static int
set_start_bits(const uschar *code, uschar *start_bits, BOOL caseless,
BOOL utf8, compile_data *cd)
{
register int c;
int yield = SSB_DONE;
int table_limit = utf8? 16:32;
#if 0
/* ========================================================================= */
/* The following comment and code was inserted in January 1999. In May 2006,
when it was observed to cause compiler warnings about unused values, I took it
out again. If anybody is still using OS/2, they will have to put it back
manually. */
/* This next statement and the later reference to dummy are here in order to
trick the optimizer of the IBM C compiler for OS/2 into generating correct
code. Apparently IBM isn't going to fix the problem, and we would rather not
disable optimization (in this module it actually makes a big difference, and
the pcre module can use all the optimization it can get). */
volatile int dummy;
/* ========================================================================= */
#endif
do
{
const uschar *tcode = code + (((int)*code == OP_CBRA)? 3:1) + LINK_SIZE;
BOOL try_next = TRUE;
while (try_next) /* Loop for items in this branch */
{
int rc;
switch(*tcode)
{
/* Fail if we reach something we don't understand */
default:
return SSB_FAIL;
/* If we hit a bracket or a positive lookahead assertion, recurse to set
bits from within the subpattern. If it can't find anything, we have to
give up. If it finds some mandatory character(s), we are done for this
branch. Otherwise, carry on scanning after the subpattern. */
case OP_BRA:
case OP_SBRA:
case OP_CBRA:
case OP_SCBRA:
case OP_ONCE:
case OP_ASSERT:
rc = set_start_bits(tcode, start_bits, caseless, utf8, cd);
if (rc == SSB_FAIL) return SSB_FAIL;
if (rc == SSB_DONE) try_next = FALSE; else
{
do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
tcode += 1 + LINK_SIZE;
}
break;
/* If we hit ALT or KET, it means we haven't found anything mandatory in
this branch, though we might have found something optional. For ALT, we
continue with the next alternative, but we have to arrange that the final
result from subpattern is SSB_CONTINUE rather than SSB_DONE. For KET,
return SSB_CONTINUE: if this is the top level, that indicates failure,
but after a nested subpattern, it causes scanning to continue. */
case OP_ALT:
yield = SSB_CONTINUE;
try_next = FALSE;
break;
case OP_KET:
case OP_KETRMAX:
case OP_KETRMIN:
return SSB_CONTINUE;
/* Skip over callout */
case OP_CALLOUT:
tcode += 2 + 2*LINK_SIZE;
break;
/* Skip over lookbehind and negative lookahead assertions */
case OP_ASSERT_NOT:
case OP_ASSERTBACK:
case OP_ASSERTBACK_NOT:
do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
tcode += 1 + LINK_SIZE;
break;
/* Skip over an option setting, changing the caseless flag */
case OP_OPT:
caseless = (tcode[1] & PCRE_CASELESS) != 0;
tcode += 2;
break;
/* BRAZERO does the bracket, but carries on. */
case OP_BRAZERO:
case OP_BRAMINZERO:
if (set_start_bits(++tcode, start_bits, caseless, utf8, cd) == SSB_FAIL)
return SSB_FAIL;
/* =========================================================================
See the comment at the head of this function concerning the next line,
which was an old fudge for the benefit of OS/2.
dummy = 1;
========================================================================= */
do tcode += GET(tcode,1); while (*tcode == OP_ALT);
tcode += 1 + LINK_SIZE;
break;
/* SKIPZERO skips the bracket. */
case OP_SKIPZERO:
tcode++;
do tcode += GET(tcode,1); while (*tcode == OP_ALT);
tcode += 1 + LINK_SIZE;
break;
/* Single-char * or ? sets the bit and tries the next item */
case OP_STAR:
case OP_MINSTAR:
case OP_POSSTAR:
case OP_QUERY:
case OP_MINQUERY:
case OP_POSQUERY:
tcode = set_table_bit(start_bits, tcode + 1, caseless, cd, utf8);
break;
/* Single-char upto sets the bit and tries the next */
case OP_UPTO:
case OP_MINUPTO:
case OP_POSUPTO:
tcode = set_table_bit(start_bits, tcode + 3, caseless, cd, utf8);
break;
/* At least one single char sets the bit and stops */
case OP_EXACT: /* Fall through */
tcode += 2;
case OP_CHAR:
case OP_CHARNC:
case OP_PLUS:
case OP_MINPLUS:
case OP_POSPLUS:
(void)set_table_bit(start_bits, tcode + 1, caseless, cd, utf8);
try_next = FALSE;
break;
/* Special spacing and line-terminating items. These recognize specific
lists of characters. The difference between VSPACE and ANYNL is that the
latter can match the two-character CRLF sequence, but that is not
relevant for finding the first character, so their code here is
identical. */
case OP_HSPACE:
SET_BIT(0x09);
SET_BIT(0x20);
if (utf8)
{
SET_BIT(0xC2); /* For U+00A0 */
SET_BIT(0xE1); /* For U+1680, U+180E */
SET_BIT(0xE2); /* For U+2000 - U+200A, U+202F, U+205F */
SET_BIT(0xE3); /* For U+3000 */
}
else SET_BIT(0xA0);
try_next = FALSE;
break;
case OP_ANYNL:
case OP_VSPACE:
SET_BIT(0x0A);
SET_BIT(0x0B);
SET_BIT(0x0C);
SET_BIT(0x0D);
if (utf8)
{
SET_BIT(0xC2); /* For U+0085 */
SET_BIT(0xE2); /* For U+2028, U+2029 */
}
else SET_BIT(0x85);
try_next = FALSE;
break;
/* Single character types set the bits and stop. Note that if PCRE_UCP
is set, we do not see these op codes because \d etc are converted to
properties. Therefore, these apply in the case when only characters less
than 256 are recognized to match the types. */
case OP_NOT_DIGIT:
set_nottype_bits(start_bits, cbit_digit, table_limit, cd);
try_next = FALSE;
break;
case OP_DIGIT:
set_type_bits(start_bits, cbit_digit, table_limit, cd);
try_next = FALSE;
break;
/* The cbit_space table has vertical tab as whitespace; we have to
ensure it is set as not whitespace. */
case OP_NOT_WHITESPACE:
set_nottype_bits(start_bits, cbit_space, table_limit, cd);
start_bits[1] |= 0x08;
try_next = FALSE;
break;
/* The cbit_space table has vertical tab as whitespace; we have to
not set it from the table. */
case OP_WHITESPACE:
c = start_bits[1]; /* Save in case it was already set */
set_type_bits(start_bits, cbit_space, table_limit, cd);
start_bits[1] = (start_bits[1] & ~0x08) | c;
try_next = FALSE;
break;
case OP_NOT_WORDCHAR:
set_nottype_bits(start_bits, cbit_word, table_limit, cd);
try_next = FALSE;
break;
case OP_WORDCHAR:
set_type_bits(start_bits, cbit_word, table_limit, cd);
try_next = FALSE;
break;
/* One or more character type fudges the pointer and restarts, knowing
it will hit a single character type and stop there. */
case OP_TYPEPLUS:
case OP_TYPEMINPLUS:
case OP_TYPEPOSPLUS:
tcode++;
break;
case OP_TYPEEXACT:
tcode += 3;
break;
/* Zero or more repeats of character types set the bits and then
try again. */
case OP_TYPEUPTO:
case OP_TYPEMINUPTO:
case OP_TYPEPOSUPTO:
tcode += 2; /* Fall through */
case OP_TYPESTAR:
case OP_TYPEMINSTAR:
case OP_TYPEPOSSTAR:
case OP_TYPEQUERY:
case OP_TYPEMINQUERY:
case OP_TYPEPOSQUERY:
switch(tcode[1])
{
default:
case OP_ANY:
case OP_ALLANY:
return SSB_FAIL;
case OP_HSPACE:
SET_BIT(0x09);
SET_BIT(0x20);
if (utf8)
{
SET_BIT(0xC2); /* For U+00A0 */
SET_BIT(0xE1); /* For U+1680, U+180E */
SET_BIT(0xE2); /* For U+2000 - U+200A, U+202F, U+205F */
SET_BIT(0xE3); /* For U+3000 */
}
else SET_BIT(0xA0);
break;
case OP_ANYNL:
case OP_VSPACE:
SET_BIT(0x0A);
SET_BIT(0x0B);
SET_BIT(0x0C);
SET_BIT(0x0D);
if (utf8)
{
SET_BIT(0xC2); /* For U+0085 */
SET_BIT(0xE2); /* For U+2028, U+2029 */
}
else SET_BIT(0x85);
break;
case OP_NOT_DIGIT:
set_nottype_bits(start_bits, cbit_digit, table_limit, cd);
break;
case OP_DIGIT:
set_type_bits(start_bits, cbit_digit, table_limit, cd);
break;
/* The cbit_space table has vertical tab as whitespace; we have to
ensure it gets set as not whitespace. */
case OP_NOT_WHITESPACE:
set_nottype_bits(start_bits, cbit_space, table_limit, cd);
start_bits[1] |= 0x08;
break;
/* The cbit_space table has vertical tab as whitespace; we have to
avoid setting it. */
case OP_WHITESPACE:
c = start_bits[1]; /* Save in case it was already set */
set_type_bits(start_bits, cbit_space, table_limit, cd);
start_bits[1] = (start_bits[1] & ~0x08) | c;
break;
case OP_NOT_WORDCHAR:
set_nottype_bits(start_bits, cbit_word, table_limit, cd);
break;
case OP_WORDCHAR:
set_type_bits(start_bits, cbit_word, table_limit, cd);
break;
}
tcode += 2;
break;
/* Character class where all the information is in a bit map: set the
bits and either carry on or not, according to the repeat count. If it was
a negative class, and we are operating with UTF-8 characters, any byte
with a value >= 0xc4 is a potentially valid starter because it starts a
character with a value > 255. */
case OP_NCLASS:
#ifdef SUPPORT_UTF8
if (utf8)
{
start_bits[24] |= 0xf0; /* Bits for 0xc4 - 0xc8 */
memset(start_bits+25, 0xff, 7); /* Bits for 0xc9 - 0xff */
}
#endif
/* Fall through */
case OP_CLASS:
{
tcode++;
/* In UTF-8 mode, the bits in a bit map correspond to character
values, not to byte values. However, the bit map we are constructing is
for byte values. So we have to do a conversion for characters whose
value is > 127. In fact, there are only two possible starting bytes for
characters in the range 128 - 255. */
#ifdef SUPPORT_UTF8
if (utf8)
{
for (c = 0; c < 16; c++) start_bits[c] |= tcode[c];
for (c = 128; c < 256; c++)
{
if ((tcode[c/8] && (1 << (c&7))) != 0)
{
int d = (c >> 6) | 0xc0; /* Set bit for this starter */
start_bits[d/8] |= (1 << (d&7)); /* and then skip on to the */
c = (c & 0xc0) + 0x40 - 1; /* next relevant character. */
}
}
}
/* In non-UTF-8 mode, the two bit maps are completely compatible. */
else
#endif
{
for (c = 0; c < 32; c++) start_bits[c] |= tcode[c];
}
/* Advance past the bit map, and act on what follows */
tcode += 32;
switch (*tcode)
{
case OP_CRSTAR:
case OP_CRMINSTAR:
case OP_CRQUERY:
case OP_CRMINQUERY:
tcode++;
break;
case OP_CRRANGE:
case OP_CRMINRANGE:
if (((tcode[1] << 8) + tcode[2]) == 0) tcode += 5;
else try_next = FALSE;
break;
default:
try_next = FALSE;
break;
}
}
break; /* End of bitmap class handling */
} /* End of switch */
} /* End of try_next loop */
static int
set_start_bits(const pcre_uchar *code, pcre_uint8 *start_bits, BOOL utf,
compile_data *cd)
{
register pcre_uint32 c;
int yield = SSB_DONE;
#if defined SUPPORT_UTF && defined COMPILE_PCRE8
int table_limit = utf? 16:32;
#else
int table_limit = 32;
#endif
#if 0
/* ========================================================================= */
/* The following comment and code was inserted in January 1999. In May 2006,
when it was observed to cause compiler warnings about unused values, I took it
out again. If anybody is still using OS/2, they will have to put it back
manually. */
/* This next statement and the later reference to dummy are here in order to
trick the optimizer of the IBM C compiler for OS/2 into generating correct
code. Apparently IBM isn't going to fix the problem, and we would rather not
disable optimization (in this module it actually makes a big difference, and
the pcre module can use all the optimization it can get). */
volatile int dummy;
/* ========================================================================= */
#endif
do
{
BOOL try_next = TRUE;
const pcre_uchar *tcode = code + 1 + LINK_SIZE;
if (*code == OP_CBRA || *code == OP_SCBRA ||
*code == OP_CBRAPOS || *code == OP_SCBRAPOS) tcode += IMM2_SIZE;
while (try_next) /* Loop for items in this branch */
{
int rc;
switch(*tcode)
{
/* If we reach something we don't understand, it means a new opcode has
been created that hasn't been added to this code. Hopefully this problem
will be discovered during testing. */
default:
return SSB_UNKNOWN;
/* Fail for a valid opcode that implies no starting bits. */
case OP_ACCEPT:
case OP_ASSERT_ACCEPT:
case OP_ALLANY:
case OP_ANY:
case OP_ANYBYTE:
case OP_CIRC:
case OP_CIRCM:
case OP_CLOSE:
case OP_COMMIT:
case OP_COND:
case OP_CREF:
case OP_DEF:
case OP_DNCREF:
case OP_DNREF:
case OP_DNREFI:
case OP_DNRREF:
case OP_DOLL:
case OP_DOLLM:
case OP_END:
case OP_EOD:
case OP_EODN:
case OP_EXTUNI:
case OP_FAIL:
case OP_MARK:
case OP_NOT:
case OP_NOTEXACT:
case OP_NOTEXACTI:
case OP_NOTI:
case OP_NOTMINPLUS:
case OP_NOTMINPLUSI:
case OP_NOTMINQUERY:
case OP_NOTMINQUERYI:
case OP_NOTMINSTAR:
case OP_NOTMINSTARI:
case OP_NOTMINUPTO:
case OP_NOTMINUPTOI:
case OP_NOTPLUS:
case OP_NOTPLUSI:
case OP_NOTPOSPLUS:
case OP_NOTPOSPLUSI:
case OP_NOTPOSQUERY:
case OP_NOTPOSQUERYI:
case OP_NOTPOSSTAR:
case OP_NOTPOSSTARI:
case OP_NOTPOSUPTO:
case OP_NOTPOSUPTOI:
case OP_NOTPROP:
case OP_NOTQUERY:
case OP_NOTQUERYI:
case OP_NOTSTAR:
case OP_NOTSTARI:
case OP_NOTUPTO:
case OP_NOTUPTOI:
case OP_NOT_HSPACE:
case OP_NOT_VSPACE:
case OP_PRUNE:
case OP_PRUNE_ARG:
case OP_RECURSE:
case OP_REF:
case OP_REFI:
case OP_REVERSE:
case OP_RREF:
case OP_SCOND:
case OP_SET_SOM:
case OP_SKIP:
case OP_SKIP_ARG:
case OP_SOD:
case OP_SOM:
case OP_THEN:
case OP_THEN_ARG:
return SSB_FAIL;
/* A "real" property test implies no starting bits, but the fake property
PT_CLIST identifies a list of characters. These lists are short, as they
are used for characters with more than one "other case", so there is no
point in recognizing them for OP_NOTPROP. */
case OP_PROP:
if (tcode[1] != PT_CLIST) return SSB_FAIL;
{
const pcre_uint32 *p = PRIV(ucd_caseless_sets) + tcode[2];
while ((c = *p++) < NOTACHAR)
{
#if defined SUPPORT_UTF && defined COMPILE_PCRE8
if (utf)
{
pcre_uchar buff[6];
(void)PRIV(ord2utf)(c, buff);
c = buff[0];
}
#endif
if (c > 0xff) SET_BIT(0xff); else SET_BIT(c);
}
}
try_next = FALSE;
break;
/* We can ignore word boundary tests. */
case OP_WORD_BOUNDARY:
case OP_NOT_WORD_BOUNDARY:
tcode++;
break;
/* If we hit a bracket or a positive lookahead assertion, recurse to set
bits from within the subpattern. If it can't find anything, we have to
give up. If it finds some mandatory character(s), we are done for this
branch. Otherwise, carry on scanning after the subpattern. */
case OP_BRA:
case OP_SBRA:
case OP_CBRA:
case OP_SCBRA:
case OP_BRAPOS:
case OP_SBRAPOS:
case OP_CBRAPOS:
case OP_SCBRAPOS:
case OP_ONCE:
case OP_ONCE_NC:
case OP_ASSERT:
rc = set_start_bits(tcode, start_bits, utf, cd);
if (rc == SSB_FAIL || rc == SSB_UNKNOWN) return rc;
if (rc == SSB_DONE) try_next = FALSE; else
{
do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
tcode += 1 + LINK_SIZE;
}
break;
/* If we hit ALT or KET, it means we haven't found anything mandatory in
this branch, though we might have found something optional. For ALT, we
continue with the next alternative, but we have to arrange that the final
result from subpattern is SSB_CONTINUE rather than SSB_DONE. For KET,
return SSB_CONTINUE: if this is the top level, that indicates failure,
but after a nested subpattern, it causes scanning to continue. */
case OP_ALT:
yield = SSB_CONTINUE;
try_next = FALSE;
break;
case OP_KET:
case OP_KETRMAX:
case OP_KETRMIN:
case OP_KETRPOS:
return SSB_CONTINUE;
/* Skip over callout */
case OP_CALLOUT:
tcode += 2 + 2*LINK_SIZE;
break;
/* Skip over lookbehind and negative lookahead assertions */
case OP_ASSERT_NOT:
case OP_ASSERTBACK:
case OP_ASSERTBACK_NOT:
do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
tcode += 1 + LINK_SIZE;
break;
/* BRAZERO does the bracket, but carries on. */
case OP_BRAZERO:
case OP_BRAMINZERO:
case OP_BRAPOSZERO:
rc = set_start_bits(++tcode, start_bits, utf, cd);
if (rc == SSB_FAIL || rc == SSB_UNKNOWN) return rc;
/* =========================================================================
See the comment at the head of this function concerning the next line,
which was an old fudge for the benefit of OS/2.
dummy = 1;
========================================================================= */
do tcode += GET(tcode,1); while (*tcode == OP_ALT);
tcode += 1 + LINK_SIZE;
break;
/* SKIPZERO skips the bracket. */
case OP_SKIPZERO:
tcode++;
do tcode += GET(tcode,1); while (*tcode == OP_ALT);
tcode += 1 + LINK_SIZE;
break;
/* Single-char * or ? sets the bit and tries the next item */
case OP_STAR:
case OP_MINSTAR:
case OP_POSSTAR:
case OP_QUERY:
case OP_MINQUERY:
case OP_POSQUERY:
tcode = set_table_bit(start_bits, tcode + 1, FALSE, cd, utf);
break;
case OP_STARI:
case OP_MINSTARI:
case OP_POSSTARI:
case OP_QUERYI:
case OP_MINQUERYI:
case OP_POSQUERYI:
tcode = set_table_bit(start_bits, tcode + 1, TRUE, cd, utf);
break;
/* Single-char upto sets the bit and tries the next */
case OP_UPTO:
case OP_MINUPTO:
case OP_POSUPTO:
tcode = set_table_bit(start_bits, tcode + 1 + IMM2_SIZE, FALSE, cd, utf);
break;
case OP_UPTOI:
case OP_MINUPTOI:
case OP_POSUPTOI:
tcode = set_table_bit(start_bits, tcode + 1 + IMM2_SIZE, TRUE, cd, utf);
break;
/* At least one single char sets the bit and stops */
case OP_EXACT:
tcode += IMM2_SIZE;
/* Fall through */
case OP_CHAR:
case OP_PLUS:
case OP_MINPLUS:
case OP_POSPLUS:
(void)set_table_bit(start_bits, tcode + 1, FALSE, cd, utf);
try_next = FALSE;
break;
case OP_EXACTI:
tcode += IMM2_SIZE;
/* Fall through */
case OP_CHARI:
case OP_PLUSI:
case OP_MINPLUSI:
case OP_POSPLUSI:
(void)set_table_bit(start_bits, tcode + 1, TRUE, cd, utf);
try_next = FALSE;
break;
/* Special spacing and line-terminating items. These recognize specific
lists of characters. The difference between VSPACE and ANYNL is that the
latter can match the two-character CRLF sequence, but that is not
relevant for finding the first character, so their code here is
identical. */
case OP_HSPACE:
SET_BIT(CHAR_HT);
SET_BIT(CHAR_SPACE);
#ifdef SUPPORT_UTF
if (utf)
{
#ifdef COMPILE_PCRE8
SET_BIT(0xC2); /* For U+00A0 */
SET_BIT(0xE1); /* For U+1680, U+180E */
SET_BIT(0xE2); /* For U+2000 - U+200A, U+202F, U+205F */
SET_BIT(0xE3); /* For U+3000 */
#elif defined COMPILE_PCRE16 || defined COMPILE_PCRE32
SET_BIT(0xA0);
SET_BIT(0xFF); /* For characters > 255 */
#endif /* COMPILE_PCRE[8|16|32] */
}
else
#endif /* SUPPORT_UTF */
{
#ifndef EBCDIC
SET_BIT(0xA0);
#endif /* Not EBCDIC */
#if defined COMPILE_PCRE16 || defined COMPILE_PCRE32
SET_BIT(0xFF); /* For characters > 255 */
#endif /* COMPILE_PCRE[16|32] */
}
try_next = FALSE;
break;
case OP_ANYNL:
case OP_VSPACE:
SET_BIT(CHAR_LF);
SET_BIT(CHAR_VT);
SET_BIT(CHAR_FF);
SET_BIT(CHAR_CR);
#ifdef SUPPORT_UTF
if (utf)
{
#ifdef COMPILE_PCRE8
SET_BIT(0xC2); /* For U+0085 */
SET_BIT(0xE2); /* For U+2028, U+2029 */
#elif defined COMPILE_PCRE16 || defined COMPILE_PCRE32
SET_BIT(CHAR_NEL);
SET_BIT(0xFF); /* For characters > 255 */
#endif /* COMPILE_PCRE[8|16|32] */
}
else
#endif /* SUPPORT_UTF */
{
SET_BIT(CHAR_NEL);
#if defined COMPILE_PCRE16 || defined COMPILE_PCRE32
SET_BIT(0xFF); /* For characters > 255 */
#endif
}
try_next = FALSE;
break;
/* Single character types set the bits and stop. Note that if PCRE_UCP
is set, we do not see these op codes because \d etc are converted to
properties. Therefore, these apply in the case when only characters less
than 256 are recognized to match the types. */
case OP_NOT_DIGIT:
set_nottype_bits(start_bits, cbit_digit, table_limit, cd);
try_next = FALSE;
break;
case OP_DIGIT:
set_type_bits(start_bits, cbit_digit, table_limit, cd);
try_next = FALSE;
break;
/* The cbit_space table has vertical tab as whitespace; we no longer
have to play fancy tricks because Perl added VT to its whitespace at
release 5.18. PCRE added it at release 8.34. */
case OP_NOT_WHITESPACE:
set_nottype_bits(start_bits, cbit_space, table_limit, cd);
try_next = FALSE;
break;
case OP_WHITESPACE:
set_type_bits(start_bits, cbit_space, table_limit, cd);
try_next = FALSE;
break;
case OP_NOT_WORDCHAR:
set_nottype_bits(start_bits, cbit_word, table_limit, cd);
try_next = FALSE;
break;
case OP_WORDCHAR:
set_type_bits(start_bits, cbit_word, table_limit, cd);
try_next = FALSE;
break;
/* One or more character type fudges the pointer and restarts, knowing
it will hit a single character type and stop there. */
case OP_TYPEPLUS:
case OP_TYPEMINPLUS:
case OP_TYPEPOSPLUS:
tcode++;
break;
case OP_TYPEEXACT:
tcode += 1 + IMM2_SIZE;
break;
/* Zero or more repeats of character types set the bits and then
try again. */
case OP_TYPEUPTO:
case OP_TYPEMINUPTO:
case OP_TYPEPOSUPTO:
tcode += IMM2_SIZE; /* Fall through */
case OP_TYPESTAR:
case OP_TYPEMINSTAR:
case OP_TYPEPOSSTAR:
case OP_TYPEQUERY:
case OP_TYPEMINQUERY:
case OP_TYPEPOSQUERY:
switch(tcode[1])
{
default:
case OP_ANY:
case OP_ALLANY:
return SSB_FAIL;
case OP_HSPACE:
SET_BIT(CHAR_HT);
SET_BIT(CHAR_SPACE);
#ifdef SUPPORT_UTF
if (utf)
{
#ifdef COMPILE_PCRE8
SET_BIT(0xC2); /* For U+00A0 */
SET_BIT(0xE1); /* For U+1680, U+180E */
SET_BIT(0xE2); /* For U+2000 - U+200A, U+202F, U+205F */
SET_BIT(0xE3); /* For U+3000 */
#elif defined COMPILE_PCRE16 || defined COMPILE_PCRE32
SET_BIT(0xA0);
SET_BIT(0xFF); /* For characters > 255 */
#endif /* COMPILE_PCRE[8|16|32] */
}
else
#endif /* SUPPORT_UTF */
#ifndef EBCDIC
SET_BIT(0xA0);
#endif /* Not EBCDIC */
break;
case OP_ANYNL:
case OP_VSPACE:
SET_BIT(CHAR_LF);
SET_BIT(CHAR_VT);
SET_BIT(CHAR_FF);
SET_BIT(CHAR_CR);
#ifdef SUPPORT_UTF
if (utf)
{
#ifdef COMPILE_PCRE8
SET_BIT(0xC2); /* For U+0085 */
SET_BIT(0xE2); /* For U+2028, U+2029 */
#elif defined COMPILE_PCRE16 || defined COMPILE_PCRE32
SET_BIT(CHAR_NEL);
SET_BIT(0xFF); /* For characters > 255 */
#endif /* COMPILE_PCRE16 */
}
else
#endif /* SUPPORT_UTF */
SET_BIT(CHAR_NEL);
break;
case OP_NOT_DIGIT:
set_nottype_bits(start_bits, cbit_digit, table_limit, cd);
break;
case OP_DIGIT:
set_type_bits(start_bits, cbit_digit, table_limit, cd);
break;
/* The cbit_space table has vertical tab as whitespace; we no longer
have to play fancy tricks because Perl added VT to its whitespace at
release 5.18. PCRE added it at release 8.34. */
case OP_NOT_WHITESPACE:
set_nottype_bits(start_bits, cbit_space, table_limit, cd);
break;
case OP_WHITESPACE:
set_type_bits(start_bits, cbit_space, table_limit, cd);
break;
case OP_NOT_WORDCHAR:
set_nottype_bits(start_bits, cbit_word, table_limit, cd);
break;
case OP_WORDCHAR:
set_type_bits(start_bits, cbit_word, table_limit, cd);
break;
}
tcode += 2;
break;
/* Character class where all the information is in a bit map: set the
bits and either carry on or not, according to the repeat count. If it was
a negative class, and we are operating with UTF-8 characters, any byte
with a value >= 0xc4 is a potentially valid starter because it starts a
character with a value > 255. */
#if defined SUPPORT_UTF || !defined COMPILE_PCRE8
case OP_XCLASS:
if ((tcode[1 + LINK_SIZE] & XCL_HASPROP) != 0)
return SSB_FAIL;
/* All bits are set. */
if ((tcode[1 + LINK_SIZE] & XCL_MAP) == 0 && (tcode[1 + LINK_SIZE] & XCL_NOT) != 0)
return SSB_FAIL;
#endif
/* Fall through */
case OP_NCLASS:
#if defined SUPPORT_UTF && defined COMPILE_PCRE8
if (utf)
{
start_bits[24] |= 0xf0; /* Bits for 0xc4 - 0xc8 */
memset(start_bits+25, 0xff, 7); /* Bits for 0xc9 - 0xff */
}
#endif
#if defined COMPILE_PCRE16 || defined COMPILE_PCRE32
SET_BIT(0xFF); /* For characters > 255 */
#endif
/* Fall through */
case OP_CLASS:
{
pcre_uint8 *map;
#if defined SUPPORT_UTF || !defined COMPILE_PCRE8
map = NULL;
if (*tcode == OP_XCLASS)
{
if ((tcode[1 + LINK_SIZE] & XCL_MAP) != 0)
map = (pcre_uint8 *)(tcode + 1 + LINK_SIZE + 1);
tcode += GET(tcode, 1);
}
else
#endif
{
tcode++;
map = (pcre_uint8 *)tcode;
tcode += 32 / sizeof(pcre_uchar);
}
/* In UTF-8 mode, the bits in a bit map correspond to character
values, not to byte values. However, the bit map we are constructing is
for byte values. So we have to do a conversion for characters whose
value is > 127. In fact, there are only two possible starting bytes for
characters in the range 128 - 255. */
#if defined SUPPORT_UTF || !defined COMPILE_PCRE8
if (map != NULL)
#endif
{
#if defined SUPPORT_UTF && defined COMPILE_PCRE8
if (utf)
{
for (c = 0; c < 16; c++) start_bits[c] |= map[c];
for (c = 128; c < 256; c++)
{
if ((map[c/8] & (1 << (c&7))) != 0)
{
int d = (c >> 6) | 0xc0; /* Set bit for this starter */
start_bits[d/8] |= (1 << (d&7)); /* and then skip on to the */
c = (c & 0xc0) + 0x40 - 1; /* next relevant character. */
}
}
}
else
#endif
{
/* In non-UTF-8 mode, the two bit maps are completely compatible. */
for (c = 0; c < 32; c++) start_bits[c] |= map[c];
}
}
/* Advance past the bit map, and act on what follows. For a zero
minimum repeat, continue; otherwise stop processing. */
switch (*tcode)
{
case OP_CRSTAR:
case OP_CRMINSTAR:
case OP_CRQUERY:
case OP_CRMINQUERY:
case OP_CRPOSSTAR:
case OP_CRPOSQUERY:
tcode++;
break;
case OP_CRRANGE:
case OP_CRMINRANGE:
case OP_CRPOSRANGE:
if (GET2(tcode, 1) == 0) tcode += 1 + 2 * IMM2_SIZE;
else try_next = FALSE;
break;
default:
try_next = FALSE;
break;
}
}
break; /* End of bitmap class handling */
} /* End of switch */
} /* End of try_next loop */
static int
set_start_bits(pcre2_real_code *re, PCRE2_SPTR code, BOOL utf)
{
uint32_t c;
int yield = SSB_DONE;
#if defined SUPPORT_UNICODE && PCRE2_CODE_UNIT_WIDTH == 8
int table_limit = utf? 16:32;
#else
int table_limit = 32;
#endif
do
{
BOOL try_next = TRUE;
PCRE2_SPTR tcode = code + 1 + LINK_SIZE;
if (*code == OP_CBRA || *code == OP_SCBRA ||
*code == OP_CBRAPOS || *code == OP_SCBRAPOS) tcode += IMM2_SIZE;
while (try_next) /* Loop for items in this branch */
{
int rc;
uint8_t *classmap = NULL;
switch(*tcode)
{
/* If we reach something we don't understand, it means a new opcode has
been created that hasn't been added to this function. Hopefully this
problem will be discovered during testing. */
default:
return SSB_UNKNOWN;
/* Fail for a valid opcode that implies no starting bits. */
case OP_ACCEPT:
case OP_ASSERT_ACCEPT:
case OP_ALLANY:
case OP_ANY:
case OP_ANYBYTE:
case OP_CIRC:
case OP_CIRCM:
case OP_CLOSE:
case OP_COMMIT:
case OP_COND:
case OP_CREF:
case OP_FALSE:
case OP_TRUE:
case OP_DNCREF:
case OP_DNREF:
case OP_DNREFI:
case OP_DNRREF:
case OP_DOLL:
case OP_DOLLM:
case OP_END:
case OP_EOD:
case OP_EODN:
case OP_EXTUNI:
case OP_FAIL:
case OP_MARK:
case OP_NOT:
case OP_NOTEXACT:
case OP_NOTEXACTI:
case OP_NOTI:
case OP_NOTMINPLUS:
case OP_NOTMINPLUSI:
case OP_NOTMINQUERY:
case OP_NOTMINQUERYI:
case OP_NOTMINSTAR:
case OP_NOTMINSTARI:
case OP_NOTMINUPTO:
case OP_NOTMINUPTOI:
case OP_NOTPLUS:
case OP_NOTPLUSI:
case OP_NOTPOSPLUS:
case OP_NOTPOSPLUSI:
case OP_NOTPOSQUERY:
case OP_NOTPOSQUERYI:
case OP_NOTPOSSTAR:
case OP_NOTPOSSTARI:
case OP_NOTPOSUPTO:
case OP_NOTPOSUPTOI:
case OP_NOTPROP:
case OP_NOTQUERY:
case OP_NOTQUERYI:
case OP_NOTSTAR:
case OP_NOTSTARI:
case OP_NOTUPTO:
case OP_NOTUPTOI:
case OP_NOT_HSPACE:
case OP_NOT_VSPACE:
case OP_PRUNE:
case OP_PRUNE_ARG:
case OP_RECURSE:
case OP_REF:
case OP_REFI:
case OP_REVERSE:
case OP_RREF:
case OP_SCOND:
case OP_SET_SOM:
case OP_SKIP:
case OP_SKIP_ARG:
case OP_SOD:
case OP_SOM:
case OP_THEN:
case OP_THEN_ARG:
return SSB_FAIL;
/* A "real" property test implies no starting bits, but the fake property
PT_CLIST identifies a list of characters. These lists are short, as they
are used for characters with more than one "other case", so there is no
point in recognizing them for OP_NOTPROP. */
case OP_PROP:
if (tcode[1] != PT_CLIST) return SSB_FAIL;
{
const uint32_t *p = PRIV(ucd_caseless_sets) + tcode[2];
while ((c = *p++) < NOTACHAR)
{
#if defined SUPPORT_UNICODE && PCRE2_CODE_UNIT_WIDTH == 8
if (utf)
{
PCRE2_UCHAR buff[6];
(void)PRIV(ord2utf)(c, buff);
c = buff[0];
}
#endif
if (c > 0xff) SET_BIT(0xff); else SET_BIT(c);
}
}
try_next = FALSE;
break;
/* We can ignore word boundary tests. */
case OP_WORD_BOUNDARY:
case OP_NOT_WORD_BOUNDARY:
tcode++;
break;
/* If we hit a bracket or a positive lookahead assertion, recurse to set
bits from within the subpattern. If it can't find anything, we have to
give up. If it finds some mandatory character(s), we are done for this
branch. Otherwise, carry on scanning after the subpattern. */
case OP_BRA:
case OP_SBRA:
case OP_CBRA:
case OP_SCBRA:
case OP_BRAPOS:
case OP_SBRAPOS:
case OP_CBRAPOS:
case OP_SCBRAPOS:
case OP_ONCE:
case OP_ONCE_NC:
case OP_ASSERT:
rc = set_start_bits(re, tcode, utf);
if (rc == SSB_FAIL || rc == SSB_UNKNOWN) return rc;
if (rc == SSB_DONE) try_next = FALSE; else
{
do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
tcode += 1 + LINK_SIZE;
}
break;
/* If we hit ALT or KET, it means we haven't found anything mandatory in
this branch, though we might have found something optional. For ALT, we
continue with the next alternative, but we have to arrange that the final
result from subpattern is SSB_CONTINUE rather than SSB_DONE. For KET,
return SSB_CONTINUE: if this is the top level, that indicates failure,
but after a nested subpattern, it causes scanning to continue. */
case OP_ALT:
yield = SSB_CONTINUE;
try_next = FALSE;
break;
case OP_KET:
case OP_KETRMAX:
case OP_KETRMIN:
case OP_KETRPOS:
return SSB_CONTINUE;
/* Skip over callout */
case OP_CALLOUT:
tcode += PRIV(OP_lengths)[OP_CALLOUT];
break;
case OP_CALLOUT_STR:
tcode += GET(tcode, 1 + 2*LINK_SIZE);
break;
/* Skip over lookbehind and negative lookahead assertions */
case OP_ASSERT_NOT:
case OP_ASSERTBACK:
case OP_ASSERTBACK_NOT:
do tcode += GET(tcode, 1); while (*tcode == OP_ALT);
tcode += 1 + LINK_SIZE;
break;
/* BRAZERO does the bracket, but carries on. */
case OP_BRAZERO:
case OP_BRAMINZERO:
case OP_BRAPOSZERO:
rc = set_start_bits(re, ++tcode, utf);
if (rc == SSB_FAIL || rc == SSB_UNKNOWN) return rc;
do tcode += GET(tcode,1); while (*tcode == OP_ALT);
tcode += 1 + LINK_SIZE;
break;
/* SKIPZERO skips the bracket. */
case OP_SKIPZERO:
tcode++;
do tcode += GET(tcode,1); while (*tcode == OP_ALT);
tcode += 1 + LINK_SIZE;
break;
/* Single-char * or ? sets the bit and tries the next item */
case OP_STAR:
case OP_MINSTAR:
case OP_POSSTAR:
case OP_QUERY:
case OP_MINQUERY:
case OP_POSQUERY:
tcode = set_table_bit(re, tcode + 1, FALSE, utf);
break;
case OP_STARI:
case OP_MINSTARI:
case OP_POSSTARI:
case OP_QUERYI:
case OP_MINQUERYI:
case OP_POSQUERYI:
tcode = set_table_bit(re, tcode + 1, TRUE, utf);
break;
/* Single-char upto sets the bit and tries the next */
case OP_UPTO:
case OP_MINUPTO:
case OP_POSUPTO:
tcode = set_table_bit(re, tcode + 1 + IMM2_SIZE, FALSE, utf);
break;
case OP_UPTOI:
case OP_MINUPTOI:
case OP_POSUPTOI:
tcode = set_table_bit(re, tcode + 1 + IMM2_SIZE, TRUE, utf);
break;
/* At least one single char sets the bit and stops */
case OP_EXACT:
tcode += IMM2_SIZE;
/* Fall through */
case OP_CHAR:
case OP_PLUS:
case OP_MINPLUS:
case OP_POSPLUS:
(void)set_table_bit(re, tcode + 1, FALSE, utf);
try_next = FALSE;
break;
case OP_EXACTI:
tcode += IMM2_SIZE;
/* Fall through */
case OP_CHARI:
case OP_PLUSI:
case OP_MINPLUSI:
case OP_POSPLUSI:
(void)set_table_bit(re, tcode + 1, TRUE, utf);
try_next = FALSE;
break;
/* Special spacing and line-terminating items. These recognize specific
lists of characters. The difference between VSPACE and ANYNL is that the
latter can match the two-character CRLF sequence, but that is not
relevant for finding the first character, so their code here is
identical. */
case OP_HSPACE:
SET_BIT(CHAR_HT);
SET_BIT(CHAR_SPACE);
/* For the 16-bit and 32-bit libraries (which can never be EBCDIC), set
the bits for 0xA0 and for code units >= 255, independently of UTF. */
#if PCRE2_CODE_UNIT_WIDTH != 8
SET_BIT(0xA0);
SET_BIT(0xFF);
#else
/* For the 8-bit library in UTF-8 mode, set the bits for the first code
units of horizontal space characters. */
#ifdef SUPPORT_UNICODE
if (utf)
{
SET_BIT(0xC2); /* For U+00A0 */
SET_BIT(0xE1); /* For U+1680, U+180E */
SET_BIT(0xE2); /* For U+2000 - U+200A, U+202F, U+205F */
SET_BIT(0xE3); /* For U+3000 */
}
else
#endif
/* For the 8-bit library not in UTF-8 mode, set the bit for 0xA0, unless
the code is EBCDIC. */
{
#ifndef EBCDIC
SET_BIT(0xA0);
#endif /* Not EBCDIC */
}
#endif /* 8-bit support */
try_next = FALSE;
break;
case OP_ANYNL:
case OP_VSPACE:
SET_BIT(CHAR_LF);
SET_BIT(CHAR_VT);
SET_BIT(CHAR_FF);
SET_BIT(CHAR_CR);
/* For the 16-bit and 32-bit libraries (which can never be EBCDIC), set
the bits for NEL and for code units >= 255, independently of UTF. */
#if PCRE2_CODE_UNIT_WIDTH != 8
SET_BIT(CHAR_NEL);
SET_BIT(0xFF);
#else
/* For the 8-bit library in UTF-8 mode, set the bits for the first code
units of vertical space characters. */
#ifdef SUPPORT_UNICODE
if (utf)
{
SET_BIT(0xC2); /* For U+0085 (NEL) */
SET_BIT(0xE2); /* For U+2028, U+2029 */
}
else
#endif
/* For the 8-bit library not in UTF-8 mode, set the bit for NEL. */
{
SET_BIT(CHAR_NEL);
}
#endif /* 8-bit support */
try_next = FALSE;
break;
/* Single character types set the bits and stop. Note that if PCRE2_UCP
is set, we do not see these op codes because \d etc are converted to
properties. Therefore, these apply in the case when only characters less
than 256 are recognized to match the types. */
case OP_NOT_DIGIT:
set_nottype_bits(re, cbit_digit, table_limit);
try_next = FALSE;
break;
case OP_DIGIT:
set_type_bits(re, cbit_digit, table_limit);
try_next = FALSE;
break;
case OP_NOT_WHITESPACE:
set_nottype_bits(re, cbit_space, table_limit);
try_next = FALSE;
break;
case OP_WHITESPACE:
set_type_bits(re, cbit_space, table_limit);
try_next = FALSE;
break;
case OP_NOT_WORDCHAR:
set_nottype_bits(re, cbit_word, table_limit);
try_next = FALSE;
break;
case OP_WORDCHAR:
set_type_bits(re, cbit_word, table_limit);
try_next = FALSE;
break;
/* One or more character type fudges the pointer and restarts, knowing
it will hit a single character type and stop there. */
case OP_TYPEPLUS:
case OP_TYPEMINPLUS:
case OP_TYPEPOSPLUS:
tcode++;
break;
case OP_TYPEEXACT:
tcode += 1 + IMM2_SIZE;
break;
/* Zero or more repeats of character types set the bits and then
try again. */
case OP_TYPEUPTO:
case OP_TYPEMINUPTO:
case OP_TYPEPOSUPTO:
tcode += IMM2_SIZE; /* Fall through */
case OP_TYPESTAR:
case OP_TYPEMINSTAR:
case OP_TYPEPOSSTAR:
case OP_TYPEQUERY:
case OP_TYPEMINQUERY:
case OP_TYPEPOSQUERY:
switch(tcode[1])
{
default:
case OP_ANY:
case OP_ALLANY:
return SSB_FAIL;
case OP_HSPACE:
SET_BIT(CHAR_HT);
SET_BIT(CHAR_SPACE);
/* For the 16-bit and 32-bit libraries (which can never be EBCDIC), set
the bits for 0xA0 and for code units >= 255, independently of UTF. */
#if PCRE2_CODE_UNIT_WIDTH != 8
SET_BIT(0xA0);
SET_BIT(0xFF);
#else
/* For the 8-bit library in UTF-8 mode, set the bits for the first code
units of horizontal space characters. */
#ifdef SUPPORT_UNICODE
if (utf)
{
SET_BIT(0xC2); /* For U+00A0 */
SET_BIT(0xE1); /* For U+1680, U+180E */
SET_BIT(0xE2); /* For U+2000 - U+200A, U+202F, U+205F */
SET_BIT(0xE3); /* For U+3000 */
}
else
#endif
/* For the 8-bit library not in UTF-8 mode, set the bit for 0xA0, unless
the code is EBCDIC. */
{
#ifndef EBCDIC
SET_BIT(0xA0);
#endif /* Not EBCDIC */
}
#endif /* 8-bit support */
break;
case OP_ANYNL:
case OP_VSPACE:
SET_BIT(CHAR_LF);
SET_BIT(CHAR_VT);
SET_BIT(CHAR_FF);
SET_BIT(CHAR_CR);
/* For the 16-bit and 32-bit libraries (which can never be EBCDIC), set
the bits for NEL and for code units >= 255, independently of UTF. */
#if PCRE2_CODE_UNIT_WIDTH != 8
SET_BIT(CHAR_NEL);
SET_BIT(0xFF);
#else
/* For the 8-bit library in UTF-8 mode, set the bits for the first code
units of vertical space characters. */
#ifdef SUPPORT_UNICODE
if (utf)
{
SET_BIT(0xC2); /* For U+0085 (NEL) */
SET_BIT(0xE2); /* For U+2028, U+2029 */
}
else
#endif
/* For the 8-bit library not in UTF-8 mode, set the bit for NEL. */
{
SET_BIT(CHAR_NEL);
}
#endif /* 8-bit support */
break;
case OP_NOT_DIGIT:
set_nottype_bits(re, cbit_digit, table_limit);
break;
case OP_DIGIT:
set_type_bits(re, cbit_digit, table_limit);
break;
case OP_NOT_WHITESPACE:
set_nottype_bits(re, cbit_space, table_limit);
break;
case OP_WHITESPACE:
set_type_bits(re, cbit_space, table_limit);
break;
case OP_NOT_WORDCHAR:
set_nottype_bits(re, cbit_word, table_limit);
break;
case OP_WORDCHAR:
set_type_bits(re, cbit_word, table_limit);
break;
}
tcode += 2;
break;
/* Extended class: if there are any property checks, or if this is a
negative XCLASS without a map, give up. If there are no property checks,
there must be wide characters on the XCLASS list, because otherwise an
XCLASS would not have been created. This means that code points >= 255
are always potential starters. */
#ifdef SUPPORT_WIDE_CHARS
case OP_XCLASS:
if ((tcode[1 + LINK_SIZE] & XCL_HASPROP) != 0 ||
(tcode[1 + LINK_SIZE] & (XCL_MAP|XCL_NOT)) == XCL_NOT)
return SSB_FAIL;
/* We have a positive XCLASS or a negative one without a map. Set up the
map pointer if there is one, and fall through. */
classmap = ((tcode[1 + LINK_SIZE] & XCL_MAP) == 0)? NULL :
(uint8_t *)(tcode + 1 + LINK_SIZE + 1);
#endif
/* Enter here for a negative non-XCLASS. In the 8-bit library, if we are
in UTF mode, any byte with a value >= 0xc4 is a potentially valid starter
because it starts a character with a value > 255. In 8-bit non-UTF mode,
there is no difference between CLASS and NCLASS. In all other wide
character modes, set the 0xFF bit to indicate code units >= 255. */
case OP_NCLASS:
#if defined SUPPORT_UNICODE && PCRE2_CODE_UNIT_WIDTH == 8
if (utf)
{
re->start_bitmap[24] |= 0xf0; /* Bits for 0xc4 - 0xc8 */
memset(re->start_bitmap+25, 0xff, 7); /* Bits for 0xc9 - 0xff */
}
#elif PCRE2_CODE_UNIT_WIDTH != 8
SET_BIT(0xFF); /* For characters >= 255 */
#endif
/* Fall through */
/* Enter here for a positive non-XCLASS. If we have fallen through from
an XCLASS, classmap will already be set; just advance the code pointer.
Otherwise, set up classmap for a a non-XCLASS and advance past it. */
case OP_CLASS:
if (*tcode == OP_XCLASS) tcode += GET(tcode, 1); else
{
classmap = (uint8_t *)(++tcode);
tcode += 32 / sizeof(PCRE2_UCHAR);
}
/* When wide characters are supported, classmap may be NULL. In UTF-8
(sic) mode, the bits in a class bit map correspond to character values,
not to byte values. However, the bit map we are constructing is for byte
values. So we have to do a conversion for characters whose code point is
greater than 127. In fact, there are only two possible starting bytes for
characters in the range 128 - 255. */
if (classmap != NULL)
{
#if defined SUPPORT_UNICODE && PCRE2_CODE_UNIT_WIDTH == 8
if (utf)
{
for (c = 0; c < 16; c++) re->start_bitmap[c] |= classmap[c];
for (c = 128; c < 256; c++)
{
if ((classmap[c/8] & (1 << (c&7))) != 0)
{
int d = (c >> 6) | 0xc0; /* Set bit for this starter */
re->start_bitmap[d/8] |= (1 << (d&7)); /* and then skip on to the */
c = (c & 0xc0) + 0x40 - 1; /* next relevant character. */
}
}
}
else
#endif
/* In all modes except UTF-8, the two bit maps are compatible. */
{
for (c = 0; c < 32; c++) re->start_bitmap[c] |= classmap[c];
}
}
/* Act on what follows the class. For a zero minimum repeat, continue;
otherwise stop processing. */
switch (*tcode)
{
case OP_CRSTAR:
case OP_CRMINSTAR:
case OP_CRQUERY:
case OP_CRMINQUERY:
case OP_CRPOSSTAR:
case OP_CRPOSQUERY:
tcode++;
break;
case OP_CRRANGE:
case OP_CRMINRANGE:
case OP_CRPOSRANGE:
if (GET2(tcode, 1) == 0) tcode += 1 + 2 * IMM2_SIZE;
else try_next = FALSE;
break;
default:
try_next = FALSE;
break;
}
break; /* End of class handling case */
} /* End of switch for opcodes */
} /* End of try_next loop */
