data->flags |= (OP(scan) == MEOL
? SF_BEFORE_MEOL
: SF_BEFORE_SEOL);
+ SCAN_COMMIT(pRExC_state, data, minlenp);
+
}
else if ( PL_regkind[OP(scan)] == BRANCHJ
/* Lookbehind, or need to calculate parens/evals/stclass: */
* the original pattern needs upgrading to utf8.
*/
-bool
+static bool
S_compile_runtime_code(pTHX_ RExC_state_t * const pRExC_state,
char *pat, STRLEN plen)
{
*flagp |= HASWIDTH;
goto finish_meta_pat;
case 'w':
- switch (get_regex_charset(RExC_flags)) {
- case REGEX_LOCALE_CHARSET:
- op = ALNUML;
- break;
- case REGEX_UNICODE_CHARSET:
- op = ALNUMU;
- break;
- case REGEX_ASCII_RESTRICTED_CHARSET:
- case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
- op = ALNUMA;
- break;
- case REGEX_DEPENDS_CHARSET:
- op = ALNUM;
- break;
- default:
- goto bad_charset;
+ op = ALNUM + get_regex_charset(RExC_flags);
+ if (op > ALNUMA) { /* /aa is same as /a */
+ op = ALNUMA;
}
ret = reg_node(pRExC_state, op);
*flagp |= HASWIDTH|SIMPLE;
goto finish_meta_pat;
case 'W':
- switch (get_regex_charset(RExC_flags)) {
- case REGEX_LOCALE_CHARSET:
- op = NALNUML;
- break;
- case REGEX_UNICODE_CHARSET:
- op = NALNUMU;
- break;
- case REGEX_ASCII_RESTRICTED_CHARSET:
- case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
- op = NALNUMA;
- break;
- case REGEX_DEPENDS_CHARSET:
- op = NALNUM;
- break;
- default:
- goto bad_charset;
+ op = NALNUM + get_regex_charset(RExC_flags);
+ if (op > NALNUMA) { /* /aa is same as /a */
+ op = NALNUMA;
}
ret = reg_node(pRExC_state, op);
*flagp |= HASWIDTH|SIMPLE;
case 'b':
RExC_seen_zerolen++;
RExC_seen |= REG_SEEN_LOOKBEHIND;
- switch (get_regex_charset(RExC_flags)) {
- case REGEX_LOCALE_CHARSET:
- op = BOUNDL;
- break;
- case REGEX_UNICODE_CHARSET:
- op = BOUNDU;
- break;
- case REGEX_ASCII_RESTRICTED_CHARSET:
- case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
- op = BOUNDA;
- break;
- case REGEX_DEPENDS_CHARSET:
- op = BOUND;
- break;
- default:
- goto bad_charset;
+ op = BOUND + get_regex_charset(RExC_flags);
+ if (op > BOUNDA) { /* /aa is same as /a */
+ op = BOUNDA;
}
ret = reg_node(pRExC_state, op);
FLAGS(ret) = get_regex_charset(RExC_flags);
case 'B':
RExC_seen_zerolen++;
RExC_seen |= REG_SEEN_LOOKBEHIND;
- switch (get_regex_charset(RExC_flags)) {
- case REGEX_LOCALE_CHARSET:
- op = NBOUNDL;
- break;
- case REGEX_UNICODE_CHARSET:
- op = NBOUNDU;
- break;
- case REGEX_ASCII_RESTRICTED_CHARSET:
- case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
- op = NBOUNDA;
- break;
- case REGEX_DEPENDS_CHARSET:
- op = NBOUND;
- break;
- default:
- goto bad_charset;
+ op = NBOUND + get_regex_charset(RExC_flags);
+ if (op > NBOUNDA) { /* /aa is same as /a */
+ op = NBOUNDA;
}
ret = reg_node(pRExC_state, op);
FLAGS(ret) = get_regex_charset(RExC_flags);
*flagp |= SIMPLE;
goto finish_meta_pat;
case 's':
- switch (get_regex_charset(RExC_flags)) {
- case REGEX_LOCALE_CHARSET:
- op = SPACEL;
- break;
- case REGEX_UNICODE_CHARSET:
- op = SPACEU;
- break;
- case REGEX_ASCII_RESTRICTED_CHARSET:
- case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
- op = SPACEA;
- break;
- case REGEX_DEPENDS_CHARSET:
- op = SPACE;
- break;
- default:
- goto bad_charset;
+ op = SPACE + get_regex_charset(RExC_flags);
+ if (op > SPACEA) { /* /aa is same as /a */
+ op = SPACEA;
}
ret = reg_node(pRExC_state, op);
*flagp |= HASWIDTH|SIMPLE;
goto finish_meta_pat;
case 'S':
- switch (get_regex_charset(RExC_flags)) {
- case REGEX_LOCALE_CHARSET:
- op = NSPACEL;
- break;
- case REGEX_UNICODE_CHARSET:
- op = NSPACEU;
- break;
- case REGEX_ASCII_RESTRICTED_CHARSET:
- case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
- op = NSPACEA;
- break;
- case REGEX_DEPENDS_CHARSET:
- op = NSPACE;
- break;
- default:
- goto bad_charset;
- }
- ret = reg_node(pRExC_state, op);
- *flagp |= HASWIDTH|SIMPLE;
- goto finish_meta_pat;
- case 'd':
- switch (get_regex_charset(RExC_flags)) {
- case REGEX_LOCALE_CHARSET:
- op = DIGITL;
- break;
- case REGEX_ASCII_RESTRICTED_CHARSET:
- case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
- op = DIGITA;
- break;
- case REGEX_DEPENDS_CHARSET: /* No difference between these */
- case REGEX_UNICODE_CHARSET:
- op = DIGIT;
- break;
- default:
- goto bad_charset;
+ op = NSPACE + get_regex_charset(RExC_flags);
+ if (op > NSPACEA) { /* /aa is same as /a */
+ op = NSPACEA;
}
ret = reg_node(pRExC_state, op);
*flagp |= HASWIDTH|SIMPLE;
goto finish_meta_pat;
case 'D':
- switch (get_regex_charset(RExC_flags)) {
- case REGEX_LOCALE_CHARSET:
- op = NDIGITL;
- break;
- case REGEX_ASCII_RESTRICTED_CHARSET:
- case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
- op = NDIGITA;
- break;
- case REGEX_DEPENDS_CHARSET: /* No difference between these */
- case REGEX_UNICODE_CHARSET:
- op = NDIGIT;
- break;
- default:
- goto bad_charset;
+ op = NDIGIT;
+ goto join_D_and_d;
+ case 'd':
+ op = DIGIT;
+ join_D_and_d:
+ {
+ U8 offset = get_regex_charset(RExC_flags);
+ if (offset == REGEX_UNICODE_CHARSET) {
+ offset = REGEX_DEPENDS_CHARSET;
+ }
+ else if (offset == REGEX_ASCII_MORE_RESTRICTED_CHARSET) {
+ offset = REGEX_ASCII_RESTRICTED_CHARSET;
+ }
+ op += offset;
}
ret = reg_node(pRExC_state, op);
*flagp |= HASWIDTH|SIMPLE;
bool is_exactfu_sharp_s;
ender = 0;
- node_type = ((! FOLD) ? EXACT
- : (LOC)
- ? EXACTFL
- : (MORE_ASCII_RESTRICTED)
- ? EXACTFA
- : (AT_LEAST_UNI_SEMANTICS)
- ? EXACTFU
- : EXACTF);
+ if (! FOLD) {
+ node_type = EXACT;
+ }
+ else {
+ node_type = get_regex_charset(RExC_flags);
+ if (node_type >= REGEX_ASCII_RESTRICTED_CHARSET) {
+ node_type--; /* /a is same as /u, and map /aa's offset to
+ what /a's would have been, so there is no
+ hole */
+ }
+ node_type += EXACTF;
+ }
ret = reg_node(pRExC_state, node_type);
s = STRING(ret);
}
return(ret);
-
-/* Jumped to when an unrecognized character set is encountered */
-bad_charset:
- Perl_croak(aTHX_ "panic: Unknown regex character set encoding: %u", get_regex_charset(RExC_flags));
- return(NULL);
}
STATIC char *
} \
}
-STATIC U8
-S_set_regclass_bit_fold(pTHX_ RExC_state_t *pRExC_state, regnode* node, const U8 value, SV** invlist_ptr, AV** alternate_ptr)
-{
-
- /* Handle the setting of folds in the bitmap for non-locale ANYOF nodes.
- * Locale folding is done at run-time, so this function should not be
- * called for nodes that are for locales.
- *
- * This function sets the bit corresponding to the fold of the input
- * 'value', if not already set. The fold of 'f' is 'F', and the fold of
- * 'F' is 'f'.
- *
- * It also knows about the characters that are in the bitmap that have
- * folds that are matchable only outside it, and sets the appropriate lists
- * and flags.
- *
- * It returns the number of bits that actually changed from 0 to 1 */
-
- U8 stored = 0;
- U8 fold;
-
- PERL_ARGS_ASSERT_SET_REGCLASS_BIT_FOLD;
-
- fold = (AT_LEAST_UNI_SEMANTICS) ? PL_fold_latin1[value]
- : PL_fold[value];
-
- /* It assumes the bit for 'value' has already been set */
- if (fold != value && ! ANYOF_BITMAP_TEST(node, fold)) {
- ANYOF_BITMAP_SET(node, fold);
- stored++;
- }
- if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(value) && (! isASCII(value) || ! MORE_ASCII_RESTRICTED)) {
- /* Certain Latin1 characters have matches outside the bitmap. To get
- * here, 'value' is one of those characters. None of these matches is
- * valid for ASCII characters under /aa, which have been excluded by
- * the 'if' above. The matches fall into three categories:
- * 1) They are singly folded-to or -from an above 255 character, as
- * LATIN SMALL LETTER Y WITH DIAERESIS and LATIN CAPITAL LETTER Y
- * WITH DIAERESIS;
- * 2) They are part of a multi-char fold with another character in the
- * bitmap, only LATIN SMALL LETTER SHARP S => "ss" fits that bill;
- * 3) They are part of a multi-char fold with a character not in the
- * bitmap, such as various ligatures.
- * We aren't dealing fully with multi-char folds, except we do deal
- * with the pattern containing a character that has a multi-char fold
- * (not so much the inverse).
- * For types 1) and 3), the matches only happen when the target string
- * is utf8; that's not true for 2), and we set a flag for it.
- *
- * The code below adds to the passed in inversion list the single fold
- * closures for 'value'. The values are hard-coded here so that an
- * innocent-looking character class, like /[ks]/i won't have to go out
- * to disk to find the possible matches. XXX It would be better to
- * generate these via regen, in case a new version of the Unicode
- * standard adds new mappings, though that is not really likely. */
- switch (value) {
- case 'k':
- case 'K':
- /* KELVIN SIGN */
- *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x212A);
- break;
- case 's':
- case 'S':
- /* LATIN SMALL LETTER LONG S */
- *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x017F);
- break;
- case MICRO_SIGN:
- *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
- GREEK_SMALL_LETTER_MU);
- *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
- GREEK_CAPITAL_LETTER_MU);
- break;
- case LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE:
- case LATIN_SMALL_LETTER_A_WITH_RING_ABOVE:
- /* ANGSTROM SIGN */
- *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x212B);
- if (DEPENDS_SEMANTICS) { /* See DEPENDS comment below */
- *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
- PL_fold_latin1[value]);
- }
- break;
- case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
- *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
- LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS);
- break;
- case LATIN_SMALL_LETTER_SHARP_S:
- *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
- LATIN_CAPITAL_LETTER_SHARP_S);
-
- /* Under /a, /d, and /u, this can match the two chars "ss" */
- if (! MORE_ASCII_RESTRICTED) {
- add_alternate(alternate_ptr, (U8 *) "ss", 2);
-
- /* And under /u or /a, it can match even if the target is
- * not utf8 */
- if (AT_LEAST_UNI_SEMANTICS) {
- ANYOF_FLAGS(node) |= ANYOF_NONBITMAP_NON_UTF8;
- }
- }
- break;
- case 'F': case 'f':
- case 'I': case 'i':
- case 'L': case 'l':
- case 'T': case 't':
- case 'A': case 'a':
- case 'H': case 'h':
- case 'J': case 'j':
- case 'N': case 'n':
- case 'W': case 'w':
- case 'Y': case 'y':
- /* These all are targets of multi-character folds from code
- * points that require UTF8 to express, so they can't match
- * unless the target string is in UTF-8, so no action here is
- * necessary, as regexec.c properly handles the general case
- * for UTF-8 matching */
- break;
- default:
- /* Use deprecated warning to increase the chances of this
- * being output */
- ckWARN2regdep(RExC_parse, "Perl folding rules are not up-to-date for 0x%x; please use the perlbug utility to report;", value);
- break;
- }
- }
- else if (DEPENDS_SEMANTICS
- && ! isASCII(value)
- && PL_fold_latin1[value] != value)
- {
- /* Under DEPENDS rules, non-ASCII Latin1 characters match their
- * folds only when the target string is in UTF-8. We add the fold
- * here to the list of things to match outside the bitmap, which
- * won't be looked at unless it is UTF8 (or else if something else
- * says to look even if not utf8, but those things better not happen
- * under DEPENDS semantics. */
- *invlist_ptr = add_cp_to_invlist(*invlist_ptr, PL_fold_latin1[value]);
- }
-
- return stored;
-}
-
-
-PERL_STATIC_INLINE U8
-S_set_regclass_bit(pTHX_ RExC_state_t *pRExC_state, regnode* node, const U8 value, SV** invlist_ptr, AV** alternate_ptr)
-{
- /* This inline function sets a bit in the bitmap if not already set, and if
- * appropriate, its fold, returning the number of bits that actually
- * changed from 0 to 1 */
-
- U8 stored;
-
- PERL_ARGS_ASSERT_SET_REGCLASS_BIT;
-
- if (ANYOF_BITMAP_TEST(node, value)) { /* Already set */
- return 0;
- }
-
- ANYOF_BITMAP_SET(node, value);
- stored = 1;
-
- if (FOLD && ! LOC) { /* Locale folds aren't known until runtime */
- stored += set_regclass_bit_fold(pRExC_state, node, value, invlist_ptr, alternate_ptr);
- }
-
- return stored;
-}
-
STATIC void
S_add_alternate(pTHX_ AV** alternate_ptr, U8* string, STRLEN len)
{
Optimizations may be possible if this is tiny */
UV n;
+ /* Certain named classes have equivalents that can appear outside a
+ * character class, e.g. \w. These flags are set for these classes. The
+ * first flag indicates the op depends on the character set modifier, like
+ * /d, /u.... The second is for those that don't have this dependency. */
+ bool has_special_charset_op = FALSE;
+ bool has_special_non_charset_op = FALSE;
+
/* Unicode properties are stored in a swash; this holds the current one
* being parsed. If this swash is the only above-latin1 component of the
* character class, an optimization is to pass it directly on to the
* on to the engine */
UV has_user_defined_property = 0;
- /* code points this node matches that can't be stored in the bitmap */
- SV* nonbitmap = NULL;
-
/* inversion list of code points this node matches only when the target
* string is in UTF-8. (Because is under /d) */
SV* depends_list = NULL;
- /* The items that are to match that aren't stored in the bitmap, but are a
- * result of things that are stored there. This is the fold closure of
- * such a character, either because it has DEPENDS semantics and shouldn't
- * be matched unless the target string is utf8, or is a code point that is
- * too large for the bit map, as for example, the fold of the MICRO SIGN is
- * above 255. This all is solely for performance reasons. By having this
- * code know the outside-the-bitmap folds that the bitmapped characters are
- * involved with, we don't have to go out to disk to find the list of
- * matches, unless the character class includes code points that aren't
- * storable in the bit map. That means that a character class with an 's'
- * in it, for example, doesn't need to go out to disk to find everything
- * that matches. A 2nd list is used so that the 'nonbitmap' list is kept
- * empty unless there is something whose fold we don't know about, and will
- * have to go out to the disk to find. */
- SV* l1_fold_invlist = NULL;
+ /* inversion list of code points this node matches. For much of the
+ * function, it includes only those that match regardless of the utf8ness
+ * of the target string */
+ SV* cp_list = NULL;
/* List of multi-character folds that are matched by this node */
AV* unicode_alternate = NULL;
#ifdef EBCDIC
+ /* In a range, counts how many 0-2 of the ends of it came from literals,
+ * not escapes. Thus we can tell if 'A' was input vs \x{C1} */
UV literal_endpoint = 0;
#endif
UV stored = 0; /* how many chars stored in the bitmap */
ckWARN4reg(RExC_parse,
"False [] range \"%*.*s\"",
w, w, rangebegin);
-
- nonbitmap = add_cp_to_invlist(nonbitmap, '-');
- nonbitmap = add_cp_to_invlist(nonbitmap, prevvalue);
+ cp_list = add_cp_to_invlist(cp_list, '-');
+ cp_list = add_cp_to_invlist(cp_list, prevvalue);
}
range = 0; /* this was not a true range */
+ element_count += 2; /* So counts for three values */
}
- if (!SIZE_ONLY) {
+ if (SIZE_ONLY) {
+
+ /* In the first pass, do a little extra work so below can
+ * possibly optimize the whole node to one of the nodes that
+ * correspond to the classes given below */
+
+ /* The optimization will only take place if there is a single
+ * element in the class, so can skip if there is more than one
+ */
+ if (element_count == 1) {
/* Possible truncation here but in some 64-bit environments
* the compiler gets heartburn about switch on 64-bit values.
* A similar issue a little earlier when switching on value.
* --jhi */
+ switch ((I32)namedclass) {
+ case ANYOF_ALNUM:
+ case ANYOF_NALNUM:
+ case ANYOF_DIGIT:
+ case ANYOF_NDIGIT:
+ case ANYOF_SPACE:
+ case ANYOF_NSPACE:
+ has_special_charset_op = TRUE;
+ break;
+
+ case ANYOF_HORIZWS:
+ case ANYOF_NHORIZWS:
+ case ANYOF_VERTWS:
+ case ANYOF_NVERTWS:
+ has_special_non_charset_op = TRUE;
+ break;
+ }
+ }
+ }
+ else {
switch ((I32)namedclass) {
case ANYOF_ALNUMC: /* C's alnum, in contrast to \w */
* them */
DO_POSIX_LATIN1_ONLY_KNOWN_L1_RESOLVED(ret, namedclass, properties,
PL_PosixDigit, "XPosixDigit", listsv);
+ has_special_charset_op = TRUE;
break;
case ANYOF_NDIGIT:
DO_N_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, properties,
PL_PosixDigit, PL_PosixDigit, "XPosixDigit", listsv);
+ has_special_charset_op = TRUE;
break;
case ANYOF_GRAPH:
DO_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, properties,
PL_PosixGraph, PL_L1PosixGraph, "XPosixGraph", listsv);
break;
case ANYOF_HORIZWS:
- /* For these, we use the nonbitmap, as /d doesn't make a
+ /* For these, we use the cp_list, as /d doesn't make a
* difference in what these match. There would be problems
* if these characters had folds other than themselves, as
- * nonbitmap is subject to folding. It turns out that \h
+ * cp_list is subject to folding. It turns out that \h
* is just a synonym for XPosixBlank */
- _invlist_union(nonbitmap, PL_XPosixBlank, &nonbitmap);
+ _invlist_union(cp_list, PL_XPosixBlank, &cp_list);
+ has_special_non_charset_op = TRUE;
break;
case ANYOF_NHORIZWS:
- _invlist_union_complement_2nd(nonbitmap,
- PL_XPosixBlank, &nonbitmap);
+ _invlist_union_complement_2nd(cp_list,
+ PL_XPosixBlank, &cp_list);
+ has_special_non_charset_op = TRUE;
break;
case ANYOF_LOWER:
case ANYOF_NLOWER:
case ANYOF_SPACE:
DO_POSIX(ret, namedclass, properties,
PL_PerlSpace, PL_XPerlSpace);
+ has_special_charset_op = TRUE;
break;
case ANYOF_NSPACE:
DO_N_POSIX(ret, namedclass, properties,
PL_PerlSpace, PL_XPerlSpace);
+ has_special_charset_op = TRUE;
break;
case ANYOF_UPPER: /* Same as LOWER, above */
case ANYOF_NUPPER:
case ANYOF_ALNUM: /* Really is 'Word' */
DO_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, properties,
PL_PosixWord, PL_L1PosixWord, "XPosixWord", listsv);
+ has_special_charset_op = TRUE;
break;
case ANYOF_NALNUM:
DO_N_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, properties,
PL_PosixWord, PL_L1PosixWord, "XPosixWord", listsv);
+ has_special_charset_op = TRUE;
break;
case ANYOF_VERTWS:
- /* For these, we use the nonbitmap, as /d doesn't make a
+ /* For these, we use the cp_list, as /d doesn't make a
* difference in what these match. There would be problems
* if these characters had folds other than themselves, as
- * nonbitmap is subject to folding */
- _invlist_union(nonbitmap, PL_VertSpace, &nonbitmap);
+ * cp_list is subject to folding */
+ _invlist_union(cp_list, PL_VertSpace, &cp_list);
+ has_special_non_charset_op = TRUE;
break;
case ANYOF_NVERTWS:
- _invlist_union_complement_2nd(nonbitmap,
- PL_VertSpace, &nonbitmap);
+ _invlist_union_complement_2nd(cp_list,
+ PL_VertSpace, &cp_list);
+ has_special_non_charset_op = TRUE;
break;
case ANYOF_XDIGIT:
DO_POSIX(ret, namedclass, properties,
"False [] range \"%*.*s\"",
w, w, rangebegin);
}
- if (!SIZE_ONLY)
- nonbitmap = add_cp_to_invlist(nonbitmap, '-');
+ if (!SIZE_ONLY)
+ cp_list = add_cp_to_invlist(cp_list, '-');
} else
range = 1; /* yeah, it's a range! */
continue; /* but do it the next time */
/* now is the next time */
if (!SIZE_ONLY) {
#ifndef EBCDIC
- nonbitmap = _add_range_to_invlist(nonbitmap, prevvalue, value);
+ cp_list = _add_range_to_invlist(cp_list, prevvalue, value);
#else
UV* this_range = _new_invlist(1);
_append_range_to_invlist(this_range, prevvalue, value);
{
_invlist_intersection(this_range, PL_ASCII, &this_range, );
_invlist_intersection(this_range, PL_Alpha, &this_range, );
-
}
- _invlist_union(nonbitmap, this_range, &nonbitmap);
+ _invlist_union(cp_list, this_range, &cp_list);
literal_endpoint = 0;
#endif
}
range = 0; /* this range (if it was one) is done now */
}
+ /* [\w] can be optimized into \w, but not if there is anything else in the
+ * brackets (except for an initial '^' which indictes omplementing). We
+ * also can optimize the common special case /[0-9]/ into /\d/a */
+ if (element_count == 1 &&
+ (has_special_charset_op
+ || has_special_non_charset_op
+ || (prevvalue == '0' && value == '9')))
+ {
+ U8 op;
+ bool invert = ANYOF_FLAGS(ret) & ANYOF_INVERT;
+ const char * cur_parse = RExC_parse;
+
+ if (has_special_charset_op) {
+ U8 offset = get_regex_charset(RExC_flags);
+
+ /* /aa is the same as /a for these */
+ if (offset == REGEX_ASCII_MORE_RESTRICTED_CHARSET) {
+ offset = REGEX_ASCII_RESTRICTED_CHARSET;
+ }
+ switch ((I32)namedclass) {
+ case ANYOF_NALNUM:
+ invert = ! invert;
+ /* FALLTHROUGH */
+ case ANYOF_ALNUM:
+ op = ALNUM;
+ break;
+ case ANYOF_NSPACE:
+ invert = ! invert;
+ /* FALLTHROUGH */
+ case ANYOF_SPACE:
+ op = SPACE;
+ break;
+ case ANYOF_NDIGIT:
+ invert = ! invert;
+ /* FALLTHROUGH */
+ case ANYOF_DIGIT:
+ op = DIGIT;
+
+ /* There is no DIGITU */
+ if (offset == REGEX_UNICODE_CHARSET) {
+ offset = REGEX_DEPENDS_CHARSET;
+ }
+ break;
+ default:
+ Perl_croak(aTHX_ "panic: Named character class %"IVdf" is not expected to have a non-[...] version", namedclass);
+ }
+
+ /* The number of varieties of each of these is the same, hence, so
+ * is the delta between the normal and complemented nodes */
+ if (invert) {
+ offset += NALNUM - ALNUM;
+ }
+
+ op += offset;
+ }
+ else if (has_special_non_charset_op) {
+ switch ((I32)namedclass) {
+ case ANYOF_NHORIZWS:
+ invert = ! invert;
+ /* FALLTHROUGH */
+ case ANYOF_HORIZWS:
+ op = HORIZWS;
+ break;
+ case ANYOF_NVERTWS:
+ invert = ! invert;
+ /* FALLTHROUGH */
+ case ANYOF_VERTWS:
+ op = VERTWS;
+ break;
+ default:
+ Perl_croak(aTHX_ "panic: Named character class %"IVdf" is not expected to have a non-[...] version", namedclass);
+ }
+
+ /* The complement version of each of these nodes is adjacently next
+ * */
+ if (invert) {
+ op++;
+ }
+ }
+ else { /* The remaining possibility is [0-9] */
+ op = (invert) ? NDIGITA : DIGITA;
+ }
+
+ /* Throw away this ANYOF regnode, and emit the calculated one, which
+ * should correspond to the beginning, not current, state of the parse
+ */
+ RExC_parse = (char *)orig_parse;
+ RExC_emit = (regnode *)orig_emit;
+ ret = reg_node(pRExC_state, op);
+ RExC_parse = (char *) cur_parse;
+ SvREFCNT_dec(listsv);
+ return ret;
+ }
if (SIZE_ONLY)
return ret;
/* If folding, we calculate all characters that could fold to or from the
* ones already on the list */
- if (FOLD && nonbitmap) {
+ if (FOLD && cp_list) {
UV start, end; /* End points of code point ranges */
SV* fold_intersection = NULL;
- const UV highest_index = invlist_len(nonbitmap) - 1;
+ const UV highest_index = invlist_len(cp_list) - 1;
/* In the Latin1 range, the characters that can be folded-to or -from
* are precisely the alphabetic characters. If the highest code point
* Otherwise, it starts a range that isn't in the set, so the max is
* one less than it */
if (! ELEMENT_RANGE_MATCHES_INVLIST(highest_index)
- && invlist_array(nonbitmap)[highest_index] <= 256)
+ && invlist_array(cp_list)[highest_index] <= 256)
{
- _invlist_intersection(PL_L1PosixAlpha, nonbitmap, &fold_intersection);
+ _invlist_intersection(PL_L1PosixAlpha, cp_list, &fold_intersection);
}
else {
+ /* This is a list of all the characters that participate in folds
+ * (except marks, etc in multi-char folds */
+ if (! PL_utf8_foldable) {
+ SV* swash = swash_init("utf8", "Cased", &PL_sv_undef, 1, 0);
+ PL_utf8_foldable = _swash_to_invlist(swash);
+ SvREFCNT_dec(swash);
+ }
- /* This is a list of all the characters that participate in folds
- * (except marks, etc in multi-char folds */
- if (! PL_utf8_foldable) {
- SV* swash = swash_init("utf8", "Cased", &PL_sv_undef, 1, 0);
- PL_utf8_foldable = _swash_to_invlist(swash);
- SvREFCNT_dec(swash);
- }
-
- /* This is a hash that for a particular fold gives all characters
- * that are involved in it */
- if (! PL_utf8_foldclosures) {
-
- /* If we were unable to find any folds, then we likely won't be
- * able to find the closures. So just create an empty list.
- * Folding will effectively be restricted to the non-Unicode rules
- * hard-coded into Perl. (This case happens legitimately during
- * compilation of Perl itself before the Unicode tables are
- * generated) */
- if (invlist_len(PL_utf8_foldable) == 0) {
- PL_utf8_foldclosures = newHV();
- } else {
- /* If the folds haven't been read in, call a fold function
- * to force that */
- if (! PL_utf8_tofold) {
- U8 dummy[UTF8_MAXBYTES+1];
- STRLEN dummy_len;
-
- /* This particular string is above \xff in both UTF-8 and
- * UTFEBCDIC */
- to_utf8_fold((U8*) "\xC8\x80", dummy, &dummy_len);
- assert(PL_utf8_tofold); /* Verify that worked */
- }
- PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold);
- }
- }
+ /* This is a hash that for a particular fold gives all characters
+ * that are involved in it */
+ if (! PL_utf8_foldclosures) {
+
+ /* If we were unable to find any folds, then we likely won't be
+ * able to find the closures. So just create an empty list.
+ * Folding will effectively be restricted to the non-Unicode
+ * rules hard-coded into Perl. (This case happens legitimately
+ * during compilation of Perl itself before the Unicode tables
+ * are generated) */
+ if (invlist_len(PL_utf8_foldable) == 0) {
+ PL_utf8_foldclosures = newHV();
+ }
+ else {
+ /* If the folds haven't been read in, call a fold function
+ * to force that */
+ if (! PL_utf8_tofold) {
+ U8 dummy[UTF8_MAXBYTES+1];
+ STRLEN dummy_len;
+
+ /* This particular string is above \xff in both UTF-8
+ * and UTFEBCDIC */
+ to_utf8_fold((U8*) "\xC8\x80", dummy, &dummy_len);
+ assert(PL_utf8_tofold); /* Verify that worked */
+ }
+ PL_utf8_foldclosures =
+ _swash_inversion_hash(PL_utf8_tofold);
+ }
+ }
- /* Only the characters in this class that participate in folds need be
- * checked. Get the intersection of this class and all the possible
- * characters that are foldable. This can quickly narrow down a large
- * class */
- _invlist_intersection(PL_utf8_foldable, nonbitmap, &fold_intersection);
+ /* Only the characters in this class that participate in folds need
+ * be checked. Get the intersection of this class and all the
+ * possible characters that are foldable. This can quickly narrow
+ * down a large class */
+ _invlist_intersection(PL_utf8_foldable, cp_list,
+ &fold_intersection);
}
/* Now look at the foldable characters in this class individually */
/* ASCII is always matched; non-ASCII is matched only
* under Unicode rules */
if (isASCII(j) || AT_LEAST_UNI_SEMANTICS) {
- nonbitmap =
- add_cp_to_invlist(nonbitmap, PL_fold_latin1[j]);
+ cp_list =
+ add_cp_to_invlist(cp_list, PL_fold_latin1[j]);
}
else {
depends_list =
case 'k':
case 'K':
/* KELVIN SIGN */
- nonbitmap =
- add_cp_to_invlist(nonbitmap, 0x212A);
+ cp_list =
+ add_cp_to_invlist(cp_list, 0x212A);
break;
case 's':
case 'S':
/* LATIN SMALL LETTER LONG S */
- nonbitmap =
- add_cp_to_invlist(nonbitmap, 0x017F);
+ cp_list =
+ add_cp_to_invlist(cp_list, 0x017F);
break;
case MICRO_SIGN:
- nonbitmap = add_cp_to_invlist(nonbitmap,
+ cp_list = add_cp_to_invlist(cp_list,
GREEK_SMALL_LETTER_MU);
- nonbitmap = add_cp_to_invlist(nonbitmap,
+ cp_list = add_cp_to_invlist(cp_list,
GREEK_CAPITAL_LETTER_MU);
break;
case LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE:
case LATIN_SMALL_LETTER_A_WITH_RING_ABOVE:
/* ANGSTROM SIGN */
- nonbitmap =
- add_cp_to_invlist(nonbitmap, 0x212B);
+ cp_list =
+ add_cp_to_invlist(cp_list, 0x212B);
break;
case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
- nonbitmap = add_cp_to_invlist(nonbitmap,
+ cp_list = add_cp_to_invlist(cp_list,
LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS);
break;
case LATIN_SMALL_LETTER_SHARP_S:
- nonbitmap = add_cp_to_invlist(nonbitmap,
+ cp_list = add_cp_to_invlist(cp_list,
LATIN_CAPITAL_LETTER_SHARP_S);
/* Under /a, /d, and /u, this can match the two
/* Here is an above Latin1 character. We don't have the rules
* hard-coded for it. First, get its fold */
f = _to_uni_fold_flags(j, foldbuf, &foldlen,
- ((allow_full_fold) ? FOLD_FLAGS_FULL : 0);
+ ((allow_full_fold) ? FOLD_FLAGS_FULL : 0)
+ | ((LOC)
+ ? FOLD_FLAGS_LOCALE
+ : (MORE_ASCII_RESTRICTED)
+ ? FOLD_FLAGS_NOMIX_ASCII
+ : 0));
if (foldlen > (STRLEN)UNISKIP(f)) {
/* If any of the folded characters of this are in the
* Latin1 range, tell the regex engine that this can
* match a non-utf8 target string. */
- while (loc < e) {
-
- /* Can't mix ascii with non- under /aa */
- if (MORE_ASCII_RESTRICTED
- && (isASCII(*loc) != isASCII(j)))
- {
- goto end_multi_fold;
- }
- if (UTF8_IS_INVARIANT(*loc)
- || UTF8_IS_DOWNGRADEABLE_START(*loc))
- {
- /* Can't mix above and below 256 under LOC
- */
- if (LOC) {
- goto end_multi_fold;
- }
- ANYOF_FLAGS(ret)
- |= ANYOF_NONBITMAP_NON_UTF8;
- break;
- }
- loc += UTF8SKIP(loc);
- }
+ while (loc < e) {
+ if (UTF8_IS_INVARIANT(*loc)
+ || UTF8_IS_DOWNGRADEABLE_START(*loc))
+ {
+ ANYOF_FLAGS(ret)
+ |= ANYOF_NONBITMAP_NON_UTF8;
+ break;
+ }
+ loc += UTF8SKIP(loc);
+ }
add_alternate(&unicode_alternate, foldbuf, foldlen);
- end_multi_fold: ;
}
}
else {
/* /aa doesn't allow folds between ASCII and non-;
* /l doesn't allow them between above and below
* 256 */
- if ((MORE_ASCII_RESTRICTED
- && (isASCII(c) != isASCII(j)))
- || (LOC && ((c < 256) != (j < 256))))
+ if ((MORE_ASCII_RESTRICTED && (isASCII(c) != isASCII(j)))
+ || (LOC && ((c < 256) != (j < 256))))
{
continue;
}
/* Folds involving non-ascii Latin1 characters
* under /d are added to a separate list */
- if (isASCII(c) || c > 255 || AT_LEAST_UNI_SEMANTICS) {
- nonbitmap = add_cp_to_invlist(nonbitmap, c);
+ if (isASCII(c) || c > 255 || AT_LEAST_UNI_SEMANTICS)
+ {
+ cp_list = add_cp_to_invlist(cp_list, c);
}
else {
depends_list = add_cp_to_invlist(depends_list, c);
}
}
}
- }
+ }
}
SvREFCNT_dec(fold_intersection);
}
- /* Combine the two lists into one. */
- if (l1_fold_invlist) {
- if (nonbitmap) {
- _invlist_union(nonbitmap, l1_fold_invlist, &nonbitmap);
- SvREFCNT_dec(l1_fold_invlist);
- }
- else {
- nonbitmap = l1_fold_invlist;
- }
- }
-
/* And combine the result (if any) with any inversion list from properties.
* The lists are kept separate up to now because we don't want to fold the
* properties */
if (properties) {
if (AT_LEAST_UNI_SEMANTICS) {
- if (nonbitmap) {
- _invlist_union(nonbitmap, properties, &nonbitmap);
- SvREFCNT_dec(properties);
- }
- else {
- nonbitmap = properties;
- }
+ if (cp_list) {
+ _invlist_union(cp_list, properties, &cp_list);
+ SvREFCNT_dec(properties);
+ }
+ else {
+ cp_list = properties;
+ }
}
else {
/* Under /d, we put the things that match only when the target
* string is utf8, into a separate list */
SV* nonascii_but_latin1_properties = NULL;
- _invlist_intersection(properties, PL_Latin1, &nonascii_but_latin1_properties);
- _invlist_subtract(nonascii_but_latin1_properties, PL_ASCII, &nonascii_but_latin1_properties);
- _invlist_subtract(properties, nonascii_but_latin1_properties, &properties);
- if (nonbitmap) {
- _invlist_union(nonbitmap, properties, &nonbitmap);
+ _invlist_intersection(properties, PL_Latin1,
+ &nonascii_but_latin1_properties);
+ _invlist_subtract(nonascii_but_latin1_properties, PL_ASCII,
+ &nonascii_but_latin1_properties);
+ _invlist_subtract(properties, nonascii_but_latin1_properties,
+ &properties);
+ if (cp_list) {
+ _invlist_union(cp_list, properties, &cp_list);
SvREFCNT_dec(properties);
}
else {
- nonbitmap = properties;
+ cp_list = properties;
}
if (depends_list) {
}
}
- /* Here, <nonbitmap> contains all the code points we can determine at
+ /* Here, we have calculated what code points should be in the character
+ * class.
+ *
+ * Now we can see about various optimizations. Fold calculation (which we
+ * did above) needs to take place before inversion. Otherwise /[^k]/i
+ * would invert to include K, which under /i would match k, which it
+ * shouldn't. */
+
+ /* Optimize inverted simple patterns (e.g. [^a-z]). Note that we haven't
+ * set the FOLD flag yet, so this does optimize those. It doesn't
+ * optimize locale. Doing so perhaps could be done as long as there is
+ * nothing like \w in it; some thought also would have to be given to the
+ * interaction with above 0x100 chars */
+ if ((ANYOF_FLAGS(ret) & ANYOF_INVERT)
+ && ! LOC
+ && ! depends_list
+ && ! unicode_alternate
+ && SvCUR(listsv) == initial_listsv_len)
+ {
+ _invlist_invert(cp_list);
+
+ /* Any swash can't be used as-is, because we've inverted things */
+ if (swash) {
+ SvREFCNT_dec(swash);
+ swash = NULL;
+ }
+
+ /* Clear the invert flag since have just done it here */
+ ANYOF_FLAGS(ret) &= ~ANYOF_INVERT;
+ }
+
+ /* Here, <cp_list> contains all the code points we can determine at
* compile time that match under all conditions. Go through it, and
* for things that belong in the bitmap, put them there, and delete from
- * <nonbitmap> */
- if (nonbitmap) {
+ * <cp_list> */
+ if (cp_list) {
/* This gets set if we actually need to modify things */
bool change_invlist = FALSE;
UV start, end;
- /* Start looking through <nonbitmap> */
- invlist_iterinit(nonbitmap);
- while (invlist_iternext(nonbitmap, &start, &end)) {
+ /* Start looking through <cp_list> */
+ invlist_iterinit(cp_list);
+ while (invlist_iternext(cp_list, &start, &end)) {
UV high;
int i;
}
/* Done with loop; remove any code points that are in the bitmap from
- * <nonbitmap> */
+ * <cp_list> */
if (change_invlist) {
- _invlist_subtract(nonbitmap, PL_Latin1, &nonbitmap);
+ _invlist_subtract(cp_list, PL_Latin1, &cp_list);
}
/* If have completely emptied it, remove it completely */
- if (invlist_len(nonbitmap) == 0) {
- SvREFCNT_dec(nonbitmap);
- nonbitmap = NULL;
+ if (invlist_len(cp_list) == 0) {
+ SvREFCNT_dec(cp_list);
+ cp_list = NULL;
}
}
/* Combine the two lists into one. */
if (depends_list) {
- if (nonbitmap) {
- _invlist_union(nonbitmap, depends_list, &nonbitmap);
+ if (cp_list) {
+ _invlist_union(cp_list, depends_list, &cp_list);
SvREFCNT_dec(depends_list);
}
else {
- nonbitmap = depends_list;
- }
- }
-
- /* Here, we have calculated what code points should be in the character
- * class. <nonbitmap> does not overlap the bitmap except possibly in the
- * case of DEPENDS rules.
- *
- * Now we can see about various optimizations. Fold calculation (which we
- * did above) needs to take place before inversion. Otherwise /[^k]/i
- * would invert to include K, which under /i would match k, which it
- * shouldn't. */
-
- /* Optimize inverted simple patterns (e.g. [^a-z]). Note that we haven't
- * set the FOLD flag yet, so this does optimize those. It doesn't
- * optimize locale. Doing so perhaps could be done as long as there is
- * nothing like \w in it; some thought also would have to be given to the
- * interaction with above 0x100 chars */
- if ((ANYOF_FLAGS(ret) & ANYOF_INVERT)
- && ! LOC
- && ! unicode_alternate
- /* In case of /d, there are some things that should match only when in
- * not in the bitmap, i.e., they require UTF8 to match. These are
- * listed in nonbitmap, but if ANYOF_NONBITMAP_NON_UTF8 is set in this
- * case, they don't require UTF8, so can invert here */
- && (! nonbitmap
- || ! DEPENDS_SEMANTICS
- || (ANYOF_FLAGS(ret) & ANYOF_NONBITMAP_NON_UTF8))
- && SvCUR(listsv) == initial_listsv_len)
- {
- int i;
- if (! nonbitmap) {
- for (i = 0; i < 256; ++i) {
- if (ANYOF_BITMAP_TEST(ret, i)) {
- ANYOF_BITMAP_CLEAR(ret, i);
- }
- else {
- ANYOF_BITMAP_SET(ret, i);
- prevvalue = value;
- value = i;
- }
- }
- /* The inversion means that everything above 255 is matched */
- ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL;
- }
- else {
- /* Here, also has things outside the bitmap that may overlap with
- * the bitmap. We have to sync them up, so that they get inverted
- * in both places. Earlier, we removed all overlaps except in the
- * case of /d rules, so no syncing is needed except for this case
- */
- SV *remove_list = NULL;
-
- if (DEPENDS_SEMANTICS) {
- UV start, end;
-
- /* Set the bits that correspond to the ones that aren't in the
- * bitmap. Otherwise, when we invert, we'll miss these.
- * Earlier, we removed from the nonbitmap all code points
- * < 128, so there is no extra work here */
- invlist_iterinit(nonbitmap);
- while (invlist_iternext(nonbitmap, &start, &end)) {
- if (start > 255) { /* The bit map goes to 255 */
- break;
- }
- if (end > 255) {
- end = 255;
- }
- for (i = start; i <= (int) end; ++i) {
- ANYOF_BITMAP_SET(ret, i);
- prevvalue = value;
- value = i;
- }
- }
- }
-
- /* Now invert both the bitmap and the nonbitmap. Anything in the
- * bitmap has to also be removed from the non-bitmap, but again,
- * there should not be overlap unless is /d rules. */
- _invlist_invert(nonbitmap);
-
- /* Any swash can't be used as-is, because we've inverted things */
- if (swash) {
- SvREFCNT_dec(swash);
- swash = NULL;
- }
-
- for (i = 0; i < 256; ++i) {
- if (ANYOF_BITMAP_TEST(ret, i)) {
- ANYOF_BITMAP_CLEAR(ret, i);
- if (DEPENDS_SEMANTICS) {
- if (! remove_list) {
- remove_list = _new_invlist(2);
- }
- remove_list = add_cp_to_invlist(remove_list, i);
- }
- }
- else {
- ANYOF_BITMAP_SET(ret, i);
- prevvalue = value;
- value = i;
- }
- }
-
- /* And do the removal */
- if (DEPENDS_SEMANTICS) {
- if (remove_list) {
- _invlist_subtract(nonbitmap, remove_list, &nonbitmap);
- SvREFCNT_dec(remove_list);
- }
- }
- else {
- /* There is no overlap for non-/d, so just delete anything
- * below 256 */
- _invlist_intersection(nonbitmap, PL_AboveLatin1, &nonbitmap);
- }
+ cp_list = depends_list;
}
-
- stored = 256 - stored;
-
- /* Clear the invert flag since have just done it here */
- ANYOF_FLAGS(ret) &= ~ANYOF_INVERT;
}
/* Folding in the bitmap is taken care of above, but not for locale (for
* run-time fold flag for these */
if (FOLD && (LOC
|| (DEPENDS_SEMANTICS
- && nonbitmap
+ && cp_list
&& ! (ANYOF_FLAGS(ret) & ANYOF_NONBITMAP_NON_UTF8))
|| unicode_alternate))
{
* characters which only have the two folds; so things like 'fF' and 'Ii'
* wouldn't work because they are part of the fold of 'LATIN SMALL LIGATURE
* FI'. */
- if (! nonbitmap
+ if (! cp_list
&& ! unicode_alternate
&& SvCUR(listsv) == initial_listsv_len
&& ! (ANYOF_FLAGS(ret) & (ANYOF_INVERT|ANYOF_UNICODE_ALL))
SvREFCNT_dec(swash);
swash = NULL;
}
- if (! nonbitmap
+ if (! cp_list
&& SvCUR(listsv) == initial_listsv_len
&& ! unicode_alternate)
{
* swash is stored there now.
* av[2] stores the multicharacter foldings, used later in
* regexec.c:S_reginclass().
- * av[3] stores the nonbitmap inversion list for use in addition or
+ * av[3] stores the cp_list inversion list for use in addition or
* instead of av[0]; not used if av[1] isn't NULL
* av[4] is set if any component of the class is from a user-defined
* property; not used if av[1] isn't NULL */
: listsv);
if (swash) {
av_store(av, 1, swash);
- SvREFCNT_dec(nonbitmap);
+ SvREFCNT_dec(cp_list);
}
else {
av_store(av, 1, NULL);
- if (nonbitmap) {
- av_store(av, 3, nonbitmap);
+ if (cp_list) {
+ av_store(av, 3, cp_list);
av_store(av, 4, newSVuv(has_user_defined_property));
}
}
https://perl5.git.perl.org / perl5.git / blobdiff