生物技术

开发平台：

C/C++

pcre.c：源码内容

}
}
length++;
/* A back reference needs an additional 2 bytes, plus either one or 5
bytes for a repeat. We also need to keep the value of the highest
back reference. */
if (c <= -ESC_REF)
{
int refnum = -c - ESC_REF;
if (refnum > top_backref) top_backref = refnum;
length += 2; /* For single back reference */
if (ptr[1] == '{' && is_counted_repeat(ptr+2, &compile_block))
{
ptr = read_repeat_counts(ptr+2, &min, &max, errorptr, &compile_block);
if (*errorptr != NULL) goto PCRE_ERROR_RETURN;
if ((min == 0 && (max == 1 || max == -1)) ||
(min == 1 && max == -1))
length++;
else length += 5;
if (ptr[1] == '?') ptr++;
}
}
continue;
case '^':
case '.':
case '$':
case '*': /* These repeats won't be after brackets; */
case '+': /* those are handled separately */
case '?':
length++;
continue;
/* This covers the cases of repeats after a single char, metachar, class,
or back reference. */
case '{':
if (!is_counted_repeat(ptr+1, &compile_block)) goto NORMAL_CHAR;
ptr = read_repeat_counts(ptr+1, &min, &max, errorptr, &compile_block);
if (*errorptr != NULL) goto PCRE_ERROR_RETURN;
if ((min == 0 && (max == 1 || max == -1)) ||
(min == 1 && max == -1))
length++;
else
{
length--; /* Uncount the original char or metachar */
if (min == 1) length++; else if (min > 0) length += 4;
if (max > 0) length += 4; else length += 2;
}
if (ptr[1] == '?') ptr++;
continue;
/* An alternation contains an offset to the next branch or ket. If any ims
options changed in the previous branch(es), and/or if we are in a
lookbehind assertion, extra space will be needed at the start of the
branch. This is handled by branch_extra. */
case '|':
length += 3 + branch_extra;
continue;
/* A character class uses 33 characters. Don't worry about character types
that aren't allowed in classes - they'll get picked up during the compile.
A character class that contains only one character uses 2 or 3 bytes,
depending on whether it is negated or not. Notice this where we can. */
case '[':
class_charcount = 0;
if (*(++ptr) == '^') ptr++;
do
{
if (*ptr == '\')
{
int ch = check_escape(&ptr, errorptr, bracount, options, TRUE,
&compile_block);
if (*errorptr != NULL) goto PCRE_ERROR_RETURN;
if (-ch == ESC_b) class_charcount++; else class_charcount = 10;
}
else class_charcount++;
ptr++;
}
while (*ptr != 0 && *ptr != ']');
/* Repeats for negated single chars are handled by the general code */
if (class_charcount == 1) length += 3; else
{
length += 33;
/* A repeat needs either 1 or 5 bytes. */
if (*ptr != 0 && ptr[1] == '{' && is_counted_repeat(ptr+2, &compile_block))
{
ptr = read_repeat_counts(ptr+2, &min, &max, errorptr, &compile_block);
if (*errorptr != NULL) goto PCRE_ERROR_RETURN;
if ((min == 0 && (max == 1 || max == -1)) ||
(min == 1 && max == -1))
length++;
else length += 5;
if (ptr[1] == '?') ptr++;
}
}
continue;
/* Brackets may be genuine groups or special things */
case '(':
branch_newextra = 0;
bracket_length = 3;
/* Handle special forms of bracket, which all start (? */
if (ptr[1] == '?')
{
int set, unset;
int *optset;
switch (c = ptr[2])
{
/* Skip over comments entirely */
case '#':
ptr += 3;
while (*ptr != 0 && *ptr != ')') ptr++;
if (*ptr == 0)
{
*errorptr = ERR18;
goto PCRE_ERROR_RETURN;
}
continue;
/* Non-referencing groups and lookaheads just move the pointer on, and
then behave like a non-special bracket, except that they don't increment
the count of extracting brackets. Ditto for the "once only" bracket,
which is in Perl from version 5.005. */
case ':':
case '=':
case '!':
case '>':
ptr += 2;
break;
/* A recursive call to the regex is an extension, to provide the
facility which can be obtained by $(?p{perl-code}) in Perl 5.6. */
case 'R':
if (ptr[3] != ')')
{
*errorptr = ERR29;
goto PCRE_ERROR_RETURN;
}
ptr += 3;
length += 1;
break;
/* Lookbehinds are in Perl from version 5.005 */
case '<':
if (ptr[3] == '=' || ptr[3] == '!')
{
ptr += 3;
branch_newextra = 3;
length += 3; /* For the first branch */
break;
}
*errorptr = ERR24;
goto PCRE_ERROR_RETURN;
/* Conditionals are in Perl from version 5.005. The bracket must either
be followed by a number (for bracket reference) or by an assertion
group. */
case '(':
if ((compile_block.ctypes[ptr[3]] & ctype_digit) != 0)
{
ptr += 4;
length += 3;
while ((compile_block.ctypes[*ptr] & ctype_digit) != 0) ptr++;
if (*ptr != ')')
{
*errorptr = ERR26;
goto PCRE_ERROR_RETURN;
}
}
else /* An assertion must follow */
{
ptr++; /* Can treat like ':' as far as spacing is concerned */
if (ptr[2] != '?' ||
(ptr[3] != '=' && ptr[3] != '!' && ptr[3] != '<') )
{
ptr += 2; /* To get right offset in message */
*errorptr = ERR28;
goto PCRE_ERROR_RETURN;
}
}
break;
/* Else loop checking valid options until ) is met. Anything else is an
error. If we are without any brackets, i.e. at top level, the settings
act as if specified in the options, so massage the options immediately.
This is for backward compatibility with Perl 5.004. */
default:
set = unset = 0;
optset = &set;
ptr += 2;
for (;; ptr++)
{
c = *ptr;
switch (c)
{
case 'i':
*optset |= PCRE_CASELESS;
continue;
case 'm':
*optset |= PCRE_MULTILINE;
continue;
case 's':
*optset |= PCRE_DOTALL;
continue;
case 'x':
*optset |= PCRE_EXTENDED;
continue;
case 'X':
*optset |= PCRE_EXTRA;
continue;
case 'U':
*optset |= PCRE_UNGREEDY;
continue;
case '-':
optset = &unset;
continue;
/* A termination by ')' indicates an options-setting-only item;
this is global at top level; otherwise nothing is done here and
it is handled during the compiling process on a per-bracket-group
basis. */
case ')':
if (brastackptr == 0)
{
options = (options | set) & (~unset);
set = unset = 0; /* To save length */
}
/* Fall through */
/* A termination by ':' indicates the start of a nested group with
the given options set. This is again handled at compile time, but
we must allow for compiled space if any of the ims options are
set. We also have to allow for resetting space at the end of
the group, which is why 4 is added to the length and not just 2.
If there are several changes of options within the same group, this
will lead to an over-estimate on the length, but this shouldn't
matter very much. We also have to allow for resetting options at
the start of any alternations, which we do by setting
branch_newextra to 2. Finally, we record whether the case-dependent
flag ever changes within the regex. This is used by the "required
character" code. */
case ':':
if (((set|unset) & PCRE_IMS) != 0)
{
length += 4;
branch_newextra = 2;
if (((set|unset) & PCRE_CASELESS) != 0) options |= PCRE_ICHANGED;
}
goto END_OPTIONS;
/* Unrecognized option character */
default:
*errorptr = ERR12;
goto PCRE_ERROR_RETURN;
}
}
/* If we hit a closing bracket, that's it - this is a freestanding
option-setting. We need to ensure that branch_extra is updated if
necessary. The only values branch_newextra can have here are 0 or 2.
If the value is 2, then branch_extra must either be 2 or 5, depending
on whether this is a lookbehind group or not. */
END_OPTIONS:
if (c == ')')
{
if (branch_newextra == 2 && (branch_extra == 0 || branch_extra == 3))
branch_extra += branch_newextra;
continue;
}
/* If options were terminated by ':' control comes here. Fall through
to handle the group below. */
}
}
/* Extracting brackets must be counted so we can process escapes in a
Perlish way. If the number exceeds EXTRACT_BASIC_MAX we are going to
need an additional 3 bytes of store per extracting bracket. */
else
{
bracount++;
if (bracount > EXTRACT_BASIC_MAX) bracket_length += 3;
}
/* Save length for computing whole length at end if there's a repeat that
requires duplication of the group. Also save the current value of
branch_extra, and start the new group with the new value. If non-zero, this
will either be 2 for a (?imsx: group, or 3 for a lookbehind assertion. */
if (brastackptr >= sizeof(brastack)/sizeof(int))
{
*errorptr = ERR19;
goto PCRE_ERROR_RETURN;
}
bralenstack[brastackptr] = branch_extra;
branch_extra = branch_newextra;
brastack[brastackptr++] = length;
length += bracket_length;
continue;
/* Handle ket. Look for subsequent max/min; for certain sets of values we
have to replicate this bracket up to that many times. If brastackptr is
0 this is an unmatched bracket which will generate an error, but take care
not to try to access brastack[-1] when computing the length and restoring
the branch_extra value. */
case ')':
length += 3;
{
int minval = 1;
int maxval = 1;
int duplength;
if (brastackptr > 0)
{
duplength = length - brastack[--brastackptr];
branch_extra = bralenstack[brastackptr];
}
else duplength = 0;
/* Leave ptr at the final char; for read_repeat_counts this happens
automatically; for the others we need an increment. */
if ((c = ptr[1]) == '{' && is_counted_repeat(ptr+2, &compile_block))
{
ptr = read_repeat_counts(ptr+2, &minval, &maxval, errorptr,
&compile_block);
if (*errorptr != NULL) goto PCRE_ERROR_RETURN;
}
else if (c == '*') { minval = 0; maxval = -1; ptr++; }
else if (c == '+') { maxval = -1; ptr++; }
else if (c == '?') { minval = 0; ptr++; }
/* If the minimum is zero, we have to allow for an OP_BRAZERO before the
group, and if the maximum is greater than zero, we have to replicate
maxval-1 times; each replication acquires an OP_BRAZERO plus a nesting
bracket set - hence the 7. */
if (minval == 0)
{
length++;
if (maxval > 0) length += (maxval - 1) * (duplength + 7);
}
/* When the minimum is greater than zero, 1 we have to replicate up to
minval-1 times, with no additions required in the copies. Then, if
there is a limited maximum we have to replicate up to maxval-1 times
allowing for a BRAZERO item before each optional copy and nesting
brackets for all but one of the optional copies. */
else
{
length += (minval - 1) * duplength;
if (maxval > minval) /* Need this test as maxval=-1 means no limit */
length += (maxval - minval) * (duplength + 7) - 6;
}
}
continue;
/* Non-special character. For a run of such characters the length required
is the number of characters + 2, except that the maximum run length is 255.
We won't get a skipped space or a non-data escape or the start of a #
comment as the first character, so the length can't be zero. */
NORMAL_CHAR:
default:
length += 2;
runlength = 0;
do
{
if ((options & PCRE_EXTENDED) != 0)
{
if ((compile_block.ctypes[c] & ctype_space) != 0) continue;
if (c == '#')
{
/* The space before the ; is to avoid a warning on a silly compiler
on the Macintosh. */
while ((c = *(++ptr)) != 0 && c != NEWLINE) ;
continue;
}
}
/* Backslash may introduce a data char or a metacharacter; stop the
string before the latter. */
if (c == '\')
{
const uschar *saveptr = ptr;
c = check_escape(&ptr, errorptr, bracount, options, FALSE,
&compile_block);
if (*errorptr != NULL) goto PCRE_ERROR_RETURN;
if (c < 0) { ptr = saveptr; break; }
#ifdef SUPPORT_UTF8
if (c > 127 && (options & PCRE_UTF8) != 0)
{
int i;
for (i = 0; i < sizeof(utf8_table1)/sizeof(int); i++)
if (c <= utf8_table1[i]) break;
runlength += i;
}
#endif
}
/* Ordinary character or single-char escape */
runlength++;
}
/* This "while" is the end of the "do" above. */
while (runlength < MAXLIT &&
(compile_block.ctypes[c = *(++ptr)] & ctype_meta) == 0);
ptr--;
length += runlength;
continue;
}
}
length += 4; /* For final KET and END */
if (length > 65539)
{
*errorptr = ERR20;
return NULL;
}
/* Compute the size of data block needed and get it, either from malloc or
externally provided function. We specify "code[0]" in the offsetof() expression
rather than just "code", because it has been reported that one broken compiler
fails on "code" because it is also an independent variable. It should make no
difference to the value of the offsetof(). */
size = length + offsetof(real_pcre, code[0]);
re = (real_pcre *)(pcre_malloc)(size);
if (re == NULL)
{
*errorptr = ERR21;
return NULL;
}
/* Put in the magic number, and save the size, options, and table pointer */
re->magic_number = MAGIC_NUMBER;
re->size = size;
re->options = options;
re->tables = tables;
/* Set up a starting, non-extracting bracket, then compile the expression. On
error, *errorptr will be set non-NULL, so we don't need to look at the result
of the function here. */
ptr = (const uschar *)pattern;
code = re->code;
*code = OP_BRA;
bracount = 0;
(void)compile_regex(options, -1, &bracount, &code, &ptr, errorptr, FALSE, 0,
&reqchar, &countlits, &compile_block);
re->top_bracket = bracount;
re->top_backref = top_backref;
/* If not reached end of pattern on success, there's an excess bracket. */
if (*errorptr == NULL && *ptr != 0) *errorptr = ERR22;
/* Fill in the terminating state and check for disastrous overflow, but
if debugging, leave the test till after things are printed out. */
*code++ = OP_END;
#ifndef DEBUG
if (code - re->code > length) *errorptr = ERR23;
#endif
/* Give an error if there's back reference to a non-existent capturing
subpattern. */
if (top_backref > re->top_bracket) *errorptr = ERR15;
/* Failed to compile */
if (*errorptr != NULL)
{
(pcre_free)(re);
PCRE_ERROR_RETURN:
*erroroffset = ptr - (const uschar *)pattern;
return NULL;
}
/* If the anchored option was not passed, set flag if we can determine that the
pattern is anchored by virtue of ^ characters or A or anything else (such as
starting with .* when DOTALL is set).
Otherwise, see if we can determine what the first character has to be, because
that speeds up unanchored matches no end. If not, see if we can set the
PCRE_STARTLINE flag. This is helpful for multiline matches when all branches
start with ^. and also when all branches start with .* for non-DOTALL matches.
*/
if ((options & PCRE_ANCHORED) == 0)
{
int temp_options = options;
if (is_anchored(re->code, &temp_options))
re->options |= PCRE_ANCHORED;
else
{
int ch = find_firstchar(re->code, &temp_options);
if (ch >= 0)
{
re->first_char = ch;
re->options |= PCRE_FIRSTSET;
}
else if (is_startline(re->code))
re->options |= PCRE_STARTLINE;
}
}
/* Save the last required character if there are at least two literal
characters on all paths, or if there is no first character setting. */
if (reqchar >= 0 && (countlits > 1 || (re->options & PCRE_FIRSTSET) == 0))
{
re->req_char = reqchar;
re->options |= PCRE_REQCHSET;
}
/* Print out the compiled data for debugging */
#ifdef DEBUG
printf("Length = %d top_bracket = %d top_backref = %dn",
length, re->top_bracket, re->top_backref);
if (re->options != 0)
{
printf("%s%s%s%s%s%s%s%s%sn",
((re->options & PCRE_ANCHORED) != 0)? "anchored " : "",
((re->options & PCRE_CASELESS) != 0)? "caseless " : "",
((re->options & PCRE_ICHANGED) != 0)? "case state changed " : "",
((re->options & PCRE_EXTENDED) != 0)? "extended " : "",
((re->options & PCRE_MULTILINE) != 0)? "multiline " : "",
((re->options & PCRE_DOTALL) != 0)? "dotall " : "",
((re->options & PCRE_DOLLAR_ENDONLY) != 0)? "endonly " : "",
((re->options & PCRE_EXTRA) != 0)? "extra " : "",
((re->options & PCRE_UNGREEDY) != 0)? "ungreedy " : "");
}
if ((re->options & PCRE_FIRSTSET) != 0)
{
if (isprint(re->first_char)) printf("First char = %cn", re->first_char);
else printf("First char = \x%02xn", re->first_char);
}
if ((re->options & PCRE_REQCHSET) != 0)
{
if (isprint(re->req_char)) printf("Req char = %cn", re->req_char);
else printf("Req char = \x%02xn", re->req_char);
}
code_end = code;
code_base = code = re->code;
while (code < code_end)
{
int charlength;
printf("%3d ", code - code_base);
if (*code >= OP_BRA)
{
if (*code - OP_BRA > EXTRACT_BASIC_MAX)
printf("%3d Bra extra", (code[1] << 8) + code[2]);
else
printf("%3d Bra %d", (code[1] << 8) + code[2], *code - OP_BRA);
code += 2;
}
else switch(*code)
{
case OP_OPT:
printf(" %.2x %s", code[1], OP_names[*code]);
code++;
break;
case OP_CHARS:
charlength = *(++code);
printf("%3d ", charlength);
while (charlength-- > 0)
if (isprint(c = *(++code))) printf("%c", c); else printf("\x%02x", c);
break;
case OP_KETRMAX:
case OP_KETRMIN:
case OP_ALT:
case OP_KET:
case OP_ASSERT:
case OP_ASSERT_NOT:
case OP_ASSERTBACK:
case OP_ASSERTBACK_NOT:
case OP_ONCE:
case OP_REVERSE:
case OP_BRANUMBER:
case OP_COND:
case OP_CREF:
printf("%3d %s", (code[1] << 8) + code[2], OP_names[*code]);
code += 2;
break;
case OP_STAR:
case OP_MINSTAR:
case OP_PLUS:
case OP_MINPLUS:
case OP_QUERY:
case OP_MINQUERY:
case OP_TYPESTAR:
case OP_TYPEMINSTAR:
case OP_TYPEPLUS:
case OP_TYPEMINPLUS:
case OP_TYPEQUERY:
case OP_TYPEMINQUERY:
if (*code >= OP_TYPESTAR)
printf(" %s", OP_names[code[1]]);
else if (isprint(c = code[1])) printf(" %c", c);
else printf(" \x%02x", c);
printf("%s", OP_names[*code++]);
break;
case OP_EXACT:
case OP_UPTO:
case OP_MINUPTO:
if (isprint(c = code[3])) printf(" %c{", c);
else printf(" \x%02x{", c);
if (*code != OP_EXACT) printf("0,");
printf("%d}", (code[1] << 8) + code[2]);
if (*code == OP_MINUPTO) printf("?");
code += 3;
break;
case OP_TYPEEXACT:
case OP_TYPEUPTO:
case OP_TYPEMINUPTO:
printf(" %s{", OP_names[code[3]]);
if (*code != OP_TYPEEXACT) printf(",");
printf("%d}", (code[1] << 8) + code[2]);
if (*code == OP_TYPEMINUPTO) printf("?");
code += 3;
break;
case OP_NOT:
if (isprint(c = *(++code))) printf(" [^%c]", c);
else printf(" [^\x%02x]", c);
break;
case OP_NOTSTAR:
case OP_NOTMINSTAR:
case OP_NOTPLUS:
case OP_NOTMINPLUS:
case OP_NOTQUERY:
case OP_NOTMINQUERY:
if (isprint(c = code[1])) printf(" [^%c]", c);
else printf(" [^\x%02x]", c);
printf("%s", OP_names[*code++]);
break;
case OP_NOTEXACT:
case OP_NOTUPTO:
case OP_NOTMINUPTO:
if (isprint(c = code[3])) printf(" [^%c]{", c);
else printf(" [^\x%02x]{", c);
if (*code != OP_NOTEXACT) printf(",");
printf("%d}", (code[1] << 8) + code[2]);
if (*code == OP_NOTMINUPTO) printf("?");
code += 3;
break;
case OP_REF:
printf(" \%d", (code[1] << 8) | code[2]);
code += 3;
goto CLASS_REF_REPEAT;
case OP_CLASS:
{
int i, min, max;
code++;
printf(" [");
for (i = 0; i < 256; i++)
{
if ((code[i/8] & (1 << (i&7))) != 0)
{
int j;
for (j = i+1; j < 256; j++)
if ((code[j/8] & (1 << (j&7))) == 0) break;
if (i == '-' || i == ']') printf("\");
if (isprint(i)) printf("%c", i); else printf("\x%02x", i);
if (--j > i)
{
printf("-");
if (j == '-' || j == ']') printf("\");
if (isprint(j)) printf("%c", j); else printf("\x%02x", j);
}
i = j;
}
}
printf("]");
code += 32;
CLASS_REF_REPEAT:
switch(*code)
{
case OP_CRSTAR:
case OP_CRMINSTAR:
case OP_CRPLUS:
case OP_CRMINPLUS:
case OP_CRQUERY:
case OP_CRMINQUERY:
printf("%s", OP_names[*code]);
break;
case OP_CRRANGE:
case OP_CRMINRANGE:
min = (code[1] << 8) + code[2];
max = (code[3] << 8) + code[4];
if (max == 0) printf("{%d,}", min);
else printf("{%d,%d}", min, max);
if (*code == OP_CRMINRANGE) printf("?");
code += 4;
break;
default:
code--;
}
}
break;
/* Anything else is just a one-node item */
default:
printf(" %s", OP_names[*code]);
break;
}
code++;
printf("n");
}
printf("------------------------------------------------------------------n");
/* This check is done here in the debugging case so that the code that
was compiled can be seen. */
if (code - re->code > length)
{
*errorptr = ERR23;
(pcre_free)(re);
*erroroffset = ptr - (uschar *)pattern;
return NULL;
}
#endif
return (pcre *)re;
}
/*************************************************
* Match a back-reference *
*************************************************/
/* If a back reference hasn't been set, the length that is passed is greater
than the number of characters left in the string, so the match fails.
Arguments:
offset index into the offset vector
eptr points into the subject
length length to be matched
md points to match data block
ims the ims flags
Returns: TRUE if matched
*/
static BOOL
match_ref(int offset, register const uschar *eptr, int length, match_data *md,
unsigned long int ims)
{
const uschar *p = md->start_subject + md->offset_vector[offset];
#ifdef 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 not enough characters left */
if (length > md->end_subject - eptr) return FALSE;
/* Separate the caselesss case for speed */
if ((ims & PCRE_CASELESS) != 0)
{
while (length-- > 0)
if (md->lcc[*p++] != md->lcc[*eptr++]) return FALSE;
}
else
{ while (length-- > 0) if (*p++ != *eptr++) return FALSE; }
return TRUE;
}
/*************************************************
* Match from current position *
*************************************************/
/* On entry ecode points to the first opcode, and eptr to the first character
in the subject string, while eptrb holds the value of eptr at the start of the
last bracketed group - used for breaking infinite loops matching zero-length
strings.
Arguments:
eptr pointer in subject
ecode position in code
offset_top current top pointer
md pointer to "static" info for the match
ims current /i, /m, and /s options
eptrb pointer to chain of blocks containing eptr at start of
brackets - for testing for empty matches
flags can contain
match_condassert - this is an assertion condition
match_isgroup - this is the start of a bracketed group
Returns: TRUE if matched
*/
static BOOL
match(register const uschar *eptr, register const uschar *ecode,
int offset_top, match_data *md, unsigned long int ims, eptrblock *eptrb,
int flags)
{
unsigned long int original_ims = ims; /* Save for resetting on ')' */
eptrblock newptrb;
/* At the start of a bracketed group, add the current subject pointer to the
stack of such pointers, to be re-instated at the end of the group when we hit
the closing ket. When match() is called in other circumstances, we don't add to
the stack. */
if ((flags & match_isgroup) != 0)
{
newptrb.prev = eptrb;
newptrb.saved_eptr = eptr;
eptrb = &newptrb;
}
/* Now start processing the operations. */
for (;;)
{
int op = (int)*ecode;
int min, max, ctype;
register int i;
register int c;
BOOL minimize = FALSE;
/* Opening capturing bracket. 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.
If the bracket fails to match, we need to restore this 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. */
if (op > OP_BRA)
{
int offset;
int number = op - OP_BRA;
/* For extended extraction brackets (large number), we have to fish out the
number from a dummy opcode at the start. */
if (number > EXTRACT_BASIC_MAX) number = (ecode[4] << 8) | ecode[5];
offset = number << 1;
#ifdef DEBUG
printf("start bracket %d subject=", number);
pchars(eptr, 16, TRUE, md);
printf("n");
#endif
if (offset < md->offset_max)
{
int save_offset1 = md->offset_vector[offset];
int save_offset2 = md->offset_vector[offset+1];
int save_offset3 = md->offset_vector[md->offset_end - number];
DPRINTF(("saving %d %d %dn", save_offset1, save_offset2, save_offset3));
md->offset_vector[md->offset_end - number] = eptr - md->start_subject;
do
{
if (match(eptr, ecode+3, offset_top, md, ims, eptrb, match_isgroup))
return TRUE;
ecode += (ecode[1] << 8) + ecode[2];
}
while (*ecode == OP_ALT);
DPRINTF(("bracket %d failedn", number));
md->offset_vector[offset] = save_offset1;
md->offset_vector[offset+1] = save_offset2;
md->offset_vector[md->offset_end - number] = save_offset3;
return FALSE;
}
/* Insufficient room for saving captured contents */
else op = OP_BRA;
}
/* Other types of node can be handled by a switch */
switch(op)
{
case OP_BRA: /* Non-capturing bracket: optimized */
DPRINTF(("start bracket 0n"));
do
{
if (match(eptr, ecode+3, offset_top, md, ims, eptrb, match_isgroup))
return TRUE;
ecode += (ecode[1] << 8) + ecode[2];
}
while (*ecode == OP_ALT);
DPRINTF(("bracket 0 failedn"));
return FALSE;
/* Conditional group: compilation checked that there are no more than
two branches. If the condition is false, skipping the first branch takes us
past the end if there is only one branch, but that's OK because that is
exactly what going to the ket would do. */
case OP_COND:
if (ecode[3] == OP_CREF) /* Condition is extraction test */
{
int offset = (ecode[4] << 9) | (ecode[5] << 1); /* Doubled ref number */
return match(eptr,
ecode + ((offset < offset_top && md->offset_vector[offset] >= 0)?
6 : 3 + (ecode[1] << 8) + ecode[2]),
offset_top, md, ims, eptrb, match_isgroup);
}
/* The condition is an assertion. Call match() to evaluate it - setting
the final argument TRUE causes it to stop at the end of an assertion. */
else
{
if (match(eptr, ecode+3, offset_top, md, ims, NULL,
match_condassert | match_isgroup))
{
ecode += 3 + (ecode[4] << 8) + ecode[5];
while (*ecode == OP_ALT) ecode += (ecode[1] << 8) + ecode[2];
}
else ecode += (ecode[1] << 8) + ecode[2];
return match(eptr, ecode+3, offset_top, md, ims, eptrb, match_isgroup);
}
/* Control never reaches here */
/* Skip over conditional reference or large extraction number data if
encountered. */
case OP_CREF:
case OP_BRANUMBER:
ecode += 3;
break;
/* End of the pattern. If PCRE_NOTEMPTY is set, fail if we have matched
an empty string - recursion will then try other alternatives, if any. */
case OP_END:
if (md->notempty && eptr == md->start_match) return FALSE;
md->end_match_ptr = eptr; /* Record where we ended */
md->end_offset_top = offset_top; /* and how many extracts were taken */
return TRUE;
/* Change option settings */
case OP_OPT:
ims = ecode[1];
ecode += 2;
DPRINTF(("ims set to %02lxn", ims));
break;
/* Assertion brackets. Check the alternative branches in turn - the
matching won't pass the KET for an assertion. If any one branch matches,
the assertion is true. Lookbehind assertions have an OP_REVERSE item at the
start of each branch to move the current point backwards, so the code at
this level is identical to the lookahead case. */
case OP_ASSERT:
case OP_ASSERTBACK:
do
{
if (match(eptr, ecode+3, offset_top, md, ims, NULL, match_isgroup)) break;
ecode += (ecode[1] << 8) + ecode[2];
}
while (*ecode == OP_ALT);
if (*ecode == OP_KET) return FALSE;
/* If checking an assertion for a condition, return TRUE. */
if ((flags & match_condassert) != 0) return TRUE;
/* Continue from after the assertion, updating the offsets high water
mark, since extracts may have been taken during the assertion. */
do ecode += (ecode[1] << 8) + ecode[2]; while (*ecode == OP_ALT);
ecode += 3;
offset_top = md->end_offset_top;
continue;
/* Negative assertion: all branches must fail to match */
case OP_ASSERT_NOT:
case OP_ASSERTBACK_NOT:
do
{
if (match(eptr, ecode+3, offset_top, md, ims, NULL, match_isgroup))
return FALSE;
ecode += (ecode[1] << 8) + ecode[2];
}
while (*ecode == OP_ALT);
if ((flags & match_condassert) != 0) return TRUE;
ecode += 3;
continue;
/* Move the subject pointer back. This occurs only at the start of
each branch of a lookbehind assertion. If we are too close to the start to
move back, this match function fails. When working with UTF-8 we move
back a number of characters, not bytes. */
case OP_REVERSE:
#ifdef SUPPORT_UTF8
c = (ecode[1] << 8) + ecode[2];
for (i = 0; i < c; i++)
{
eptr--;
BACKCHAR(eptr)
}
#else
eptr -= (ecode[1] << 8) + ecode[2];
#endif
if (eptr < md->start_subject) return FALSE;
ecode += 3;
break;
/* Recursion matches the current regex, nested. If there are any capturing
brackets started but not finished, we have to save their starting points
and reinstate them after the recursion. However, we don't know how many
such there are (offset_top records the completed total) so we just have
to save all the potential data. There may be up to 99 such values, which
is a bit large to put on the stack, but using malloc for small numbers
seems expensive. As a compromise, the stack is used when there are fewer
than 16 values to store; otherwise malloc is used. A problem is what to do
if the malloc fails ... there is no way of returning to the top level with
an error. Save the top 15 values on the stack, and accept that the rest
may be wrong. */
case OP_RECURSE:
{
BOOL rc;
int *save;
int stacksave[15];
c = md->offset_max;
if (c < 16) save = stacksave; else
{
save = (int *)(pcre_malloc)((c+1) * sizeof(int));
if (save == NULL)
{
save = stacksave;
c = 15;
}
}
for (i = 1; i <= c; i++)
save[i] = md->offset_vector[md->offset_end - i];
rc = match(eptr, md->start_pattern, offset_top, md, ims, eptrb,
match_isgroup);
for (i = 1; i <= c; i++)
md->offset_vector[md->offset_end - i] = save[i];
if (save != stacksave) (pcre_free)(save);
if (!rc) return FALSE;
/* In case the recursion has set more capturing values, save the final
number, then move along the subject till after the recursive match,
and advance one byte in the pattern code. */
offset_top = md->end_offset_top;
eptr = md->end_match_ptr;
ecode++;
}
break;
/* "Once" brackets are like assertion brackets except that after a match,
the point in the subject string is not moved back. Thus there can never be
a move back into the brackets. 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. */
case OP_ONCE:
{
const uschar *prev = ecode;
const uschar *saved_eptr = eptr;
do
{
if (match(eptr, ecode+3, offset_top, md, ims, eptrb, match_isgroup))
break;
ecode += (ecode[1] << 8) + ecode[2];
}
while (*ecode == OP_ALT);
/* If hit the end of the group (which could be repeated), fail */
if (*ecode != OP_ONCE && *ecode != OP_ALT) return FALSE;
/* Continue as from after the assertion, updating the offsets high water
mark, since extracts may have been taken. */
do ecode += (ecode[1] << 8) + ecode[2]; 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 there is an options reset, it will get obeyed in the normal
course of events. */
if (*ecode == OP_KET || eptr == saved_eptr)
{
ecode += 3;
break;
}
/* The repeating kets try the rest of the pattern or restart from the
preceding bracket, in the appropriate order. We need to reset any options
that changed within the bracket before re-running it, so check the next
opcode. */
if (ecode[3] == OP_OPT)
{
ims = (ims & ~PCRE_IMS) | ecode[4];
DPRINTF(("ims set to %02lx at group repeatn", ims));
}
if (*ecode == OP_KETRMIN)
{
if (match(eptr, ecode+3, offset_top, md, ims, eptrb, 0) ||
match(eptr, prev, offset_top, md, ims, eptrb, match_isgroup))
return TRUE;
}
else /* OP_KETRMAX */
{
if (match(eptr, prev, offset_top, md, ims, eptrb, match_isgroup) ||
match(eptr, ecode+3, offset_top, md, ims, eptrb, 0)) return TRUE;
}
}
return FALSE;
/* An alternation is the end of a branch; scan along to find the end of the
bracketed group and go to there. */
case OP_ALT:
do ecode += (ecode[1] << 8) + ecode[2]; while (*ecode == OP_ALT);
break;
/* BRAZERO and BRAMINZERO occur just before a bracket group, indicating
that it may occur zero times. It may repeat infinitely, or not at all -
i.e. it could be ()* or ()? in the pattern. Brackets with fixed upper
repeat limits are compiled as a number of copies, with the optional ones
preceded by BRAZERO or BRAMINZERO. */
case OP_BRAZERO:
{
const uschar *next = ecode+1;
if (match(eptr, next, offset_top, md, ims, eptrb, match_isgroup))
return TRUE;
do next += (next[1] << 8) + next[2]; while (*next == OP_ALT);
ecode = next + 3;
}
break;
case OP_BRAMINZERO:
{
const uschar *next = ecode+1;
do next += (next[1] << 8) + next[2]; while (*next == OP_ALT);
if (match(eptr, next+3, offset_top, md, ims, eptrb, match_isgroup))
return TRUE;
ecode++;
}
break;
/* End of a group, repeated or non-repeating. If we are at the end of
an assertion "group", stop matching and return TRUE, but record the
current high water mark for use by positive assertions. Do this also
for the "once" (not-backup up) groups. */
case OP_KET:
case OP_KETRMIN:
case OP_KETRMAX:
{
const uschar *prev = ecode - (ecode[1] << 8) - ecode[2];
const uschar *saved_eptr = eptrb->saved_eptr;
eptrb = eptrb->prev; /* Back up the stack of bracket start pointers */
if (*prev == OP_ASSERT || *prev == OP_ASSERT_NOT ||
*prev == OP_ASSERTBACK || *prev == OP_ASSERTBACK_NOT ||
*prev == OP_ONCE)
{
md->end_match_ptr = eptr; /* For ONCE */
md->end_offset_top = offset_top;
return TRUE;
}
/* In all other cases except a conditional group we have to check the
group number back at the start and if necessary complete handling an
extraction by setting the offsets and bumping the high water mark. */
if (*prev != OP_COND)
{
int offset;
int number = *prev - OP_BRA;
/* For extended extraction brackets (large number), we have to fish out
the number from a dummy opcode at the start. */
if (number > EXTRACT_BASIC_MAX) number = (prev[4] << 8) | prev[5];
offset = number << 1;
#ifdef DEBUG
printf("end bracket %d", number);
printf("n");
#endif
if (number > 0)
{
if (offset >= md->offset_max) md->offset_overflow = TRUE; else
{
md->offset_vector[offset] =
md->offset_vector[md->offset_end - number];
md->offset_vector[offset+1] = eptr - md->start_subject;
if (offset_top <= offset) offset_top = offset + 2;
}
}
}
/* Reset the value of the ims flags, in case they got changed during
the group. */
ims = original_ims;
DPRINTF(("ims reset to %02lxn", ims));
/* 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 there is an options reset, it will get obeyed in the normal
course of events. */
if (*ecode == OP_KET || eptr == saved_eptr)
{
ecode += 3;
break;
}
/* The repeating kets try the rest of the pattern or restart from the
preceding bracket, in the appropriate order. */
if (*ecode == OP_KETRMIN)
{
if (match(eptr, ecode+3, offset_top, md, ims, eptrb, 0) ||
match(eptr, prev, offset_top, md, ims, eptrb, match_isgroup))
return TRUE;
}
else /* OP_KETRMAX */
{
if (match(eptr, prev, offset_top, md, ims, eptrb, match_isgroup) ||
match(eptr, ecode+3, offset_top, md, ims, eptrb, 0)) return TRUE;
}
}
return FALSE;
/* Start of subject unless notbol, or after internal newline if multiline */
case OP_CIRC:
if (md->notbol && eptr == md->start_subject) return FALSE;
if ((ims & PCRE_MULTILINE) != 0)
{
if (eptr != md->start_subject && eptr[-1] != NEWLINE) return FALSE;
ecode++;
break;
}
/* ... else fall through */
/* Start of subject assertion */
case OP_SOD:
if (eptr != md->start_subject) return FALSE;
ecode++;
break;
/* Assert before internal newline if multiline, or before a terminating
newline unless endonly is set, else end of subject unless noteol is set. */
case OP_DOLL:
if ((ims & PCRE_MULTILINE) != 0)
{
if (eptr < md->end_subject) { if (*eptr != NEWLINE) return FALSE; }
else { if (md->noteol) return FALSE; }
ecode++;
break;
}
else
{
if (md->noteol) return FALSE;
if (!md->endonly)
{
if (eptr < md->end_subject - 1 ||
(eptr == md->end_subject - 1 && *eptr != NEWLINE)) return FALSE;
ecode++;
break;
}
}
/* ... else fall through */
/* End of subject assertion (z) */
case OP_EOD:
if (eptr < md->end_subject) return FALSE;
ecode++;
break;
/* End of subject or ending n assertion (Z) */
case OP_EODN:
if (eptr < md->end_subject - 1 ||
(eptr == md->end_subject - 1 && *eptr != NEWLINE)) return FALSE;
ecode++;
break;
/* Word boundary assertions */
case OP_NOT_WORD_BOUNDARY:
case OP_WORD_BOUNDARY:
{
BOOL prev_is_word = (eptr != md->start_subject) &&
((md->ctypes[eptr[-1]] & ctype_word) != 0);
BOOL cur_is_word = (eptr < md->end_subject) &&
((md->ctypes[*eptr] & ctype_word) != 0);
if ((*ecode++ == OP_WORD_BOUNDARY)?
cur_is_word == prev_is_word : cur_is_word != prev_is_word)
return FALSE;
}
break;
/* Match a single character type; inline for speed */
case OP_ANY:
if ((ims & PCRE_DOTALL) == 0 && eptr < md->end_subject && *eptr == NEWLINE)
return FALSE;
if (eptr++ >= md->end_subject) return FALSE;
#ifdef SUPPORT_UTF8
if (md->utf8)
while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
#endif
ecode++;
break;
case OP_NOT_DIGIT:
if (eptr >= md->end_subject ||
(md->ctypes[*eptr++] & ctype_digit) != 0)
return FALSE;
ecode++;
break;
case OP_DIGIT:
if (eptr >= md->end_subject ||
(md->ctypes[*eptr++] & ctype_digit) == 0)
return FALSE;
ecode++;
break;
case OP_NOT_WHITESPACE:
if (eptr >= md->end_subject ||
(md->ctypes[*eptr++] & ctype_space) != 0)
return FALSE;
ecode++;
break;
case OP_WHITESPACE:
if (eptr >= md->end_subject ||
(md->ctypes[*eptr++] & ctype_space) == 0)
return FALSE;
ecode++;
break;
case OP_NOT_WORDCHAR:
if (eptr >= md->end_subject ||
(md->ctypes[*eptr++] & ctype_word) != 0)
return FALSE;
ecode++;
break;
case OP_WORDCHAR:
if (eptr >= md->end_subject ||
(md->ctypes[*eptr++] & ctype_word) == 0)
return FALSE;
ecode++;
break;
/* Match a back reference, possibly repeatedly. Look past the end of the
item to see if there is repeat information following. The code is similar
to that for character classes, but repeated for efficiency. Then obey
similar code to character type repeats - written out again for speed.
However, if the referenced string is the empty string, always treat
it as matched, any number of times (otherwise there could be infinite
loops). */
case OP_REF:
{
int length;
int offset = (ecode[1] << 9) | (ecode[2] << 1); /* Doubled ref number */
ecode += 3; /* Advance past item */
/* If the reference is unset, set the length to be longer than the amount
of subject left; this ensures that every attempt at a match fails. We
can't just fail here, because of the possibility of quantifiers with zero
minima. */
length = (offset >= offset_top || md->offset_vector[offset] < 0)?
md->end_subject - eptr + 1 :
md->offset_vector[offset+1] - md->offset_vector[offset];
/* Set up for repetition, or handle the non-repeated case */
switch (*ecode)
{
case OP_CRSTAR:
case OP_CRMINSTAR:
case OP_CRPLUS:
case OP_CRMINPLUS:
case OP_CRQUERY:
case OP_CRMINQUERY:
c = *ecode++ - OP_CRSTAR;
minimize = (c & 1) != 0;
min = rep_min[c]; /* Pick up values from tables; */
max = rep_max[c]; /* zero for max => infinity */
if (max == 0) max = INT_MAX;
break;
case OP_CRRANGE:
case OP_CRMINRANGE:
minimize = (*ecode == OP_CRMINRANGE);
min = (ecode[1] << 8) + ecode[2];
max = (ecode[3] << 8) + ecode[4];
if (max == 0) max = INT_MAX;
ecode += 5;
break;
default: /* No repeat follows */
if (!match_ref(offset, eptr, length, md, ims)) return FALSE;
eptr += length;
continue; /* With the main loop */
}
/* If the length of the reference is zero, just continue with the
main loop. */
if (length == 0) continue;
/* First, ensure the minimum number of matches are present. We get back
the length of the reference string explicitly rather than passing the
address of eptr, so that eptr can be a register variable. */
for (i = 1; i <= min; i++)
{
if (!match_ref(offset, eptr, length, md, ims)) return FALSE;
eptr += length;
}
/* If min = max, continue at the same level without recursion.
They are not both allowed to be zero. */
if (min == max) continue;
/* If minimizing, keep trying and advancing the pointer */
if (minimize)
{
for (i = min;; i++)
{
if (match(eptr, ecode, offset_top, md, ims, eptrb, 0))
return TRUE;
if (i >= max || !match_ref(offset, eptr, length, md, ims))
return FALSE;
eptr += length;
}
/* Control never gets here */
}
/* If maximizing, find the longest string and work backwards */
else
{
const uschar *pp = eptr;
for (i = min; i < max; i++)
{
if (!match_ref(offset, eptr, length, md, ims)) break;
eptr += length;
}
while (eptr >= pp)
{
if (match(eptr, ecode, offset_top, md, ims, eptrb, 0))
return TRUE;
eptr -= length;
}
return FALSE;
}
}
/* Control never gets here */
/* Match a character class, possibly repeatedly. Look past the end of the
item to see if there is repeat information following. Then obey similar
code to character type repeats - written out again for speed. */
case OP_CLASS:
{
const uschar *data = ecode + 1; /* Save for matching */
ecode += 33; /* Advance past the item */
switch (*ecode)
{
case OP_CRSTAR:
case OP_CRMINSTAR:
case OP_CRPLUS:
case OP_CRMINPLUS:
case OP_CRQUERY:
case OP_CRMINQUERY:
c = *ecode++ - OP_CRSTAR;
minimize = (c & 1) != 0;
min = rep_min[c]; /* Pick up values from tables; */
max = rep_max[c]; /* zero for max => infinity */
if (max == 0) max = INT_MAX;
break;
case OP_CRRANGE:
case OP_CRMINRANGE:
minimize = (*ecode == OP_CRMINRANGE);
min = (ecode[1] << 8) + ecode[2];
max = (ecode[3] << 8) + ecode[4];
if (max == 0) max = INT_MAX;
ecode += 5;
break;
default: /* No repeat follows */
min = max = 1;
break;
}
/* First, ensure the minimum number of matches are present. */
for (i = 1; i <= min; i++)
{
if (eptr >= md->end_subject) return FALSE;
GETCHARINC(c, eptr) /* Get character; increment eptr */
#ifdef SUPPORT_UTF8
/* We do not yet support class members > 255 */
if (c > 255) return FALSE;
#endif
if ((data[c/8] & (1 << (c&7))) != 0) continue;
return FALSE;
}
/* If max == min we can continue with the main loop without the
need to recurse. */
if (min == max) continue;
/* If minimizing, keep testing the rest of the expression and advancing
the pointer while it matches the class. */
if (minimize)
{
for (i = min;; i++)
{
if (match(eptr, ecode, offset_top, md, ims, eptrb, 0))
return TRUE;
if (i >= max || eptr >= md->end_subject) return FALSE;
GETCHARINC(c, eptr) /* Get character; increment eptr */
#ifdef SUPPORT_UTF8
/* We do not yet support class members > 255 */
if (c > 255) return FALSE;
#endif
if ((data[c/8] & (1 << (c&7))) != 0) continue;
return FALSE;
}
/* Control never gets here */
}
/* If maximizing, find the longest possible run, then work backwards. */
else
{
const uschar *pp = eptr;
int len = 1;
for (i = min; i < max; i++)
{
if (eptr >= md->end_subject) break;
GETCHARLEN(c, eptr, len) /* Get character, set length if UTF-8 */
#ifdef SUPPORT_UTF8
/* We do not yet support class members > 255 */
if (c > 255) break;
#endif
if ((data[c/8] & (1 << (c&7))) == 0) break;
eptr += len;
}
while (eptr >= pp)
{
if (match(eptr--, ecode, offset_top, md, ims, eptrb, 0))
return TRUE;
#ifdef SUPPORT_UTF8
BACKCHAR(eptr)
#endif
}
return FALSE;
}
}
/* Control never gets here */
/* Match a run of characters */
case OP_CHARS:
{
register int length = ecode[1];
ecode += 2;
#ifdef DEBUG /* Sigh. Some compilers never learn. */
if (eptr >= md->end_subject)
printf("matching subject <null> against pattern ");
else
{
printf("matching subject ");
pchars(eptr, length, TRUE, md);
printf(" against pattern ");
}
pchars(ecode, length, FALSE, md);
printf("n");
#endif
if (length > md->end_subject - eptr) return FALSE;
if ((ims & PCRE_CASELESS) != 0)
{
while (length-- > 0)
if (md->lcc[*ecode++] != md->lcc[*eptr++])
return FALSE;
}
else
{
while (length-- > 0) if (*ecode++ != *eptr++) return FALSE;
}
}
break;
/* Match a single character repeatedly; different opcodes share code. */
case OP_EXACT:
min = max = (ecode[1] << 8) + ecode[2];
ecode += 3;
goto REPEATCHAR;
case OP_UPTO:
case OP_MINUPTO:
min = 0;
max = (ecode[1] << 8) + ecode[2];
minimize = *ecode == OP_MINUPTO;
ecode += 3;
goto REPEATCHAR;
case OP_STAR:
case OP_MINSTAR:
case OP_PLUS:
case OP_MINPLUS:
case OP_QUERY:
case OP_MINQUERY:
c = *ecode++ - OP_STAR;
minimize = (c & 1) != 0;
min = rep_min[c]; /* Pick up values from tables; */
max = rep_max[c]; /* zero for max => infinity */
if (max == 0) max = INT_MAX;
/* Common code for all repeated single-character matches. We can give
up quickly if there are fewer than the minimum number of characters left in
the subject. */
REPEATCHAR:
if (min > md->end_subject - eptr) return FALSE;
c = *ecode++;
/* The code is duplicated for the caseless and caseful cases, for speed,
since matching characters is likely to be quite common. First, ensure the
minimum number of matches are present. If min = max, continue at the same
level without recursing. Otherwise, if minimizing, keep trying the rest of
the expression and advancing one matching character if failing, up to the
maximum. Alternatively, if maximizing, find the maximum number of
characters and work backwards. */
DPRINTF(("matching %c{%d,%d} against subject %.*sn", c, min, max,
max, eptr));
if ((ims & PCRE_CASELESS) != 0)
{
c = md->lcc[c];
for (i = 1; i <= min; i++)
if (c != md->lcc[*eptr++]) return FALSE;
if (min == max) continue;
if (minimize)
{
for (i = min;; i++)
{
if (match(eptr, ecode, offset_top, md, ims, eptrb, 0))
return TRUE;
if (i >= max || eptr >= md->end_subject ||
c != md->lcc[*eptr++])
return FALSE;
}
/* Control never gets here */
}
else
{
const uschar *pp = eptr;
for (i = min; i < max; i++)
{
if (eptr >= md->end_subject || c != md->lcc[*eptr]) break;
eptr++;
}
while (eptr >= pp)
if (match(eptr--, ecode, offset_top, md, ims, eptrb, 0))
return TRUE;
return FALSE;
}
/* Control never gets here */
}
/* Caseful comparisons */
else
{
for (i = 1; i <= min; i++) if (c != *eptr++) return FALSE;
if (min == max) continue;
if (minimize)
{
for (i = min;; i++)
{
if (match(eptr, ecode, offset_top, md, ims, eptrb, 0))
return TRUE;
if (i >= max || eptr >= md->end_subject || c != *eptr++) return FALSE;
}
/* Control never gets here */
}
else
{
const uschar *pp = eptr;
for (i = min; i < max; i++)
{
if (eptr >= md->end_subject || c != *eptr) break;
eptr++;
}
while (eptr >= pp)
if (match(eptr--, ecode, offset_top, md, ims, eptrb, 0))
return TRUE;
return FALSE;
}
}
/* Control never gets here */
/* Match a negated single character */
case OP_NOT:
if (eptr >= md->end_subject) return FALSE;
ecode++;
if ((ims & PCRE_CASELESS) != 0)
{
if (md->lcc[*ecode++] == md->lcc[*eptr++]) return FALSE;
}
else
{
if (*ecode++ == *eptr++) return FALSE;
}
break;
/* Match a negated single character repeatedly. This is almost a repeat of
the code for a repeated single character, but I haven't found a nice way of
commoning these up that doesn't require a test of the positive/negative
option for each character match. Maybe that wouldn't add very much to the
time taken, but character matching *is* what this is all about... */
case OP_NOTEXACT:
min = max = (ecode[1] << 8) + ecode[2];
ecode += 3;
goto REPEATNOTCHAR;
case OP_NOTUPTO:
case OP_NOTMINUPTO:
min = 0;
max = (ecode[1] << 8) + ecode[2];
minimize = *ecode == OP_NOTMINUPTO;
ecode += 3;
goto REPEATNOTCHAR;
case OP_NOTSTAR:
case OP_NOTMINSTAR:
case OP_NOTPLUS:
case OP_NOTMINPLUS:
case OP_NOTQUERY:
case OP_NOTMINQUERY:
c = *ecode++ - OP_NOTSTAR;
minimize = (c & 1) != 0;
min = rep_min[c]; /* Pick up values from tables; */
max = rep_max[c]; /* zero for max => infinity */
if (max == 0) max = INT_MAX;
/* Common code for all repeated single-character matches. We can give
up quickly if there are fewer than the minimum number of characters left in
the subject. */
REPEATNOTCHAR:
if (min > md->end_subject - eptr) return FALSE;
c = *ecode++;
/* The code is duplicated for the caseless and caseful cases, for speed,
since matching characters is likely to be quite common. First, ensure the
minimum number of matches are present. If min = max, continue at the same
level without recursing. Otherwise, if minimizing, keep trying the rest of
the expression and advancing one matching character if failing, up to the
maximum. Alternatively, if maximizing, find the maximum number of
characters and work backwards. */
DPRINTF(("negative matching %c{%d,%d} against subject %.*sn", c, min, max,
max, eptr));
if ((ims & PCRE_CASELESS) != 0)
{
c = md->lcc[c];
for (i = 1; i <= min; i++)
if (c == md->lcc[*eptr++]) return FALSE;
if (min == max) continue;
if (minimize)
{
for (i = min;; i++)
{
if (match(eptr, ecode, offset_top, md, ims, eptrb, 0))
return TRUE;
if (i >= max || eptr >= md->end_subject ||
c == md->lcc[*eptr++])
return FALSE;
}
/* Control never gets here */
}
else
{
const uschar *pp = eptr;
for (i = min; i < max; i++)
{
if (eptr >= md->end_subject || c == md->lcc[*eptr]) break;
eptr++;
}
while (eptr >= pp)
if (match(eptr--, ecode, offset_top, md, ims, eptrb, 0))
return TRUE;
return FALSE;
}
/* Control never gets here */
}
/* Caseful comparisons */
else
{
for (i = 1; i <= min; i++) if (c == *eptr++) return FALSE;
if (min == max) continue;
if (minimize)
{
for (i = min;; i++)
{
if (match(eptr, ecode, offset_top, md, ims, eptrb, 0))
return TRUE;
if (i >= max || eptr >= md->end_subject || c == *eptr++) return FALSE;
}
/* Control never gets here */
}
else
{
const uschar *pp = eptr;
for (i = min; i < max; i++)
{
if (eptr >= md->end_subject || c == *eptr) break;
eptr++;
}
while (eptr >= pp)
if (match(eptr--, ecode, offset_top, md, ims, eptrb, 0))
return TRUE;
return FALSE;
}
}
/* Control never gets here */
/* Match a single character type repeatedly; several different opcodes
share code. This is very similar to the code for single characters, but we
repeat it in the interests of efficiency. */
case OP_TYPEEXACT:
min = max = (ecode[1] << 8) + ecode[2];
minimize = TRUE;
ecode += 3;
goto REPEATTYPE;
case OP_TYPEUPTO:
case OP_TYPEMINUPTO:
min = 0;
max = (ecode[1] << 8) + ecode[2];
minimize = *ecode == OP_TYPEMINUPTO;
ecode += 3;
goto REPEATTYPE;
case OP_TYPESTAR:
case OP_TYPEMINSTAR:
case OP_TYPEPLUS:
case OP_TYPEMINPLUS:
case OP_TYPEQUERY:
case OP_TYPEMINQUERY:
c = *ecode++ - OP_TYPESTAR;
minimize = (c & 1) != 0;
min = rep_min[c]; /* Pick up values from tables; */
max = rep_max[c]; /* zero for max => infinity */
if (max == 0) max = INT_MAX;
/* Common code for all repeated single character type matches */
REPEATTYPE:
ctype = *ecode++; /* Code for the character type */
/* First, ensure the minimum number of matches are present. Use inline
code for maximizing the speed, and do the type test once at the start
(i.e. keep it out of the loop). Also we can test that there are at least
the minimum number of bytes before we start, except when doing '.' in
UTF8 mode. Leave the test in in all cases; in the special case we have
to test after each character. */
if (min > md->end_subject - eptr) return FALSE;
if (min > 0) switch(ctype)
{
case OP_ANY:
#ifdef SUPPORT_UTF8
if (md->utf8)
{
for (i = 1; i <= min; i++)
{
if (eptr >= md->end_subject ||
(*eptr++ == NEWLINE && (ims & PCRE_DOTALL) == 0))
return FALSE;
while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
}
break;
}
#endif
/* Non-UTF8 can be faster */
if ((ims & PCRE_DOTALL) == 0)
{ for (i = 1; i <= min; i++) if (*eptr++ == NEWLINE) return FALSE; }
else eptr += min;
break;
case OP_NOT_DIGIT:
for (i = 1; i <= min; i++)
if ((md->ctypes[*eptr++] & ctype_digit) != 0) return FALSE;
break;
case OP_DIGIT:
for (i = 1; i <= min; i++)
if ((md->ctypes[*eptr++] & ctype_digit) == 0) return FALSE;
break;
case OP_NOT_WHITESPACE:
for (i = 1; i <= min; i++)
if ((md->ctypes[*eptr++] & ctype_space) != 0) return FALSE;
break;
case OP_WHITESPACE:
for (i = 1; i <= min; i++)
if ((md->ctypes[*eptr++] & ctype_space) == 0) return FALSE;
break;
case OP_NOT_WORDCHAR:
for (i = 1; i <= min; i++)
if ((md->ctypes[*eptr++] & ctype_word) != 0)
return FALSE;
break;
case OP_WORDCHAR:
for (i = 1; i <= min; i++)
if ((md->ctypes[*eptr++] & ctype_word) == 0)
return FALSE;
break;
}
/* If min = max, continue at the same level without recursing */
if (min == max) continue;
/* If minimizing, we have to test the rest of the pattern before each
subsequent match. */
if (minimize)
{
for (i = min;; i++)
{
if (match(eptr, ecode, offset_top, md, ims, eptrb, 0)) return TRUE;
if (i >= max || eptr >= md->end_subject) return FALSE;
c = *eptr++;
switch(ctype)
{
case OP_ANY:
if ((ims & PCRE_DOTALL) == 0 && c == NEWLINE) return FALSE;
#ifdef SUPPORT_UTF8
if (md->utf8)
while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
#endif
break;
case OP_NOT_DIGIT:
if ((md->ctypes[c] & ctype_digit) != 0) return FALSE;
break;
case OP_DIGIT:
if ((md->ctypes[c] & ctype_digit) == 0) return FALSE;
break;
case OP_NOT_WHITESPACE:
if ((md->ctypes[c] & ctype_space) != 0) return FALSE;
break;
case OP_WHITESPACE:
if ((md->ctypes[c] & ctype_space) == 0) return FALSE;
break;
case OP_NOT_WORDCHAR:
if ((md->ctypes[c] & ctype_word) != 0) return FALSE;
break;
case OP_WORDCHAR:
if ((md->ctypes[c] & ctype_word) == 0) return FALSE;
break;
}
}
/* Control never gets here */
}
/* If maximizing it is worth using inline code for speed, doing the type
test once at the start (i.e. keep it out of the loop). */
else
{
const uschar *pp = eptr;
switch(ctype)
{
case OP_ANY:
/* Special code is required for UTF8, but when the maximum is unlimited
we don't need it. */
#ifdef SUPPORT_UTF8
if (md->utf8 && max < INT_MAX)
{
if ((ims & PCRE_DOTALL) == 0)
{
for (i = min; i < max; i++)
{
if (eptr >= md->end_subject || *eptr++ == NEWLINE) break;
while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
}
}
else
{
for (i = min; i < max; i++)
{
eptr++;
while (eptr < md->end_subject && (*eptr & 0xc0) == 0x80) eptr++;
}
}
break;
}
#endif
/* Non-UTF8 can be faster */
if ((ims & PCRE_DOTALL) == 0)
{
for (i = min; i < max; i++)
{
if (eptr >= md->end_subject || *eptr == NEWLINE) break;
eptr++;
}
}
else
{
c = max - min;
if (c > md->end_subject - eptr) c = md->end_subject - eptr;
eptr += c;
}
break;
case OP_NOT_DIGIT:
for (i = min; i < max; i++)
{
if (eptr >= md->end_subject || (md->ctypes[*eptr] & ctype_digit) != 0)
break;
eptr++;
}
break;
case OP_DIGIT:
for (i = min; i < max; i++)
{
if (eptr >= md->end_subject || (md->ctypes[*eptr] & ctype_digit) == 0)
break;
eptr++;
}
break;
case OP_NOT_WHITESPACE:
for (i = min; i < max; i++)
{
if (eptr >= md->end_subject || (md->ctypes[*eptr] & ctype_space) != 0)
break;
eptr++;
}
break;
case OP_WHITESPACE:
for (i = min; i < max; i++)
{
if (eptr >= md->end_subject || (md->ctypes[*eptr] & ctype_space) == 0)
break;
eptr++;
}
break;
case OP_NOT_WORDCHAR:
for (i = min; i < max; i++)
{
if (eptr >= md->end_subject || (md->ctypes[*eptr] & ctype_word) != 0)
break;
eptr++;
}
break;
case OP_WORDCHAR:
for (i = min; i < max; i++)
{
if (eptr >= md->end_subject || (md->ctypes[*eptr] & ctype_word) == 0)
break;
eptr++;
}
break;
}
while (eptr >= pp)
{
if (match(eptr--, ecode, offset_top, md, ims, eptrb, 0))
return TRUE;
#ifdef SUPPORT_UTF8
if (md->utf8)
while (eptr > pp && (*eptr & 0xc0) == 0x80) eptr--;
#endif
}
return FALSE;
}
/* Control never gets here */
/* There's been some horrible disaster. */
default:
DPRINTF(("Unknown opcode %dn", *ecode));
md->errorcode = PCRE_ERROR_UNKNOWN_NODE;
return FALSE;
}
/* Do not stick any code in here without much thought; it is assumed
that "continue" in the code above comes out to here to repeat the main
loop. */
} /* End of main loop */
/* Control never reaches here */
}
/*************************************************
* Execute a Regular Expression *
*************************************************/
/* This function applies a compiled re to a subject string and picks out
portions of the string if it matches. Two elements in the vector are set for
each substring: the offsets to the start and end of the substring.
Arguments:
external_re points to the compiled expression
external_extra points to "hints" from pcre_study() or is NULL
subject points to the subject string
length length of subject string (may contain binary zeros)
start_offset where to start in the subject string
options option bits
offsets points to a vector of ints to be filled in with offsets
offsetcount the number of elements in the vector
Returns: > 0 => success; value is the number of elements filled in
= 0 => success, but offsets is not big enough
-1 => failed to match
< -1 => some kind of unexpected problem
*/
int
pcre_exec(const pcre *external_re, const pcre_extra *external_extra,
const char *subject, int length, int start_offset, int options, int *offsets,
int offsetcount)
{
int resetcount, ocount;
int first_char = -1;
int req_char = -1;
int req_char2 = -1;
unsigned long int ims = 0;
match_data match_block;
const uschar *start_bits = NULL;
const uschar *start_match = (const uschar *)subject + start_offset;
const uschar *end_subject;
const uschar *req_char_ptr = start_match - 1;
const real_pcre *re = (const real_pcre *)external_re;
const real_pcre_extra *extra = (const real_pcre_extra *)external_extra;
BOOL using_temporary_offsets = FALSE;
BOOL anchored;
BOOL startline;
if ((options & ~PUBLIC_EXEC_OPTIONS) != 0) return PCRE_ERROR_BADOPTION;
if (re == NULL || subject == NULL ||
(offsets == NULL && offsetcount > 0)) return PCRE_ERROR_NULL;
if (re->magic_number != MAGIC_NUMBER) return PCRE_ERROR_BADMAGIC;
anchored = ((re->options | options) & PCRE_ANCHORED) != 0;
startline = (re->options & PCRE_STARTLINE) != 0;
match_block.start_pattern = re->code;
match_block.start_subject = (const uschar *)subject;
match_block.end_subject = match_block.start_subject + length;
end_subject = match_block.end_subject;
match_block.endonly = (re->options & PCRE_DOLLAR_ENDONLY) != 0;
match_block.utf8 = (re->options & PCRE_UTF8) != 0;
match_block.notbol = (options & PCRE_NOTBOL) != 0;
match_block.noteol = (options & PCRE_NOTEOL) != 0;
match_block.notempty = (options & PCRE_NOTEMPTY) != 0;
match_block.errorcode = PCRE_ERROR_NOMATCH; /* Default error */
match_block.lcc = re->tables + lcc_offset;
match_block.ctypes = re->tables + ctypes_offset;
/* The ims options can vary during the matching as a result of the presence
of (?ims) items in the pattern. They are kept in a local variable so that
restoring at the exit of a group is easy. */
ims = re->options & (PCRE_CASELESS|PCRE_MULTILINE|PCRE_DOTALL);
/* If the expression has got more back references than the offsets supplied can
hold, we get a temporary bit of working store to use during the matching.
Otherwise, we can use the vector supplied, rounding down its size to a multiple
of 3. */
ocount = offsetcount - (offsetcount % 3);
if (re->top_backref > 0 && re->top_backref >= ocount/3)
{
ocount = re->top_backref * 3 + 3;
match_block.offset_vector = (int *)(pcre_malloc)(ocount * sizeof(int));
if (match_block.offset_vector == NULL) return PCRE_ERROR_NOMEMORY;
using_temporary_offsets = TRUE;
DPRINTF(("Got memory to hold back referencesn"));
}
else match_block.offset_vector = offsets;
match_block.offset_end = ocount;
match_block.offset_max = (2*ocount)/3;
match_block.offset_overflow = FALSE;
/* Compute the minimum number of offsets that we need to reset each time. Doing
this makes a huge difference to execution time when there aren't many brackets
in the pattern. */
resetcount = 2 + re->top_bracket * 2;
if (resetcount > offsetcount) resetcount = ocount;
/* Reset the working variable associated with each extraction. These should
never be used unless previously set, but they get saved and restored, and so we
initialize them to avoid reading uninitialized locations. */
if (match_block.offset_vector != NULL)
{
register int *iptr = match_block.offset_vector + ocount;
register int *iend = iptr - resetcount/2 + 1;
while (--iptr >= iend) *iptr = -1;
}
/* Set up the first character to match, if available. The first_char value is
never set for an anchored regular expression, but the anchoring may be forced
at run time, so we have to test for anchoring. The first char may be unset for
an unanchored pattern, of course. If there's no first char and the pattern was
studied, there may be a bitmap of possible first characters. */
if (!anchored)
{
if ((re->options & PCRE_FIRSTSET) != 0)
{
first_char = re->first_char;
if ((ims & PCRE_CASELESS) != 0) first_char = match_block.lcc[first_char];
}
else
if (!startline && extra != NULL &&
(extra->options & PCRE_STUDY_MAPPED) != 0)
start_bits = extra->start_bits;
}
/* For anchored or unanchored matches, there may be a "last known required
character" set. If the PCRE_CASELESS is set, implying that the match starts
caselessly, or if there are any changes of this flag within the regex, set up
both cases of the character. Otherwise set the two values the same, which will
avoid duplicate testing (which takes significant time). This covers the vast
majority of cases. It will be suboptimal when the case flag changes in a regex
and the required character in fact is caseful. */
if ((re->options & PCRE_REQCHSET) != 0)
{
req_char = re->req_char;
req_char2 = ((re->options & (PCRE_CASELESS | PCRE_ICHANGED)) != 0)?
(re->tables + fcc_offset)[req_char] : req_char;
}
/* Loop for handling unanchored repeated matching attempts; for anchored regexs
the loop runs just once. */
do
{
int rc;
register int *iptr = match_block.offset_vector;
register int *iend = iptr + resetcount;
/* Reset the maximum number of extractions we might see. */
while (iptr < iend) *iptr++ = -1;
/* Advance to a unique first char if possible */
if (first_char >= 0)
{
if ((ims & PCRE_CASELESS) != 0)
while (start_match < end_subject &&
match_block.lcc[*start_match] != first_char)
start_match++;
else
while (start_match < end_subject && *start_match != first_char)
start_match++;
}
/* Or to just after n for a multiline match if possible */
else if (startline)
{
if (start_match > match_block.start_subject + start_offset)
{
while (start_match < end_subject && start_match[-1] != NEWLINE)
start_match++;
}
}
/* Or to a non-unique first char after study */
else if (start_bits != NULL)
{
while (start_match < end_subject)
{
register int c = *start_match;
if ((start_bits[c/8] & (1 << (c&7))) == 0) start_match++; else break;
}
}
#ifdef DEBUG /* Sigh. Some compilers never learn. */
printf(">>>> Match against: ");
pchars(start_match, end_subject - start_match, TRUE, &match_block);
printf("n");
#endif
/* If req_char is set, we know that that character must appear in the subject
for the match to succeed. If the first character is set, req_char must be
later in the subject; otherwise the test starts at the match point. This
optimization can save a huge amount of backtracking in patterns with nested
unlimited repeats that aren't going to match. We don't know what the state of
case matching may be when this character is hit, so test for it in both its
cases if necessary. However, the different cased versions will not be set up
unless PCRE_CASELESS was given or the casing state changes within the regex.
Writing separate code makes it go faster, as does using an autoincrement and
backing off on a match. */
if (req_char >= 0)
{
register const uschar *p = start_match + ((first_char >= 0)? 1 : 0);
/* We don't need to repeat the search if we haven't yet reached the
place we found it at last time. */
if (p > req_char_ptr)
{
/* Do a single test if no case difference is set up */
if (req_char == req_char2)
{
while (p < end_subject)
{
if (*p++ == req_char) { p--; break; }
}
}
/* Otherwise test for either case */
else
{
while (p < end_subject)
{
register int pp = *p++;
if (pp == req_char || pp == req_char2) { p--; break; }
}
}
/* If we can't find the required character, break the matching loop */
if (p >= end_subject) break;
/* If we have found the required character, save the point where we
found it, so that we don't search again next time round the loop if
the start hasn't passed this character yet. */
req_char_ptr = p;
}
}
/* When a match occurs, substrings will be set for all internal extractions;
we just need to set up the whole thing as substring 0 before returning. If
there were too many extractions, set the return code to zero. In the case
where we had to get some local store to hold offsets for backreferences, copy
those back references that we can. In this case there need not be overflow
if certain parts of the pattern were not used. */
match_block.start_match = start_match;
if (!match(start_match, re->code, 2, &match_block, ims, NULL, match_isgroup))
continue;
/* Copy the offset information from temporary store if necessary */
if (using_temporary_offsets)
{
if (offsetcount >= 4)
{
memcpy(offsets + 2, match_block.offset_vector + 2,
(offsetcount - 2) * sizeof(int));
DPRINTF(("Copied offsets from temporary memoryn"));
}
if (match_block.end_offset_top > offsetcount)
match_block.offset_overflow = TRUE;
DPRINTF(("Freeing temporary memoryn"));
(pcre_free)(match_block.offset_vector);
}
rc = match_block.offset_overflow? 0 : match_block.end_offset_top/2;
if (offsetcount < 2) rc = 0; else
{
offsets[0] = start_match - match_block.start_subject;
offsets[1] = match_block.end_match_ptr - match_block.start_subject;
}
DPRINTF((">>>> returning %dn", rc));
return rc;
}
/* This "while" is the end of the "do" above */
while (!anchored &&
match_block.errorcode == PCRE_ERROR_NOMATCH &&
start_match++ < end_subject);
if (using_temporary_offsets)
{
DPRINTF(("Freeing temporary memoryn"));
(pcre_free)(match_block.offset_vector);
}
DPRINTF((">>>> returning %dn", match_block.errorcode));
return match_block.errorcode;
}
/* End of pcre.c */