Git-Raw
view release on metacpan or search on metacpan
deps/libgit2/deps/pcre/pcre_exec.c view on Meta::CPAN
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
deps/libgit2/deps/pcre/pcre_exec.c view on Meta::CPAN
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
( run in 0.820 second using v1.01-cache-2.11-cpan-8f98c5d2c55 )