strend);
if (t < other_last) /* These positions already checked */
t = other_last;
last2 = last1 = HOP3c(strend, -prog->minlen, strbeg);
if (last < last1)
last1 = last;
/* XXXX It is not documented what units *_offsets are in.
We assume bytes, but this is clearly wrong.
Meaning this code needs to be carefully reviewed for errors.
dmq.
*/
/* On end-of-str: see comment below. */
must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
if (must == &PL_sv_undef) {
s = (char*)NULL;
DEBUG_r(must = prog->anchored_utf8); /* for debug */
}
else
s = fbm_instr(
(unsigned char*)t,
HOP3(HOP3(last1, prog->anchored_offset, strend)
+ SvCUR(must), -(SvTAIL(must)!=0), strbeg),
must,
multiline ? FBMrf_MULTILINE : 0
);
DEBUG_EXECUTE_r({
RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s",
(s ? "Found" : "Contradicts"),
quoted, RE_SV_TAIL(must));
});
if (!s) {
if (last1 >= last2) {
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
", giving up...\n"));
goto fail_finish;
}
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
", trying floating at offset %ld...\n",
(long)(HOP3c(saved_s, 1, strend) - i_strpos)));
other_last = HOP3c(last1, prog->anchored_offset+1, strend);
s = HOP3c(last, 1, strend);
goto restart;
}
else {
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
(long)(s - i_strpos)));
t = HOP3c(s, -prog->anchored_offset, strbeg);
other_last = HOP3c(s, 1, strend);
s = saved_s;
if (t == strpos)
goto try_at_start;
goto try_at_offset;
}
}
}
else { /* Take into account the floating substring. */
char *last, *last1;
char * const saved_s = s;
SV* must;
t = HOP3c(s, -start_shift, strbeg);
last1 = last =
HOP3c(strend, -prog->minlen + prog->float_min_offset, strbeg);
if (CHR_DIST((U8*)last, (U8*)t) > prog->float_max_offset)
last = HOP3c(t, prog->float_max_offset, strend);
s = HOP3c(t, prog->float_min_offset, strend);
if (s < other_last)
s = other_last;
/* XXXX It is not documented what units *_offsets are in. Assume bytes. */
must = utf8_target ? prog->float_utf8 : prog->float_substr;
/* fbm_instr() takes into account exact value of end-of-str
if the check is SvTAIL(ed). Since false positives are OK,
and end-of-str is not later than strend we are OK. */
if (must == &PL_sv_undef) {
s = (char*)NULL;
DEBUG_r(must = prog->float_utf8); /* for debug message */
}
else
s = fbm_instr((unsigned char*)s,
(unsigned char*)last + SvCUR(must)
- (SvTAIL(must)!=0),
must, multiline ? FBMrf_MULTILINE : 0);
DEBUG_EXECUTE_r({
RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
PerlIO_printf(Perl_debug_log, "%s floating substr %s%s",
(s ? "Found" : "Contradicts"),
quoted, RE_SV_TAIL(must));
});
if (!s) {
if (last1 == last) {
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
", giving up...\n"));
goto fail_finish;
}
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
", trying anchored starting at offset %ld...\n",
(long)(saved_s + 1 - i_strpos)));
other_last = last;
s = HOP3c(t, 1, strend);
goto restart;
}
else {
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
(long)(s - i_strpos)));
other_last = s; /* Fix this later. --Hugo */
s = saved_s;
if (t == strpos)
goto try_at_start;
goto try_at_offset;
}
}
}
t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos);
DEBUG_OPTIMISE_MORE_r(
PerlIO_printf(Perl_debug_log,
"Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n",
(IV)prog->check_offset_min,
(IV)prog->check_offset_max,
(IV)(s-strpos),
(IV)(t-strpos),
(IV)(t-s),
(IV)(strend-strpos)
)
);
if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */
&& (!utf8_target
|| ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos)))
&& t > strpos)))
{
/* Fixed substring is found far enough so that the match
cannot start at strpos. */
try_at_offset:
if (ml_anch && t[-1] != '\n') {
/* Eventually fbm_*() should handle this, but often
anchored_offset is not 0, so this check will not be wasted. */
/* XXXX In the code below we prefer to look for "^" even in
presence of anchored substrings. And we search even
beyond the found float position. These pessimizations
are historical artefacts only. */
find_anchor:
while (t < strend - prog->minlen) {
if (*t == '\n') {
if (t < check_at - prog->check_offset_min) {
if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
/* Since we moved from the found position,
we definitely contradict the found anchored
substr. Due to the above check we do not
contradict "check" substr.
Thus we can arrive here only if check substr
is float. Redo checking for "other"=="fixed".
*/
strpos = t + 1;
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
PL_colors[0], PL_colors[1], (long)(strpos - i_strpos), (long)(strpos - i_strpos + prog->anchored_offset)));
goto do_other_anchored;
}
/* We don't contradict the found floating substring. */
/* XXXX Why not check for STCLASS? */
s = t + 1;
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m at offset %ld...\n",
PL_colors[0], PL_colors[1], (long)(s - i_strpos)));
goto set_useful;
}
/* Position contradicts check-string */
/* XXXX probably better to look for check-string
than for "\n", so one should lower the limit for t? */
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
PL_colors[0], PL_colors[1], (long)(t + 1 - i_strpos)));
other_last = strpos = s = t + 1;
goto restart;
}
t++;
}
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Did not find /%s^%s/m...\n",
PL_colors[0], PL_colors[1]));
goto fail_finish;
}
else {
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Starting position does not contradict /%s^%s/m...\n",
PL_colors[0], PL_colors[1]));
}
s = t;
set_useful:
++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
}
else {
/* The found string does not prohibit matching at strpos,
- no optimization of calling REx engine can be performed,
unless it was an MBOL and we are not after MBOL,
or a future STCLASS check will fail this. */
try_at_start:
/* Even in this situation we may use MBOL flag if strpos is offset
wrt the start of the string. */
if (ml_anch && sv && !SvROK(sv) /* See prev comment on SvROK */
&& (strpos != strbeg) && strpos[-1] != '\n'
/* May be due to an implicit anchor of m{.*foo} */
&& !(prog->intflags & PREGf_IMPLICIT))
{
t = strpos;
goto find_anchor;
}
DEBUG_EXECUTE_r( if (ml_anch)
PerlIO_printf(Perl_debug_log, "Position at offset %ld does not contradict /%s^%s/m...\n",
(long)(strpos - i_strpos), PL_colors[0], PL_colors[1]);
);
success_at_start:
if (!(prog->intflags & PREGf_NAUGHTY) /* XXXX If strpos moved? */
&& (utf8_target ? (
prog->check_utf8 /* Could be deleted already */
&& --BmUSEFUL(prog->check_utf8) < 0
&& (prog->check_utf8 == prog->float_utf8)
) : (
prog->check_substr /* Could be deleted already */
&& --BmUSEFUL(prog->check_substr) < 0
&& (prog->check_substr == prog->float_substr)
)))
{
/* If flags & SOMETHING - do not do it many times on the same match */
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "... Disabling check substring...\n"));
/* XXX Does the destruction order has to change with utf8_target? */
SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
prog->check_substr = prog->check_utf8 = NULL; /* disable */
prog->float_substr = prog->float_utf8 = NULL; /* clear */
check = NULL; /* abort */
s = strpos;
/* XXXX If the check string was an implicit check MBOL, then we need to unset the relevant flag
see http://bugs.activestate.com/show_bug.cgi?id=87173 */
if (prog->intflags & PREGf_IMPLICIT)
prog->extflags &= ~RXf_ANCH_MBOL;
/* XXXX This is a remnant of the old implementation. It
looks wasteful, since now INTUIT can use many
other heuristics. */
prog->extflags &= ~RXf_USE_INTUIT;
/* XXXX What other flags might need to be cleared in this branch? */
}
else
s = strpos;
}
/* Last resort... */
/* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */
/* trie stclasses are too expensive to use here, we are better off to
leave it to regmatch itself */
if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
/* minlen == 0 is possible if regstclass is \b or \B,
and the fixed substr is ''$.
Since minlen is already taken into account, s+1 is before strend;
accidentally, minlen >= 1 guaranties no false positives at s + 1
even for \b or \B. But (minlen? 1 : 0) below assumes that
regstclass does not come from lookahead... */
/* If regstclass takes bytelength more than 1: If charlength==1, OK.
This leaves EXACTF-ish only, which are dealt with in find_byclass(). */
const U8* const str = (U8*)STRING(progi->regstclass);
const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
? CHR_DIST(str+STR_LEN(progi->regstclass), str)
: 1);
char * endpos;
if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend);
else if (prog->float_substr || prog->float_utf8)
endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend);
else
endpos= strend;
if (checked_upto < s)
checked_upto = s;
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" s: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
(IV)start_shift, (IV)(check_at - strbeg), (IV)(s - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
t = s;
s = find_byclass(prog, progi->regstclass, checked_upto, endpos, NULL);
if (s) {
checked_upto = s;
} else {
#ifdef DEBUGGING
const char *what = NULL;
#endif
if (endpos == strend) {
DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
"Could not match STCLASS...\n") );
goto fail;
}
DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
"This position contradicts STCLASS...\n") );
if ((prog->extflags & RXf_ANCH) && !ml_anch)
goto fail;
checked_upto = HOPBACKc(endpos, start_shift);
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %"IVdf" endpos: %"IVdf" checked_upto: %"IVdf"\n",
(IV)start_shift, (IV)(check_at - strbeg), (IV)(endpos - strbeg), (IV)(checked_upto- strbeg)));
/* Contradict one of substrings */
if (prog->anchored_substr || prog->anchored_utf8) {
if ((utf8_target ? prog->anchored_utf8 : prog->anchored_substr) == check) {
DEBUG_EXECUTE_r( what = "anchored" );
hop_and_restart:
s = HOP3c(t, 1, strend);
if (s + start_shift + end_shift > strend) {
/* XXXX Should be taken into account earlier? */
DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
"Could not match STCLASS...\n") );
goto fail;
}
if (!check)
goto giveup;
DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
"Looking for %s substr starting at offset %ld...\n",
what, (long)(s + start_shift - i_strpos)) );
goto restart;
}
/* Have both, check_string is floating */
if (t + start_shift >= check_at) /* Contradicts floating=check */
goto retry_floating_check;
/* Recheck anchored substring, but not floating... */
s = check_at;
if (!check)
goto giveup;
DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
"Looking for anchored substr starting at offset %ld...\n",
(long)(other_last - i_strpos)) );
goto do_other_anchored;
}
/* Another way we could have checked stclass at the
current position only: */
if (ml_anch) {
s = t = t + 1;
if (!check)
goto giveup;
DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
"Looking for /%s^%s/m starting at offset %ld...\n",
PL_colors[0], PL_colors[1], (long)(t - i_strpos)) );
goto try_at_offset;
}
if (!(utf8_target ? prog->float_utf8 : prog->float_substr)) /* Could have been deleted */
goto fail;
/* Check is floating substring. */
retry_floating_check:
t = check_at - start_shift;
DEBUG_EXECUTE_r( what = "floating" );
goto hop_and_restart;
}
if (t != s) {
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
"By STCLASS: moving %ld --> %ld\n",
(long)(t - i_strpos), (long)(s - i_strpos))
);
}
else {
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
"Does not contradict STCLASS...\n");
);
}
}
giveup:
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s%s:%s match at offset %ld\n",
PL_colors[4], (check ? "Guessed" : "Giving up"),
PL_colors[5], (long)(s - i_strpos)) );
return s;
fail_finish: /* Substring not found */
if (prog->check_substr || prog->check_utf8) /* could be removed already */
BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
fail:
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
PL_colors[4], PL_colors[5]));
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) \
? (utf8_target ? trie_utf8 : trie_plain) \
: (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
#define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
STRLEN skiplen; \
switch (trie_type) { \
case trie_utf8_fold: \
if ( foldlen>0 ) { \
uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
foldlen -= len; \
uscan += len; \
len=0; \
} else { \
uvc = to_utf8_fold( (const U8*) uc, foldbuf, &foldlen ); \
len = UTF8SKIP(uc); \
skiplen = UNISKIP( uvc ); \
foldlen -= skiplen; \
uscan = foldbuf + skiplen; \
} \
break; \
case trie_latin_utf8_fold: \
if ( foldlen>0 ) { \
uvc = utf8n_to_uvuni( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
foldlen -= len; \
uscan += len; \
len=0; \
} else { \
len = 1; \
uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, 1); \
skiplen = UNISKIP( uvc ); \
foldlen -= skiplen; \
uscan = foldbuf + skiplen; \
} \
break; \
case trie_utf8: \
uvc = utf8n_to_uvuni( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
break; \
case trie_plain: \
uvc = (UV)*uc; \
len = 1; \
} \
if (uvc < 256) { \
charid = trie->charmap[ uvc ]; \
} \
else { \
charid = 0; \
if (widecharmap) { \
SV** const svpp = hv_fetch(widecharmap, \
(char*)&uvc, sizeof(UV), 0); \
if (svpp) \
charid = (U16)SvIV(*svpp); \
} \
} \
} STMT_END
#define REXEC_FBC_EXACTISH_SCAN(CoNd) \
STMT_START { \
while (s <= e) { \
if ( (CoNd) \
&& (ln == 1 || folder(s, pat_string, ln)) \
&& (!reginfo || regtry(reginfo, &s)) ) \
goto got_it; \
s++; \
} \
} STMT_END
#define REXEC_FBC_UTF8_SCAN(CoDe) \
STMT_START { \
while (s + (uskip = UTF8SKIP(s)) <= strend) { \
CoDe \
s += uskip; \
} \
} STMT_END
#define REXEC_FBC_SCAN(CoDe) \
STMT_START { \
while (s < strend) { \
CoDe \
s++; \
} \
} STMT_END
#define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \
REXEC_FBC_UTF8_SCAN( \
if (CoNd) { \
if (tmp && (!reginfo || regtry(reginfo, &s))) \
goto got_it; \
else \
tmp = doevery; \
} \
else \
tmp = 1; \
)
#define REXEC_FBC_CLASS_SCAN(CoNd) \
REXEC_FBC_SCAN( \
if (CoNd) { \
if (tmp && (!reginfo || regtry(reginfo, &s))) \
goto got_it; \
else \
tmp = doevery; \
} \
else \
tmp = 1; \
)
#define REXEC_FBC_TRYIT \
if ((!reginfo || regtry(reginfo, &s))) \
goto got_it
#define REXEC_FBC_CSCAN(CoNdUtF8,CoNd) \
if (utf8_target) { \
REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
} \
else { \
REXEC_FBC_CLASS_SCAN(CoNd); \
}
#define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \
if (utf8_target) { \
UtFpReLoAd; \
REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
} \
else { \
REXEC_FBC_CLASS_SCAN(CoNd); \
}
#define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \
PL_reg_flags |= RF_tainted; \
if (utf8_target) { \
REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \
} \
else { \
REXEC_FBC_CLASS_SCAN(CoNd); \
}
#define DUMP_EXEC_POS(li,s,doutf8) \
dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8)
#define UTF8_NOLOAD(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
tmp = TEST_NON_UTF8(tmp); \
REXEC_FBC_UTF8_SCAN( \
if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
tmp = !tmp; \
IF_SUCCESS; \
} \
else { \
IF_FAIL; \
} \
); \
#define UTF8_LOAD(TeSt1_UtF8, TeSt2_UtF8, IF_SUCCESS, IF_FAIL) \
if (s == PL_bostr) { \
tmp = '\n'; \
} \
else { \
U8 * const r = reghop3((U8*)s, -1, (U8*)PL_bostr); \
tmp = utf8n_to_uvchr(r, UTF8SKIP(r), 0, UTF8_ALLOW_DEFAULT); \
} \
tmp = TeSt1_UtF8; \
LOAD_UTF8_CHARCLASS_ALNUM(); \
REXEC_FBC_UTF8_SCAN( \
if (tmp == ! (TeSt2_UtF8)) { \
tmp = !tmp; \
IF_SUCCESS; \
} \
else { \
IF_FAIL; \
} \
); \
/* The only difference between the BOUND and NBOUND cases is that
* REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
* NBOUND. This is accomplished by passing it in either the if or else clause,
* with the other one being empty */
#define FBC_BOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
#define FBC_BOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
#define FBC_NBOUND(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
FBC_BOUND_COMMON(UTF8_LOAD(TEST1_UTF8, TEST2_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
#define FBC_NBOUND_NOLOAD(TEST_NON_UTF8, TEST1_UTF8, TEST2_UTF8) \
FBC_BOUND_COMMON(UTF8_NOLOAD(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
/* Common to the BOUND and NBOUND cases. Unfortunately the UTF8 tests need to
* be passed in completely with the variable name being tested, which isn't
* such a clean interface, but this is easier to read than it was before. We
* are looking for the boundary (or non-boundary between a word and non-word
* character. The utf8 and non-utf8 cases have the same logic, but the details
* must be different. Find the "wordness" of the character just prior to this
* one, and compare it with the wordness of this one. If they differ, we have
* a boundary. At the beginning of the string, pretend that the previous
* character was a new-line */
#define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
if (utf8_target) { \
UTF8_CODE \
} \
else { /* Not utf8 */ \
tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; \
tmp = TEST_NON_UTF8(tmp); \
REXEC_FBC_SCAN( \
if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
tmp = !tmp; \
IF_SUCCESS; \
} \
else { \
IF_FAIL; \
} \
); \
} \
if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, &s))) \
goto got_it;
/* We know what class REx starts with. Try to find this position... */
/* if reginfo is NULL, its a dryrun */
/* annoyingly all the vars in this routine have different names from their counterparts
in regmatch. /grrr */
STATIC char *
S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
const char *strend, regmatch_info *reginfo)
{
dVAR;
const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
char *pat_string; /* The pattern's exactish string */
char *pat_end; /* ptr to end char of pat_string */
re_fold_t folder; /* Function for computing non-utf8 folds */
const U8 *fold_array; /* array for folding ords < 256 */
STRLEN ln;
STRLEN lnc;
STRLEN uskip;
U8 c1;
U8 c2;
char *e;
I32 tmp = 1; /* Scratch variable? */
const bool utf8_target = PL_reg_match_utf8;
UV utf8_fold_flags = 0;
RXi_GET_DECL(prog,progi);
PERL_ARGS_ASSERT_FIND_BYCLASS;
/* We know what class it must start with. */
switch (OP(c)) {
case ANYOFV:
case ANYOF:
if (utf8_target || OP(c) == ANYOFV) {
STRLEN inclasslen = strend - s;
REXEC_FBC_UTF8_CLASS_SCAN(
reginclass(prog, c, (U8*)s, &inclasslen, utf8_target));
}
else {
REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
}
break;
case CANY:
REXEC_FBC_SCAN(
if (tmp && (!reginfo || regtry(reginfo, &s)))
goto got_it;
else
tmp = doevery;
);
break;
case EXACTFA:
if (UTF_PATTERN || utf8_target) {
utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
goto do_exactf_utf8;
}
fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
folder = foldEQ_latin1; /* /a, except the sharp s one which */
goto do_exactf_non_utf8; /* isn't dealt with by these */
case EXACTF:
if (utf8_target) {
/* regcomp.c already folded this if pattern is in UTF-8 */
utf8_fold_flags = 0;
goto do_exactf_utf8;
}
fold_array = PL_fold;
folder = foldEQ;
goto do_exactf_non_utf8;
case EXACTFL:
if (UTF_PATTERN || utf8_target) {
utf8_fold_flags = FOLDEQ_UTF8_LOCALE;
goto do_exactf_utf8;
}
fold_array = PL_fold_locale;
folder = foldEQ_locale;
goto do_exactf_non_utf8;
case EXACTFU_SS:
if (UTF_PATTERN) {
utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
}
goto do_exactf_utf8;
case EXACTFU_TRICKYFOLD:
case EXACTFU:
if (UTF_PATTERN || utf8_target) {
utf8_fold_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0;
goto do_exactf_utf8;
}
/* Any 'ss' in the pattern should have been replaced by regcomp,
* so we don't have to worry here about this single special case
* in the Latin1 range */
fold_array = PL_fold_latin1;
folder = foldEQ_latin1;
/* FALL THROUGH */
do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
are no glitches with fold-length differences
between the target string and pattern */
/* The idea in the non-utf8 EXACTF* cases is to first find the
* first character of the EXACTF* node and then, if necessary,
* case-insensitively compare the full text of the node. c1 is the
* first character. c2 is its fold. This logic will not work for
* Unicode semantics and the german sharp ss, which hence should
* not be compiled into a node that gets here. */
pat_string = STRING(c);
ln = STR_LEN(c); /* length to match in octets/bytes */
/* We know that we have to match at least 'ln' bytes (which is the
* same as characters, since not utf8). If we have to match 3
* characters, and there are only 2 availabe, we know without
* trying that it will fail; so don't start a match past the
* required minimum number from the far end */
e = HOP3c(strend, -((I32)ln), s);
if (!reginfo && e < s) {
e = s; /* Due to minlen logic of intuit() */
}
c1 = *pat_string;
c2 = fold_array[c1];
if (c1 == c2) { /* If char and fold are the same */
REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
}
else {
REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
}
break;
do_exactf_utf8:
{
unsigned expansion;
/* If one of the operands is in utf8, we can't use the simpler
* folding above, due to the fact that many different characters
* can have the same fold, or portion of a fold, or different-
* length fold */
pat_string = STRING(c);
ln = STR_LEN(c); /* length to match in octets/bytes */
pat_end = pat_string + ln;
lnc = (UTF_PATTERN) /* length to match in characters */
? utf8_length((U8 *) pat_string, (U8 *) pat_end)
: ln;
/* We have 'lnc' characters to match in the pattern, but because of
* multi-character folding, each character in the target can match
* up to 3 characters (Unicode guarantees it will never exceed
* this) if it is utf8-encoded; and up to 2 if not (based on the
* fact that the Latin 1 folds are already determined, and the
* only multi-char fold in that range is the sharp-s folding to
* 'ss'. Thus, a pattern character can match as little as 1/3 of a
* string character. Adjust lnc accordingly, rounding up, so that
* if we need to match at least 4+1/3 chars, that really is 5. */
expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
lnc = (lnc + expansion - 1) / expansion;
/* As in the non-UTF8 case, if we have to match 3 characters, and
* only 2 are left, it's guaranteed to fail, so don't start a
* match that would require us to go beyond the end of the string
*/
e = HOP3c(strend, -((I32)lnc), s);
if (!reginfo && e < s) {
e = s; /* Due to minlen logic of intuit() */
}
/* XXX Note that we could recalculate e to stop the loop earlier,
* as the worst case expansion above will rarely be met, and as we
* go along we would usually find that e moves further to the left.
* This would happen only after we reached the point in the loop
* where if there were no expansion we should fail. Unclear if
* worth the expense */
while (s <= e) {
char *my_strend= (char *)strend;
if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
pat_string, NULL, ln, cBOOL(UTF_PATTERN), utf8_fold_flags)
&& (!reginfo || regtry(reginfo, &s)) )
{
goto got_it;
}
s += (utf8_target) ? UTF8SKIP(s) : 1;
}
break;
}
case BOUNDL:
PL_reg_flags |= RF_tainted;
FBC_BOUND(isALNUM_LC,
isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
isALNUM_LC_utf8((U8*)s));
break;
case NBOUNDL:
PL_reg_flags |= RF_tainted;
FBC_NBOUND(isALNUM_LC,
isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp)),
isALNUM_LC_utf8((U8*)s));
break;
case BOUND:
FBC_BOUND(isWORDCHAR,
isALNUM_uni(tmp),
cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
break;
case BOUNDA:
FBC_BOUND_NOLOAD(isWORDCHAR_A,
isWORDCHAR_A(tmp),
isWORDCHAR_A((U8*)s));
break;
case NBOUND:
FBC_NBOUND(isWORDCHAR,
isALNUM_uni(tmp),
cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
break;
case NBOUNDA:
FBC_NBOUND_NOLOAD(isWORDCHAR_A,
isWORDCHAR_A(tmp),
isWORDCHAR_A((U8*)s));
break;
case BOUNDU:
FBC_BOUND(isWORDCHAR_L1,
isALNUM_uni(tmp),
cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
break;
case NBOUNDU:
FBC_NBOUND(isWORDCHAR_L1,
isALNUM_uni(tmp),
cBOOL(swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target)));
break;
case ALNUML:
REXEC_FBC_CSCAN_TAINT(
isALNUM_LC_utf8((U8*)s),
isALNUM_LC(*s)
);
break;
case ALNUMU:
REXEC_FBC_CSCAN_PRELOAD(
LOAD_UTF8_CHARCLASS_ALNUM(),
swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
isWORDCHAR_L1((U8) *s)
);
break;
case ALNUM:
REXEC_FBC_CSCAN_PRELOAD(
LOAD_UTF8_CHARCLASS_ALNUM(),
swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
isWORDCHAR((U8) *s)
);
break;
case ALNUMA:
/* Don't need to worry about utf8, as it can match only a single
* byte invariant character */
REXEC_FBC_CLASS_SCAN( isWORDCHAR_A(*s));
break;
case NALNUMU:
REXEC_FBC_CSCAN_PRELOAD(
LOAD_UTF8_CHARCLASS_ALNUM(),
!swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target),
! isWORDCHAR_L1((U8) *s)
);
break;
case NALNUM:
REXEC_FBC_CSCAN_PRELOAD(
LOAD_UTF8_CHARCLASS_ALNUM(),
!swash_fetch(PL_utf8_alnum, (U8*)s, utf8_target),
! isALNUM(*s)
);
break;
case NALNUMA:
REXEC_FBC_CSCAN(
!isWORDCHAR_A(*s),
!isWORDCHAR_A(*s)
);
break;
case NALNUML:
REXEC_FBC_CSCAN_TAINT(
!isALNUM_LC_utf8((U8*)s),
!isALNUM_LC(*s)
);
break;
case SPACEU:
REXEC_FBC_CSCAN_PRELOAD(
LOAD_UTF8_CHARCLASS_SPACE(),
*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
isSPACE_L1((U8) *s)
);
break;
case SPACE:
REXEC_FBC_CSCAN_PRELOAD(
LOAD_UTF8_CHARCLASS_SPACE(),
*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target),
isSPACE((U8) *s)
);
break;
case SPACEA:
/* Don't need to worry about utf8, as it can match only a single
* byte invariant character */
REXEC_FBC_CLASS_SCAN( isSPACE_A(*s));
break;
case SPACEL:
REXEC_FBC_CSCAN_TAINT(
isSPACE_LC_utf8((U8*)s),
isSPACE_LC(*s)
);
break;
case NSPACEU:
REXEC_FBC_CSCAN_PRELOAD(
LOAD_UTF8_CHARCLASS_SPACE(),
!( *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
! isSPACE_L1((U8) *s)
);
break;
case NSPACE:
REXEC_FBC_CSCAN_PRELOAD(
LOAD_UTF8_CHARCLASS_SPACE(),
!(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, utf8_target)),
! isSPACE((U8) *s)
);
break;
case NSPACEA:
REXEC_FBC_CSCAN(
!isSPACE_A(*s),
!isSPACE_A(*s)
);
break;
case NSPACEL:
REXEC_FBC_CSCAN_TAINT(
!isSPACE_LC_utf8((U8*)s),
!isSPACE_LC(*s)
);
break;
case DIGIT:
REXEC_FBC_CSCAN_PRELOAD(
LOAD_UTF8_CHARCLASS_DIGIT(),
swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
isDIGIT(*s)
);
break;
case DIGITA:
/* Don't need to worry about utf8, as it can match only a single
* byte invariant character */
REXEC_FBC_CLASS_SCAN( isDIGIT_A(*s));
break;
case DIGITL:
REXEC_FBC_CSCAN_TAINT(
isDIGIT_LC_utf8((U8*)s),
isDIGIT_LC(*s)
);
break;
case NDIGIT:
REXEC_FBC_CSCAN_PRELOAD(
LOAD_UTF8_CHARCLASS_DIGIT(),
!swash_fetch(PL_utf8_digit,(U8*)s, utf8_target),
!isDIGIT(*s)
);
break;
case NDIGITA:
REXEC_FBC_CSCAN(
!isDIGIT_A(*s),
!isDIGIT_A(*s)
);
break;
case NDIGITL:
REXEC_FBC_CSCAN_TAINT(
!isDIGIT_LC_utf8((U8*)s),
!isDIGIT_LC(*s)
);
break;
case LNBREAK:
REXEC_FBC_CSCAN(
is_LNBREAK_utf8(s),
is_LNBREAK_latin1(s)
);
break;
case VERTWS:
REXEC_FBC_CSCAN(
is_VERTWS_utf8(s),
is_VERTWS_latin1(s)
);
break;
case NVERTWS:
REXEC_FBC_CSCAN(
!is_VERTWS_utf8(s),
!is_VERTWS_latin1(s)
);
break;
case HORIZWS:
REXEC_FBC_CSCAN(
is_HORIZWS_utf8(s),
is_HORIZWS_latin1(s)
);
break;
case NHORIZWS:
REXEC_FBC_CSCAN(
!is_HORIZWS_utf8(s),
!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:
{
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 = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
const char *last_start = strend - trie->minlen;
#ifdef DEBUGGING
const char *real_start = s;
#endif
STRLEN maxlen = trie->maxlen;
SV *sv_points;
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. */
U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
U8 *bitmap=NULL;
GET_RE_DEBUG_FLAGS_DECL;
/* We can't just allocate points here. We need to wrap it in
* an SV so it gets freed properly if there is a croak while
* running the match */
ENTER;
SAVETMPS;
sv_points=newSV(maxlen * sizeof(U8 *));
SvCUR_set(sv_points,
maxlen * sizeof(U8 *));
SvPOK_on(sv_points);
sv_2mortal(sv_points);
points=(U8**)SvPV_nolen(sv_points );
if ( trie_type != trie_utf8_fold
&& (trie->bitmap || OP(c)==AHOCORASICKC) )
{
if (trie->bitmap)
bitmap=(U8*)trie->bitmap;
else
bitmap=(U8*)ANYOF_BITMAP(c);
}
/* this is the Aho-Corasick algorithm modified a touch
to include special handling for long "unknown char"
sequences. The basic idea being that we use AC as long
as we are dealing with a possible matching char, when
we encounter an unknown char (and we have not encountered
an accepting state) we scan forward until we find a legal
starting char.
AC matching is basically that of trie matching, except
that when we encounter a failing transition, we fall back
to the current states "fail state", and try the current char
again, a process we repeat until we reach the root state,
state 1, or a legal transition. If we fail on the root state
then we can either terminate if we have reached an accepting
state previously, or restart the entire process from the beginning
if we have not.
*/
while (s <= last_start) {
const U32 uniflags = UTF8_ALLOW_DEFAULT;
U8 *uc = (U8*)s;
U16 charid = 0;
U32 base = 1;
U32 state = 1;
UV uvc = 0;
STRLEN len = 0;
STRLEN foldlen = 0;
U8 *uscan = (U8*)NULL;
U8 *leftmost = NULL;
#ifdef DEBUGGING
U32 accepted_word= 0;
#endif
U32 pointpos = 0;
while ( state && uc <= (U8*)strend ) {
int failed=0;
U32 word = aho->states[ state ].wordnum;
if( state==1 ) {
if ( bitmap ) {
DEBUG_TRIE_EXECUTE_r(
if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
dump_exec_pos( (char *)uc, c, strend, real_start,
(char *)uc, utf8_target );
PerlIO_printf( Perl_debug_log,
" Scanning for legal start char...\n");
}
);
if (utf8_target) {
while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
uc += UTF8SKIP(uc);
}
} else {
while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
uc++;
}
}
s= (char *)uc;
}
if (uc >(U8*)last_start) break;
}
if ( word ) {
U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
if (!leftmost || lpos < leftmost) {
DEBUG_r(accepted_word=word);
leftmost= lpos;
}
if (base==0) break;
}
points[pointpos++ % maxlen]= uc;
REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
uscan, len, uvc, charid, foldlen,
foldbuf, uniflags);
DEBUG_TRIE_EXECUTE_r({
dump_exec_pos( (char *)uc, c, strend, real_start,
s, utf8_target );
PerlIO_printf(Perl_debug_log,
" Charid:%3u CP:%4"UVxf" ",
charid, uvc);
});
do {
#ifdef DEBUGGING
word = aho->states[ state ].wordnum;
#endif
base = aho->states[ state ].trans.base;
DEBUG_TRIE_EXECUTE_r({
if (failed)
dump_exec_pos( (char *)uc, c, strend, real_start,
s, utf8_target );
PerlIO_printf( Perl_debug_log,
"%sState: %4"UVxf", word=%"UVxf,
failed ? " Fail transition to " : "",
(UV)state, (UV)word);
});
if ( base ) {
U32 tmp;
I32 offset;
if (charid &&
( ((offset = base + charid
- 1 - trie->uniquecharcount)) >= 0)
&& ((U32)offset < trie->lasttrans)
&& trie->trans[offset].check == state
&& (tmp=trie->trans[offset].next))
{
DEBUG_TRIE_EXECUTE_r(
PerlIO_printf( Perl_debug_log," - legal\n"));
state = tmp;
break;
}
else {
DEBUG_TRIE_EXECUTE_r(
PerlIO_printf( Perl_debug_log," - fail\n"));
failed = 1;
state = aho->fail[state];
}
}
else {
/* we must be accepting here */
DEBUG_TRIE_EXECUTE_r(
PerlIO_printf( Perl_debug_log," - accepting\n"));
failed = 1;
break;
}
} while(state);
uc += len;
if (failed) {
if (leftmost)
break;
if (!state) state = 1;
}
}
if ( aho->states[ state ].wordnum ) {
U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
if (!leftmost || lpos < leftmost) {
DEBUG_r(accepted_word=aho->states[ state ].wordnum);
leftmost = lpos;
}
}
if (leftmost) {
s = (char*)leftmost;
DEBUG_TRIE_EXECUTE_r({
PerlIO_printf(
Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
(UV)accepted_word, (IV)(s - real_start)
);
});
if (!reginfo || regtry(reginfo, &s)) {
FREETMPS;
LEAVE;
goto got_it;
}
s = HOPc(s,1);
DEBUG_TRIE_EXECUTE_r({
PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
});
} else {
DEBUG_TRIE_EXECUTE_r(
PerlIO_printf( Perl_debug_log,"No match.\n"));
break;
}
}
FREETMPS;
LEAVE;
}
break;
default:
Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
break;
}
return 0;
got_it:
return s;
}
/*
- regexec_flags - match a regexp against a string
*/
I32
Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *strend,
char *strbeg, I32 minend, SV *sv, void *data, U32 flags)
/* 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;
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 */
I32 end_shift = 0; /* Same for the end. */ /* CC */
I32 scream_pos = -1; /* Internal iterator of scream. */
char *scream_olds = NULL;
const bool utf8_target = cBOOL(DO_UTF8(sv));
I32 multiline;
RXi_GET_DECL(prog,progi);
regmatch_info reginfo; /* create some info to pass to regtry etc */
regexp_paren_pair *swap = NULL;
GET_RE_DEBUG_FLAGS_DECL;
PERL_ARGS_ASSERT_REGEXEC_FLAGS;
PERL_UNUSED_ARG(data);
/* Be paranoid... */
if (prog == NULL || startpos == NULL) {
Perl_croak(aTHX_ "NULL regexp parameter");
return 0;
}
multiline = prog->extflags & RXf_PMf_MULTILINE;
reginfo.prog = rx; /* Yes, sorry that this is confusing. */
RX_MATCH_UTF8_set(rx, utf8_target);
DEBUG_EXECUTE_r(
debug_start_match(rx, utf8_target, startpos, strend,
"Matching");
);
minlen = prog->minlen;
if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
"String too short [regexec_flags]...\n"));
goto phooey;
}
/* Check validity of program. */
if (UCHARAT(progi->program) != REG_MAGIC) {
Perl_croak(aTHX_ "corrupted regexp program");
}
PL_reg_flags = 0;
PL_reg_state.re_state_eval_setup_done = FALSE;
PL_reg_maxiter = 0;
if (RX_UTF8(rx))
PL_reg_flags |= RF_utf8;
/* Mark beginning of line for ^ and lookbehind. */
reginfo.bol = startpos; /* XXX not used ??? */
PL_bostr = strbeg;
reginfo.sv = sv;
/* Mark end of line for $ (and such) */
PL_regeol = strend;
/* see how far we have to get to not match where we matched before */
reginfo.till = startpos+minend;
/* If there is a "must appear" string, look for it. */
s = startpos;
if (prog->extflags & RXf_GPOS_SEEN) { /* Need to set reginfo->ganch */
MAGIC *mg;
if (flags & REXEC_IGNOREPOS){ /* Means: check only at start */
reginfo.ganch = startpos + prog->gofs;
DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
"GPOS IGNOREPOS: reginfo.ganch = startpos + %"UVxf"\n",(UV)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",(IV)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",(UV)prog->gofs));
if (s < strbeg)
goto phooey;
}
}
else if (data) {
reginfo.ganch = strbeg + PTR2UV(data);
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) == 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);
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;
d.scream_olds = &scream_olds;
d.scream_pos = &scream_pos;
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 */
}
}
/* Simplest case: anchored match need be tried only once. */
/* [unless only anchor is BOL and multiline is set] */
if (prog->extflags & (RXf_ANCH & ~RXf_ANCH_GPOS)) {
if (s == startpos && regtry(&reginfo, &startpos))
goto got_it;
else if (multiline || (prog->intflags & PREGf_IMPLICIT)
|| (prog->extflags & RXf_ANCH_MBOL)) /* XXXX SBOL? */
{
char *end;
if (minlen)
dontbother = minlen - 1;
end = HOP3c(strend, -dontbother, strbeg) - 1;
/* for multiline we only have to try after newlines */
if (prog->check_substr || prog->check_utf8) {
/* because of the goto we can not easily reuse the macros for bifurcating the
unicode/non-unicode match modes here like we do elsewhere - demerphq */
if (utf8_target) {
if (s == startpos)
goto after_try_utf8;
while (1) {
if (regtry(&reginfo, &s)) {
goto got_it;
}
after_try_utf8:
if (s > end) {
goto phooey;
}
if (prog->extflags & RXf_USE_INTUIT) {
s = re_intuit_start(rx, sv, s + UTF8SKIP(s), strend, flags, NULL);
if (!s) {
goto phooey;
}
}
else {
s += UTF8SKIP(s);
}
}
} /* end search for check string in unicode */
else {
if (s == startpos) {
goto after_try_latin;
}
while (1) {
if (regtry(&reginfo, &s)) {
goto got_it;
}
after_try_latin:
if (s > end) {
goto phooey;
}
if (prog->extflags & RXf_USE_INTUIT) {
s = re_intuit_start(rx, sv, s + 1, strend, flags, NULL);
if (!s) {
goto phooey;
}
}
else {
s++;
}
}
} /* end search for check string in latin*/
} /* end search for check string */
else { /* search for newline */
if (s > startpos) {
/*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
s--;
}
/* We can use a more efficient search as newlines are the same in unicode as they are in latin */
while (s <= end) { /* note it could be possible to match at the end of the string */
if (*s++ == '\n') { /* don't need PL_utf8skip here */
if (regtry(&reginfo, &s))
goto got_it;
}
}
} /* end search for newline */
} /* end anchored/multiline check string search */
goto phooey;
} else if (RXf_GPOS_CHECK == (prog->extflags & RXf_GPOS_CHECK))
{
/* the warning about reginfo.ganch being used without initialization
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 (tmp_s >= strbeg && regtry(&reginfo, &tmp_s))
goto got_it;
goto phooey;
}
/* Messy cases: unanchored match. */
if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
/* we have /x+whatever/ */
/* it must be a one character string (XXXX Except UTF_PATTERN?) */
char ch;
#ifdef DEBUGGING
int did_match = 0;
#endif
if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
ch = SvPVX_const(utf8_target ? prog->anchored_utf8 : prog->anchored_substr)[0];
if (utf8_target) {
REXEC_FBC_SCAN(
if (*s == ch) {
DEBUG_EXECUTE_r( did_match = 1 );
if (regtry(&reginfo, &s)) goto got_it;
s += UTF8SKIP(s);
while (s < strend && *s == ch)
s += UTF8SKIP(s);
}
);
}
else {
REXEC_FBC_SCAN(
if (*s == ch) {
DEBUG_EXECUTE_r( did_match = 1 );
if (regtry(&reginfo, &s)) goto got_it;
s++;
while (s < strend && *s == ch)
s++;
}
);
}
DEBUG_EXECUTE_r(if (!did_match)
PerlIO_printf(Perl_debug_log,
"Did not find anchored character...\n")
);
}
else if (prog->anchored_substr != NULL
|| prog->anchored_utf8 != NULL
|| ((prog->float_substr != NULL || prog->float_utf8 != NULL)
&& prog->float_max_offset < strend - s)) {
SV *must;
I32 back_max;
I32 back_min;
char *last;
char *last1; /* Last position checked before */
#ifdef DEBUGGING
int did_match = 0;
#endif
if (prog->anchored_substr || prog->anchored_utf8) {
if (!(utf8_target ? prog->anchored_utf8 : prog->anchored_substr))
utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
must = utf8_target ? prog->anchored_utf8 : prog->anchored_substr;
back_max = back_min = prog->anchored_offset;
} else {
if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
must = utf8_target ? prog->float_utf8 : prog->float_substr;
back_max = prog->float_max_offset;
back_min = prog->float_min_offset;
}
if (must == &PL_sv_undef)
/* could not downgrade utf8 check substring, so must fail */
goto phooey;
if (back_min<0) {
last = strend;
} else {
last = HOP3c(strend, /* Cannot start after this */
-(I32)(CHR_SVLEN(must)
- (SvTAIL(must) != 0) + back_min), strbeg);
}
if (s > PL_bostr)
last1 = HOPc(s, -1);
else
last1 = s - 1; /* bogus */
/* XXXX check_substr already used to find "s", can optimize if
check_substr==must. */
scream_pos = -1;
dontbother = end_shift;
strend = HOPc(strend, -dontbother);
while ( (s <= last) &&
(s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)),
(unsigned char*)strend, must,
multiline ? FBMrf_MULTILINE : 0)) ) {
DEBUG_EXECUTE_r( did_match = 1 );
if (HOPc(s, -back_max) > last1) {
last1 = HOPc(s, -back_min);
s = HOPc(s, -back_max);
}
else {
char * const t = (last1 >= PL_bostr) ? HOPc(last1, 1) : last1 + 1;
last1 = HOPc(s, -back_min);
s = t;
}
if (utf8_target) {
while (s <= last1) {
if (regtry(&reginfo, &s))
goto got_it;
s += UTF8SKIP(s);
}
}
else {
while (s <= last1) {
if (regtry(&reginfo, &s))
goto got_it;
s++;
}
}
}
DEBUG_EXECUTE_r(if (!did_match) {
RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
((must == prog->anchored_substr || must == prog->anchored_utf8)
? "anchored" : "floating"),
quoted, RE_SV_TAIL(must));
});
goto phooey;
}
else if ( (c = progi->regstclass) ) {
if (minlen) {
const OPCODE op = OP(progi->regstclass);
/* don't bother with what can't match */
if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
strend = HOPc(strend, -(minlen - 1));
}
DEBUG_EXECUTE_r({
SV * const prop = sv_newmortal();
regprop(prog, prop, c);
{
RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
s,strend-s,60);
PerlIO_printf(Perl_debug_log,
"Matching stclass %.*s against %s (%d bytes)\n",
(int)SvCUR(prop), SvPVX_const(prop),
quoted, (int)(strend - s));
}
});
if (find_byclass(prog, c, s, strend, &reginfo))
goto got_it;
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
}
else {
dontbother = 0;
if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
/* Trim the end. */
char *last= NULL;
SV* float_real;
STRLEN len;
const char *little;
if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog);
float_real = utf8_target ? prog->float_utf8 : prog->float_substr;
little = SvPV_const(float_real, len);
if (SvTAIL(float_real)) {
/* This means that float_real contains an artificial \n on the end
* due to the presence of something like this: /foo$/
* where we can match both "foo" and "foo\n" at the end of the string.
* So we have to compare the end of the string first against the float_real
* without the \n and then against the full float_real with the string.
* We have to watch out for cases where the string might be smaller
* than the float_real or the float_real without the \n.
*/
char *checkpos= strend - len;
DEBUG_OPTIMISE_r(
PerlIO_printf(Perl_debug_log,
"%sChecking for float_real.%s\n",
PL_colors[4], PL_colors[5]));
if (checkpos + 1 < strbeg) {
/* can't match, even if we remove the trailing \n string is too short to match */
DEBUG_EXECUTE_r(
PerlIO_printf(Perl_debug_log,
"%sString shorter than required trailing substring, cannot match.%s\n",
PL_colors[4], PL_colors[5]));
goto phooey;
} else if (memEQ(checkpos + 1, little, len - 1)) {
/* can match, the end of the string matches without the "\n" */
last = checkpos + 1;
} else if (checkpos < strbeg) {
/* cant match, string is too short when the "\n" is included */
DEBUG_EXECUTE_r(
PerlIO_printf(Perl_debug_log,
"%sString does not contain required trailing substring, cannot match.%s\n",
PL_colors[4], PL_colors[5]));
goto phooey;
} else if (!multiline) {
/* non multiline match, so compare with the "\n" at the end of the string */
if (memEQ(checkpos, little, len)) {
last= checkpos;
} else {
DEBUG_EXECUTE_r(
PerlIO_printf(Perl_debug_log,
"%sString does not contain required trailing substring, cannot match.%s\n",
PL_colors[4], PL_colors[5]));
goto phooey;
}
} else {
/* multiline match, so we have to search for a place where the full string is located */
goto find_last;
}
} else {
find_last:
if (len)
last = rninstr(s, strend, little, little + len);
else
last = strend; /* matching "$" */
}
if (!last) {
/* at one point this block contained a comment which was probably
* incorrect, which said that this was a "should not happen" case.
* Even if it was true when it was written I am pretty sure it is
* not anymore, so I have removed the comment and replaced it with
* this one. Yves */
DEBUG_EXECUTE_r(
PerlIO_printf(Perl_debug_log,
"String does not contain required substring, cannot match.\n"
));
goto phooey;
}
dontbother = strend - last + prog->float_min_offset;
}
if (minlen && (dontbother < minlen))
dontbother = minlen - 1;
strend -= dontbother; /* this one's always in bytes! */
/* We don't know much -- general case. */
if (utf8_target) {
for (;;) {
if (regtry(&reginfo, &s))
goto got_it;
if (s >= strend)
break;
s += UTF8SKIP(s);
};
}
else {
do {
if (regtry(&reginfo, &s))
goto got_it;
} while (s++ < strend);
}
}
/* Failure. */
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);
if (PL_reg_state.re_state_eval_setup_done)
restore_pos(aTHX_ prog);
if (RXp_PAREN_NAMES(prog))
(void)hv_iterinit(RXp_PAREN_NAMES(prog));
/* make sure $`, $&, $', and $digit will work later */
if ( !(flags & REXEC_NOT_FIRST) ) {
if (flags & REXEC_COPY_STR) {
#ifdef PERL_OLD_COPY_ON_WRITE
if ((SvIsCOW(sv)
|| (SvFLAGS(sv) & CAN_COW_MASK) == CAN_COW_FLAGS)) {
if (DEBUG_C_TEST) {
PerlIO_printf(Perl_debug_log,
"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
{
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 */
}
}
return 1;
phooey:
DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
PL_colors[4], PL_colors[5]));
if (PL_reg_state.re_state_eval_setup_done)
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;
}
return 0;
}
/* 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 **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;
PERL_ARGS_ASSERT_REGTRY;
reginfo->cutpoint=NULL;
if ((prog->extflags & RXf_EVAL_SEEN)
&& !PL_reg_state.re_state_eval_setup_done)
{
MAGIC *mg;
PL_reg_state.re_state_eval_setup_done = TRUE;
if (reginfo->sv) {
/* Make $_ available to executed code. */
if (reginfo->sv != DEFSV) {
SAVE_DEFSV;
DEFSV_set(reginfo->sv);
}
if (!(SvTYPE(reginfo->sv) >= SVt_PVMG && SvMAGIC(reginfo->sv)
&& (mg = mg_find(reginfo->sv, PERL_MAGIC_regex_global)))) {
/* prepare for quick setting of pos */
#ifdef PERL_OLD_COPY_ON_WRITE
if (SvIsCOW(reginfo->sv))
sv_force_normal_flags(reginfo->sv, 0);
#endif
mg = sv_magicext(reginfo->sv, NULL, PERL_MAGIC_regex_global,
&PL_vtbl_mglob, NULL, 0);
mg->mg_len = -1;
}
PL_reg_magic = mg;
PL_reg_oldpos = mg->mg_len;
SAVEDESTRUCTOR_X(restore_pos, prog);
}
if (!PL_reg_curpm) {
Newxz(PL_reg_curpm, 1, PMOP);
#ifdef USE_ITHREADS
{
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
}
SET_reg_curpm(rx);
PL_reg_oldcurpm = PL_curpm;
PL_curpm = PL_reg_curpm;
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... */
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
RXp_MATCH_COPIED_off(prog);
}
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 */
}
#ifdef DEBUGGING
PL_reg_starttry = *startposp;
#endif
prog->offs[0].start = *startposp - PL_bostr;
prog->lastparen = 0;
prog->lastcloseparen = 0;
PL_regsize = 0;
/* XXXX What this code is doing here?!!! There should be no need
to do this again and again, prog->lastparen should take care of
this! --ilya*/
/* 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
* prog->lastparen), is not needed at all by the test suite
* (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 updated by dapm */
#if 1
if (prog->nparens) {
regexp_paren_pair *pp = prog->offs;
I32 i;
for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
++pp;
pp->start = -1;
pp->end = -1;
}
}
#endif
REGCP_SET(lastcp);
result = regmatch(reginfo, *startposp, progi->program + 1);
if (result != -1) {
prog->offs[0].end = result;
return 1;
}
if (reginfo->cutpoint)
*startposp= reginfo->cutpoint;
REGCP_UNWIND(lastcp);
return 0;
}
#define sayYES goto yes
#define sayNO goto no
#define sayNO_SILENT goto no_silent
/* we dont use STMT_START/END here because it leads to
"unreachable code" warnings, which are bogus, but distracting. */
#define CACHEsayNO \
if (ST.cache_mask) \
PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \
sayNO
/* this is used to determine how far from the left messages like
'failed...' are printed. It should be set such that messages
are inline with the regop output that created them.
*/
#define REPORT_CODE_OFF 32
#define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
#define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
#define SLAB_FIRST(s) (&(s)->states[0])
#define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
/* grab a new slab and return the first slot in it */
STATIC regmatch_state *
S_push_slab(pTHX)
{
#if PERL_VERSION < 9 && !defined(PERL_CORE)
dMY_CXT;
#endif
regmatch_slab *s = PL_regmatch_slab->next;
if (!s) {
Newx(s, 1, regmatch_slab);
s->prev = PL_regmatch_slab;
s->next = NULL;
PL_regmatch_slab->next = s;
}
PL_regmatch_slab = s;
return SLAB_FIRST(s);
}
/* push a new state then goto it */
#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, input) \
pushinput = input; \
scan = node; \
st->resume_state = state; \
goto push_yes_state;
/*
regmatch() - main matching routine
This is basically one big switch statement in a loop. We execute an op,
set 'next' to point the next op, and continue. If we come to a point which
we may need to backtrack to on failure such as (A|B|C), we push a
backtrack state onto the backtrack stack. On failure, we pop the top
state, and re-enter the loop at the state indicated. If there are no more
states to pop, we return failure.
Sometimes we also need to backtrack on success; for example /A+/, where
after successfully matching one A, we need to go back and try to
match another one; similarly for lookahead assertions: if the assertion
completes successfully, we backtrack to the state just before the assertion
and then carry on. In these cases, the pushed state is marked as
'backtrack on success too'. This marking is in fact done by a chain of
pointers, each pointing to the previous 'yes' state. On success, we pop to
the nearest yes state, discarding any intermediate failure-only states.
Sometimes a yes state is pushed just to force some cleanup code to be
called at the end of a successful match or submatch; e.g. (??{$re}) uses
it to free the inner regex.
Note that failure backtracking rewinds the cursor position, while
success backtracking leaves it alone.
A pattern is complete when the END op is executed, while a subpattern
such as (?=foo) is complete when the SUCCESS op is executed. Both of these
ops trigger the "pop to last yes state if any, otherwise return true"
behaviour.
A common convention in this function is to use A and B to refer to the two
subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
the subpattern to be matched possibly multiple times, while B is the entire
rest of the pattern. Variable and state names reflect this convention.
The states in the main switch are the union of ops and failure/success of
substates associated with with that op. For example, IFMATCH is the op
that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
successfully matched A and IFMATCH_A_fail is a state saying that we have
just failed to match A. Resume states always come in pairs. The backtrack
state we push is marked as 'IFMATCH_A', but when that is popped, we resume
at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
on success or failure.
The struct that holds a backtracking state is actually a big union, with
one variant for each major type of op. The variable st points to the
top-most backtrack struct. To make the code clearer, within each
block of code we #define ST to alias the relevant union.
Here's a concrete example of a (vastly oversimplified) IFMATCH
implementation:
switch (state) {
....
#define ST st->u.ifmatch
case IFMATCH: // we are executing the IFMATCH op, (?=A)B
ST.foo = ...; // some state we wish to save
...
// 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, newinput);
// NOTREACHED
case IFMATCH_A: // we have successfully executed A; now continue with B
next = B;
bar = ST.foo; // do something with the preserved value
break;
case IFMATCH_A_fail: // A failed, so the assertion failed
...; // do some housekeeping, then ...
sayNO; // propagate the failure
#undef ST
...
}
For any old-timers reading this who are familiar with the old recursive
approach, the code above is equivalent to:
case IFMATCH: // we are executing the IFMATCH op, (?=A)B
{
int foo = ...
...
if (regmatch(A)) {
next = B;
bar = foo;
break;
}
...; // do some housekeeping, then ...
sayNO; // propagate the failure
}
The topmost backtrack state, pointed to by st, is usually free. If you
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, 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'.
On backtracking, the free slot is popped, and the saved state becomes the
new free state. An ST.foo field in this new top state can be temporarily
accessed to retrieve values, but once the main loop is re-entered, it
becomes available for reuse.
Note that the depth of the backtrack stack constantly increases during the
left-to-right execution of the pattern, rather than going up and down with
the pattern nesting. For example the stack is at its maximum at Z at the
end of the pattern, rather than at X in the following:
/(((X)+)+)+....(Y)+....Z/
The only exceptions to this are lookahead/behind assertions and the cut,
(?>A), which pop all the backtrack states associated with A before
continuing.
Backtrack state structs are allocated in slabs of about 4K in size.
PL_regmatch_state and st always point to the currently active state,
and PL_regmatch_slab points to the slab currently containing
PL_regmatch_state. The first time regmatch() is called, the first slab is
allocated, and is never freed until interpreter destruction. When the slab
is full, a new one is allocated and chained to the end. At exit from
regmatch(), slabs allocated since entry are freed.
*/
#define DEBUG_STATE_pp(pp) \
DEBUG_STATE_r({ \
DUMP_EXEC_POS(locinput, scan, utf8_target); \
PerlIO_printf(Perl_debug_log, \
" %*s"pp" %s%s%s%s%s\n", \
depth*2, "", \
PL_reg_name[st->resume_state], \
((st==yes_state||st==mark_state) ? "[" : ""), \
((st==yes_state) ? "Y" : ""), \
((st==mark_state) ? "M" : ""), \
((st==yes_state||st==mark_state) ? "]" : "") \
); \
});
#define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
#ifdef DEBUGGING
STATIC void
S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
const char *start, const char *end, const char *blurb)
{
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),
RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
start, end - start, 60);
PerlIO_printf(Perl_debug_log,
"%s%s REx%s %s against %s\n",
PL_colors[4], blurb, PL_colors[5], s0, s1);
if (utf8_target||utf8_pat)
PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
utf8_pat ? "pattern" : "",
utf8_pat && utf8_target ? " and " : "",
utf8_target ? "string" : ""
);
}
}
STATIC void
S_dump_exec_pos(pTHX_ const char *locinput,
const regnode *scan,
const char *loc_regeol,
const char *loc_bostr,
const char *loc_reg_starttry,
const bool utf8_target)
{
const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
/* The part of the string before starttry has one color
(pref0_len chars), between starttry and current
position another one (pref_len - pref0_len chars),
after the current position the third one.
We assume that pref0_len <= pref_len, otherwise we
decrease pref0_len. */
int pref_len = (locinput - loc_bostr) > (5 + taill) - l
? (5 + taill) - l : locinput - loc_bostr;
int pref0_len;
PERL_ARGS_ASSERT_DUMP_EXEC_POS;
while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
pref_len++;
pref0_len = pref_len - (locinput - loc_reg_starttry);
if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
l = ( loc_regeol - locinput > (5 + taill) - pref_len
? (5 + taill) - pref_len : loc_regeol - locinput);
while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
l--;
if (pref0_len < 0)
pref0_len = 0;
if (pref0_len > pref_len)
pref0_len = pref_len;
{
const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
(locinput - pref_len),pref0_len, 60, 4, 5);
RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
(locinput - pref_len + pref0_len),
pref_len - pref0_len, 60, 2, 3);
RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
locinput, loc_regeol - locinput, 10, 0, 1);
const STRLEN tlen=len0+len1+len2;
PerlIO_printf(Perl_debug_log,
"%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
(IV)(locinput - loc_bostr),
len0, s0,
len1, s1,
(docolor ? "" : "> <"),
len2, s2,
(int)(tlen > 19 ? 0 : 19 - tlen),
"");
}
}
#endif
/* reg_check_named_buff_matched()
* Checks to see if a named buffer has matched. The data array of
* buffer numbers corresponding to the buffer is expected to reside
* in the regexp->data->data array in the slot stored in the ARG() of
* node involved. Note that this routine doesn't actually care about the
* name, that information is not preserved from compilation to execution.
* Returns the index of the leftmost defined buffer with the given name
* or 0 if non of the buffers matched.
*/
STATIC I32
S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan)
{
I32 n;
RXi_GET_DECL(rex,rexi);
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)rex->lastparen >= nums[n] &&
rex->offs[nums[n]].end != -1)
{
return nums[n];
}
}
return 0;
}
/* free all slabs above current one - called during LEAVE_SCOPE */
STATIC void
S_clear_backtrack_stack(pTHX_ void *p)
{
regmatch_slab *s = PL_regmatch_slab->next;
PERL_UNUSED_ARG(p);
if (!s)
return;
PL_regmatch_slab->next = NULL;
while (s) {
regmatch_slab * const osl = s;
s = s->next;
Safefree(osl);
}
}
/* 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;
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 */
regmatch_state *st;
/* cache heavy used fields of st in registers */
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 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
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;
bool no_final = 0; /* prevent failure from backtracking? */
bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
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
during a successful 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
*/
bool sw = 0; /* the condition value in (?(cond)a|b) */
bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
int logical = 0; /* the following EVAL is:
0: (?{...})
1: (?(?{...})X|Y)
2: (??{...})
or the following IFMATCH/UNLESSM is:
false: plain (?=foo)
true: used as a condition: (?(?=foo))
*/
PAD* last_pad = NULL;
dMULTICALL;
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;
#endif
/* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
multicall_oldcatch = 0;
multicall_cv = NULL;
cx = NULL;
PERL_UNUSED_VAR(multicall_cop);
PERL_UNUSED_VAR(newsp);
PERL_ARGS_ASSERT_REGMATCH;
DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
PerlIO_printf(Perl_debug_log,"regmatch start\n");
}));
/* on first ever call to regmatch, allocate first slab */
if (!PL_regmatch_slab) {
Newx(PL_regmatch_slab, 1, regmatch_slab);
PL_regmatch_slab->prev = NULL;
PL_regmatch_slab->next = NULL;
PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
}
oldsave = PL_savestack_ix;
SAVEDESTRUCTOR_X(S_clear_backtrack_stack, NULL);
SAVEVPTR(PL_regmatch_slab);
SAVEVPTR(PL_regmatch_state);
/* grab next free state slot */
st = ++PL_regmatch_state;
if (st > SLAB_LAST(PL_regmatch_slab))
st = PL_regmatch_state = S_push_slab(aTHX);
/* Note that nextchr is a byte even in UTF */
nextchr = UCHARAT(locinput);
scan = prog;
while (scan != NULL) {
DEBUG_EXECUTE_r( {
SV * const prop = sv_newmortal();
regnode *rnext=regnext(scan);
DUMP_EXEC_POS( locinput, scan, utf8_target );
regprop(rex, prop, scan);
PerlIO_printf(Perl_debug_log,
"%3"IVdf":%*s%s(%"IVdf")\n",
(IV)(scan - rexi->program), depth*2, "",
SvPVX_const(prop),
(PL_regkind[OP(scan)] == END || !rnext) ?
0 : (IV)(rnext - rexi->program));
});
next = scan + NEXT_OFF(scan);
if (next == scan)
next = NULL;
state_num = OP(scan);
reenter_switch:
switch (state_num) {
case BOL:
if (locinput == PL_bostr)
{
/* reginfo->till = reginfo->bol; */
break;
}
sayNO;
case MBOL:
if (locinput == PL_bostr ||
((nextchr || locinput < PL_regeol) && locinput[-1] == '\n'))
{
break;
}
sayNO;
case SBOL:
if (locinput == PL_bostr)
break;
sayNO;
case GPOS:
if (locinput == reginfo->ganch)
break;
sayNO;
case KEEPS:
/* update the startpoint */
st->u.keeper.val = rex->offs[0].start;
rex->offs[0].start = locinput - PL_bostr;
PUSH_STATE_GOTO(KEEPS_next, next, locinput);
/*NOT-REACHED*/
case KEEPS_next_fail:
/* rollback the start point change */
rex->offs[0].start = st->u.keeper.val;
sayNO_SILENT;
/*NOT-REACHED*/
case EOL:
goto seol;
case MEOL:
if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
sayNO;
break;
case SEOL:
seol:
if ((nextchr || locinput < PL_regeol) && nextchr != '\n')
sayNO;
if (PL_regeol - locinput > 1)
sayNO;
break;
case EOS:
if (PL_regeol != locinput)
sayNO;
break;
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;
case CANY:
if (!nextchr && locinput >= PL_regeol)
sayNO;
nextchr = UCHARAT(++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;
#undef ST
#define ST st->u.trie
case TRIEC:
/* 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)) {
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;
assert(0); /* NOTREACHED */
}
/* FALL THROUGH */
case TRIE:
/* the basic plan of execution of the trie is:
* At the beginning, run though all the states, and
* find the longest-matching word. Also remember the position
* of the shortest matching word. For example, this pattern:
* 1 2 3 4 5
* ab|a|x|abcd|abc
* when matched against the string "abcde", will generate
* accept states for all words except 3, with the longest
* matching word being 4, and the shortest being 1 (with
* the position being after char 1 of the string).
*
* Then for each matching word, in word order (i.e. 1,2,4,5),
* we run the remainder of the pattern; on each try setting
* the current position to the character following the word,
* returning to try the next word on failure.
*
* We avoid having to build a list of words at runtime by
* using a compile-time structure, wordinfo[].prev, which
* gives, for each word, the previous accepting word (if any).
* In the case above it would contain the mappings 1->2, 2->0,
* 3->0, 4->5, 5->1. We can use this table to generate, from
* the longest word (4 above), a list of all words, by
* following the list of prev pointers; this gives us the
* unordered list 4,5,1,2. Then given the current word we have
* just tried, we can go through the list and find the
* next-biggest word to try (so if we just failed on word 2,
* the next in the list is 4).
*
* Since at runtime we don't record the matching position in
* the string for each word, we have to work that out for
* each word we're about to process. The wordinfo table holds
* the character length of each word; given that we recorded
* at the start: the position of the shortest word and its
* length in chars, we just need to move the pointer the
* difference between the two char lengths. Depending on
* Unicode status and folding, that's cheap or expensive.
*
* This algorithm is optimised for the case where are only a
* small number of accept states, i.e. 0,1, or maybe 2.
* With lots of accepts states, and having to try all of them,
* it becomes quadratic on number of accept states to find all
* the next words.
*/
{
/* what type of TRIE am I? (utf8 makes this contextual) */
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 = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
U32 state = trie->startstate;
if (trie->bitmap && !TRIE_BITMAP_TEST(trie,*locinput) ) {
if (trie->states[ state ].wordnum) {
DEBUG_EXECUTE_r(
PerlIO_printf(Perl_debug_log,
"%*s %smatched empty string...%s\n",
REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
);
if (!trie->jump)
break;
} else {
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;
}
}
{
U8 *uc = ( U8* )locinput;
STRLEN len = 0;
STRLEN foldlen = 0;
U8 *uscan = (U8*)NULL;
U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
U32 charcount = 0; /* how many input chars we have matched */
U32 accepted = 0; /* have we seen any accepting states? */
ST.jump = trie->jump;
ST.me = scan;
ST.firstpos = NULL;
ST.longfold = FALSE; /* char longer if folded => it's harder */
ST.nextword = 0;
/* fully traverse the TRIE; note the position of the
shortest accept state and the wordnum of the longest
accept state */
while ( state && uc <= (U8*)PL_regeol ) {
U32 base = trie->states[ state ].trans.base;
UV uvc = 0;
U16 charid = 0;
U16 wordnum;
wordnum = trie->states[ state ].wordnum;
if (wordnum) { /* it's an accept state */
if (!accepted) {
accepted = 1;
/* record first match position */
if (ST.longfold) {
ST.firstpos = (U8*)locinput;
ST.firstchars = 0;
}
else {
ST.firstpos = uc;
ST.firstchars = charcount;
}
}
if (!ST.nextword || wordnum < ST.nextword)
ST.nextword = wordnum;
ST.topword = wordnum;
}
DEBUG_TRIE_EXECUTE_r({
DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
PerlIO_printf( Perl_debug_log,
"%*s %sState: %4"UVxf" Accepted: %c ",
2+depth * 2, "", PL_colors[4],
(UV)state, (accepted ? 'Y' : 'N'));
});
/* read a char and goto next state */
if ( base ) {
I32 offset;
REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
uscan, len, uvc, charid, foldlen,
foldbuf, uniflags);
charcount++;
if (foldlen>0)
ST.longfold = TRUE;
if (charid &&
( ((offset =
base + charid - 1 - trie->uniquecharcount)) >= 0)
&& ((U32)offset < trie->lasttrans)
&& trie->trans[offset].check == state)
{
state = trie->trans[offset].next;
}
else {
state = 0;
}
uc += len;
}
else {
state = 0;
}
DEBUG_TRIE_EXECUTE_r(
PerlIO_printf( Perl_debug_log,
"Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
charid, uvc, (UV)state, PL_colors[5] );
);
}
if (!accepted)
sayNO;
/* calculate total number of accept states */
{
U16 w = ST.topword;
accepted = 0;
while (w) {
w = trie->wordinfo[w].prev;
accepted++;
}
ST.accepted = accepted;
}
DEBUG_EXECUTE_r(
PerlIO_printf( Perl_debug_log,
"%*s %sgot %"IVdf" possible matches%s\n",
REPORT_CODE_OFF + depth * 2, "",
PL_colors[4], (IV)ST.accepted, PL_colors[5] );
);
goto trie_first_try; /* jump into the fail handler */
}}
assert(0); /* NOTREACHED */
case TRIE_next_fail: /* we failed - try next alternative */
{
U8 *uc;
if ( ST.jump) {
REGCP_UNWIND(ST.cp);
UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
}
if (!--ST.accepted) {
DEBUG_EXECUTE_r({
PerlIO_printf( Perl_debug_log,
"%*s %sTRIE failed...%s\n",
REPORT_CODE_OFF+depth*2, "",
PL_colors[4],
PL_colors[5] );
});
sayNO_SILENT;
}
{
/* Find next-highest word to process. Note that this code
* is O(N^2) per trie run (O(N) per branch), so keep tight */
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))
min = word;
}
ST.nextword = min;
}
trie_first_try:
if (do_cutgroup) {
do_cutgroup = 0;
no_final = 0;
}
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 */
reg_trie_data * const trie
= (reg_trie_data*)rexi->data->data[ARG(ST.me)];
assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
>= 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
* its folded length (which may be different */
U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
STRLEN foldlen;
STRLEN len;
UV uvc;
U8 *uscan;
while (chars) {
if (utf8_target) {
uvc = utf8n_to_uvuni((U8*)uc, UTF8_MAXLEN, &len,
uniflags);
uc += len;
}
else {
uvc = *uc;
uc++;
}
uvc = to_uni_fold(uvc, foldbuf, &foldlen);
uscan = foldbuf;
while (foldlen) {
if (!--chars)
break;
uvc = utf8n_to_uvuni(uscan, UTF8_MAXLEN, &len,
uniflags);
uscan += len;
foldlen -= len;
}
}
}
else {
if (utf8_target)
while (chars--)
uc += UTF8SKIP(uc);
else
uc += chars;
}
}
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,
"%*s %sTRIE matched word #%d, continuing%s\n",
REPORT_CODE_OFF+depth*2, "",
PL_colors[4],
ST.nextword,
PL_colors[5]
);
});
if (ST.accepted > 1 || has_cutgroup) {
PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
assert(0); /* NOTREACHED */
}
/* only one choice left - just continue */
DEBUG_EXECUTE_r({
AV *const trie_words
= MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
SV ** const tmp = av_fetch( trie_words,
ST.nextword-1, 0 );
SV *sv= tmp ? sv_newmortal() : NULL;
PerlIO_printf( Perl_debug_log,
"%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
REPORT_CODE_OFF+depth*2, "", PL_colors[4],
ST.nextword,
tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
PL_colors[0], PL_colors[1],
(SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
)
: "not compiled under -Dr",
PL_colors[5] );
});
locinput = (char*)uc;
nextchr = UCHARAT(locinput);
continue; /* execute rest of RE */
assert(0); /* NOTREACHED */
}
#undef ST
case EXACT: {
char *s = STRING(scan);
ln = STR_LEN(scan);
if (utf8_target != UTF_PATTERN) {
/* The target and the pattern have differing utf8ness. */
char *l = locinput;