pcre/pcre_exec.c

7174 lines
213 KiB
C
Raw Permalink Normal View History

2023-02-15 17:00:09 +03:00
/*************************************************
* Perl-Compatible Regular Expressions *
*************************************************/
/* PCRE is a library of functions to support regular expressions whose syntax
and semantics are as close as possible to those of the Perl 5 language.
Written by Philip Hazel
Copyright (c) 1997-2021 University of Cambridge
-----------------------------------------------------------------------------
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:
* Redistributions of source code must retain the above copyright notice,
this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
* Neither the name of the University of Cambridge nor the names of its
contributors may be used to endorse or promote products derived from
this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
POSSIBILITY OF SUCH DAMAGE.
-----------------------------------------------------------------------------
*/
/* This module contains pcre_exec(), the externally visible function that does
pattern matching using an NFA algorithm, trying to mimic Perl as closely as
possible. There are also some static supporting functions. */
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif
#define NLBLOCK md /* Block containing newline information */
#define PSSTART start_subject /* Field containing processed string start */
#define PSEND end_subject /* Field containing processed string end */
#include "pcre_internal.h"
/* Undefine some potentially clashing cpp symbols */
#undef min
#undef max
/* The md->capture_last field uses the lower 16 bits for the last captured
substring (which can never be greater than 65535) and a bit in the top half
to mean "capture vector overflowed". This odd way of doing things was
implemented when it was realized that preserving and restoring the overflow bit
whenever the last capture number was saved/restored made for a neater
interface, and doing it this way saved on (a) another variable, which would
have increased the stack frame size (a big NO-NO in PCRE) and (b) another
separate set of save/restore instructions. The following defines are used in
implementing this. */
#define CAPLMASK 0x0000ffff /* The bits used for last_capture */
#define OVFLMASK 0xffff0000 /* The bits used for the overflow flag */
#define OVFLBIT 0x00010000 /* The bit that is set for overflow */
/* Values for setting in md->match_function_type to indicate two special types
of call to match(). We do it this way to save on using another stack variable,
as stack usage is to be discouraged. */
#define MATCH_CONDASSERT 1 /* Called to check a condition assertion */
#define MATCH_CBEGROUP 2 /* Could-be-empty unlimited repeat group */
/* Non-error returns from the match() function. Error returns are externally
defined PCRE_ERROR_xxx codes, which are all negative. */
#define MATCH_MATCH 1
#define MATCH_NOMATCH 0
/* Special internal returns from the match() function. Make them sufficiently
negative to avoid the external error codes. */
#define MATCH_ACCEPT (-999)
#define MATCH_KETRPOS (-998)
#define MATCH_ONCE (-997)
/* The next 5 must be kept together and in sequence so that a test that checks
for any one of them can use a range. */
#define MATCH_COMMIT (-996)
#define MATCH_PRUNE (-995)
#define MATCH_SKIP (-994)
#define MATCH_SKIP_ARG (-993)
#define MATCH_THEN (-992)
#define MATCH_BACKTRACK_MAX MATCH_THEN
#define MATCH_BACKTRACK_MIN MATCH_COMMIT
/* Maximum number of ints of offset to save on the stack for recursive calls.
If the offset vector is bigger, malloc is used. This should be a multiple of 3,
because the offset vector is always a multiple of 3 long. */
#define REC_STACK_SAVE_MAX 30
/* Min and max values for the common repeats; for the maxima, 0 => infinity */
static const char rep_min[] = { 0, 0, 1, 1, 0, 0, 0, 0, 0, 1, 0, };
static const char rep_max[] = { 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 1, };
#ifdef PCRE_DEBUG
/*************************************************
* Debugging function to print chars *
*************************************************/
/* Print a sequence of chars in printable format, stopping at the end of the
subject if the requested.
Arguments:
p points to characters
length number to print
is_subject TRUE if printing from within md->start_subject
md pointer to matching data block, if is_subject is TRUE
Returns: nothing
*/
static void
pchars(const pcre_uchar *p, int length, BOOL is_subject, match_data *md)
{
pcre_uint32 c;
BOOL utf = md->utf;
if (is_subject && length > md->end_subject - p) length = md->end_subject - p;
while (length-- > 0)
if (isprint(c = UCHAR21INCTEST(p))) printf("%c", (char)c); else printf("\\x{%02x}", c);
}
#endif
/*************************************************
* Match a back-reference *
*************************************************/
/* Normally, if a back reference hasn't been set, the length that is passed is
negative, so the match always fails. However, in JavaScript compatibility mode,
the length passed is zero. Note that in caseless UTF-8 mode, the number of
subject bytes matched may be different to the number of reference bytes.
Arguments:
offset index into the offset vector
eptr pointer into the subject
length length of reference to be matched (number of bytes)
md points to match data block
caseless TRUE if caseless
Returns: >= 0 the number of subject bytes matched
-1 no match
-2 partial match; always given if at end subject
*/
static int
match_ref(int offset, register PCRE_PUCHAR eptr, int length, match_data *md,
BOOL caseless)
{
PCRE_PUCHAR eptr_start = eptr;
register PCRE_PUCHAR p = md->start_subject + md->offset_vector[offset];
#if defined SUPPORT_UTF && defined SUPPORT_UCP
BOOL utf = md->utf;
#endif
#ifdef PCRE_DEBUG
if (eptr >= md->end_subject)
printf("matching subject <null>");
else
{
printf("matching subject ");
pchars(eptr, length, TRUE, md);
}
printf(" against backref ");
pchars(p, length, FALSE, md);
printf("\n");
#endif
/* Always fail if reference not set (and not JavaScript compatible - in that
case the length is passed as zero). */
if (length < 0) return -1;
/* Separate the caseless case for speed. In UTF-8 mode we can only do this
properly if Unicode properties are supported. Otherwise, we can check only
ASCII characters. */
if (caseless)
{
#if defined SUPPORT_UTF && defined SUPPORT_UCP
if (utf)
{
/* Match characters up to the end of the reference. NOTE: the number of
data units matched may differ, because in UTF-8 there are some characters
whose upper and lower case versions code have different numbers of bytes.
For example, U+023A (2 bytes in UTF-8) is the upper case version of U+2C65
(3 bytes in UTF-8); a sequence of 3 of the former uses 6 bytes, as does a
sequence of two of the latter. It is important, therefore, to check the
length along the reference, not along the subject (earlier code did this
wrong). */
PCRE_PUCHAR endptr = p + length;
while (p < endptr)
{
pcre_uint32 c, d;
const ucd_record *ur;
if (eptr >= md->end_subject) return -2; /* Partial match */
GETCHARINC(c, eptr);
GETCHARINC(d, p);
ur = GET_UCD(d);
if (c != d && c != d + ur->other_case)
{
const pcre_uint32 *pp = PRIV(ucd_caseless_sets) + ur->caseset;
for (;;)
{
if (c < *pp) return -1;
if (c == *pp++) break;
}
}
}
}
else
#endif
/* The same code works when not in UTF-8 mode and in UTF-8 mode when there
is no UCP support. */
{
while (length-- > 0)
{
pcre_uint32 cc, cp;
if (eptr >= md->end_subject) return -2; /* Partial match */
cc = UCHAR21TEST(eptr);
cp = UCHAR21TEST(p);
if (TABLE_GET(cp, md->lcc, cp) != TABLE_GET(cc, md->lcc, cc)) return -1;
p++;
eptr++;
}
}
}
/* In the caseful case, we can just compare the bytes, whether or not we
are in UTF-8 mode. */
else
{
while (length-- > 0)
{
if (eptr >= md->end_subject) return -2; /* Partial match */
if (UCHAR21INCTEST(p) != UCHAR21INCTEST(eptr)) return -1;
}
}
return (int)(eptr - eptr_start);
}
/***************************************************************************
****************************************************************************
RECURSION IN THE match() FUNCTION
The match() function is highly recursive, though not every recursive call
increases the recursive depth. Nevertheless, some regular expressions can cause
it to recurse to a great depth. I was writing for Unix, so I just let it call
itself recursively. This uses the stack for saving everything that has to be
saved for a recursive call. On Unix, the stack can be large, and this works
fine.
It turns out that on some non-Unix-like systems there are problems with
programs that use a lot of stack. (This despite the fact that every last chip
has oodles of memory these days, and techniques for extending the stack have
been known for decades.) So....
There is a fudge, triggered by defining NO_RECURSE, which avoids recursive
calls by keeping local variables that need to be preserved in blocks of memory
obtained from malloc() instead instead of on the stack. Macros are used to
achieve this so that the actual code doesn't look very different to what it
always used to.
The original heap-recursive code used longjmp(). However, it seems that this
can be very slow on some operating systems. Following a suggestion from Stan
Switzer, the use of longjmp() has been abolished, at the cost of having to
provide a unique number for each call to RMATCH. There is no way of generating
a sequence of numbers at compile time in C. I have given them names, to make
them stand out more clearly.
Crude tests on x86 Linux show a small speedup of around 5-8%. However, on
FreeBSD, avoiding longjmp() more than halves the time taken to run the standard
tests. Furthermore, not using longjmp() means that local dynamic variables
don't have indeterminate values; this has meant that the frame size can be
reduced because the result can be "passed back" by straight setting of the
variable instead of being passed in the frame.
****************************************************************************
***************************************************************************/
/* Numbers for RMATCH calls. When this list is changed, the code at HEAP_RETURN
below must be updated in sync. */
enum { RM1=1, RM2, RM3, RM4, RM5, RM6, RM7, RM8, RM9, RM10,
RM11, RM12, RM13, RM14, RM15, RM16, RM17, RM18, RM19, RM20,
RM21, RM22, RM23, RM24, RM25, RM26, RM27, RM28, RM29, RM30,
RM31, RM32, RM33, RM34, RM35, RM36, RM37, RM38, RM39, RM40,
RM41, RM42, RM43, RM44, RM45, RM46, RM47, RM48, RM49, RM50,
RM51, RM52, RM53, RM54, RM55, RM56, RM57, RM58, RM59, RM60,
RM61, RM62, RM63, RM64, RM65, RM66, RM67 };
/* These versions of the macros use the stack, as normal. There are debugging
versions and production versions. Note that the "rw" argument of RMATCH isn't
actually used in this definition. */
#ifndef NO_RECURSE
#define REGISTER register
#ifdef PCRE_DEBUG
#define RMATCH(ra,rb,rc,rd,re,rw) \
{ \
printf("match() called in line %d\n", __LINE__); \
rrc = match(ra,rb,mstart,rc,rd,re,rdepth+1); \
printf("to line %d\n", __LINE__); \
}
#define RRETURN(ra) \
{ \
printf("match() returned %d from line %d\n", ra, __LINE__); \
return ra; \
}
#else
#define RMATCH(ra,rb,rc,rd,re,rw) \
rrc = match(ra,rb,mstart,rc,rd,re,rdepth+1)
#define RRETURN(ra) return ra
#endif
#else
/* These versions of the macros manage a private stack on the heap. Note that
the "rd" argument of RMATCH isn't actually used in this definition. It's the md
argument of match(), which never changes. */
#define REGISTER
#define RMATCH(ra,rb,rc,rd,re,rw)\
{\
heapframe *newframe = frame->Xnextframe;\
if (newframe == NULL)\
{\
newframe = (heapframe *)(PUBL(stack_malloc))(sizeof(heapframe));\
if (newframe == NULL) RRETURN(PCRE_ERROR_NOMEMORY);\
newframe->Xnextframe = NULL;\
frame->Xnextframe = newframe;\
}\
frame->Xwhere = rw;\
newframe->Xeptr = ra;\
newframe->Xecode = rb;\
newframe->Xmstart = mstart;\
newframe->Xoffset_top = rc;\
newframe->Xeptrb = re;\
newframe->Xrdepth = frame->Xrdepth + 1;\
newframe->Xprevframe = frame;\
frame = newframe;\
DPRINTF(("restarting from line %d\n", __LINE__));\
goto HEAP_RECURSE;\
L_##rw:\
DPRINTF(("jumped back to line %d\n", __LINE__));\
}
#define RRETURN(ra)\
{\
heapframe *oldframe = frame;\
frame = oldframe->Xprevframe;\
if (frame != NULL)\
{\
rrc = ra;\
goto HEAP_RETURN;\
}\
return ra;\
}
/* Structure for remembering the local variables in a private frame */
typedef struct heapframe {
struct heapframe *Xprevframe;
struct heapframe *Xnextframe;
/* Function arguments that may change */
PCRE_PUCHAR Xeptr;
const pcre_uchar *Xecode;
PCRE_PUCHAR Xmstart;
int Xoffset_top;
eptrblock *Xeptrb;
unsigned int Xrdepth;
/* Function local variables */
PCRE_PUCHAR Xcallpat;
#ifdef SUPPORT_UTF
PCRE_PUCHAR Xcharptr;
#endif
PCRE_PUCHAR Xdata;
PCRE_PUCHAR Xnext;
PCRE_PUCHAR Xpp;
PCRE_PUCHAR Xprev;
PCRE_PUCHAR Xsaved_eptr;
recursion_info Xnew_recursive;
BOOL Xcur_is_word;
BOOL Xcondition;
BOOL Xprev_is_word;
#ifdef SUPPORT_UCP
int Xprop_type;
unsigned int Xprop_value;
int Xprop_fail_result;
int Xoclength;
pcre_uchar Xocchars[6];
#endif
int Xcodelink;
int Xctype;
unsigned int Xfc;
int Xfi;
int Xlength;
int Xmax;
int Xmin;
unsigned int Xnumber;
int Xoffset;
unsigned int Xop;
pcre_int32 Xsave_capture_last;
int Xsave_offset1, Xsave_offset2, Xsave_offset3;
int Xstacksave[REC_STACK_SAVE_MAX];
eptrblock Xnewptrb;
/* Where to jump back to */
int Xwhere;
} heapframe;
#endif
/***************************************************************************
***************************************************************************/
/*************************************************
* Match from current position *
*************************************************/
/* This function is called recursively in many circumstances. Whenever it
returns a negative (error) response, the outer incarnation must also return the
same response. */
/* These macros pack up tests that are used for partial matching, and which
appear several times in the code. We set the "hit end" flag if the pointer is
at the end of the subject and also past the start of the subject (i.e.
something has been matched). For hard partial matching, we then return
immediately. The second one is used when we already know we are past the end of
the subject. */
#define CHECK_PARTIAL()\
if (md->partial != 0 && eptr >= md->end_subject && \
eptr > md->start_used_ptr) \
{ \
md->hitend = TRUE; \
if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL); \
}
#define SCHECK_PARTIAL()\
if (md->partial != 0 && eptr > md->start_used_ptr) \
{ \
md->hitend = TRUE; \
if (md->partial > 1) RRETURN(PCRE_ERROR_PARTIAL); \
}
/* Performance note: It might be tempting to extract commonly used fields from
the md structure (e.g. utf, end_subject) into individual variables to improve
performance. Tests using gcc on a SPARC disproved this; in the first case, it
made performance worse.
Arguments:
eptr pointer to current character in subject
ecode pointer to current position in compiled code
mstart pointer to the current match start position (can be modified
by encountering \K)
offset_top current top pointer
md pointer to "static" info for the match
eptrb pointer to chain of blocks containing eptr at start of
brackets - for testing for empty matches
rdepth the recursion depth
Returns: MATCH_MATCH if matched ) these values are >= 0
MATCH_NOMATCH if failed to match )
a negative MATCH_xxx value for PRUNE, SKIP, etc
a negative PCRE_ERROR_xxx value if aborted by an error condition
(e.g. stopped by repeated call or recursion limit)
*/
static int
match(REGISTER PCRE_PUCHAR eptr, REGISTER const pcre_uchar *ecode,
PCRE_PUCHAR mstart, int offset_top, match_data *md, eptrblock *eptrb,
unsigned int rdepth)
{
/* These variables do not need to be preserved over recursion in this function,
so they can be ordinary variables in all cases. Mark some of them with
"register" because they are used a lot in loops. */
register int rrc; /* Returns from recursive calls */
register int i; /* Used for loops not involving calls to RMATCH() */
register pcre_uint32 c; /* Character values not kept over RMATCH() calls */
register BOOL utf; /* Local copy of UTF flag for speed */
BOOL minimize, possessive; /* Quantifier options */
BOOL caseless;
int condcode;
/* When recursion is not being used, all "local" variables that have to be
preserved over calls to RMATCH() are part of a "frame". We set up the top-level
frame on the stack here; subsequent instantiations are obtained from the heap
whenever RMATCH() does a "recursion". See the macro definitions above. Putting
the top-level on the stack rather than malloc-ing them all gives a performance
boost in many cases where there is not much "recursion". */
#ifdef NO_RECURSE
heapframe *frame = (heapframe *)md->match_frames_base;
/* Copy in the original argument variables */
frame->Xeptr = eptr;
frame->Xecode = ecode;
frame->Xmstart = mstart;
frame->Xoffset_top = offset_top;
frame->Xeptrb = eptrb;
frame->Xrdepth = rdepth;
/* This is where control jumps back to to effect "recursion" */
HEAP_RECURSE:
/* Macros make the argument variables come from the current frame */
#define eptr frame->Xeptr
#define ecode frame->Xecode
#define mstart frame->Xmstart
#define offset_top frame->Xoffset_top
#define eptrb frame->Xeptrb
#define rdepth frame->Xrdepth
/* Ditto for the local variables */
#ifdef SUPPORT_UTF
#define charptr frame->Xcharptr
#endif
#define callpat frame->Xcallpat
#define codelink frame->Xcodelink
#define data frame->Xdata
#define next frame->Xnext
#define pp frame->Xpp
#define prev frame->Xprev
#define saved_eptr frame->Xsaved_eptr
#define new_recursive frame->Xnew_recursive
#define cur_is_word frame->Xcur_is_word
#define condition frame->Xcondition
#define prev_is_word frame->Xprev_is_word
#ifdef SUPPORT_UCP
#define prop_type frame->Xprop_type
#define prop_value frame->Xprop_value
#define prop_fail_result frame->Xprop_fail_result
#define oclength frame->Xoclength
#define occhars frame->Xocchars
#endif
#define ctype frame->Xctype
#define fc frame->Xfc
#define fi frame->Xfi
#define length frame->Xlength
#define max frame->Xmax
#define min frame->Xmin
#define number frame->Xnumber
#define offset frame->Xoffset
#define op frame->Xop
#define save_capture_last frame->Xsave_capture_last
#define save_offset1 frame->Xsave_offset1
#define save_offset2 frame->Xsave_offset2
#define save_offset3 frame->Xsave_offset3
#define stacksave frame->Xstacksave
#define newptrb frame->Xnewptrb
/* When recursion is being used, local variables are allocated on the stack and
get preserved during recursion in the normal way. In this environment, fi and
i, and fc and c, can be the same variables. */
#else /* NO_RECURSE not defined */
#define fi i
#define fc c
/* Many of the following variables are used only in small blocks of the code.
My normal style of coding would have declared them within each of those blocks.
However, in order to accommodate the version of this code that uses an external
"stack" implemented on the heap, it is easier to declare them all here, so the
declarations can be cut out in a block. The only declarations within blocks
below are for variables that do not have to be preserved over a recursive call
to RMATCH(). */
#ifdef SUPPORT_UTF
const pcre_uchar *charptr;
#endif
const pcre_uchar *callpat;
const pcre_uchar *data;
const pcre_uchar *next;
PCRE_PUCHAR pp;
const pcre_uchar *prev;
PCRE_PUCHAR saved_eptr;
recursion_info new_recursive;
BOOL cur_is_word;
BOOL condition;
BOOL prev_is_word;
#ifdef SUPPORT_UCP
int prop_type;
unsigned int prop_value;
int prop_fail_result;
int oclength;
pcre_uchar occhars[6];
#endif
int codelink;
int ctype;
int length;
int max;
int min;
unsigned int number;
int offset;
unsigned int op;
pcre_int32 save_capture_last;
int save_offset1, save_offset2, save_offset3;
int stacksave[REC_STACK_SAVE_MAX];
eptrblock newptrb;
/* There is a special fudge for calling match() in a way that causes it to
measure the size of its basic stack frame when the stack is being used for
recursion. The second argument (ecode) being NULL triggers this behaviour. It
cannot normally ever be NULL. The return is the negated value of the frame
size. */
if (ecode == NULL)
{
if (rdepth == 0)
return match((PCRE_PUCHAR)&rdepth, NULL, NULL, 0, NULL, NULL, 1);
else
{
int len = (int)((char *)&rdepth - (char *)eptr);
return (len > 0)? -len : len;
}
}
#endif /* NO_RECURSE */
/* To save space on the stack and in the heap frame, I have doubled up on some
of the local variables that are used only in localised parts of the code, but
still need to be preserved over recursive calls of match(). These macros define
the alternative names that are used. */
#define allow_zero cur_is_word
#define cbegroup condition
#define code_offset codelink
#define condassert condition
#define matched_once prev_is_word
#define foc number
#define save_mark data
/* These statements are here to stop the compiler complaining about unitialized
variables. */
#ifdef SUPPORT_UCP
prop_value = 0;
prop_fail_result = 0;
#endif
/* This label is used for tail recursion, which is used in a few cases even
when NO_RECURSE is not defined, in order to reduce the amount of stack that is
used. Thanks to Ian Taylor for noticing this possibility and sending the
original patch. */
TAIL_RECURSE:
/* OK, now we can get on with the real code of the function. Recursive calls
are specified by the macro RMATCH and RRETURN is used to return. When
NO_RECURSE is *not* defined, these just turn into a recursive call to match()
and a "return", respectively (possibly with some debugging if PCRE_DEBUG is
defined). However, RMATCH isn't like a function call because it's quite a
complicated macro. It has to be used in one particular way. This shouldn't,
however, impact performance when true recursion is being used. */
#ifdef SUPPORT_UTF
utf = md->utf; /* Local copy of the flag */
#else
utf = FALSE;
#endif
/* First check that we haven't called match() too many times, or that we
haven't exceeded the recursive call limit. */
if (md->match_call_count++ >= md->match_limit) RRETURN(PCRE_ERROR_MATCHLIMIT);
if (rdepth >= md->match_limit_recursion) RRETURN(PCRE_ERROR_RECURSIONLIMIT);
/* At the start of a group with an unlimited repeat that may match an empty
string, the variable md->match_function_type is set to MATCH_CBEGROUP. It is
done this way to save having to use another function argument, which would take
up space on the stack. See also MATCH_CONDASSERT below.
When MATCH_CBEGROUP is set, add the current subject pointer to the chain of
such remembered pointers, to be checked when we hit the closing ket, in order
to break infinite loops that match no characters. When match() is called in
other circumstances, don't add to the chain. The MATCH_CBEGROUP feature must
NOT be used with tail recursion, because the memory block that is used is on
the stack, so a new one may be required for each match(). */
if (md->match_function_type == MATCH_CBEGROUP)
{
newptrb.epb_saved_eptr = eptr;
newptrb.epb_prev = eptrb;
eptrb = &newptrb;
md->match_function_type = 0;
}
/* Now start processing the opcodes. */
for (;;)
{
minimize = possessive = FALSE;
op = *ecode;
switch(op)
{
case OP_MARK:
md->nomatch_mark = ecode + 2;
md->mark = NULL; /* In case previously set by assertion */
RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top, md,
eptrb, RM55);
if ((rrc == MATCH_MATCH || rrc == MATCH_ACCEPT || rrc == MATCH_KETRPOS) &&
md->mark == NULL) md->mark = ecode + 2;
/* A return of MATCH_SKIP_ARG means that matching failed at SKIP with an
argument, and we must check whether that argument matches this MARK's
argument. It is passed back in md->start_match_ptr (an overloading of that
variable). If it does match, we reset that variable to the current subject
position and return MATCH_SKIP. Otherwise, pass back the return code
unaltered. */
else if (rrc == MATCH_SKIP_ARG &&
STRCMP_UC_UC_TEST(ecode + 2, md->start_match_ptr) == 0)
{
md->start_match_ptr = eptr;
RRETURN(MATCH_SKIP);
}
RRETURN(rrc);
case OP_FAIL:
RRETURN(MATCH_NOMATCH);
case OP_COMMIT:
RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
eptrb, RM52);
if (rrc != MATCH_NOMATCH) RRETURN(rrc);
RRETURN(MATCH_COMMIT);
case OP_PRUNE:
RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
eptrb, RM51);
if (rrc != MATCH_NOMATCH) RRETURN(rrc);
RRETURN(MATCH_PRUNE);
case OP_PRUNE_ARG:
md->nomatch_mark = ecode + 2;
md->mark = NULL; /* In case previously set by assertion */
RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top, md,
eptrb, RM56);
if ((rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) &&
md->mark == NULL) md->mark = ecode + 2;
if (rrc != MATCH_NOMATCH) RRETURN(rrc);
RRETURN(MATCH_PRUNE);
case OP_SKIP:
RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
eptrb, RM53);
if (rrc != MATCH_NOMATCH) RRETURN(rrc);
md->start_match_ptr = eptr; /* Pass back current position */
RRETURN(MATCH_SKIP);
/* Note that, for Perl compatibility, SKIP with an argument does NOT set
nomatch_mark. When a pattern match ends with a SKIP_ARG for which there was
not a matching mark, we have to re-run the match, ignoring the SKIP_ARG
that failed and any that precede it (either they also failed, or were not
triggered). To do this, we maintain a count of executed SKIP_ARGs. If a
SKIP_ARG gets to top level, the match is re-run with md->ignore_skip_arg
set to the count of the one that failed. */
case OP_SKIP_ARG:
md->skip_arg_count++;
if (md->skip_arg_count <= md->ignore_skip_arg)
{
ecode += PRIV(OP_lengths)[*ecode] + ecode[1];
break;
}
RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top, md,
eptrb, RM57);
if (rrc != MATCH_NOMATCH) RRETURN(rrc);
/* Pass back the current skip name by overloading md->start_match_ptr and
returning the special MATCH_SKIP_ARG return code. This will either be
caught by a matching MARK, or get to the top, where it causes a rematch
with md->ignore_skip_arg set to the value of md->skip_arg_count. */
md->start_match_ptr = ecode + 2;
RRETURN(MATCH_SKIP_ARG);
/* For THEN (and THEN_ARG) we pass back the address of the opcode, so that
the branch in which it occurs can be determined. Overload the start of
match pointer to do this. */
case OP_THEN:
RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
eptrb, RM54);
if (rrc != MATCH_NOMATCH) RRETURN(rrc);
md->start_match_ptr = ecode;
RRETURN(MATCH_THEN);
case OP_THEN_ARG:
md->nomatch_mark = ecode + 2;
md->mark = NULL; /* In case previously set by assertion */
RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode] + ecode[1], offset_top,
md, eptrb, RM58);
if ((rrc == MATCH_MATCH || rrc == MATCH_ACCEPT) &&
md->mark == NULL) md->mark = ecode + 2;
if (rrc != MATCH_NOMATCH) RRETURN(rrc);
md->start_match_ptr = ecode;
RRETURN(MATCH_THEN);
/* Handle an atomic group that does not contain any capturing parentheses.
This can be handled like an assertion. Prior to 8.13, all atomic groups
were handled this way. In 8.13, the code was changed as below for ONCE, so
that backups pass through the group and thereby reset captured values.
However, this uses a lot more stack, so in 8.20, atomic groups that do not
contain any captures generate OP_ONCE_NC, which can be handled in the old,
less stack intensive way.
Check the alternative branches in turn - the matching won't pass the KET
for this kind of subpattern. If any one branch matches, we carry on as at
the end of a normal bracket, leaving the subject pointer, but resetting
the start-of-match value in case it was changed by \K. */
case OP_ONCE_NC:
prev = ecode;
saved_eptr = eptr;
save_mark = md->mark;
do
{
RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM64);
if (rrc == MATCH_MATCH) /* Note: _not_ MATCH_ACCEPT */
{
mstart = md->start_match_ptr;
break;
}
if (rrc == MATCH_THEN)
{
next = ecode + GET(ecode,1);
if (md->start_match_ptr < next &&
(*ecode == OP_ALT || *next == OP_ALT))
rrc = MATCH_NOMATCH;
}
if (rrc != MATCH_NOMATCH) RRETURN(rrc);
ecode += GET(ecode,1);
md->mark = save_mark;
}
while (*ecode == OP_ALT);
/* If hit the end of the group (which could be repeated), fail */
if (*ecode != OP_ONCE_NC && *ecode != OP_ALT) RRETURN(MATCH_NOMATCH);
/* Continue as from after the group, updating the offsets high water
mark, since extracts may have been taken. */
do ecode += GET(ecode, 1); while (*ecode == OP_ALT);
offset_top = md->end_offset_top;
eptr = md->end_match_ptr;
/* For a non-repeating ket, just continue at this level. This also
happens for a repeating ket if no characters were matched in the group.
This is the forcible breaking of infinite loops as implemented in Perl
5.005. */
if (*ecode == OP_KET || eptr == saved_eptr)
{
ecode += 1+LINK_SIZE;
break;
}
/* The repeating kets try the rest of the pattern or restart from the
preceding bracket, in the appropriate order. The second "call" of match()
uses tail recursion, to avoid using another stack frame. */
if (*ecode == OP_KETRMIN)
{
RMATCH(eptr, ecode + 1 + LINK_SIZE, offset_top, md, eptrb, RM65);
if (rrc != MATCH_NOMATCH) RRETURN(rrc);
ecode = prev;
goto TAIL_RECURSE;
}
else /* OP_KETRMAX */
{
RMATCH(eptr, prev, offset_top, md, eptrb, RM66);
if (rrc != MATCH_NOMATCH) RRETURN(rrc);
ecode += 1 + LINK_SIZE;
goto TAIL_RECURSE;
}
/* Control never gets here */
/* Handle a capturing bracket, other than those that are possessive with an
unlimited repeat. If there is space in the offset vector, save the current
subject position in the working slot at the top of the vector. We mustn't
change the current values of the data slot, because they may be set from a
previous iteration of this group, and be referred to by a reference inside
the group. A failure to match might occur after the group has succeeded,
if something later on doesn't match. For this reason, we need to restore
the working value and also the values of the final offsets, in case they
were set by a previous iteration of the same bracket.
If there isn't enough space in the offset vector, treat this as if it were
a non-capturing bracket. Don't worry about setting the flag for the error
case here; that is handled in the code for KET. */
case OP_CBRA:
case OP_SCBRA:
number = GET2(ecode, 1+LINK_SIZE);
offset = number << 1;
#ifdef PCRE_DEBUG
printf("start bracket %d\n", number);
printf("subject=");
pchars(eptr, 16, TRUE, md);
printf("\n");
#endif
if (offset < md->offset_max)
{
save_offset1 = md->offset_vector[offset];
save_offset2 = md->offset_vector[offset+1];
save_offset3 = md->offset_vector[md->offset_end - number];
save_capture_last = md->capture_last;
save_mark = md->mark;
DPRINTF(("saving %d %d %d\n", save_offset1, save_offset2, save_offset3));
md->offset_vector[md->offset_end - number] =
(int)(eptr - md->start_subject);
for (;;)
{
if (op >= OP_SBRA) md->match_function_type = MATCH_CBEGROUP;
RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md,
eptrb, RM1);
if (rrc == MATCH_ONCE) break; /* Backing up through an atomic group */
/* If we backed up to a THEN, check whether it is within the current
branch by comparing the address of the THEN that is passed back with
the end of the branch. If it is within the current branch, and the
branch is one of two or more alternatives (it either starts or ends
with OP_ALT), we have reached the limit of THEN's action, so convert
the return code to NOMATCH, which will cause normal backtracking to
happen from now on. Otherwise, THEN is passed back to an outer
alternative. This implements Perl's treatment of parenthesized groups,
where a group not containing | does not affect the current alternative,
that is, (X) is NOT the same as (X|(*F)). */
if (rrc == MATCH_THEN)
{
next = ecode + GET(ecode,1);
if (md->start_match_ptr < next &&
(*ecode == OP_ALT || *next == OP_ALT))
rrc = MATCH_NOMATCH;
}
/* Anything other than NOMATCH is passed back. */
if (rrc != MATCH_NOMATCH) RRETURN(rrc);
md->capture_last = save_capture_last;
ecode += GET(ecode, 1);
md->mark = save_mark;
if (*ecode != OP_ALT) break;
}
DPRINTF(("bracket %d failed\n", number));
md->offset_vector[offset] = save_offset1;
md->offset_vector[offset+1] = save_offset2;
md->offset_vector[md->offset_end - number] = save_offset3;
/* At this point, rrc will be one of MATCH_ONCE or MATCH_NOMATCH. */
RRETURN(rrc);
}
/* FALL THROUGH ... Insufficient room for saving captured contents. Treat
as a non-capturing bracket. */
/* VVVVVVVVVVVVVVVVVVVVVVVVV */
/* VVVVVVVVVVVVVVVVVVVVVVVVV */
DPRINTF(("insufficient capture room: treat as non-capturing\n"));
/* VVVVVVVVVVVVVVVVVVVVVVVVV */
/* VVVVVVVVVVVVVVVVVVVVVVVVV */
/* Non-capturing or atomic group, except for possessive with unlimited
repeat and ONCE group with no captures. Loop for all the alternatives.
When we get to the final alternative within the brackets, we used to return
the result of a recursive call to match() whatever happened so it was
possible to reduce stack usage by turning this into a tail recursion,
except in the case of a possibly empty group. However, now that there is
the possiblity of (*THEN) occurring in the final alternative, this
optimization is no longer always possible.
We can optimize if we know there are no (*THEN)s in the pattern; at present
this is the best that can be done.
MATCH_ONCE is returned when the end of an atomic group is successfully
reached, but subsequent matching fails. It passes back up the tree (causing
captured values to be reset) until the original atomic group level is
reached. This is tested by comparing md->once_target with the start of the
group. At this point, the return is converted into MATCH_NOMATCH so that
previous backup points can be taken. */
case OP_ONCE:
case OP_BRA:
case OP_SBRA:
DPRINTF(("start non-capturing bracket\n"));
for (;;)
{
if (op >= OP_SBRA || op == OP_ONCE)
md->match_function_type = MATCH_CBEGROUP;
/* If this is not a possibly empty group, and there are no (*THEN)s in
the pattern, and this is the final alternative, optimize as described
above. */
else if (!md->hasthen && ecode[GET(ecode, 1)] != OP_ALT)
{
ecode += PRIV(OP_lengths)[*ecode];
goto TAIL_RECURSE;
}
/* In all other cases, we have to make another call to match(). */
save_mark = md->mark;
save_capture_last = md->capture_last;
RMATCH(eptr, ecode + PRIV(OP_lengths)[*ecode], offset_top, md, eptrb,
RM2);
/* See comment in the code for capturing groups above about handling
THEN. */
if (rrc == MATCH_THEN)
{
next = ecode + GET(ecode,1);