I32 sawback; /* Did we see \1, ...? */
U32 seen;
SSize_t size; /* Code size. */
- I32 npar; /* Capture buffer count, (OPEN) plus
+ I32 npar; /* Capture buffer count, (OPEN) plus
one. ("par" 0 is the whole
pattern)*/
I32 nestroot; /* root parens we are in - used by
bool seen_unfolded_sharp_s;
bool strict;
bool study_started;
+ bool in_script_run;
};
#define RExC_flags (pRExC_state->flags)
#define RExC_strict (pRExC_state->strict)
#define RExC_study_started (pRExC_state->study_started)
#define RExC_warn_text (pRExC_state->warn_text)
+#define RExC_in_script_run (pRExC_state->in_script_run)
/* Heuristic check on the complexity of the pattern: if TOO_NAUGHTY, we set
* a flag to disable back-off on the fixed/floating substrings - if it's
/* Change from /d into /u rules, and restart the parse if we've already seen
* something whose size would increase as a result, by setting *flagp and
* returning 'restart_retval'. RExC_uni_semantics is a flag that indicates
- * we've change to /u during the parse. */
+ * we've changed to /u during the parse. */
#define REQUIRE_UNI_RULES(flagp, restart_retval) \
STMT_START { \
if (DEPENDS_SEMANTICS) { \
} \
} STMT_END
+/* Executes a return statement with the value 'X', if 'flags' contains any of
+ * 'RESTART_PASS1', 'NEED_UTF8', or 'extra'. If so, *flagp is set to those
+ * flags */
+#define RETURN_X_ON_RESTART_OR_FLAGS(X, flags, flagp, extra) \
+ STMT_START { \
+ if ((flags) & (RESTART_PASS1|NEED_UTF8|(extra))) { \
+ *(flagp) = (flags) & (RESTART_PASS1|NEED_UTF8|(extra)); \
+ return X; \
+ } \
+ } STMT_END
+
+#define RETURN_NULL_ON_RESTART_OR_FLAGS(flags,flagp,extra) \
+ RETURN_X_ON_RESTART_OR_FLAGS(NULL,flags,flagp,extra)
+
+#define RETURN_X_ON_RESTART(X, flags,flagp) \
+ RETURN_X_ON_RESTART_OR_FLAGS( X, flags, flagp, 0)
+
+
+#define RETURN_NULL_ON_RESTART_FLAGP_OR_FLAGS(flagp,extra) \
+ if (*(flagp) & (RESTART_PASS1|(extra))) return NULL
+
+#define MUST_RESTART(flags) ((flags) & (RESTART_PASS1))
+
+#define RETURN_NULL_ON_RESTART(flags,flagp) \
+ RETURN_X_ON_RESTART(NULL, flags,flagp)
+#define RETURN_NULL_ON_RESTART_FLAGP(flagp) \
+ RETURN_NULL_ON_RESTART_FLAGP_OR_FLAGS(flagp,0)
+
/* This converts the named class defined in regcomp.h to its equivalent class
* number defined in handy.h. */
#define namedclass_to_classnum(class) ((int) ((class) / 2))
UTF8fARG(UTF, \
(xI(xC) > eC) /* Don't run off end */ \
? eC - sC /* Length before the <--HERE */ \
- : ( __ASSERT_(xI_offset(xC) >= 0) xI_offset(xC) ), \
+ : ((xI_offset(xC) >= 0) \
+ ? xI_offset(xC) \
+ : (Perl_croak(aTHX_ "panic: %s: %d: negative offset: %" \
+ IVdf " trying to output message for " \
+ " pattern %.*s", \
+ __FILE__, __LINE__, xI_offset(xC), \
+ ((int) (eC - sC)), sC), 0)), \
sC), /* The input pattern printed up to the <--HERE */ \
UTF8fARG(UTF, \
(xI(xC) > eC) ? 0 : eC - xI(xC), /* Length after <--HERE */ \
switch (flags) {
case EXACT: case EXACTL: break;
- case EXACTFA:
+ case EXACTFAA:
case EXACTFU_SS:
case EXACTFU:
case EXACTFLU8: folder = PL_fold_latin1; break;
* XXX khw thinks this should be enhanced to fill EXACT (at least) nodes as full
* as possible, even if that means splitting an existing node so that its first
* part is moved to the preceeding node. This would maximise the efficiency of
- * memEQ during matching. Elsewhere in this file, khw proposes splitting
- * EXACTFish nodes into portions that don't change under folding vs those that
- * do. Those portions that don't change may be the only things in the pattern that
- * could be used to find fixed and floating strings.
+ * memEQ during matching.
*
* If a node is to match under /i (folded), the number of characters it matches
* can be different than its character length if it contains a multi-character
* input nodes.
*
* And *unfolded_multi_char is set to indicate whether or not the node contains
- * an unfolded multi-char fold. This happens when whether the fold is valid or
- * not won't be known until runtime; namely for EXACTF nodes that contain LATIN
- * SMALL LETTER SHARP S, as only if the target string being matched against
- * turns out to be UTF-8 is that fold valid; and also for EXACTFL nodes whose
- * folding rules depend on the locale in force at runtime. (Multi-char folds
- * whose components are all above the Latin1 range are not run-time locale
- * dependent, and have already been folded by the time this function is
- * called.)
+ * an unfolded multi-char fold. This happens when it won't be known until
+ * runtime whether the fold is valid or not; namely
+ * 1) for EXACTF nodes that contain LATIN SMALL LETTER SHARP S, as only if the
+ * target string being matched against turns out to be UTF-8 is that fold
+ * valid; or
+ * 2) for EXACTFL nodes whose folding rules depend on the locale in force at
+ * runtime.
+ * (Multi-char folds whose components are all above the Latin1 range are not
+ * run-time locale dependent, and have already been folded by the time this
+ * function is called.)
*
* This is as good a place as any to discuss the design of handling these
* multi-character fold sequences. It's been wrong in Perl for a very long
* described in the next item.
* 3) A problem remains for unfolded multi-char folds. (These occur when the
* validity of the fold won't be known until runtime, and so must remain
- * unfolded for now. This happens for the sharp s in EXACTF and EXACTFA
+ * unfolded for now. This happens for the sharp s in EXACTF and EXACTFAA
* nodes when the pattern isn't in UTF-8. (Note, BTW, that there cannot
* be an EXACTF node with a UTF-8 pattern.) They also occur for various
* folds in EXACTFL nodes, regardless of the UTF-ness of the pattern.)
* character in the target string. (And I do mean character, and not byte
* here, unlike other parts of the documentation that have never been
* updated to account for multibyte Unicode.) sharp s in EXACTF and
- * EXACTFL nodes can match the two character string 'ss'; in EXACTFA nodes
- * it can match "\x{17F}\x{17F}". These, along with other ones in EXACTFL
- * nodes, violate the assumption, and they are the only instances where it
- * is violated. I'm reluctant to try to change the assumption, as the
- * code involved is impenetrable to me (khw), so instead the code here
- * punts. This routine examines EXACTFL nodes, and (when the pattern
- * isn't UTF-8) EXACTF and EXACTFA for such unfolded folds, and returns a
+ * EXACTFL nodes can match the two character string 'ss'; in EXACTFAA
+ * nodes it can match "\x{17F}\x{17F}". These, along with other ones in
+ * EXACTFL nodes, violate the assumption, and they are the only instances
+ * where it is violated. I'm reluctant to try to change the assumption,
+ * as the code involved is impenetrable to me (khw), so instead the code
+ * here punts. This routine examines EXACTFL nodes, and (when the pattern
+ * isn't UTF-8) EXACTF and EXACTFAA for such unfolded folds, and returns a
* boolean indicating whether or not the node contains such a fold. When
* it is true, the caller sets a flag that later causes the optimizer in
* this file to not set values for the floating and fixed string lengths,
* and thus avoids the optimizer code in regexec.c that makes the invalid
* assumption. Thus, there is no optimization based on string lengths for
* EXACTFL nodes that contain these few folds, nor for non-UTF8-pattern
- * EXACTF and EXACTFA nodes that contain the sharp s. (The reason the
+ * EXACTF and EXACTFAA nodes that contain the sharp s. (The reason the
* assumption is wrong only in these cases is that all other non-UTF-8
* folds are 1-1; and, for UTF-8 patterns, we pre-fold all other folds to
* their expanded versions. (Again, we can't prefold sharp s to 'ss' in
* EXACTF nodes because we don't know at compile time if it actually
* matches 'ss' or not. For EXACTF nodes it will match iff the target
* string is in UTF-8. This is in contrast to EXACTFU nodes, where it
- * always matches; and EXACTFA where it never does. In an EXACTFA node in
- * a UTF-8 pattern, sharp s is folded to "\x{17F}\x{17F}, avoiding the
+ * always matches; and EXACTFAA where it never does. In an EXACTFAA node
+ * in a UTF-8 pattern, sharp s is folded to "\x{17F}\x{17F}, avoiding the
* problem; but in a non-UTF8 pattern, folding it to that above-Latin1
* string would require the pattern to be forced into UTF-8, the overhead
* of which we want to avoid. Similarly the unfolded multi-char folds in
*
* Similarly, the code that generates tries doesn't currently handle
* not-already-folded multi-char folds, and it looks like a pain to change
- * that. Therefore, trie generation of EXACTFA nodes with the sharp s
- * doesn't work. Instead, such an EXACTFA is turned into a new regnode,
- * EXACTFA_NO_TRIE, which the trie code knows not to handle. Most people
+ * that. Therefore, trie generation of EXACTFAA nodes with the sharp s
+ * doesn't work. Instead, such an EXACTFAA is turned into a new regnode,
+ * EXACTFAA_NO_TRIE, which the trie code knows not to handle. Most people
* using /iaa matching will be doing so almost entirely with ASCII
* strings, so this should rarely be encountered in practice */
}
/* Nodes with 'ss' require special handling, except for
- * EXACTFA-ish for which there is no multi-char fold to this */
+ * EXACTFAA-ish for which there is no multi-char fold to this */
if (len == 2 && *s == 's' && *(s+1) == 's'
- && OP(scan) != EXACTFA
- && OP(scan) != EXACTFA_NO_TRIE)
+ && OP(scan) != EXACTFAA
+ && OP(scan) != EXACTFAA_NO_TRIE)
{
count = 2;
if (OP(scan) != EXACTFL) {
/* Count how many characters are in it. In the case of
* /aa, no folds which contain ASCII code points are
* allowed, so check for those, and skip if found. */
- if (OP(scan) != EXACTFA && OP(scan) != EXACTFA_NO_TRIE) {
+ if (OP(scan) != EXACTFAA && OP(scan) != EXACTFAA_NO_TRIE) {
count = utf8_length(s, multi_end);
s = multi_end;
}
*min_subtract += total_count_delta;
Safefree(folded);
}
- else if (OP(scan) == EXACTFA) {
+ else if (OP(scan) == EXACTFAA) {
- /* Non-UTF-8 pattern, EXACTFA node. There can't be a multi-char
+ /* Non-UTF-8 pattern, EXACTFAA node. There can't be a multi-char
* fold to the ASCII range (and there are no existing ones in the
* upper latin1 range). But, as outlined in the comments preceding
* this function, we need to flag any occurrences of the sharp s.
|| UNICODE_DOT_DOT_VERSION > 0)
while (s < s_end) {
if (*s == LATIN_SMALL_LETTER_SHARP_S) {
- OP(scan) = EXACTFA_NO_TRIE;
+ OP(scan) = EXACTFAA_NO_TRIE;
*unfolded_multi_char = TRUE;
break;
}
}
else {
- /* Non-UTF-8 pattern, not EXACTFA node. Look for the multi-char
+ /* Non-UTF-8 pattern, not EXACTFAA node. Look for the multi-char
* folds that are all Latin1. As explained in the comments
* preceding this function, we look also for the sharp s in EXACTF
* and EXACTFL nodes; it can be in the final position. Otherwise
} while (f);
}
-
+/* the return from this sub is the minimum length that could possibly match */
STATIC SSize_t
S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
SSize_t *minlenp, SSize_t *deltap,
EXACT | EXACT
EXACTFU | EXACTFU
EXACTFU_SS | EXACTFU
- EXACTFA | EXACTFA
+ EXACTFAA | EXACTFAA
EXACTL | EXACTL
EXACTFLU8 | EXACTFLU8
? EXACT \
: ( EXACTFU == (X) || EXACTFU_SS == (X) ) \
? EXACTFU \
- : ( EXACTFA == (X) ) \
- ? EXACTFA \
+ : ( EXACTFAA == (X) ) \
+ ? EXACTFAA \
: ( EXACTL == (X) ) \
? EXACTL \
: ( EXACTFLU8 == (X) ) \
/* Cannot expect anything... */
scan_commit(pRExC_state, data, minlenp, is_inf);
data->pos_min += 1;
- data->pos_delta += 1;
+ if (data->pos_delta != SSize_t_MAX) {
+ data->pos_delta += 1;
+ }
data->cur_is_floating = 1; /* float */
}
}
(regnode_charclass *) scan);
break;
+ case ANYOFM:
+ {
+ SV* cp_list = get_ANYOFM_contents(scan);
+
+ if (flags & SCF_DO_STCLASS_OR) {
+ ssc_union(data->start_class,
+ cp_list,
+ FALSE /* don't invert */
+ );
+ }
+ else if (flags & SCF_DO_STCLASS_AND) {
+ ssc_intersection(data->start_class,
+ cp_list,
+ FALSE /* don't invert */
+ );
+ }
+
+ SvREFCNT_dec_NN(cp_list);
+ break;
+ }
+
case NPOSIXL:
invert = 1;
/* FALLTHROUGH */
}
break;
+ case NASCII:
+ invert = 1;
+ /* FALLTHROUGH */
+ case ASCII:
+ my_invlist = invlist_clone(PL_Posix_ptrs[_CC_ASCII]);
+
+ /* This can be handled as a Posix class */
+ goto join_posix_and_ascii;
+
case NPOSIXA: /* For these, we always know the exact set of
what's matched */
invert = 1;
/* FALLTHROUGH */
case POSIXA:
- if (FLAGS(scan) == _CC_ASCII) {
- my_invlist = invlist_clone(PL_XPosix_ptrs[_CC_ASCII]);
- }
- else {
- _invlist_intersection(PL_XPosix_ptrs[FLAGS(scan)],
- PL_XPosix_ptrs[_CC_ASCII],
- &my_invlist);
- }
- goto join_posix;
+ assert(FLAGS(scan) != _CC_ASCII);
+ my_invlist = invlist_clone(PL_Posix_ptrs[FLAGS(scan)]);
+ goto join_posix_and_ascii;
case NPOSIXD:
case NPOSIXU:
&my_invlist);
}
- join_posix:
+ join_posix_and_ascii:
if (flags & SCF_DO_STCLASS_AND) {
ssc_intersection(data->start_class, my_invlist, invert);
data->cur_is_floating = 1; /* float */
}
min += min1;
- if (delta != SSize_t_MAX)
- delta += max1 - min1;
+ if (delta != SSize_t_MAX) {
+ if (SSize_t_MAX - (max1 - min1) >= delta)
+ delta += max1 - min1;
+ else
+ delta = SSize_t_MAX;
+ }
if (flags & SCF_DO_STCLASS_OR) {
ssc_or(pRExC_state, data->start_class, (regnode_charclass *) &accum);
if (min1) {
* it is properly null terminated or we will fail asserts
* later. In theory we probably shouldn't get such SV's,
* but if we do we should handle it gracefully. */
- if ( SvTYPE(msv) != SVt_PV || (SvLEN(msv) > SvCUR(msv) && *(SvEND(msv)) == 0) ) {
+ if ( SvTYPE(msv) != SVt_PV || (SvLEN(msv) > SvCUR(msv) && *(SvEND(msv)) == 0) || SvIsCOW_shared_hash(msv) ) {
/* not a string, or a string with a trailing null */
pat = msv;
} else {
/* a string with no trailing null, we need to copy it
- * so it we have a trailing null */
- pat = newSVsv(msv);
+ * so it has a trailing null */
+ pat = sv_2mortal(newSVsv(msv));
}
}
/* Initialize these here instead of as-needed, as is quick and avoids
* having to test them each time otherwise */
- if (! PL_AboveLatin1) {
+ if (! PL_InBitmap) {
#ifdef DEBUGGING
char * dump_len_string;
#endif
- PL_AboveLatin1 = _new_invlist_C_array(AboveLatin1_invlist);
- PL_Latin1 = _new_invlist_C_array(Latin1_invlist);
- PL_UpperLatin1 = _new_invlist_C_array(UpperLatin1_invlist);
- PL_utf8_foldable = _new_invlist_C_array(_Perl_Any_Folds_invlist);
- PL_HasMultiCharFold =
- _new_invlist_C_array(_Perl_Folds_To_Multi_Char_invlist);
-
/* This is calculated here, because the Perl program that generates the
* static global ones doesn't currently have access to
* NUM_ANYOF_CODE_POINTS */
RExC_seen_unfolded_sharp_s = 0;
RExC_contains_locale = 0;
RExC_strict = cBOOL(pm_flags & RXf_PMf_STRICT);
+ RExC_in_script_run = 0;
RExC_study_started = 0;
pRExC_state->runtime_code_qr = NULL;
RExC_frame_head= NULL;
at least some part of the pattern, and therefore must convert the whole
thing.
-- dmq */
- if (flags & RESTART_PASS1) {
+ if (MUST_RESTART(flags)) {
if (flags & NEED_UTF8) {
S_pat_upgrade_to_utf8(aTHX_ pRExC_state, &exp, &plen,
pRExC_state->code_blocks ? pRExC_state->code_blocks->count : 0);
+ DEBUG_PARSE_r(Perl_re_printf( aTHX_ "Need to redo pass 1 after upgrade\n"));
}
else {
- DEBUG_PARSE_r(Perl_re_printf( aTHX_
- "Need to redo pass 1\n"));
+ DEBUG_PARSE_r(Perl_re_printf( aTHX_ "Need to redo pass 1\n"));
}
goto redo_first_pass;
PERL_UNUSED_CONTEXT;
PERL_ARGS_ASSERT_INVLIST_SET_LEN;
- assert(SvTYPE(invlist) == SVt_INVLIST);
+ assert(is_invlist(invlist));
SvCUR_set(invlist,
(len == 0)
PERL_ARGS_ASSERT_INVLIST_REPLACE_LIST_DESTROYS_SRC;
- assert(SvTYPE(src) == SVt_INVLIST);
- assert(SvTYPE(dest) == SVt_INVLIST);
+ assert(is_invlist(src));
+ assert(is_invlist(dest));
assert(! invlist_is_iterating(src));
assert(SvCUR(src) == 0 || SvCUR(src) < SvLEN(src));
* */
PERL_ARGS_ASSERT_GET_INVLIST_PREVIOUS_INDEX_ADDR;
- assert(SvTYPE(invlist) == SVt_INVLIST);
+ assert(is_invlist(invlist));
return &(((XINVLIST*) SvANY(invlist))->prev_index);
}
PERL_ARGS_ASSERT_INVLIST_TRIM;
- assert(SvTYPE(invlist) == SVt_INVLIST);
+ assert(is_invlist(invlist));
SvPV_renew(invlist, MAX(min_size, SvCUR(invlist) + 1));
}
{
PERL_ARGS_ASSERT_INVLIST_CLEAR;
- assert(SvTYPE(invlist) == SVt_INVLIST);
+ assert(is_invlist(invlist));
invlist_set_len(invlist, 0, 0);
invlist_trim(invlist);
PERL_ARGS_ASSERT_INVLIST_MAX;
- assert(SvTYPE(invlist) == SVt_INVLIST);
+ assert(is_invlist(invlist));
/* Assumes worst case, in which the 0 element is not counted in the
* inversion list, so subtracts 1 for that */
invlist_set_len(new_list, 0, 0);
/* Force iterinit() to be used to get iteration to work */
- *get_invlist_iter_addr(new_list) = (STRLEN) UV_MAX;
+ invlist_iterfinish(new_list);
*get_invlist_previous_index_addr(new_list) = 0;
PERL_ARGS_ASSERT_INVLIST_EXTEND;
- assert(SvTYPE(invlist) == SVt_INVLIST);
+ assert(is_invlist(invlist));
/* Add one to account for the zero element at the beginning which may not
* be counted by the calling parameters */
PERL_ARGS_ASSERT__INVLIST_UNION_MAYBE_COMPLEMENT_2ND;
assert(a != b);
- assert(*output == NULL || SvTYPE(*output) == SVt_INVLIST);
+ assert(*output == NULL || is_invlist(*output));
len_b = _invlist_len(b);
if (len_b == 0) {
PERL_ARGS_ASSERT__INVLIST_INTERSECTION_MAYBE_COMPLEMENT_2ND;
assert(a != b);
- assert(*i == NULL || SvTYPE(*i) == SVt_INVLIST);
+ assert(*i == NULL || is_invlist(*i));
/* Special case if either one is empty */
len_a = (a == NULL) ? 0 : _invlist_len(a);
PERL_ARGS_ASSERT_GET_INVLIST_ITER_ADDR;
- assert(SvTYPE(invlist) == SVt_INVLIST);
+ assert(is_invlist(invlist));
return &(((XINVLIST*) SvANY(invlist))->iterator);
}
}
}
-void
-Perl__load_PL_utf8_foldclosures (pTHX)
-{
- assert(! PL_utf8_foldclosures);
-
- /* If the folds haven't been read in, call a fold function
- * to force that */
- if (! PL_utf8_tofold) {
- U8 dummy[UTF8_MAXBYTES_CASE+1];
- const U8 hyphen[] = HYPHEN_UTF8;
-
- /* This string is just a short named one above \xff */
- toFOLD_utf8_safe(hyphen, hyphen + sizeof(hyphen) - 1, dummy, NULL);
- assert(PL_utf8_tofold); /* Verify that worked */
- }
- PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold);
-}
#endif
#if defined(PERL_ARGS_ASSERT__INVLISTEQ) && !defined(PERL_IN_XSUB_RE)
/* Some characters match above-Latin1 ones under /i. This
* is true of EXACTFL ones when the locale is UTF-8 */
if (HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(uc)
- && (! isASCII(uc) || (OP(node) != EXACTFA
- && OP(node) != EXACTFA_NO_TRIE)))
+ && (! isASCII(uc) || (OP(node) != EXACTFAA
+ && OP(node) != EXACTFAA_NO_TRIE)))
{
add_above_Latin1_folds(pRExC_state, (U8) uc, &invlist);
}
}
else { /* Pattern is UTF-8 */
U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' };
- STRLEN foldlen = UTF8SKIP(s);
const U8* e = s + bytelen;
- SV** listp;
+ IV fc;
- uc = utf8_to_uvchr_buf(s, s + bytelen, NULL);
+ fc = uc = utf8_to_uvchr_buf(s, s + bytelen, NULL);
/* The only code points that aren't folded in a UTF EXACTFish
* node are are the problematic ones in EXACTFL nodes */
U8 *d = folded;
int i;
+ fc = -1;
for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < e; i++) {
if (isASCII(*s)) {
*(d++) = (U8) toFOLD(*s);
+ if (fc < 0) { /* Save the first fold */
+ fc = *(d-1);
+ }
s++;
}
else {
STRLEN len;
- toFOLD_utf8_safe(s, e, d, &len);
+ UV fold = toFOLD_utf8_safe(s, e, d, &len);
+ if (fc < 0) { /* Save the first fold */
+ fc = fold;
+ }
d += len;
s += UTF8SKIP(s);
}
/* And set up so the code below that looks in this folded
* buffer instead of the node's string */
e = d;
- foldlen = UTF8SKIP(folded);
s = folded;
}
/* When we reach here 's' points to the fold of the first
* character(s) of the node; and 'e' points to far enough along
* the folded string to be just past any possible multi-char
- * fold. 'foldlen' is the length in bytes of the first
- * character in 's'
+ * fold.
*
* Unlike the non-UTF-8 case, the macro for determining if a
* string is a multi-char fold requires all the characters to
invlist = _add_range_to_invlist(invlist, 0, UV_MAX);
}
else { /* Single char fold */
-
- /* It matches all the things that fold to it, which are
- * found in PL_utf8_foldclosures (including itself) */
- invlist = add_cp_to_invlist(invlist, uc);
- if (! PL_utf8_foldclosures)
- _load_PL_utf8_foldclosures();
- if ((listp = hv_fetch(PL_utf8_foldclosures,
- (char *) s, foldlen, FALSE)))
- {
- AV* list = (AV*) *listp;
- IV k;
- for (k = 0; k <= av_tindex_skip_len_mg(list); k++) {
- SV** c_p = av_fetch(list, k, FALSE);
- UV c;
- assert(c_p);
-
- c = SvUV(*c_p);
-
- /* /aa doesn't allow folds between ASCII and non- */
- if ((OP(node) == EXACTFA || OP(node) == EXACTFA_NO_TRIE)
- && isASCII(c) != isASCII(uc))
- {
- continue;
- }
-
- invlist = add_cp_to_invlist(invlist, c);
+ unsigned int k;
+ unsigned int first_folds_to;
+ const unsigned int * remaining_folds_to_list;
+ Size_t folds_to_count;
+
+ /* It matches itself */
+ invlist = add_cp_to_invlist(invlist, fc);
+
+ /* ... plus all the things that fold to it, which are found in
+ * PL_utf8_foldclosures */
+ folds_to_count = _inverse_folds(fc, &first_folds_to,
+ &remaining_folds_to_list);
+ for (k = 0; k < folds_to_count; k++) {
+ UV c = (k == 0) ? first_folds_to : remaining_folds_to_list[k-1];
+
+ /* /aa doesn't allow folds between ASCII and non- */
+ if ( (OP(node) == EXACTFAA || OP(node) == EXACTFAA_NO_TRIE)
+ && isASCII(c) != isASCII(fc))
+ {
+ continue;
}
+
+ invlist = add_cp_to_invlist(invlist, c);
}
}
}
* RExC_parse beyond the '('. Things like '(?' are indivisible tokens, and
* this flag alerts us to the need to check for that */
{
- regnode *ret; /* Will be the head of the group. */
+ regnode *ret = NULL; /* Will be the head of the group. */
regnode *br;
regnode *lastbr;
regnode *ender = NULL;
* here (if paren ==2). The forms '(*VERB' and '(?...' disallow such
* intervening space, as the sequence is a token, and a token should be
* indivisible */
- bool has_intervening_patws = paren == 2 && *(RExC_parse - 1) != '(';
+ bool has_intervening_patws = (paren == 2)
+ && *(RExC_parse - 1) != '(';
if (RExC_parse >= RExC_end) {
vFAIL("Unmatched (");
}
- if ( *RExC_parse == '*') { /* (*VERB:ARG) */
+ if (paren == 'r') { /* Atomic script run */
+ paren = '>';
+ goto parse_rest;
+ }
+ else if ( *RExC_parse == '*') { /* (*VERB:ARG), (*construct:...) */
char *start_verb = RExC_parse + 1;
STRLEN verb_len;
char *start_arg = NULL;
unsigned char op = 0;
int arg_required = 0;
int internal_argval = -1; /* if >-1 we are not allowed an argument*/
+ bool has_upper = FALSE;
if (has_intervening_patws) {
RExC_parse++; /* past the '*' */
- vFAIL("In '(*VERB...)', the '(' and '*' must be adjacent");
+
+ /* For strict backwards compatibility, don't change the message
+ * now that we also have lowercase operands */
+ if (isUPPER(*RExC_parse)) {
+ vFAIL("In '(*VERB...)', the '(' and '*' must be adjacent");
+ }
+ else {
+ vFAIL("In '(*...)', the '(' and '*' must be adjacent");
+ }
}
while (RExC_parse < RExC_end && *RExC_parse != ')' ) {
if ( *RExC_parse == ':' ) {
start_arg = RExC_parse + 1;
break;
}
- RExC_parse += UTF ? UTF8SKIP(RExC_parse) : 1;
+ else if (! UTF) {
+ if (isUPPER(*RExC_parse)) {
+ has_upper = TRUE;
+ }
+ RExC_parse++;
+ }
+ else {
+ RExC_parse += UTF8SKIP(RExC_parse);
+ }
}
verb_len = RExC_parse - start_verb;
if ( start_arg ) {
if (RExC_parse >= RExC_end) {
goto unterminated_verb_pattern;
}
+
RExC_parse += UTF ? UTF8SKIP(RExC_parse) : 1;
- while ( RExC_parse < RExC_end && *RExC_parse != ')' )
+ while ( RExC_parse < RExC_end && *RExC_parse != ')' ) {
RExC_parse += UTF ? UTF8SKIP(RExC_parse) : 1;
- if ( RExC_parse >= RExC_end || *RExC_parse != ')' )
+ }
+ if ( RExC_parse >= RExC_end || *RExC_parse != ')' ) {
unterminated_verb_pattern:
- vFAIL("Unterminated verb pattern argument");
- if ( RExC_parse == start_arg )
- start_arg = NULL;
+ if (has_upper) {
+ vFAIL("Unterminated verb pattern argument");
+ }
+ else {
+ vFAIL("Unterminated '(*...' argument");
+ }
+ }
} else {
- if ( RExC_parse >= RExC_end || *RExC_parse != ')' )
- vFAIL("Unterminated verb pattern");
+ if ( RExC_parse >= RExC_end || *RExC_parse != ')' ) {
+ if (has_upper) {
+ vFAIL("Unterminated verb pattern");
+ }
+ else {
+ vFAIL("Unterminated '(*...' construct");
+ }
+ }
}
/* Here, we know that RExC_parse < RExC_end */
RExC_seen |= REG_CUTGROUP_SEEN;
}
break;
- }
+ case 'a':
+ if ( memEQs(start_verb, verb_len, "asr")
+ || memEQs(start_verb, verb_len, "atomic_script_run"))
+ {
+ paren = 'r'; /* Mnemonic: recursed run */
+ goto script_run;
+ }
+ else if (memEQs(start_verb, verb_len, "atomic")) {
+ paren = 't'; /* AtOMIC */
+ goto alpha_assertions;
+ }
+ break;
+ case 'p':
+ if ( memEQs(start_verb, verb_len, "plb")
+ || memEQs(start_verb, verb_len, "positive_lookbehind"))
+ {
+ paren = 'b';
+ goto lookbehind_alpha_assertions;
+ }
+ else if ( memEQs(start_verb, verb_len, "pla")
+ || memEQs(start_verb, verb_len, "positive_lookahead"))
+ {
+ paren = 'a';
+ goto alpha_assertions;
+ }
+ break;
+ case 'n':
+ if ( memEQs(start_verb, verb_len, "nlb")
+ || memEQs(start_verb, verb_len, "negative_lookbehind"))
+ {
+ paren = 'B';
+ goto lookbehind_alpha_assertions;
+ }
+ else if ( memEQs(start_verb, verb_len, "nla")
+ || memEQs(start_verb, verb_len, "negative_lookahead"))
+ {
+ paren = 'A';
+ goto alpha_assertions;
+ }
+ break;
+ case 's':
+ if ( memEQs(start_verb, verb_len, "sr")
+ || memEQs(start_verb, verb_len, "script_run"))
+ {
+ regnode * atomic;
+
+ paren = 's';
+
+ script_run:
+
+ /* This indicates Unicode rules. */
+ REQUIRE_UNI_RULES(flagp, NULL);
+
+ if (! start_arg) {
+ goto no_colon;
+ }
+
+ RExC_parse = start_arg;
+
+ if (RExC_in_script_run) {
+
+ /* Nested script runs are treated as no-ops, because
+ * if the nested one fails, the outer one must as
+ * well. It could fail sooner, and avoid (??{} with
+ * side effects, but that is explicitly documented as
+ * undefined behavior. */
+
+ ret = NULL;
+
+ if (paren == 's') {
+ paren = ':';
+ goto parse_rest;
+ }
+
+ /* But, the atomic part of a nested atomic script run
+ * isn't a no-op, but can be treated just like a '(?>'
+ * */
+ paren = '>';
+ goto parse_rest;
+ }
+
+ /* By doing this here, we avoid extra warnings for nested
+ * script runs */
+ if (PASS2) {
+ Perl_ck_warner_d(aTHX_
+ packWARN(WARN_EXPERIMENTAL__SCRIPT_RUN),
+ "The script_run feature is experimental"
+ REPORT_LOCATION, REPORT_LOCATION_ARGS(RExC_parse));
+
+ }
+
+ if (paren == 's') {
+ /* Here, we're starting a new regular script run */
+ ret = reg_node(pRExC_state, SROPEN);
+ RExC_in_script_run = 1;
+ is_open = 1;
+ goto parse_rest;
+ }
+
+ /* Here, we are starting an atomic script run. This is
+ * handled by recursing to deal with the atomic portion
+ * separately, enclosed in SROPEN ... SRCLOSE nodes */
+
+ ret = reg_node(pRExC_state, SROPEN);
+
+ RExC_in_script_run = 1;
+
+ atomic = reg(pRExC_state, 'r', &flags, depth);
+ if (flags & (RESTART_PASS1|NEED_UTF8)) {
+ *flagp = flags & (RESTART_PASS1|NEED_UTF8);
+ return NULL;
+ }
+
+ REGTAIL(pRExC_state, ret, atomic);
+
+ REGTAIL(pRExC_state, atomic,
+ reg_node(pRExC_state, SRCLOSE));
+
+ RExC_in_script_run = 0;
+ return ret;
+ }
+
+ break;
+
+ lookbehind_alpha_assertions:
+ RExC_seen |= REG_LOOKBEHIND_SEEN;
+ RExC_in_lookbehind++;
+ /*FALLTHROUGH*/
+
+ alpha_assertions:
+
+ if (PASS2) {
+ Perl_ck_warner_d(aTHX_
+ packWARN(WARN_EXPERIMENTAL__ALPHA_ASSERTIONS),
+ "The alpha_assertions feature is experimental"
+ REPORT_LOCATION, REPORT_LOCATION_ARGS(RExC_parse));
+ }
+
+ RExC_seen_zerolen++;
+
+ if (! start_arg) {
+ goto no_colon;
+ }
+
+ /* An empty negative lookahead assertion simply is failure */
+ if (paren == 'A' && RExC_parse == start_arg) {
+ ret=reganode(pRExC_state, OPFAIL, 0);
+ nextchar(pRExC_state);
+ return ret;
+ }
+
+ RExC_parse = start_arg;
+ goto parse_rest;
+
+ no_colon:
+ vFAIL2utf8f(
+ "'(*%" UTF8f "' requires a terminating ':'",
+ UTF8fARG(UTF, verb_len, start_verb));
+ NOT_REACHED; /*NOTREACHED*/
+
+ } /* End of switch */
if ( ! op ) {
RExC_parse += UTF ? UTF8SKIP(RExC_parse) : 1;
- vFAIL2utf8f(
+ if (has_upper || verb_len == 0) {
+ vFAIL2utf8f(
"Unknown verb pattern '%" UTF8f "'",
UTF8fARG(UTF, verb_len, start_verb));
+ }
+ else {
+ vFAIL2utf8f(
+ "Unknown '(*...)' construct '%" UTF8f "'",
+ UTF8fARG(UTF, verb_len, start_verb));
+ }
}
+ if ( RExC_parse == start_arg ) {
+ start_arg = NULL;
+ }
if ( arg_required && !start_arg ) {
vFAIL3("Verb pattern '%.*s' has a mandatory argument",
verb_len, start_verb);
paren = 1;
goto capturing_parens;
}
+
RExC_seen |= REG_LOOKBEHIND_SEEN;
RExC_in_lookbehind++;
RExC_parse++;
RExC_parse++;
is_neg = TRUE;
}
+ endptr = RExC_end;
if (grok_atoUV(RExC_parse, &unum, &endptr)
&& unum <= I32_MAX
) {
{
int is_define= 0;
const int DEFINE_len = sizeof("DEFINE") - 1;
- if (RExC_parse[0] == '?') { /* (?(?...)) */
- if ( RExC_parse < RExC_end - 1
- && ( RExC_parse[1] == '='
- || RExC_parse[1] == '!'
- || RExC_parse[1] == '<'
- || RExC_parse[1] == '{')
- ) { /* Lookahead or eval. */
- I32 flag;
- regnode *tail;
-
- ret = reg_node(pRExC_state, LOGICAL);
- if (!SIZE_ONLY)
- ret->flags = 1;
-
- tail = reg(pRExC_state, 1, &flag, depth+1);
- if (flag & (RESTART_PASS1|NEED_UTF8)) {
- *flagp = flag & (RESTART_PASS1|NEED_UTF8);
- return NULL;
- }
- REGTAIL(pRExC_state, ret, tail);
- goto insert_if;
- }
- /* Fall through to ‘Unknown switch condition’ at the
- end of the if/else chain. */
- }
- else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
+ if ( RExC_parse < RExC_end - 1
+ && ( ( RExC_parse[0] == '?' /* (?(?...)) */
+ && ( RExC_parse[1] == '='
+ || RExC_parse[1] == '!'
+ || RExC_parse[1] == '<'
+ || RExC_parse[1] == '{'))
+ || ( RExC_parse[0] == '*' /* (?(*...)) */
+ && ( memBEGINs(RExC_parse + 1,
+ (Size_t) (RExC_end - (RExC_parse + 1)),
+ "pla:")
+ || memBEGINs(RExC_parse + 1,
+ (Size_t) (RExC_end - (RExC_parse + 1)),
+ "plb:")
+ || memBEGINs(RExC_parse + 1,
+ (Size_t) (RExC_end - (RExC_parse + 1)),
+ "nla:")
+ || memBEGINs(RExC_parse + 1,
+ (Size_t) (RExC_end - (RExC_parse + 1)),
+ "nlb:")
+ || memBEGINs(RExC_parse + 1,
+ (Size_t) (RExC_end - (RExC_parse + 1)),
+ "positive_lookahead:")
+ || memBEGINs(RExC_parse + 1,
+ (Size_t) (RExC_end - (RExC_parse + 1)),
+ "positive_lookbehind:")
+ || memBEGINs(RExC_parse + 1,
+ (Size_t) (RExC_end - (RExC_parse + 1)),
+ "negative_lookahead:")
+ || memBEGINs(RExC_parse + 1,
+ (Size_t) (RExC_end - (RExC_parse + 1)),
+ "negative_lookbehind:"))))
+ ) { /* Lookahead or eval. */
+ I32 flag;
+ regnode *tail;
+
+ ret = reg_node(pRExC_state, LOGICAL);
+ if (!SIZE_ONLY)
+ ret->flags = 1;
+
+ tail = reg(pRExC_state, 1, &flag, depth+1);
+ RETURN_NULL_ON_RESTART(flag,flagp);
+ REGTAIL(pRExC_state, ret, tail);
+ goto insert_if;
+ }
+ else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
|| RExC_parse[0] == '\'' ) /* (?('NAME')...) */
{
char ch = RExC_parse[0] == '<' ? '>' : '\'';
}
else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
UV uv;
+ endptr = RExC_end;
if (grok_atoUV(RExC_parse, &uv, &endptr)
&& uv <= I32_MAX
) {
/* (?(1)...) */
char c;
UV uv;
+ endptr = RExC_end;
if (grok_atoUV(RExC_parse, &uv, &endptr)
&& uv <= I32_MAX
) {
REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
br = regbranch(pRExC_state, &flags, 1,depth+1);
if (br == NULL) {
- if (flags & (RESTART_PASS1|NEED_UTF8)) {
- *flagp = flags & (RESTART_PASS1|NEED_UTF8);
- return NULL;
- }
+ RETURN_NULL_ON_RESTART(flags,flagp);
FAIL2("panic: regbranch returned NULL, flags=%#" UVxf,
(UV) flags);
} else
lastbr = reganode(pRExC_state, IFTHEN, 0);
if (!regbranch(pRExC_state, &flags, 1,depth+1)) {
- if (flags & (RESTART_PASS1|NEED_UTF8)) {
- *flagp = flags & (RESTART_PASS1|NEED_UTF8);
- return NULL;
- }
+ RETURN_NULL_ON_RESTART(flags,flagp);
FAIL2("panic: regbranch returned NULL, flags=%#" UVxf,
(UV) flags);
}
vFAIL("Unknown switch condition (?(...))");
}
case '[': /* (?[ ... ]) */
- return handle_regex_sets(pRExC_state, NULL, flagp, depth,
+ return handle_regex_sets(pRExC_state, NULL, flagp, depth+1,
oregcomp_parse);
case 0: /* A NUL */
RExC_parse--; /* for vFAIL to print correctly */
goto parse_rest;
} /* end switch */
}
- else if (!(RExC_flags & RXf_PMf_NOCAPTURE)) { /* (...) */
+ else {
+ if (*RExC_parse == '{' && PASS2) {
+ ckWARNregdep(RExC_parse + 1,
+ "Unescaped left brace in regex is "
+ "deprecated here (and will be fatal "
+ "in Perl 5.32), passed through");
+ }
+ /* Not bothering to indent here, as the above 'else' is temporary
+ * */
+ if (!(RExC_flags & RXf_PMf_NOCAPTURE)) { /* (...) */
capturing_parens:
parno = RExC_npar;
RExC_npar++;
paren = ':';
ret = NULL;
}
+ }
}
else /* ! paren */
ret = NULL;
/* branch_len = (paren != 0); */
if (br == NULL) {
- if (flags & (RESTART_PASS1|NEED_UTF8)) {
- *flagp = flags & (RESTART_PASS1|NEED_UTF8);
- return NULL;
- }
+ RETURN_NULL_ON_RESTART(flags,flagp);
FAIL2("panic: regbranch returned NULL, flags=%#" UVxf, (UV) flags);
}
if (*RExC_parse == '|') {
br = regbranch(pRExC_state, &flags, 0, depth+1);
if (br == NULL) {
- if (flags & (RESTART_PASS1|NEED_UTF8)) {
- *flagp = flags & (RESTART_PASS1|NEED_UTF8);
- return NULL;
- }
+ RETURN_NULL_ON_RESTART(flags,flagp);
FAIL2("panic: regbranch returned NULL, flags=%#" UVxf, (UV) flags);
}
REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
Set_Node_Offset(ender,RExC_parse+1); /* MJD */
Set_Node_Length(ender,1); /* MJD */
break;
+ case 's':
+ ender = reg_node(pRExC_state, SRCLOSE);
+ RExC_in_script_run = 0;
+ break;
case '<':
+ case 'a':
+ case 'A':
+ case 'b':
+ case 'B':
case ',':
case '=':
case '!':
*flagp &= ~HASWIDTH;
/* FALLTHROUGH */
+ case 't': /* aTomic */
case '>':
ender = reg_node(pRExC_state, SUCCEED);
break;
{
const char *p;
- static const char parens[] = "=!<,>";
+ /* Even/odd or x=don't care: 010101x10x */
+ static const char parens[] = "=!aA<,>Bbt";
+ /* flag below is set to 0 up through 'A'; 1 for larger */
if (paren && (p = strchr(parens, paren))) {
U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
- int flag = (p - parens) > 1;
+ int flag = (p - parens) > 3;
- if (paren == '>')
+ if (paren == '>' || paren == 't') {
node = SUSPEND, flag = 0;
+ }
+
reginsert(pRExC_state, node,ret, depth+1);
Set_Node_Cur_Length(ret, parse_start);
Set_Node_Offset(ret, parse_start + 1);
if (latest == NULL) {
if (flags & TRYAGAIN)
continue;
- if (flags & (RESTART_PASS1|NEED_UTF8)) {
- *flagp = flags & (RESTART_PASS1|NEED_UTF8);
- return NULL;
- }
+ RETURN_NULL_ON_RESTART(flags,flagp);
FAIL2("panic: regpiece returned NULL, flags=%#" UVxf, (UV) flags);
}
else if (ret == NULL)
ret = regatom(pRExC_state, &flags,depth+1);
if (ret == NULL) {
- if (flags & (TRYAGAIN|RESTART_PASS1|NEED_UTF8))
- *flagp |= flags & (TRYAGAIN|RESTART_PASS1|NEED_UTF8);
- else
- FAIL2("panic: regatom returned NULL, flags=%#" UVxf, (UV) flags);
- return(NULL);
+ RETURN_NULL_ON_RESTART_OR_FLAGS(flags,flagp,TRYAGAIN);
+ FAIL2("panic: regatom returned NULL, flags=%#" UVxf, (UV) flags);
}
op = *RExC_parse;
maxpos = next;
RExC_parse++;
if (isDIGIT(*RExC_parse)) {
+ endptr = RExC_end;
if (!grok_atoUV(RExC_parse, &uv, &endptr))
vFAIL("Invalid quantifier in {,}");
if (uv >= REG_INFTY)
else
maxpos = RExC_parse;
if (isDIGIT(*maxpos)) {
+ endptr = RExC_end;
if (!grok_atoUV(maxpos, &uv, &endptr))
vFAIL("Invalid quantifier in {,}");
if (uv >= REG_INFTY)
* *node_p, nor *code_point_p, nor *flagp.
*
* If <cp_count> is not NULL, the caller wants to know the length (in code
- * points) that this \N sequence matches. This is set even if the function
- * returns FALSE, as detailed below.
+ * points) that this \N sequence matches. This is set, and the input is
+ * parsed for errors, even if the function returns FALSE, as detailed below.
*
* There are 5 possibilities here, as detailed in the next 5 paragraphs.
*
*/
char * endbrace; /* points to '}' following the name */
- char *endchar; /* Points to '.' or '}' ending cur char in the input
- stream */
char* p = RExC_parse; /* Temporary */
+ SV * substitute_parse = NULL;
+ char *orig_end;
+ char *save_start;
+ I32 flags;
+ Size_t count = 0; /* code point count kept internally by this function */
+
GET_RE_DEBUG_FLAGS_DECL;
PERL_ARGS_ASSERT_GROK_BSLASH_N;
* [^\n]. The latter is assumed when the {...} following the \N is a legal
* quantifier, or there is no '{' at all */
if (*p != '{' || regcurly(p)) {
- RExC_parse = p;
+ RExC_parse = p;
if (cp_count) {
*cp_count = -1;
}
- if (! node_p) {
+ if (! node_p) {
return FALSE;
}
- *node_p = reg_node(pRExC_state, REG_ANY);
- *flagp |= HASWIDTH|SIMPLE;
- MARK_NAUGHTY(1);
+ *node_p = reg_node(pRExC_state, REG_ANY);
+ *flagp |= HASWIDTH|SIMPLE;
+ MARK_NAUGHTY(1);
Set_Node_Length(*node_p, 1); /* MJD */
- return TRUE;
+ return TRUE;
}
- /* Here, we have decided it should be a named character or sequence */
-
/* The test above made sure that the next real character is a '{', but
* under the /x modifier, it could be separated by space (or a comment and
* \n) and this is not allowed (for consistency with \x{...} and the
* tokenizer handling of \N{NAME}). */
if (*RExC_parse != '{') {
- vFAIL("Missing braces on \\N{}");
+ vFAIL("Missing braces on \\N{}");
}
- RExC_parse++; /* Skip past the '{' */
+ RExC_parse++; /* Skip past the '{' */
endbrace = (char *) memchr(RExC_parse, '}', RExC_end - RExC_parse);
if (! endbrace) { /* no trailing brace */
vFAIL2("Missing right brace on \\%c{}", 'N');
}
- else if (!( endbrace == RExC_parse /* nothing between the {} */
- || memBEGINs(RExC_parse, /* U+ (bad hex is checked below
- for a better error msg) */
- (STRLEN) (RExC_end - RExC_parse),
- "U+")))
- {
- RExC_parse = endbrace; /* position msg's '<--HERE' */
- vFAIL("\\N{NAME} must be resolved by the lexer");
- }
+ /* Here, we have decided it should be a named character or sequence */
REQUIRE_UNI_RULES(flagp, FALSE); /* Unicode named chars imply Unicode
semantics */
*cp_count = 0;
}
nextchar(pRExC_state);
- if (! node_p) {
+ if (! node_p) {
return FALSE;
}
return TRUE;
}
- RExC_parse += 2; /* Skip past the 'U+' */
+ /* If we haven't got something that begins with 'U+', then it didn't get lexed. */
+ if ( endbrace - RExC_parse < 2
+ || strnNE(RExC_parse, "U+", 2))
+ {
+ RExC_parse = endbrace; /* position msg's '<--HERE' */
+ vFAIL("\\N{NAME} must be resolved by the lexer");
+ }
- /* Because toke.c has generated a special construct for us guaranteed not
- * to have NULs, we can use a str function */
- endchar = RExC_parse + strcspn(RExC_parse, ".}");
+ /* This code purposely indented below because of future changes coming */
- /* Code points are separated by dots. If none, there is only one code
- * point, and is terminated by the brace */
+ /* We can get to here when the input is \N{U+...} or when toke.c has
+ * converted a name to the \N{U+...} form. This include changing a
+ * name that evaluates to multiple code points to \N{U+c1.c2.c3 ...} */
- if (endchar >= endbrace) {
- STRLEN length_of_hex;
- I32 grok_hex_flags;
+ RExC_parse += 2; /* Skip past the 'U+' */
- /* Here, exactly one code point. If that isn't what is wanted, fail */
- if (! code_point_p) {
- RExC_parse = p;
- return FALSE;
- }
+ /* Code points are separated by dots. The '}' terminates the whole
+ * thing. */
- /* Convert code point from hex */
- length_of_hex = (STRLEN)(endchar - RExC_parse);
- grok_hex_flags = PERL_SCAN_ALLOW_UNDERSCORES
- | PERL_SCAN_DISALLOW_PREFIX
-
- /* No errors in the first pass (See [perl
- * #122671].) We let the code below find the
- * errors when there are multiple chars. */
- | ((SIZE_ONLY)
- ? PERL_SCAN_SILENT_ILLDIGIT
- : 0);
-
- /* This routine is the one place where both single- and double-quotish
- * \N{U+xxxx} are evaluated. The value is a Unicode code point which
- * must be converted to native. */
- *code_point_p = UNI_TO_NATIVE(grok_hex(RExC_parse,
- &length_of_hex,
- &grok_hex_flags,
- NULL));
-
- /* The tokenizer should have guaranteed validity, but it's possible to
- * bypass it by using single quoting, so check. Don't do the check
- * here when there are multiple chars; we do it below anyway. */
- if (length_of_hex == 0
- || length_of_hex != (STRLEN)(endchar - RExC_parse) )
- {
- RExC_parse += length_of_hex; /* Includes all the valid */
- RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */
- ? UTF8SKIP(RExC_parse)
- : 1;
- /* Guard against malformed utf8 */
- if (RExC_parse >= endchar) {
- RExC_parse = endchar;
+ do { /* Loop until the ending brace */
+ UV cp = 0;
+ char * start_digit; /* The first of the current code point */
+ if (! isXDIGIT(*RExC_parse)) {
+ RExC_parse++;
+ vFAIL("Invalid hexadecimal number in \\N{U+...}");
}
- vFAIL("Invalid hexadecimal number in \\N{U+...}");
- }
- RExC_parse = endbrace + 1;
- return TRUE;
- }
- else { /* Is a multiple character sequence */
- SV * substitute_parse;
- STRLEN len;
- char *orig_end = RExC_end;
- char *save_start = RExC_start;
- I32 flags;
+ start_digit = RExC_parse;
+ count++;
- /* Count the code points, if desired, in the sequence */
- if (cp_count) {
- *cp_count = 0;
- while (RExC_parse < endbrace) {
- /* Point to the beginning of the next character in the sequence. */
- RExC_parse = endchar + 1;
- endchar = RExC_parse + strcspn(RExC_parse, ".}");
- (*cp_count)++;
+ /* Loop through the hex digits of the current code point */
+ do {
+ /* Adding this digit will shift the result 4 bits. If that
+ * result would be above the legal max, it's overflow */
+ if (cp > MAX_LEGAL_CP >> 4) {
+
+ /* Find the end of the code point */
+ do {
+ RExC_parse ++;
+ } while (isXDIGIT(*RExC_parse) || *RExC_parse == '_');
+
+ /* Be sure to synchronize this message with the similar one
+ * in utf8.c */
+ vFAIL4("Use of code point 0x%.*s is not allowed; the"
+ " permissible max is 0x%" UVxf,
+ (int) (RExC_parse - start_digit), start_digit,
+ MAX_LEGAL_CP);
+ }
+
+ /* Accumulate this (valid) digit into the running total */
+ cp = (cp << 4) + READ_XDIGIT(RExC_parse);
+
+ /* READ_XDIGIT advanced the input pointer. Ignore a single
+ * underscore separator */
+ if (*RExC_parse == '_' && isXDIGIT(RExC_parse[1])) {
+ RExC_parse++;
+ }
+ } while (isXDIGIT(*RExC_parse));
+
+ /* Here, have accumulated the next code point */
+ if (RExC_parse >= endbrace) { /* If done ... */
+ if (count != 1) {
+ goto do_concat;
+ }
+
+ /* Here, is a single code point; fail if doesn't want that */
+ if (! code_point_p) {
+ RExC_parse = p;
+ return FALSE;
+ }
+
+ /* A single code point is easy to handle; just return it */
+ *code_point_p = UNI_TO_NATIVE(cp);
+ RExC_parse = endbrace;
+ nextchar(pRExC_state);
+ return TRUE;
}
- }
- /* Fail if caller doesn't want to handle a multi-code-point sequence.
- * But don't backup up the pointer if the caller wants to know how many
- * code points there are (they can then handle things) */
- if (! node_p) {
- if (! cp_count) {
- RExC_parse = p;
+ /* Here, the only legal thing would be a multiple character
+ * sequence (of the form "\N{U+c1.c2. ... }". So the next
+ * character must be a dot (and the one after that can't be the
+ * endbrace, or we'd have something like \N{U+100.} ) */
+ if (*RExC_parse != '.' || RExC_parse + 1 >= endbrace) {
+ RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */
+ ? UTF8SKIP(RExC_parse)
+ : 1;
+ if (RExC_parse >= endbrace) { /* Guard against malformed utf8 */
+ RExC_parse = endbrace;
+ }
+ vFAIL("Invalid hexadecimal number in \\N{U+...}");
}
- return FALSE;
- }
- /* What is done here is to convert this to a sub-pattern of the form
- * \x{char1}\x{char2}... and then call reg recursively to parse it
- * (enclosing in "(?: ... )" ). That way, it retains its atomicness,
- * while not having to worry about special handling that some code
- * points may have. */
+ /* Here, looks like its really a multiple character sequence. Fail
+ * if that's not what the caller wants. But continue with counting
+ * and error checking if they still want a count */
+ if (! node_p && ! cp_count) {
+ return FALSE;
+ }
- substitute_parse = newSVpvs("?:");
+ /* What is done here is to convert this to a sub-pattern of the
+ * form \x{char1}\x{char2}... and then call reg recursively to
+ * parse it (enclosing in "(?: ... )" ). That way, it retains its
+ * atomicness, while not having to worry about special handling
+ * that some code points may have. We don't create a subpattern,
+ * but go through the motions of code point counting and error
+ * checking, if the caller doesn't want a node returned. */
- while (RExC_parse < endbrace) {
+ if (node_p && count == 1) {
+ substitute_parse = newSVpvs("?:");
+ }
- /* Convert to notation the rest of the code understands */
- sv_catpv(substitute_parse, "\\x{");
- sv_catpvn(substitute_parse, RExC_parse, endchar - RExC_parse);
- sv_catpv(substitute_parse, "}");
+ do_concat:
- /* Point to the beginning of the next character in the sequence. */
- RExC_parse = endchar + 1;
- endchar = RExC_parse + strcspn(RExC_parse, ".}");
+ if (node_p) {
+ /* Convert to notation the rest of the code understands */
+ sv_catpvs(substitute_parse, "\\x{");
+ sv_catpvn(substitute_parse, start_digit,
+ RExC_parse - start_digit);
+ sv_catpvs(substitute_parse, "}");
+ }
- }
- sv_catpv(substitute_parse, ")");
+ /* Move to after the dot (or ending brace the final time through.)
+ * */
+ RExC_parse++;
+ count++;
- len = SvCUR(substitute_parse);
+ } while (RExC_parse < endbrace);
- /* Don't allow empty number */
- if (len < (STRLEN) 8) {
- RExC_parse = endbrace;
- vFAIL("Invalid hexadecimal number in \\N{U+...}");
- }
+ if (! node_p) { /* Doesn't want the node */
+ assert (cp_count);
- RExC_parse = RExC_start = RExC_adjusted_start
- = SvPV_nolen(substitute_parse);
- RExC_end = RExC_parse + len;
+ *cp_count = count;
+ return FALSE;
+ }
+
+ sv_catpvs(substitute_parse, ")");
- /* The values are Unicode, and therefore not subject to recoding, but
- * have to be converted to native on a non-Unicode (meaning non-ASCII)
- * platform. */
#ifdef EBCDIC
+ /* The values are Unicode, and therefore have to be converted to native
+ * on a non-Unicode (meaning non-ASCII) platform. */
RExC_recode_x_to_native = 1;
#endif
- *node_p = reg(pRExC_state, 1, &flags, depth+1);
+ /* Here, we have the string the name evaluates to, ready to be parsed,
+ * stored in 'substitute_parse' as a series of valid "\x{...}\x{...}"
+ * constructs. This can be called from within a substitute parse already.
+ * The error reporting mechanism doesn't work for 2 levels of this, but the
+ * code above has validated this new construct, so there should be no
+ * errors generated by the below.*/
+ save_start = RExC_start;
+ orig_end = RExC_end;
- /* Restore the saved values */
- RExC_start = RExC_adjusted_start = save_start;
- RExC_parse = endbrace;
- RExC_end = orig_end;
+ RExC_parse = RExC_start = SvPVX(substitute_parse);
+ RExC_end = RExC_parse + SvCUR(substitute_parse);
+
+ *node_p = reg(pRExC_state, 1, &flags, depth+1);
+
+ /* Restore the saved values */
+ RExC_start = save_start;
+ RExC_parse = endbrace;
+ RExC_end = orig_end;
#ifdef EBCDIC
- RExC_recode_x_to_native = 0;
+ RExC_recode_x_to_native = 0;
#endif
- SvREFCNT_dec_NN(substitute_parse);
-
- if (! *node_p) {
- if (flags & (RESTART_PASS1|NEED_UTF8)) {
- *flagp = flags & (RESTART_PASS1|NEED_UTF8);
- return FALSE;
- }
- FAIL2("panic: reg returned NULL to grok_bslash_N, flags=%#" UVxf,
- (UV) flags);
- }
- *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
- nextchar(pRExC_state);
+ SvREFCNT_dec_NN(substitute_parse);
- return TRUE;
+ if (! *node_p) {
+ RETURN_X_ON_RESTART(FALSE, flags,flagp);
+ FAIL2("panic: reg returned NULL to grok_bslash_N, flags=%#" UVxf,
+ (UV) flags);
}
+ *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
+
+ nextchar(pRExC_state);
+
+ return TRUE;
}
* in which case return I32_MAX (rather than possibly 32-bit wrapping) */
static I32
-S_backref_value(char *p)
+S_backref_value(char *p, char *e)
{
- const char* endptr;
+ const char* endptr = e;
UV val;
if (grok_atoUV(p, &val, &endptr) && val <= I32_MAX)
return (I32)val;
NULL,
NULL);
if (ret == NULL) {
- if (*flagp & (RESTART_PASS1|NEED_UTF8))
- return NULL;
+ RETURN_NULL_ON_RESTART_FLAGP_OR_FLAGS(flagp,NEED_UTF8);
FAIL2("panic: regclass returned NULL to regatom, flags=%#" UVxf,
(UV) *flagp);
}
}
goto tryagain;
}
- if (flags & (RESTART_PASS1|NEED_UTF8)) {
- *flagp = flags & (RESTART_PASS1|NEED_UTF8);
- return NULL;
- }
+ RETURN_NULL_ON_RESTART(flags,flagp);
FAIL2("panic: reg returned NULL to regatom, flags=%#" UVxf,
(UV) flags);
}
TRUE, /* Allow an optimized regnode result */
NULL,
NULL);
- if (*flagp & RESTART_PASS1)
- return NULL;
+ RETURN_NULL_ON_RESTART_FLAGP(flagp);
/* regclass() can only return RESTART_PASS1 and NEED_UTF8 if
* multi-char folds are allowed. */
if (!ret)
break;
}
- if (*flagp & RESTART_PASS1)
- return NULL;
+ RETURN_NULL_ON_RESTART_FLAGP(flagp);
/* Here, evaluates to a single code point. Go get that */
RExC_parse = parse_start;
if (RExC_parse >= RExC_end) {
goto unterminated_g;
}
- num = S_backref_value(RExC_parse);
+ num = S_backref_value(RExC_parse, RExC_end);
if (num == 0)
vFAIL("Reference to invalid group 0");
else if (num == I32_MAX) {
}
}
else {
- num = S_backref_value(RExC_parse);
+ num = S_backref_value(RExC_parse, RExC_end);
/* bare \NNN might be backref or octal - if it is larger
* than or equal RExC_npar then it is assumed to be an
* octal escape. Note RExC_npar is +1 from the actual
UV ender = 0;
char *p;
char *s;
-#define MAX_NODE_STRING_SIZE 127
- char foldbuf[MAX_NODE_STRING_SIZE+UTF8_MAXBYTES_CASE];
+
+/* This allows us to fill a node with just enough spare so that if the final
+ * character folds, its expansion is guaranteed to fit */
+#define MAX_NODE_STRING_SIZE (255-UTF8_MAXBYTES_CASE)
+ char foldbuf[MAX_NODE_STRING_SIZE+UTF8_MAXBYTES_CASE+1];
+
char *s0;
U8 upper_parse = MAX_NODE_STRING_SIZE;
- U8 node_type = compute_EXACTish(pRExC_state);
+
+ /* We start out as an EXACT node, even if under /i, until we find a
+ * character which is in a fold. The algorithm now segregates into
+ * separate nodes, characters that fold from those that don't under
+ * /i. (This hopefull will create nodes that are fixed strings
+ * even under /i, giving the optimizer something to grab onto to.)
+ * So, if a node has something in it and the next character is in
+ * the opposite category, that node is closed up, and the function
+ * returns. Then regatom is called again, and a new node is
+ * created for the new category. */
+ U8 node_type = EXACT;
+
bool next_is_quantifier;
char * oldp = NULL;
* which don't participate in folds with Latin1-range characters,
* as the latter's folds aren't known until runtime. (We don't
* need to figure this out until pass 2) */
- bool maybe_exactfu = PASS2
- && (node_type == EXACTF || node_type == EXACTFL);
+ bool maybe_exactfu = PASS2;
- /* If a folding node contains only code points that don't
- * participate in folds, it can be changed into an EXACT node,
- * which allows the optimizer more things to look for */
- bool maybe_exact;
+ /* To see if RExC_uni_semantics changes during parsing of the node.
+ * */
+ bool uni_semantics_at_node_start;
+ /* The node_type may change below, but since the size of the node
+ * doesn't change, it works */
ret = reg_node(pRExC_state, node_type);
/* In pass1, folded, we use a temporary buffer instead of the
reparse:
- /* We look for the EXACTFish to EXACT node optimizaton only if
- * folding. (And we don't need to figure this out until pass 2).
- * XXX It might actually make sense to split the node into portions
- * that are exact and ones that aren't, so that we could later use
- * the exact ones to find the longest fixed and floating strings.
- * One would want to join them back into a larger node. One could
- * use a pseudo regnode like 'EXACT_ORIG_FOLD' */
- maybe_exact = FOLD && PASS2;
-
- /* XXX The node can hold up to 255 bytes, yet this only goes to
- * 127. I (khw) do not know why. Keeping it somewhat less than
- * 255 allows us to not have to worry about overflow due to
- * converting to utf8 and fold expansion, but that value is
- * 255-UTF8_MAXBYTES_CASE. join_exact() may join adjacent nodes
- * split up by this limit into a single one using the real max of
- * 255. Even at 127, this breaks under rare circumstances. If
- * folding, we do not want to split a node at a character that is a
- * non-final in a multi-char fold, as an input string could just
- * happen to want to match across the node boundary. The join
- * would solve that problem if the join actually happens. But a
- * series of more than two nodes in a row each of 127 would cause
- * the first join to succeed to get to 254, but then there wouldn't
- * be room for the next one, which could at be one of those split
- * multi-char folds. I don't know of any fool-proof solution. One
- * could back off to end with only a code point that isn't such a
- * non-final, but it is possible for there not to be any in the
- * entire node. */
-
- assert( ! UTF /* Is at the beginning of a character */
+ /* This breaks under rare circumstances. If folding, we do not
+ * want to split a node at a character that is a non-final in a
+ * multi-char fold, as an input string could just happen to want to
+ * match across the node boundary. The code at the end of the loop
+ * looks for this, and backs off until it finds not such a
+ * character, but it is possible (though extremely, extremely
+ * unlikely) for all characters in the node to be non-final fold
+ * ones, in which case we just leave the node fully filled, and
+ * hope that it doesn't match the string in just the wrong place */
+
+ assert( ! UTF /* Is at the beginning of a character */
|| UTF8_IS_INVARIANT(UCHARAT(RExC_parse))
|| UTF8_IS_START(UCHARAT(RExC_parse)));
+ uni_semantics_at_node_start = RExC_uni_semantics;
+
/* Here, we have a literal character. Find the maximal string of
* them in the input that we can fit into a single EXACTish node.
- * We quit at the first non-literal or when the node gets full */
- for (p = RExC_parse;
- len < upper_parse && p < RExC_end;
- len++)
- {
+ * We quit at the first non-literal or when the node gets full, or
+ * under /i the categorization of folding/non-folding character
+ * changes */
+ for (p = RExC_parse; len < upper_parse && p < RExC_end; ) {
+
+ /* In most cases each iteration adds one byte to the output.
+ * The exceptions override this */
+ Size_t added_len = 1;
+
oldp = p;
/* White space has already been ignored */
) {
if (*flagp & NEED_UTF8)
FAIL("panic: grok_bslash_N set NEED_UTF8");
- if (*flagp & RESTART_PASS1)
- return NULL;
+ RETURN_NULL_ON_RESTART_FLAGP(flagp);
/* Here, it wasn't a single code point. Go close
* up this EXACTish node. The switch() prior to
/* NOTE, RExC_npar is 1 more than the actual number of
* parens we have seen so far, hence the < RExC_npar below. */
- if ( !isDIGIT(p[1]) || S_backref_value(p) < RExC_npar)
+ if ( !isDIGIT(p[1]) || S_backref_value(p, RExC_end) < RExC_npar)
{ /* Not to be treated as an octal constant, go
find backref */
--p;
} /* End of switch on '\' */
break;
case '{':
- /* Currently we allow an lbrace at the start of a construct
- * without raising a warning. This is because we think we
- * will never want such a brace to be meant to be other
- * than taken literally. */
+ /* Trying to gain new uses for '{' without breaking too
+ * much existing code is hard. The solution currently
+ * adopted is:
+ * 1) If there is no ambiguity that a '{' should always
+ * be taken literally, at the start of a construct, we
+ * just do so.
+ * 2) If the literal '{' conflicts with our desired use
+ * of it as a metacharacter, we die. The deprecation
+ * cycles for this have come and gone.
+ * 3) If there is ambiguity, we raise a simple warning.
+ * This could happen, for example, if the user
+ * intended it to introduce a quantifier, but slightly
+ * misspelled the quantifier. Without this warning,
+ * the quantifier would silently be taken as a literal
+ * string of characters instead of a meta construct */
if (len || (p > RExC_start && isALPHA_A(*(p - 1)))) {
-
- /* But, we raise a fatal warning otherwise, as the
- * deprecation cycle has come and gone. Except that it
- * turns out that some heavily-relied on upstream
- * software, notably GNU Autoconf, have failed to fix
- * their uses. For these, don't make it fatal unless
- * we anticipate using the '{' for something else.
- * This happens after any alpha, and for a looser {m,n}
- * quantifier specification */
if ( RExC_strict
|| ( p > parse_start + 1
&& isALPHA_A(*(p - 1))
"illegal here");
}
if (PASS2) {
- ckWARNregdep(p + 1,
- "Unescaped left brace in regex is "
- "deprecated here (and will be fatal "
- "in Perl 5.30), passed through");
+ ckWARNreg(p + 1, "Unescaped left brace in regex is"
+ " passed through");
}
}
goto normal_default;
break;
} /* End of switch on the literal */
- /* Here, have looked at the literal character and <ender>
- * contains its ordinal
, <p> points to the character after it.
+ /* Here, have looked at the literal character, and <ender>
+ * contains its ordinal
; <p> points to the character after it.
* We need to check if the next non-ignored thing is a
* quantifier. Move <p> to after anything that should be
* ignored, which, as a side effect, positions <p> for the next
if (UTF && ! UVCHR_IS_INVARIANT(ender)) {
const STRLEN unilen = UVCHR_SKIP(ender);
s += unilen;
-
- /* We have to subtract 1 just below (and again in
- * the corresponding PASS2 code) because the loop
- * increments <len> each time, as all but this path
- * (and one other) through it add a single byte to
- * the EXACTish node. But these paths would change
- * len to be the correct final value, so cancel out
- * the increment that follows */
- len += unilen - 1;
+ added_len = unilen;
}
else {
s++;
not_fold_common:
if (UTF && ! UVCHR_IS_INVARIANT(ender)) {
U8 * new_s = uvchr_to_utf8((U8*)s, ender);
- len += (char *) new_s - s - 1;
+ added_len = (char *) new_s - s;
s = (char *) new_s;
}
else {
else if (LOC && is_PROBLEMATIC_LOCALE_FOLD_cp(ender)) {
/* Here are folding under /l, and the code point is
- * problematic. First, we know we can't simplify things */
- maybe_exact = FALSE;
- maybe_exactfu = FALSE;
-
- /* A problematic code point in this context means that its
- * fold isn't known until runtime, so we can't fold it now.
+ * problematic. If this is the first character in the
+ * node, change the node type to folding. Otherwise, if
+ * this is the first problematic character, close up the
+ * existing node, so can start a new node with this one */
+ if (! len) {
+ node_type = EXACTFL;
+ }
+ else if (node_type == EXACT) {
+ p = oldp;
+ goto loopdone;
+ }
+
+ /* This code point means we can't simplify things */
+ maybe_exactfu = FALSE;
+
+ /* A problematic code point in this context means that its
+ * fold isn't known until runtime, so we can't fold it now.
* (The non-problematic code points are the above-Latin1
* ones that fold to also all above-Latin1. Their folds
* don't vary no matter what the locale is.) But here we
* do for both passes is the PASS2 code for non-folding */
goto not_fold_common;
}
- else /* A regular FOLD code point */
- if (! ( UTF
-#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
- || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
- || UNICODE_DOT_DOT_VERSION > 0)
- /* See comments for join_exact() as to why we fold
- * this non-UTF at compile time */
- || ( node_type == EXACTFU
- && ender == LATIN_SMALL_LETTER_SHARP_S)
-#endif
- )) {
+ else /* A regular FOLD code point */
+ if (! UTF)
+ {
/* Here, are folding and are not UTF-8 encoded; therefore
- * the character must be in the range 0-255, and is not /l
+ * the character must be in the range 0-255, and is not /l.
* (Not /l because we already handled these under /l in
* is_PROBLEMATIC_LOCALE_FOLD_cp) */
- if (IS_IN_SOME_FOLD_L1(ender)) {
- maybe_exact = FALSE;
+ if (! IS_IN_SOME_FOLD_L1(ender)) {
- /* See if the character's fold differs between /d and
- * /u. This includes the multi-char fold SHARP S to
- * 'ss' */
- if (UNLIKELY(ender == LATIN_SMALL_LETTER_SHARP_S)) {
- RExC_seen_unfolded_sharp_s = 1;
- maybe_exactfu = FALSE;
+ /* Start a new node for this non-folding character if
+ * previous ones in the node were folded */
+ if (len && node_type != EXACT) {
+ p = oldp;
+ goto loopdone;
+ }
+
+ *(s++) = (char) ender;
+ }
+ else { /* Here, does participate in some fold */
+
+ /* if this is the first character in the node, change
+ * its type to folding. Otherwise, if this is the
+ * first folding character in the node, close up the
+ * existing node, so can start a new node with this
+ * one. */
+ if (! len) {
+ node_type = compute_EXACTish(pRExC_state);
+ }
+ else if (node_type == EXACT) {
+ p = oldp;
+ goto loopdone;
}
- else if (maybe_exactfu
- && (PL_fold[ender] != PL_fold_latin1[ender]
+
+ /* See if the character's fold differs between /d and
+ * /u. On non-ancient Unicode versions, this includes
+ * the multi-char fold SHARP S to 'ss' */
+
#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
|| (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
|| UNICODE_DOT_DOT_VERSION > 0)
- || ( len > 0
- && isALPHA_FOLD_EQ(ender, 's')
- && isALPHA_FOLD_EQ(*(s-1), 's'))
+
+ if (UNLIKELY(ender == LATIN_SMALL_LETTER_SHARP_S)) {
+
+ /* See comments for join_exact() as to why we fold
+ * this non-UTF at compile time */
+ if (node_type == EXACTFU) {
+ *(s++) = 's';
+
+ /* Let the code below add in the extra 's' */
+ ender = 's';
+ added_len = 2;
+ }
+ else if ( uni_semantics_at_node_start
+ != RExC_uni_semantics)
+ {
+ /* Here, we are supossed to be using Unicode
+ * rules, but this folding node is not. This
+ * happens during pass 1 when the node started
+ * out not under Unicode rules, but a \N{} was
+ * encountered during the processing of it,
+ * causing Unicode rules to be switched into.
+ * Pass 1 continues uninterrupted, as by the
+ * time we get to pass 2, we will know enough
+ * to generate the correct folds. Except in
+ * this one case, we need to restart the node,
+ * because the fold of the sharp s requires 2
+ * characters, and the sizing needs to account
+ * for that. */
+ p = oldp;
+ goto loopdone;
+ }
+ else {
+ RExC_seen_unfolded_sharp_s = 1;
+ maybe_exactfu = FALSE;
+ }
+ }
+ else if ( len
+ && isALPHA_FOLD_EQ(ender, 's')
+ && isALPHA_FOLD_EQ(*(s-1), 's'))
+ {
+ maybe_exactfu = FALSE;
+ }
+ else
#endif
- )) {
+
+ if (PL_fold[ender] != PL_fold_latin1[ender]) {
maybe_exactfu = FALSE;
}
- }
- /* Even when folding, we store just the input character, as
- * we have an array that finds its fold quickly */
- *(s++) = (char) ender;
+ /* Even when folding, we store just the input
+ * character, as we have an array that finds its fold
+ * quickly */
+ *(s++) = (char) ender;
+ }
}
- else { /* FOLD, and UTF (or sharp s) */
+ else { /* FOLD, and UTF */
/* Unlike the non-fold case, we do actually have to
- * calculate the results here in pass 1. This is for two
- * reasons, the folded length may be longer than the
- * unfolded, and we have to calculate how many EXACTish
- * nodes it will take; and we may run out of room in a node
- * in the middle of a potential multi-char fold, and have
- * to back off accordingly. */
-
- UV folded;
+ * calculate the fold in pass 1. This is for two reasons,
+ * the folded length may be longer than the unfolded, and
+ * we have to calculate how many EXACTish nodes it will
+ * take; and we may run out of room in a node in the middle
+ * of a potential multi-char fold, and have to back off
+ * accordingly. */
+
if (isASCII_uni(ender)) {
- folded = toFOLD(ender);
- *(s)++ = (U8) folded;
+
+ /* As above, we close up and start a new node if the
+ * previous characters don't match the fold/non-fold
+ * state of this one. And if this is the first
+ * character in the node, and it folds, we change the
+ * node away from being EXACT */
+ if (! IS_IN_SOME_FOLD_L1(ender)) {
+ if (len && node_type != EXACT) {
+ p = oldp;
+ goto loopdone;
+ }
+
+ *(s)++ = (U8) ender;
+ }
+ else { /* Is in a fold */
+
+ if (! len) {
+ node_type = compute_EXACTish(pRExC_state);
+ }
+ else if (node_type == EXACT) {
+ p = oldp;
+ goto loopdone;
+ }
+
+ *(s)++ = (U8) toFOLD(ender);
+ }
}
- else {
+ else { /* Not ASCII */
STRLEN foldlen;
- folded = _to_uni_fold_flags(
+ /* As above, we close up and start a new node if the
+ * previous characters don't match the fold/non-fold
+ * state of this one. And if this is the first
+ * character in the node, and it folds, we change the
+ * node away from being EXACT */
+ if (! _invlist_contains_cp(PL_utf8_foldable, ender)) {
+ if (len && node_type != EXACT) {
+ p = oldp;
+ goto loopdone;
+ }
+
+ s = (char *) uvchr_to_utf8((U8 *) s, ender);
+ added_len = UVCHR_SKIP(ender);
+ }
+ else {
+
+ if (! len) {
+ node_type = compute_EXACTish(pRExC_state);
+ }
+ else if (node_type == EXACT) {
+ p = oldp;
+ goto loopdone;
+ }
+
+ ender = _to_uni_fold_flags(
ender,
(U8 *) s,
&foldlen,
FOLD_FLAGS_FULL | ((ASCII_FOLD_RESTRICTED)
? FOLD_FLAGS_NOMIX_ASCII
: 0));
- s += foldlen;
-
- /* The loop increments <len> each time, as all but this
- * path (and one other) through it add a single byte to
- * the EXACTish node. But this one has changed len to
- * be the correct final value, so subtract one to
- * cancel out the increment that follows */
- len += foldlen - 1;
- }
- /* If this node only contains non-folding code points so
- * far, see if this new one is also non-folding */
- if (maybe_exact) {
- if (folded != ender) {
- maybe_exact = FALSE;
- }
- else {
- /* Here the fold is the original; we have to check
- * further to see if anything folds to it */
- if (_invlist_contains_cp(PL_utf8_foldable,
- ender))
- {
- maybe_exact = FALSE;
- }
+ s += foldlen;
+ added_len = foldlen;
}
}
- ender = folded;
}
+ len += added_len;
+
if (next_is_quantifier) {
/* Here, the next input is a quantifier, and to get here,
- * the current character is the only one in the node.
- * Also, here <len> doesn't include the final byte for this
- * character */
- len++;
+ * the current character is the only one in the node. */
goto loopdone;
}
len = s - s0 + 1;
}
else {
- if (! PL_NonL1NonFinalFold) {
- PL_NonL1NonFinalFold = _new_invlist_C_array(
- NonL1_Perl_Non_Final_Folds_invlist);
- }
/* Point to the first byte of the final character */
s = (char *) utf8_hop((U8 *) s, -1);
while (s >= s0) { /* Search backwards until find
- non-problematic char */
+ a non-problematic char */
if (UTF8_IS_INVARIANT(*s)) {
/* There are no ascii characters that participate
OP(ret) = NOTHING;
}
else {
- if (FOLD) {
- /* If 'maybe_exact' is still set here, means there are no
- * code points in the node that participate in folds;
- * similarly for 'maybe_exactfu' and code points that match
- * differently depending on UTF8ness of the target string
- * (for /u), or depending on locale for /l */
- if (maybe_exact) {
- OP(ret) = (LOC)
- ? EXACTL
- : EXACT;
+ OP(ret) = node_type;
+
+ /* If the node type is EXACT here, check to see if it
+ * should be EXACTL. */
+ if (node_type == EXACT) {
+ if (LOC) {
+ OP(ret) = EXACTL;
}
- else if (maybe_exactfu) {
- OP(ret) = (LOC)
- ? EXACTFLU8
- : EXACTFU;
+ }
+
+ if (FOLD) {
+ /* If 'maybe_exactfu' is set, then there are no code points
+ * that match differently depending on UTF8ness of the
+ * target string (for /u), or depending on locale for /l */
+ if (maybe_exactfu) {
+ if (node_type == EXACTF) {
+ OP(ret) = EXACTFU;
+ }
+ else if (node_type == EXACTFL) {
+ OP(ret) = EXACTFLU8;
+ }
}
}
+
alloc_maybe_populate_EXACT(pRExC_state, ret, flagp, len, ender,
FALSE /* Don't look to see if could
be turned into an EXACT
/* Position parse to next real character */
skip_to_be_ignored_text(pRExC_state, &RExC_parse,
FALSE /* Don't force to /x */ );
- if (PASS2 && *RExC_parse == '{' && OP(ret) != SBOL && ! regcurly(RExC_parse)) {
- ckWARNregdep(RExC_parse + 1, "Unescaped left brace in regex is deprecated here (and will be fatal in Perl 5.30), passed through");
+ if ( PASS2 && *RExC_parse == '{'
+ && OP(ret) != SBOL && ! regcurly(RExC_parse))
+ {
+ if (RExC_strict || new_regcurly(RExC_parse, RExC_end)) {
+ RExC_parse++;
+ vFAIL("Unescaped left brace in regex is illegal here");
+ }
+ ckWARNreg(RExC_parse + 1, "Unescaped left brace in regex is"
+ " passed through");
}
return(ret);
PERL_ARGS_ASSERT_HANDLE_REGEX_SETS;
+ DEBUG_PARSE("xcls");
+
if (in_locale) {
set_regex_charset(&RExC_flags, REGEX_UNICODE_CHARSET);
}
* these things, we need to realize that something preceded by a backslash
* is escaped, so we have to keep track of backslashes */
if (SIZE_ONLY) {
- UV depth = 0; /* how many nested (?[...]) constructs */
+ UV nest_depth = 0; /* how many nested (?[...]) constructs */
while (RExC_parse < RExC_end) {
SV* current = NULL;
TRUE /* Force /x */ );
switch (*RExC_parse) {
- case '?':
- if (RExC_parse[1] == '[') depth++, RExC_parse++;
+ case '(':
+ if (RExC_parse[1] == '?' && RExC_parse[2] == '[')
+ nest_depth++, RExC_parse+=2;
/* FALLTHROUGH */
default:
break;
}
case ']':
- if (depth--) break;
- RExC_parse++;
- if (*RExC_parse == ')') {
+ if (RExC_parse[1] == ')') {
+ RExC_parse++;
+ if (nest_depth--) break;
node = reganode(pRExC_state, ANYOF, 0);
RExC_size += ANYOF_SKIP;
nextchar(pRExC_state);
return node;
}
- goto no_close;
+ /* We output the messages even if warnings are off, because we'll fail
+ * the very next thing, and these give a likely diagnosis for that */
+ if (posix_warnings && av_tindex_skip_len_mg(posix_warnings) >= 0) {
+ output_or_return_posix_warnings(pRExC_state, posix_warnings, NULL);
+ }
+ RExC_parse++;
+ vFAIL("Unexpected ']' with no following ')' in (?[...");
}
RExC_parse += UTF ? UTF8SKIP(RExC_parse) : 1;
}
- no_close:
/* We output the messages even if warnings are off, because we'll fail
* the very next thing, and these give a likely diagnosis for that */
if (posix_warnings && av_tindex_skip_len_mg(posix_warnings) >= 0) {
output_or_return_posix_warnings(pRExC_state, posix_warnings, NULL);
}
- FAIL("Syntax error in (?[...])");
+ vFAIL("Syntax error in (?[...])");
}
/* Pass 2 only after this. */
* inversion list, and RExC_parse points to the trailing
* ']'; the next character should be the ')' */
RExC_parse++;
- assert(UCHARAT(RExC_parse) == ')');
+ if (UCHARAT(RExC_parse) != ')')
+ vFAIL("Expecting close paren for nested extended charclass");
/* Then the ')' matching the original '(' handled by this
* case: statement */
RExC_parse++;
- assert(UCHARAT(RExC_parse) == ')');
+ if (UCHARAT(RExC_parse) != ')')
+ vFAIL("Expecting close paren for wrapper for nested extended charclass");
RExC_parse++;
RExC_flags = save_flags;
* fence. Get rid of it */
fence_ptr = av_pop(fence_stack);
assert(fence_ptr);
- fence = SvIV(fence_ptr) - 1;
+ fence = SvIV(fence_ptr);
SvREFCNT_dec_NN(fence_ptr);
fence_ptr = NULL;
if (av_tindex_skip_len_mg(stack) < 0 /* Was empty */
|| ((final = av_pop(stack)) == NULL)
|| ! IS_OPERAND(final)
- || SvTYPE(final) != SVt_INVLIST
+ || ! is_invlist(final)
|| av_tindex_skip_len_mg(stack) >= 0) /* More left on stack */
{
bad_syntax:
STATIC void
S_add_above_Latin1_folds(pTHX_ RExC_state_t *pRExC_state, const U8 cp, SV** invlist)
{
- /* This hard-codes the Latin1/above-Latin1 folding rules, so that an
- * innocent-looking character class, like /[ks]/i won't have to go out to
- * disk to find the possible matches.
+ /* This adds the Latin1/above-Latin1 folding rules.
*
* This should be called only for a Latin1-range code points, cp, which is
* known to be involved in a simple fold with other code points above
* Latin1. It would give false results if /aa has been specified.
* Multi-char folds are outside the scope of this, and must be handled
- * specially.
- *
- * 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, and may be caught by the default: case of the switch
- * below. */
+ * specially. */
PERL_ARGS_ASSERT_ADD_ABOVE_LATIN1_FOLDS;
assert(HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(cp));
+ /* The rules that are valid for all Unicode versions are hard-coded in */
switch (cp) {
case 'k':
case 'K':
LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS);
break;
-#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
+ default: /* Other code points are checked against the data for the
+ current Unicode version */
+ {
+ Size_t folds_to_count;
+ unsigned int first_folds_to;
+ const unsigned int * remaining_folds_to_list;
+ UV folded_cp;
- case LATIN_SMALL_LETTER_SHARP_S:
- *invlist = add_cp_to_invlist(*invlist, LATIN_CAPITAL_LETTER_SHARP_S);
- break;
+ if (isASCII(cp)) {
+ folded_cp = toFOLD(cp);
+ }
+ else {
+ U8 dummy_fold[UTF8_MAXBYTES_CASE+1];
+ Size_t dummy_len;
+ folded_cp = _to_fold_latin1(cp, dummy_fold, &dummy_len, 0);
+ }
-#endif
+ if (folded_cp > 255) {
+ *invlist = add_cp_to_invlist(*invlist, folded_cp);
+ }
-#if UNICODE_MAJOR_VERSION < 3 \
- || (UNICODE_MAJOR_VERSION == 3 && UNICODE_DOT_VERSION == 0)
+ folds_to_count = _inverse_folds(folded_cp, &first_folds_to,
+ &remaining_folds_to_list);
+ if (folds_to_count == 0) {
- /* In 3.0 and earlier, U+0130 folded simply to 'i'; and in 3.0.1 so did
- * U+0131. */
- case 'i':
- case 'I':
- *invlist =
- add_cp_to_invlist(*invlist, LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
-# if UNICODE_DOT_DOT_VERSION == 1
- *invlist = add_cp_to_invlist(*invlist, LATIN_SMALL_LETTER_DOTLESS_I);
-# endif
- break;
-#endif
+ /* Use deprecated warning to increase the chances of this being
+ * output */
+ if (PASS2) {
+ ckWARN2reg_d(RExC_parse,
+ "Perl folding rules are not up-to-date for 0x%02X;"
+ " please use the perlbug utility to report;", cp);
+ }
+ }
+ else {
+ unsigned int i;
- default:
- /* Use deprecated warning to increase the chances of this being
- * output */
- if (PASS2) {
- ckWARN2reg_d(RExC_parse, "Perl folding rules are not up-to-date for 0x%02X; please use the perlbug utility to report;", cp);
+ if (first_folds_to > 255) {
+ *invlist = add_cp_to_invlist(*invlist, first_folds_to);
+ }
+ for (i = 0; i < folds_to_count - 1; i++) {
+ if (remaining_folds_to_list[i] > 255) {
+ *invlist = add_cp_to_invlist(*invlist,
+ remaining_folds_to_list[i]);
+ }
+ }
}
break;
+ }
}
}
do_posix_warnings ? &posix_warnings : NULL,
TRUE /* checking only */);
}
+ else if ( strict && ! skip_white
+ && ( _generic_isCC(value, _CC_VERTSPACE)
+ || is_VERTWS_cp_high(value)))
+ {
+ vFAIL("Literal vertical space in [] is illegal except under /x");
+ }
else if (value == '\\') {
/* Is a backslash; get the code point of the char after it */
if (*flagp & NEED_UTF8)
FAIL("panic: grok_bslash_N set NEED_UTF8");
- if (*flagp & RESTART_PASS1)
- return NULL;
+
+ RETURN_NULL_ON_RESTART_FLAGP(flagp);
if (cp_count < 0) {
vFAIL("\\N in a character class must be a named character: \\N{...}");
case 'P':
{
char *e;
+ char *i;
+
/* We will handle any undefined properties ourselves */
U8 swash_init_flags = _CORE_SWASH_INIT_RETURN_IF_UNDEF
* anyway, to save a little time */
|_CORE_SWASH_INIT_ACCEPT_INVLIST;
+ SvREFCNT_dec(swash); /* Free any left-overs */
if (RExC_parse >= RExC_end)
vFAIL2("Empty \\%c", (U8)value);
if (*RExC_parse == '{') {
}
RExC_parse++;
+
+ /* White space is allowed adjacent to the braces and after
+ * any '^', even when not under /x */
while (isSPACE(*RExC_parse)) {
RExC_parse++;
}
n = e - RExC_parse;
while (isSPACE(*(RExC_parse + n - 1)))
n--;
+
} /* The \p isn't immediately followed by a '{' */
else if (! isALPHA(*RExC_parse)) {
RExC_parse += (UTF) ? UTF8SKIP(RExC_parse) : 1;
n = 1;
}
if (!SIZE_ONLY) {
- SV* invlist;
- char* name;
+ char* name = RExC_parse;
char* base_name; /* name after any packages are stripped */
char* lookup_name = NULL;
const char * const colon_colon = "::";
+ bool invert;
+
+ SV* invlist;
+
+ /* Temporary workaround for [perl #133136]. For this
+ * precise input that is in the .t that is failing, load
+ * utf8.pm, which is what the test wants, so that that
+ * .t passes */
+ if ( memEQs(RExC_start, e + 1 - RExC_start,
+ "foo\\p{Alnum}")
+ && ! hv_common(GvHVn(PL_incgv),
+ NULL,
+ "utf8.pm", sizeof("utf8.pm") - 1,
+ 0, HV_FETCH_ISEXISTS, NULL, 0))
+ {
+ require_pv("utf8.pm");
+ }
+ invlist = parse_uniprop_string(name, n, FOLD, &invert);
+ if (invlist) {
+ if (invert) {
+ value ^= 'P' ^ 'p';
+ }
+ }
+ else {
/* Try to get the definition of the property into
* <invlist>. If /i is in effect, the effective property
* 2f833f5208e26b208886e51e09e2c072b5eabb46 */
name = savepv(Perl_form(aTHX_ "%.*s", (int)n, RExC_parse));
SAVEFREEPV(name);
+
+ for (i = RExC_parse; i < RExC_parse + n; i++) {
+ if (isCNTRL(*i) && *i != '\t') {
+ RExC_parse = e + 1;
+ vFAIL2("Can't find Unicode property definition \"%s\"", name);
+ }
+ }
+
if (FOLD) {
lookup_name = savepv(Perl_form(aTHX_ "__%s_i", name));
/* Look up the property name, and get its swash and
* inversion list, if the property is found */
- SvREFCNT_dec(swash); /* Free any left-overs */
swash = _core_swash_init("utf8",
(lookup_name)
? lookup_name
{
has_user_defined_property = TRUE;
}
- else if
+ }
+ }
+ if (invlist) {
+ if (! has_user_defined_property &&
/* We warn on matching an above-Unicode code point
* if the match would return true, except don't
* warn for \p{All}, which has exactly one element
* = 0 */
(_invlist_contains_cp(invlist, 0x110000)
&& (! (_invlist_len(invlist) == 1
- && *invlist_array(invlist) == 0)))
+ && *invlist_array(invlist) == 0))))
{
warn_super = TRUE;
}
-
/* Invert if asking for the complement */
if (value == 'P') {
_invlist_union_complement_2nd(properties,
/* The swash can't be used as-is, because we've
* inverted things; delay removing it to here after
* have copied its invlist above */
- SvREFCNT_dec_NN(swash);
+ if (! swash) {
+ SvREFCNT_dec_NN(invlist);
+ }
+ SvREFCNT_dec(swash);
swash = NULL;
}
else {
_invlist_union(properties, invlist, &properties);
+ if (! swash) {
+ SvREFCNT_dec_NN(invlist);
+ }
}
- }
- }
+ }
+ } /* End of actually getting the values in pass 2 */
+
RExC_parse = e + 1;
namedclass = ANYOF_UNIPROP; /* no official name, but it's
named */
}
}
else if ( UNI_SEMANTICS
+ || AT_LEAST_ASCII_RESTRICTED
|| classnum == _CC_ASCII
|| (DEPENDS_SEMANTICS && ( classnum == _CC_DIGIT
|| classnum == _CC_XDIGIT)))
{
- /* We usually have to worry about /d and /a affecting what
- * POSIX classes match, with special code needed for /d
- * because we won't know until runtime what all matches.
- * But there is no extra work needed under /u, and
- * [:ascii:] is unaffected by /a and /d; and :digit: and
- * :xdigit: don't have runtime differences under /d. So we
- * can special case these, and avoid some extra work below,
- * and at runtime. */
+ /* We usually have to worry about /d a affecting what POSIX
+ * classes match, with special code needed because we won't
+ * know until runtime what all matches. But there is no
+ * extra work needed under /u and /a; and [:ascii:] is
+ * unaffected by /d; and :digit: and :xdigit: don't have
+ * runtime differences under /d. So we can special case
+ * these, and avoid some extra work below, and at runtime.
+ * */
_invlist_union_maybe_complement_2nd(
simple_posixes,
- PL_XPosix_ptrs[classnum],
+ ((AT_LEAST_ASCII_RESTRICTED)
+ ? PL_Posix_ptrs[classnum]
+ : PL_XPosix_ptrs[classnum]),
namedclass % 2 != 0,
&simple_posixes);
}
{
/* Here <value> is indeed a multi-char fold. Get what it is */
- U8 foldbuf[UTF8_MAXBYTES_CASE];
+ U8 foldbuf[UTF8_MAXBYTES_CASE+1];
STRLEN foldlen;
UV folded = _to_uni_fold_flags(
" be some subset of \"0-9\","
" \"A-Z\", or \"a-z\"");
}
- else if (prevvalue >= 0x660) { /* ARABIC_INDIC_DIGIT_ZERO */
+ else if (prevvalue >= FIRST_NON_ASCII_DECIMAL_DIGIT) {
SSize_t index_start;
SSize_t index_final;
* can't do the same checks for above-ASCII ranges,
* except in the case of digit ones. These should
* contain only digits from the same group of 10. The
- * ASCII case is handled just above. 0x660 is the
- * first digit character beyond ASCII. Hence here, the
+ * ASCII case is handled just above. Hence here, the
* range could be a range of digits. First some
* unlikely special cases. Grandfather in that a range
* ending in 19DA (NEW TAI LUE THAM DIGIT ONE) is bad
#if 0 /* Have decided not to deal with multi-char folds in inverted classes,
because too confusing */
if (invert) {
- sv_catpv(substitute_parse, "(?:");
+ sv_catpvs(substitute_parse, "(?:");
}
#endif
&PL_sv_undef)
{
if (! first_time) {
- sv_catpv(substitute_parse, "|");
+ sv_catpvs(substitute_parse, "|");
}
first_time = FALSE;
/* If the character class contains anything else besides these
* multi-character folds, have to include it in recursive parsing */
if (element_count) {
- sv_catpv(substitute_parse, "|[");
+ sv_catpvs(substitute_parse, "|[");
prefix_end = SvCUR(substitute_parse);
sv_catpvn(substitute_parse, orig_parse, RExC_parse - orig_parse);
/* Put in a closing ']' only if not going off the end, as otherwise
* we are adding something that really isn't there */
if (RExC_parse < RExC_end) {
- sv_catpv(substitute_parse, "]");
+ sv_catpvs(substitute_parse, "]");
}
}
- sv_catpv(substitute_parse, ")");
+ sv_catpvs(substitute_parse, ")");
#if 0
if (invert) {
/* This is a way to get the parse to skip forward a whole named
* sequence instead of matching the 2nd character when it fails the
* first */
- sv_catpv(substitute_parse, "(*THEN)(*SKIP)(*FAIL)|.)");
+ sv_catpvs(substitute_parse, "(*THEN)(*SKIP)(*FAIL)|.)");
}
#endif
ret = reg(pRExC_state, 1, ®_flags, depth+1);
- *flagp |= reg_flags&(HASWIDTH|SIMPLE|SPSTART|POSTPONED|RESTART_PASS1|NEED_UTF8);
+ *flagp |= reg_flags & (HASWIDTH|SIMPLE|SPSTART|POSTPONED|RESTART_PASS1|NEED_UTF8);
/* And restore so can parse the rest of the pattern */
RExC_parse = save_parse;
/* The actual POSIXish node for all the rest depends on the
* charset modifier. The ones in the first set depend only on
* ASCII or, if available on this platform, also locale */
+
case ANYOF_ASCII:
case ANYOF_NASCII:
+
#ifdef HAS_ISASCII
- op = (LOC) ? POSIXL : POSIXA;
-#else
- op = POSIXA;
+ if (LOC) {
+ op = POSIXL;
+ goto join_posix;
+ }
#endif
- goto join_posix;
+ /* (named_class - ANYOF_ASCII) is 0 or 1. xor'ing with
+ * invert converts that to 1 or 0 */
+ op = ASCII + ((namedclass - ANYOF_ASCII) ^ invert);
+ break;
/* The following don't have any matches in the upper Latin1
* range, hence /d is equivalent to /u for them. Making it /u
TRUE /* downgradable to EXACT */
);
}
+ else {
+ *flagp |= HASWIDTH|SIMPLE;
+ }
RExC_parse = (char *) cur_parse;
_invlist_intersection(PL_utf8_foldable, cp_foldable_list,
&fold_intersection);
- /* The folds for all the Latin1 characters are hard-coded into this
- * program, but we have to go out to disk to get the others. */
- if (invlist_highest(cp_foldable_list) >= 256) {
-
- /* This is a hash that for a particular fold gives all
- * characters that are involved in it */
- if (! PL_utf8_foldclosures) {
- _load_PL_utf8_foldclosures();
- }
- }
-
/* Now look at the foldable characters in this class individually */
invlist_iterinit(fold_intersection);
while (invlist_iternext(fold_intersection, &start, &end)) {
UV j;
+ UV folded;
/* Look at every character in the range */
for (j = start; j <= end; j++) {
U8 foldbuf[UTF8_MAXBYTES_CASE+1];
STRLEN foldlen;
- SV** listp;
+ unsigned int k;
+ Size_t folds_to_count;
+ unsigned int first_folds_to;
+ const unsigned int * remaining_folds_to_list;
if (j < 256) {
* rules hard-coded for it. First, get its fold. This is
* the simple fold, as the multi-character folds have been
* handled earlier and separated out */
- _to_uni_fold_flags(j, foldbuf, &foldlen,
+ folded = _to_uni_fold_flags(j, foldbuf, &foldlen,
(ASCII_FOLD_RESTRICTED)
? FOLD_FLAGS_NOMIX_ASCII
: 0);
- /* Single character fold of above Latin1. Add everything in
- * its fold closure to the list that this node should match.
- * The fold closures data structure is a hash with the keys
- * being the UTF-8 of every character that is folded to, like
- * 'k', and the values each an array of all code points that
- * fold to its key. e.g. [ 'k', 'K', KELVIN_SIGN ].
- * Multi-character folds are not included */
- if ((listp = hv_fetch(PL_utf8_foldclosures,
- (char *) foldbuf, foldlen, FALSE)))
- {
- AV* list = (AV*) *listp;
- IV k;
- for (k = 0; k <= av_tindex_skip_len_mg(list); k++) {
- SV** c_p = av_fetch(list, k, FALSE);
- UV c;
- assert(c_p);
-
- c = SvUV(*c_p);
-
- /* /aa doesn't allow folds between ASCII and non- */
- if ((ASCII_FOLD_RESTRICTED
- && (isASCII(c) != isASCII(j))))
- {
- continue;
- }
+ /* Single character fold of above Latin1. Add everything
+ * in its fold closure to the list that this node should
+ * match. */
+ folds_to_count = _inverse_folds(folded, &first_folds_to,
+ &remaining_folds_to_list);
+ for (k = 0; k <= folds_to_count; k++) {
+ UV c = (k == 0) /* First time through use itself */
+ ? folded
+ : (k == 1) /* 2nd time use, the first fold */
+ ? first_folds_to
+
+ /* Then the remaining ones */
+ : remaining_folds_to_list[k-2];
+
+ /* /aa doesn't allow folds between ASCII and non- */
+ if (( ASCII_FOLD_RESTRICTED
+ && (isASCII(c) != isASCII(j))))
+ {
+ continue;
+ }
- /* Folds under /l which cross the 255/256 boundary
- * are added to a separate list. (These are valid
- * only when the locale is UTF-8.) */
- if (c < 256 && LOC) {
- *use_list = add_cp_to_invlist(*use_list, c);
- continue;
- }
+ /* Folds under /l which cross the 255/256 boundary are
+ * added to a separate list. (These are valid only
+ * when the locale is UTF-8.) */
+ if (c < 256 && LOC) {
+ *use_list = add_cp_to_invlist(*use_list, c);
+ continue;
+ }
- if (isASCII(c) || c > 255 || AT_LEAST_UNI_SEMANTICS)
- {
- cp_list = add_cp_to_invlist(cp_list, c);
- }
- else {
- /* Similarly folds involving non-ascii Latin1
- * characters under /d are added to their list */
- has_upper_latin1_only_utf8_matches
- = add_cp_to_invlist(
- has_upper_latin1_only_utf8_matches,
- c);
- }
+ if (isASCII(c) || c > 255 || AT_LEAST_UNI_SEMANTICS)
+ {
+ cp_list = add_cp_to_invlist(cp_list, c);
+ }
+ else {
+ /* Similarly folds involving non-ascii Latin1
+ * characters under /d are added to their list */
+ has_upper_latin1_only_utf8_matches
+ = add_cp_to_invlist(
+ has_upper_latin1_only_utf8_matches,
+ c);
}
}
}
}
}
if (posixes || nposixes) {
-
- /* We have to adjust /a and /aa */
- if (AT_LEAST_ASCII_RESTRICTED) {
-
- /* Under /a and /aa, nothing above ASCII matches these */
- if (posixes) {
- _invlist_intersection(posixes,
- PL_XPosix_ptrs[_CC_ASCII],
- &posixes);
- }
-
- /* Under /a and /aa, everything above ASCII matches these
- * complements */
- if (nposixes) {
- _invlist_union_complement_2nd(nposixes,
- PL_XPosix_ptrs[_CC_ASCII],
- &nposixes);
- }
- }
-
if (! DEPENDS_SEMANTICS) {
/* For everything but /d, we can just add the current 'posixes' and
if (_invlist_len(only_non_utf8_list) != 0) {
ANYOF_FLAGS(ret) |= ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER;
}
+ SvREFCNT_dec_NN(only_non_utf8_list);
}
else {
/* Here there were no complemented posix classes. That means
* certain common classes that are easy to test. Getting to this point in
* the code means that the class didn't get optimized there. Since this
* code is only executed in Pass 2, it is too late to save space--it has
- * been allocated in Pass 1, and currently isn't given back. But turning
- * things into an EXACTish node can allow the optimizer to join it to any
- * adjacent such nodes. And if the class is equivalent to things like /./,
- * expensive run-time swashes can be avoided. Now that we have more
- * complete information, we can find things necessarily missed by the
- * earlier code. Another possible "optimization" that isn't done is that
- * something like [Ee] could be changed into an EXACTFU. khw tried this
- * and found that the ANYOF is faster, including for code points not in the
- * bitmap. This still might make sense to do, provided it got joined with
- * an adjacent node(s) to create a longer EXACTFU one. This could be
- * accomplished by creating a pseudo ANYOF_EXACTFU node type that the join
- * routine would know is joinable. If that didn't happen, the node type
- * could then be made a straight ANYOF */
+ * been allocated in Pass 1, and currently isn't given back. XXX Why not?
+ * But turning things into an EXACTish node can allow the optimizer to join
+ * it to any adjacent such nodes. And if the class is equivalent to things
+ * like /./, expensive run-time swashes can be avoided. Now that we have
+ * more complete information, we can find things necessarily missed by the
+ * earlier code. */
if (optimizable && cp_list && ! invert) {
UV start, end;
U8 op = END; /* The optimzation node-type */
int posix_class = -1; /* Illegal value */
const char * cur_parse= RExC_parse;
+ U8 ANYOFM_mask = 0xFF;
+ U32 anode_arg = 0;
invlist_iterinit(cp_list);
if (! invlist_iternext(cp_list, &start, &end)) {
invlist_iterfinish(cp_list);
if (op == END) {
- const UV cp_list_len = _invlist_len(cp_list);
- const UV* cp_list_array = invlist_array(cp_list);
/* Here, didn't find an optimization. See if this matches any of
- * the POSIX classes. These run slightly faster for above-Unicode
- * code points, so don't bother with POSIXA ones nor the 2 that
- * have no above-Unicode matches. We can avoid these checks unless
- * the ANYOF matches at least as high as the lowest POSIX one
- * (which was manually found to be \v. The actual code point may
- * increase in later Unicode releases, if a higher code point is
- * assigned to be \v, but this code will never break. It would
- * just mean we could execute the checks for posix optimizations
- * unnecessarily) */
-
- if (cp_list_array[cp_list_len-1] > 0x2029) {
+ * the POSIX classes. First try ASCII */
+
+ if (_invlistEQ(cp_list, PL_XPosix_ptrs[_CC_ASCII], 0)) {
+ op = ASCII;
+ *flagp |= HASWIDTH|SIMPLE;
+ }
+ else if (_invlistEQ(cp_list, PL_XPosix_ptrs[_CC_ASCII], 1)) {
+ op = NASCII;
+ *flagp |= HASWIDTH|SIMPLE;
+ }
+ else if (invlist_highest(cp_list) >= 0x2029) {
+
+ /* Then try the other POSIX classes. The POSIXA ones are about
+ * the same speed as ANYOF ops, but the ones that have
+ * above-Latin1 code point matches are somewhat faster than
+ * ANYOF. So optimize those, but don't bother with the POSIXA
+ * ones nor [:cntrl:] which has no above-Latin1 matches. If
+ * this ANYOF node has a lower highest possible matching code
+ * point than any of the XPosix ones, we know that it can't
+ * possibly be the same as any of them, so we can avoid
+ * executing this code. The 0x2029 above for the lowest max
+ * was determined by manual inspection of the classes, and
+ * comes from \v. Suppose Unicode in a later version adds a
+ * higher code point to \v. All that means is that this code
+ * can be executed unnecessarily. It will still give the
+ * correct answer. */
+
for (posix_class = 0;
posix_class <= _HIGHEST_REGCOMP_DOT_H_SYNC;
posix_class++)
{
int try_inverted;
- if (posix_class == _CC_ASCII || posix_class == _CC_CNTRL) {
+
+ if (posix_class == _CC_CNTRL) {
continue;
}
+
for (try_inverted = 0; try_inverted < 2; try_inverted++) {
/* Check if matches normal or inverted */
}
found_posix: ;
}
+
+ /* If it didn't match a POSIX class, it might be able to be turned
+ * into an ANYOFM node. Compare two different bytes, bit-by-bit.
+ * In some positions, the bits in each will be 1; and in other
+ * positions both will be 0; and in some positions the bit will be
+ * 1 in one byte, and 0 in the other. Let 'n' be the number of
+ * positions where the bits differ. We create a mask which has
+ * exactly 'n' 0 bits, each in a position where the two bytes
+ * differ. Now take the set of all bytes that when ANDed with the
+ * mask yield the same result. That set has 2**n elements, and is
+ * representable by just two 8 bit numbers: the result and the
+ * mask. Importantly, matching the set can be vectorized by
+ * creating a word full of the result bytes, and a word full of the
+ * mask bytes, yielding a significant speed up. Here, see if this
+ * node matches such a set. As a concrete example consider [01],
+ * and the byte representing '0' which is 0x30 on ASCII machines.
+ * It has the bits 0011 0000. Take the mask 1111 1110. If we AND
+ * 0x31 and 0x30 with that mask we get 0x30. Any other bytes ANDed
+ * yield something else. So [01], which is a common usage, is
+ * optimizable into ANYOFM, and can benefit from the speed up. We
+ * can only do this on UTF-8 invariant bytes, because the variance
+ * would throw this off. */
+ if ( op == END
+ && invlist_highest(cp_list) <=
+#ifdef EBCDIC
+ 0xFF
+#else
+ 0x7F
+#endif
+ ) {
+ Size_t cp_count = 0;
+ bool first_time = TRUE;
+ unsigned int lowest_cp = 0xFF;
+ U8 bits_differing = 0;
+
+ /* Only needed on EBCDIC, as there, variants and non- are mixed
+ * together. Could #ifdef it out on ASCII, but probably the
+ * compiler will optimize it out */
+ bool has_variant = FALSE;
+
+ /* Go through the bytes and find the bit positions that differ */
+ invlist_iterinit(cp_list);
+ while (invlist_iternext(cp_list, &start, &end)) {
+ unsigned int i = start;
+
+ cp_count += end - start + 1;
+
+ if (first_time) {
+ if (! UVCHR_IS_INVARIANT(i)) {
+ has_variant = TRUE;
+ continue;
+ }
+
+ first_time = FALSE;
+ lowest_cp = start;
+
+ i++;
+ }
+
+ /* Find the bit positions that differ from the lowest code
+ * point in the node. Keep track of all such positions by
+ * OR'ing */
+ for (; i <= end; i++) {
+ if (! UVCHR_IS_INVARIANT(i)) {
+ has_variant = TRUE;
+ continue;
+ }
+
+ bits_differing |= i ^ lowest_cp;
+ }
+ }
+ invlist_iterfinish(cp_list);
+
+ /* At the end of the loop, we count how many bits differ from
+ * the bits in lowest code point, call the count 'd'. If the
+ * set we found contains 2**d elements, it is the closure of
+ * all code points that differ only in those bit positions. To
+ * convince yourself of that, first note that the number in the
+ * closure must be a power of 2, which we test for. The only
+ * way we could have that count and it be some differing set,
+ * is if we got some code points that don't differ from the
+ * lowest code point in any position, but do differ from each
+ * other in some other position. That means one code point has
+ * a 1 in that position, and another has a 0. But that would
+ * mean that one of them differs from the lowest code point in
+ * that position, which possibility we've already excluded. */
+ if ( ! has_variant
+ && cp_count == 1U << PL_bitcount[bits_differing])
+ {
+ assert(cp_count > 1);
+ op = ANYOFM;
+
+ /* We need to make the bits that differ be 0's */
+ ANYOFM_mask = ~ bits_differing; /* This goes into FLAGS */
+
+ /* The argument is the lowest code point */
+ anode_arg = lowest_cp;
+ *flagp |= HASWIDTH|SIMPLE;
+ }
+ }
}
if (op != END) {
RExC_emit = (regnode *)orig_emit;
if (regarglen[op]) {
- ret = reganode(pRExC_state, op, 0);
+ ret = reganode(pRExC_state, op, anode_arg);
} else {
ret = reg_node(pRExC_state, op);
}
else if (PL_regkind[op] == POSIXD || PL_regkind[op] == NPOSIXD) {
FLAGS(ret) = posix_class;
}
+ else if (PL_regkind[op] == ANYOFM) {
+ FLAGS(ret) = ANYOFM_mask;
+ }
SvREFCNT_dec_NN(cp_list);
return ret;
case EXACT:
case EXACTL:
case EXACTF:
- case EXACTFA_NO_TRIE:
- case EXACTFA:
+ case EXACTFAA_NO_TRIE:
+ case EXACTFAA:
case EXACTFU:
case EXACTFLU8:
case EXACTFU_SS:
}
#endif
+STATIC SV*
+S_get_ANYOFM_contents(pTHX_ const regnode * n) {
+
+ /* Returns an inversion list of all the code points matched by the ANYOFM
+ * node 'n' */
+
+ SV * cp_list = _new_invlist(-1);
+ const U8 lowest = (U8) ARG(n);
+ unsigned int i;
+ U8 count = 0;
+ U8 needed = 1U << PL_bitcount[ (U8) ~ FLAGS(n)];
+
+ PERL_ARGS_ASSERT_GET_ANYOFM_CONTENTS;
+
+ /* Starting with the lowest code point, any code point that ANDed with the
+ * mask yields the lowest code point is in the set */
+ for (i = lowest; i <= 0xFF; i++) {
+ if ((i & FLAGS(n)) == ARG(n)) {
+ cp_list = add_cp_to_invlist(cp_list, i);
+ count++;
+
+ /* We know how many code points (a power of two) that are in the
+ * set. No use looking once we've got that number */
+ if (count >= needed) break;
+ }
+ }
+
+ return cp_list;
+}
+
/*
- regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
*/
if ( k == REF && reginfo) {
U32 n = ARG(o); /* which paren pair */
I32 ln = prog->offs[n].start;
- if (prog->lastparen < n || ln == -1)
+ if (prog->lastparen < n || ln == -1 || prog->offs[n].end == -1)
Perl_sv_catpvf(aTHX_ sv, ": FAIL");
else if (ln == prog->offs[n].end)
Perl_sv_catpvf(aTHX_ sv, ": ACCEPT - EMPTY STRING");
SvREFCNT_dec(unresolved);
}
+ else if (k == ANYOFM) {
+ SV * cp_list = get_ANYOFM_contents(o);
+
+ Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
+ put_charclass_bitmap_innards(sv, NULL, cp_list, NULL, NULL, TRUE);
+ Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
+
+ SvREFCNT_dec(cp_list);
+ }
else if (k == POSIXD || k == NPOSIXD) {
U8 index = FLAGS(o) * 2;
if (index < C_ARRAY_LENGTH(anyofs)) {
if (*anyofs[index] != '[') {
- sv_catpv(sv, "[");
+ sv_catpvs(sv, "[");
}
sv_catpv(sv, anyofs[index]);
if (*anyofs[index] != '[') {
- sv_catpv(sv, "]");
+ sv_catpvs(sv, "]");
}
}
else {
#else
format = "\\x%02" UVXf "-\\x%02" UVXf;
#endif
- GCC_DIAG_IGNORE(-Wformat-nonliteral);
+ GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral);
Perl_sv_catpvf(aTHX_ sv, format, start, this_end);
- GCC_DIAG_RESTORE;
+ GCC_DIAG_RESTORE_STMT;
break;
}
}
{
/* Appends to 'sv' a displayable version of the innards of the bracketed
* character class defined by the other arguments:
- * 'bitmap' points to the bitmap.
+ * 'bitmap' points to the bitmap, or NULL if to ignore that.
* 'nonbitmap_invlist' is an inversion list of the code points that are in
* the bitmap range, but for some reason aren't in the bitmap; NULL if
* none. The reasons for this could be that they require some
* was not resolved at the time of the regex compilation (under /u)
* 'only_utf8_locale_invlist' is an inversion list of the code points that
* are valid only if the runtime locale is a UTF-8 one; NULL if none
- * 'node' is the regex pattern node. It is needed only when the above two
- * parameters are not null, and is passed so that this routine can
- * tease apart the various reasons for them.
+ * 'node' is the regex pattern ANYOF node. It is needed only when the
+ * above two parameters are not null, and is passed so that this
+ * routine can tease apart the various reasons for them.
* 'force_as_is_display' is TRUE if this routine should definitely NOT try
* to invert things to see if that leads to a cleaner display. If
* FALSE, this routine is free to use its judgment about doing this.
}
/* Accumulate the bit map into the unconditional match list */
- for (i = 0; i < NUM_ANYOF_CODE_POINTS; i++) {
- if (BITMAP_TEST(bitmap, i)) {
- int start = i++;
- for (; i < NUM_ANYOF_CODE_POINTS && BITMAP_TEST(bitmap, i); i++) {
- /* empty */
+ if (bitmap) {
+ for (i = 0; i < NUM_ANYOF_CODE_POINTS; i++) {
+ if (BITMAP_TEST(bitmap, i)) {
+ int start = i++;
+ for (;
+ i < NUM_ANYOF_CODE_POINTS && BITMAP_TEST(bitmap, i);
+ i++)
+ { /* empty */ }
+ invlist = _add_range_to_invlist(invlist, start, i-1);
}
- invlist = _add_range_to_invlist(invlist, start, i-1);
}
}
/* While that wasn't END last time... */
NODE_ALIGN(node);
op = OP(node);
- if (op == CLOSE || op == WHILEM)
+ if (op == CLOSE || op == SRCLOSE || op == WHILEM)
indent--;
next = regnext((regnode *)node);
node = NEXTOPER(node);
node += regarglen[(U8)op];
}
- if (op == CURLYX || op == OPEN)
+ if (op == CURLYX || op == OPEN || op == SROPEN)
indent++;
}
CLEAR_OPTSTART;
#endif /* DEBUGGING */
+#ifndef PERL_IN_XSUB_RE
+
+#include "uni_keywords.h"
+
+void
+Perl_init_uniprops(pTHX)
+{
+ /* Set up the inversion list global variables */
+
+ PL_XPosix_ptrs[_CC_ASCII] = _new_invlist_C_array(uni_prop_ptrs[UNI_ASCII]);
+ PL_XPosix_ptrs[_CC_ALPHANUMERIC] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXALNUM]);
+ PL_XPosix_ptrs[_CC_ALPHA] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXALPHA]);
+ PL_XPosix_ptrs[_CC_BLANK] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXBLANK]);
+ PL_XPosix_ptrs[_CC_CASED] = _new_invlist_C_array(uni_prop_ptrs[UNI_CASED]);
+ PL_XPosix_ptrs[_CC_CNTRL] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXCNTRL]);
+ PL_XPosix_ptrs[_CC_DIGIT] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXDIGIT]);
+ PL_XPosix_ptrs[_CC_GRAPH] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXGRAPH]);
+ PL_XPosix_ptrs[_CC_LOWER] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXLOWER]);
+ PL_XPosix_ptrs[_CC_PRINT] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXPRINT]);
+ PL_XPosix_ptrs[_CC_PUNCT] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXPUNCT]);
+ PL_XPosix_ptrs[_CC_SPACE] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXSPACE]);
+ PL_XPosix_ptrs[_CC_UPPER] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXUPPER]);
+ PL_XPosix_ptrs[_CC_VERTSPACE] = _new_invlist_C_array(uni_prop_ptrs[UNI_VERTSPACE]);
+ PL_XPosix_ptrs[_CC_WORDCHAR] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXWORD]);
+ PL_XPosix_ptrs[_CC_XDIGIT] = _new_invlist_C_array(uni_prop_ptrs[UNI_XPOSIXXDIGIT]);
+
+ PL_Posix_ptrs[_CC_ASCII] = _new_invlist_C_array(uni_prop_ptrs[UNI_ASCII]);
+ PL_Posix_ptrs[_CC_ALPHANUMERIC] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXALNUM]);
+ PL_Posix_ptrs[_CC_ALPHA] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXALPHA]);
+ PL_Posix_ptrs[_CC_BLANK] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXBLANK]);
+ PL_Posix_ptrs[_CC_CASED] = _new_invlist_C_array(uni_prop_ptrs[UNI_CASED]);
+ PL_Posix_ptrs[_CC_CNTRL] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXCNTRL]);
+ PL_Posix_ptrs[_CC_DIGIT] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXDIGIT]);
+ PL_Posix_ptrs[_CC_GRAPH] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXGRAPH]);
+ PL_Posix_ptrs[_CC_LOWER] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXLOWER]);
+ PL_Posix_ptrs[_CC_PRINT] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXPRINT]);
+ PL_Posix_ptrs[_CC_PUNCT] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXPUNCT]);
+ PL_Posix_ptrs[_CC_SPACE] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXSPACE]);
+ PL_Posix_ptrs[_CC_UPPER] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXUPPER]);
+ PL_Posix_ptrs[_CC_VERTSPACE] = _new_invlist_C_array(uni_prop_ptrs[UNI_VERTSPACE]);
+ PL_Posix_ptrs[_CC_WORDCHAR] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXWORD]);
+ PL_Posix_ptrs[_CC_XDIGIT] = _new_invlist_C_array(uni_prop_ptrs[UNI_POSIXXDIGIT]);
+
+ PL_GCB_invlist = _new_invlist_C_array(_Perl_GCB_invlist);
+ PL_SB_invlist = _new_invlist_C_array(_Perl_SB_invlist);
+ PL_WB_invlist = _new_invlist_C_array(_Perl_WB_invlist);
+ PL_LB_invlist = _new_invlist_C_array(_Perl_LB_invlist);
+ PL_SCX_invlist = _new_invlist_C_array(_Perl_SCX_invlist);
+
+ PL_AboveLatin1 = _new_invlist_C_array(AboveLatin1_invlist);
+ PL_Latin1 = _new_invlist_C_array(Latin1_invlist);
+ PL_UpperLatin1 = _new_invlist_C_array(UpperLatin1_invlist);
+
+ PL_Assigned_invlist = _new_invlist_C_array(uni_prop_ptrs[UNI_ASSIGNED]);
+
+ PL_utf8_perl_idstart = _new_invlist_C_array(uni_prop_ptrs[UNI__PERL_IDSTART]);
+ PL_utf8_perl_idcont = _new_invlist_C_array(uni_prop_ptrs[UNI__PERL_IDCONT]);
+
+ PL_utf8_charname_begin = _new_invlist_C_array(uni_prop_ptrs[UNI__PERL_CHARNAME_BEGIN]);
+ PL_utf8_charname_continue = _new_invlist_C_array(uni_prop_ptrs[UNI__PERL_CHARNAME_CONTINUE]);
+
+ PL_utf8_foldable = _new_invlist_C_array(uni_prop_ptrs[UNI__PERL_ANY_FOLDS]);
+ PL_HasMultiCharFold = _new_invlist_C_array(uni_prop_ptrs[
+ UNI__PERL_FOLDS_TO_MULTI_CHAR]);
+ PL_NonL1NonFinalFold = _new_invlist_C_array(
+ NonL1_Perl_Non_Final_Folds_invlist);
+
+ PL_utf8_toupper = _new_invlist_C_array(Uppercase_Mapping_invlist);
+ PL_utf8_tolower = _new_invlist_C_array(Lowercase_Mapping_invlist);
+ PL_utf8_totitle = _new_invlist_C_array(Titlecase_Mapping_invlist);
+ PL_utf8_tofold = _new_invlist_C_array(Case_Folding_invlist);
+ PL_utf8_tosimplefold = _new_invlist_C_array(Simple_Case_Folding_invlist);
+ PL_utf8_foldclosures = _new_invlist_C_array(_Perl_IVCF_invlist);
+ PL_utf8_mark = _new_invlist_C_array(uni_prop_ptrs[UNI_M]);
+
+ /* The below are used only by deprecated functions. They could be removed */
+ PL_utf8_xidcont = _new_invlist_C_array(uni_prop_ptrs[UNI_XIDC]);
+ PL_utf8_idcont = _new_invlist_C_array(uni_prop_ptrs[UNI_IDC]);
+ PL_utf8_xidstart = _new_invlist_C_array(uni_prop_ptrs[UNI_XIDS]);
+}
+
+SV *
+Perl_parse_uniprop_string(pTHX_ const char * const name, const Size_t name_len,
+ const bool to_fold, bool * invert)
+{
+ /* Parse the interior meat of \p{} passed to this in 'name' with length
+ * 'name_len', and return an inversion list if a property with 'name' is
+ * found, or NULL if not. 'name' point to the input with leading and
+ * trailing space trimmed. 'to_fold' indicates if /i is in effect.
+ *
+ * When the return is an inversion list, '*invert' will be set to a boolean
+ * indicating if it should be inverted or not
+ *
+ * This currently doesn't handle all cases. A NULL return indicates the
+ * caller should try a different approach
+ */
+
+ char* lookup_name;
+ bool stricter = FALSE;
+ bool is_nv_type = FALSE; /* nv= or numeric_value=, or possibly one
+ of the cjk numeric properties (though
+ it requires extra effort to compile
+ them) */
+ unsigned int i;
+ unsigned int j = 0, lookup_len;
+ int equals_pos = -1; /* Where the '=' is found, or negative if none */
+ int slash_pos = -1; /* Where the '/' is found, or negative if none */
+ int table_index = 0;
+ bool starts_with_In_or_Is = FALSE;
+ Size_t lookup_offset = 0;
+
+ PERL_ARGS_ASSERT_PARSE_UNIPROP_STRING;
+
+ /* The input will be modified into 'lookup_name' */
+ Newx(lookup_name, name_len, char);
+ SAVEFREEPV(lookup_name);
+
+ /* Parse the input. */
+ for (i = 0; i < name_len; i++) {
+ char cur = name[i];
+
+ /* These characters can be freely ignored in most situations. Later it
+ * may turn out we shouldn't have ignored them, and we have to reparse,
+ * but we don't have enough information yet to make that decision */
+ if (cur == '-' || cur == '_' || isSPACE_A(cur)) {
+ continue;
+ }
+
+ /* Case differences are also ignored. Our lookup routine assumes
+ * everything is lowercase */
+ if (isUPPER_A(cur)) {
+ lookup_name[j++] = toLOWER(cur);
+ continue;
+ }
+
+ /* A double colon is either an error, or a package qualifier to a
+ * subroutine user-defined property; neither of which do we currently
+ * handle
+ *
+ * But a single colon is a synonym for '=' */
+ if (cur == ':') {
+ if (i < name_len - 1 && name[i+1] == ':') {
+ return NULL;
+ }
+ cur = '=';
+ }
+
+ /* Otherwise, this character is part of the name. */
+ lookup_name[j++] = cur;
+
+ /* Only the equals sign needs further processing */
+ if (cur == '=') {
+ equals_pos = j; /* Note where it occurred in the input */
+ break;
+ }
+ }
+
+ /* Here, we are either done with the whole property name, if it was simple;
+ * or are positioned just after the '=' if it is compound. */
+
+ if (equals_pos >= 0) {
+ assert(! stricter); /* We shouldn't have set this yet */
+
+ /* Space immediately after the '=' is ignored */
+ i++;
+ for (; i < name_len; i++) {
+ if (! isSPACE_A(name[i])) {
+ break;
+ }
+ }
+
+ /* Certain properties need special handling. They may optionally be
+ * prefixed by 'is'. Ignore that prefix for the purposes of checking
+ * if this is one of those properties */
+ if (memBEGINPs(lookup_name, name_len, "is")) {
+ lookup_offset = 2;
+ }
+
+ /* Then check if it is one of these properties. This is hard-coded
+ * because easier this way, and the list is unlikely to change. There
+ * are several properties like this in the Unihan DB, which is unlikely
+ * to be compiled, and they all end with 'numeric'. The interiors
+ * aren't checked for the precise property. This would stop working if
+ * a cjk property were to be created that ended with 'numeric' and
+ * wasn't a numeric type */
+ is_nv_type = memEQs(lookup_name + lookup_offset,
+ j - 1 - lookup_offset, "numericvalue")
+ || memEQs(lookup_name + lookup_offset,
+ j - 1 - lookup_offset, "nv")
+ || ( memENDPs(lookup_name + lookup_offset,
+ j - 1 - lookup_offset, "numeric")
+ && ( memBEGINPs(lookup_name + lookup_offset,
+ j - 1 - lookup_offset, "cjk")
+ || memBEGINPs(lookup_name + lookup_offset,
+ j - 1 - lookup_offset, "k")));
+ if ( is_nv_type
+ || memEQs(lookup_name + lookup_offset,
+ j - 1 - lookup_offset, "canonicalcombiningclass")
+ || memEQs(lookup_name + lookup_offset,
+ j - 1 - lookup_offset, "ccc")
+ || memEQs(lookup_name + lookup_offset,
+ j - 1 - lookup_offset, "age")
+ || memEQs(lookup_name + lookup_offset,
+ j - 1 - lookup_offset, "in")
+ || memEQs(lookup_name + lookup_offset,
+ j - 1 - lookup_offset, "presentin"))
+ {
+ unsigned int k;
+
+ /* What makes these properties special is that the stuff after the
+ * '=' is a number. Therefore, we can't throw away '-'
+ * willy-nilly, as those could be a minus sign. Other stricter
+ * rules also apply. However, these properties all can have the
+ * rhs not be a number, in which case they contain at least one
+ * alphabetic. In those cases, the stricter rules don't apply.
+ * But the numeric type properties can have the alphas [Ee] to
+ * signify an exponent, and it is still a number with stricter
+ * rules. So look for an alpha that signifys not-strict */
+ stricter = TRUE;
+ for (k = i; k < name_len; k++) {
+ if ( isALPHA_A(name[k])
+ && (! is_nv_type || ! isALPHA_FOLD_EQ(name[k], 'E')))
+ {
+ stricter = FALSE;
+ break;
+ }
+ }
+ }
+
+ if (stricter) {
+
+ /* A number may have a leading '+' or '-'. The latter is retained
+ * */
+ if (name[i] == '+') {
+ i++;
+ }
+ else if (name[i] == '-') {
+ lookup_name[j++] = '-';
+ i++;
+ }
+
+ /* Skip leading zeros including single underscores separating the
+ * zeros, or between the final leading zero and the first other
+ * digit */
+ for (; i < name_len - 1; i++) {
+ if ( name[i] != '0'
+ && (name[i] != '_' || ! isDIGIT_A(name[i+1])))
+ {
+ break;
+ }
+ }
+ }
+ }
+ else { /* No '=' */
+
+ /* We are now in a position to determine if this property should have
+ * been parsed using stricter rules. Only a few are like that, and
+ * unlikely to change. */
+ if ( memBEGINPs(lookup_name, j, "perl")
+ && memNEs(lookup_name + 4, j - 4, "space")
+ && memNEs(lookup_name + 4, j - 4, "word"))
+ {
+ stricter = TRUE;
+
+ /* We set the inputs back to 0 and the code below will reparse,
+ * using strict */
+ i = j = 0;
+ }
+ }
+
+ /* Here, we have either finished the property, or are positioned to parse
+ * the remainder, and we know if stricter rules apply. Finish out, if not
+ * already done */
+ for (; i < name_len; i++) {
+ char cur = name[i];
+
+ /* In all instances, case differences are ignored, and we normalize to
+ * lowercase */
+ if (isUPPER_A(cur)) {
+ lookup_name[j++] = toLOWER(cur);
+ continue;
+ }
+
+ /* An underscore is skipped, but not under strict rules unless it
+ * separates two digits */
+ if (cur == '_') {
+ if ( stricter
+ && ( i == 0 || (int) i == equals_pos || i == name_len- 1
+ || ! isDIGIT_A(name[i-1]) || ! isDIGIT_A(name[i+1])))
+ {
+ lookup_name[j++] = '_';
+ }
+ continue;
+ }
+
+ /* Hyphens are skipped except under strict */
+ if (cur == '-' && ! stricter) {
+ continue;
+ }
+
+ /* XXX Bug in documentation. It says white space skipped adjacent to
+ * non-word char. Maybe we should, but shouldn't skip it next to a dot
+ * in a number */
+ if (isSPACE_A(cur) && ! stricter) {
+ continue;
+ }
+
+ lookup_name[j++] = cur;
+
+ /* Unless this is a non-trailing slash, we are done with it */
+ if (i >= name_len - 1 || cur != '/') {
+ continue;
+ }
+
+ slash_pos = j;
+
+ /* A slash in the 'numeric value' property indicates that what follows
+ * is a denominator. It can have a leading '+' and '0's that should be
+ * skipped. But we have never allowed a negative denominator, so treat
+ * a minus like every other character. (No need to rule out a second
+ * '/', as that won't match anything anyway */
+ if (is_nv_type) {
+ i++;
+ if (i < name_len && name[i] == '+') {
+ i++;
+ }
+
+ /* Skip leading zeros including underscores separating digits */
+ for (; i < name_len - 1; i++) {
+ if ( name[i] != '0'
+ && (name[i] != '_' || ! isDIGIT_A(name[i+1])))
+ {
+ break;
+ }
+ }
+
+ /* Store the first real character in the denominator */
+ lookup_name[j++] = name[i];
+ }
+ }
+
+ /* Here are completely done parsing the input 'name', and 'lookup_name'
+ * contains a copy, normalized.
+ *
+ * This special case is grandfathered in: 'L_' and 'GC=L_' are accepted and
+ * different from without the underscores. */
+ if ( ( UNLIKELY(memEQs(lookup_name, j, "l"))
+ || UNLIKELY(memEQs(lookup_name, j, "gc=l")))
+ && UNLIKELY(name[name_len-1] == '_'))
+ {
+ lookup_name[j++] = '&';
+ }
+ else if (name_len > 2 && name[0] == 'I' && ( name[1] == 'n'
+ || name[1] == 's'))
+ {
+
+ /* Also, if the original input began with 'In' or 'Is', it could be a
+ * subroutine call instead of a property names, which currently isn't
+ * handled by this function. Subroutine calls can't happen if there is
+ * an '=' in the name */
+ if (equals_pos < 0 && get_cvn_flags(name, name_len, GV_NOTQUAL) != NULL)
+ {
+ return NULL;
+ }
+
+ starts_with_In_or_Is = TRUE;
+ }
+
+ lookup_len = j; /* Use a more mnemonic name starting here */
+
+ /* Get the index into our pointer table of the inversion list corresponding
+ * to the property */
+ table_index = match_uniprop((U8 *) lookup_name, lookup_len);
+
+ /* If it didn't find the property */
+ if (table_index == 0) {
+
+ /* If didn't find the property, we try again stripping off any initial
+ * 'In' or 'Is' */
+ if (starts_with_In_or_Is) {
+ lookup_name += 2;
+ lookup_len -= 2;
+ equals_pos -= 2;
+ slash_pos -= 2;
+
+ table_index = match_uniprop((U8 *) lookup_name, lookup_len);
+ }
+
+ if (table_index == 0) {
+ char * canonical;
+
+ /* If not found, and not a numeric type property, isn't a legal
+ * property */
+ if (! is_nv_type) {
+ return NULL;
+ }
+
+ /* But the numeric type properties need more work to decide. What
+ * we do is make sure we have the number in canonical form and look
+ * that up. */
+
+ if (slash_pos < 0) { /* No slash */
+
+ /* When it isn't a rational, take the input, convert it to a
+ * NV, then create a canonical string representation of that
+ * NV. */
+
+ NV value;
+
+ /* Get the value */
+ if (my_atof3(lookup_name + equals_pos, &value,
+ lookup_len - equals_pos)
+ != lookup_name + lookup_len)
+ {
+ return NULL;
+ }
+
+ /* If the value is an integer, the canonical value is integral */
+ if (Perl_ceil(value) == value) {
+ canonical = Perl_form(aTHX_ "%.*s%.0" NVff,
+ equals_pos, lookup_name, value);
+ }
+ else { /* Otherwise, it is %e with a known precision */
+ char * exp_ptr;
+
+ canonical = Perl_form(aTHX_ "%.*s%.*" NVef,
+ equals_pos, lookup_name,
+ PL_E_FORMAT_PRECISION, value);
+
+ /* The exponent generated is expecting two digits, whereas
+ * %e on some systems will generate three. Remove leading
+ * zeros in excess of 2 from the exponent. We start
+ * looking for them after the '=' */
+ exp_ptr = strchr(canonical + equals_pos, 'e');
+ if (exp_ptr) {
+ char * cur_ptr = exp_ptr + 2; /* past the 'e[+-]' */
+ SSize_t excess_exponent_len = strlen(cur_ptr) - 2;
+
+ assert(*(cur_ptr - 1) == '-' || *(cur_ptr - 1) == '+');
+
+ if (excess_exponent_len > 0) {
+ SSize_t leading_zeros = strspn(cur_ptr, "0");
+ SSize_t excess_leading_zeros
+ = MIN(leading_zeros, excess_exponent_len);
+ if (excess_leading_zeros > 0) {
+ Move(cur_ptr + excess_leading_zeros,
+ cur_ptr,
+ strlen(cur_ptr) - excess_leading_zeros
+ + 1, /* Copy the NUL as well */
+ char);
+ }
+ }
+ }
+ }
+ }
+ else { /* Has a slash. Create a rational in canonical form */
+ UV numerator, denominator, gcd, trial;
+ const char * end_ptr;
+ const char * sign = "";
+
+ /* We can't just find the numerator, denominator, and do the
+ * division, then use the method above, because that is
+ * inexact. And the input could be a rational that is within
+ * epsilon (given our precision) of a valid rational, and would
+ * then incorrectly compare valid.
+ *
+ * We're only interested in the part after the '=' */
+ const char * this_lookup_name = lookup_name + equals_pos;
+ lookup_len -= equals_pos;
+ slash_pos -= equals_pos;
+
+ /* Handle any leading minus */
+ if (this_lookup_name[0] == '-') {
+ sign = "-";
+ this_lookup_name++;
+ lookup_len--;
+ slash_pos--;
+ }
+
+ /* Convert the numerator to numeric */
+ end_ptr = this_lookup_name + slash_pos;
+ if (! grok_atoUV(this_lookup_name, &numerator, &end_ptr)) {
+ return NULL;
+ }
+
+ /* It better have included all characters before the slash */
+ if (*end_ptr != '/') {
+ return NULL;
+ }
+
+ /* Set to look at just the denominator */
+ this_lookup_name += slash_pos;
+ lookup_len -= slash_pos;
+ end_ptr = this_lookup_name + lookup_len;
+
+ /* Convert the denominator to numeric */
+ if (! grok_atoUV(this_lookup_name, &denominator, &end_ptr)) {
+ return NULL;
+ }
+
+ /* It better be the rest of the characters, and don't divide by
+ * 0 */
+ if ( end_ptr != this_lookup_name + lookup_len
+ || denominator == 0)
+ {
+ return NULL;
+ }
+
+ /* Get the greatest common denominator using
+ http://en.wikipedia.org/wiki/Euclidean_algorithm */
+ gcd = numerator;
+ trial = denominator;
+ while (trial != 0) {
+ UV temp = trial;
+ trial = gcd % trial;
+ gcd = temp;
+ }
+
+ /* If already in lowest possible terms, we have already tried
+ * looking this up */
+ if (gcd == 1) {
+ return NULL;
+ }
+
+ /* Reduce the rational, which should put it in canonical form.
+ * Then look it up */
+ numerator /= gcd;
+ denominator /= gcd;
+
+ canonical = Perl_form(aTHX_ "%.*s%s%" UVuf "/%" UVuf,
+ equals_pos, lookup_name, sign, numerator, denominator);
+ }
+
+ /* Here, we have the number in canonical form. Try that */
+ table_index = match_uniprop((U8 *) canonical, strlen(canonical));
+ if (table_index == 0) {
+ return NULL;
+ }
+ }
+ }
+
+ /* The return is an index into a table of ptrs. A negative return
+ * signifies that the real index is the absolute value, but the result
+ * needs to be inverted */
+ if (table_index < 0) {
+ *invert = TRUE;
+ table_index = -table_index;
+ }
+ else {
+ *invert = FALSE;
+ }
+
+ /* Out-of band indices indicate a deprecated property. The proper index is
+ * modulo it with the table size. And dividing by the table size yields
+ * an offset into a table constructed to contain the corresponding warning
+ * message */
+ if (table_index > MAX_UNI_KEYWORD_INDEX) {
+ Size_t warning_offset = table_index / MAX_UNI_KEYWORD_INDEX;
+ table_index %= MAX_UNI_KEYWORD_INDEX;
+ Perl_ck_warner_d(aTHX_ packWARN(WARN_DEPRECATED),
+ "Use of '%.*s' in \\p{} or \\P{} is deprecated because: %s",
+ (int) name_len, name, deprecated_property_msgs[warning_offset]);
+ }
+
+ /* In a few properties, a different property is used under /i. These are
+ * unlikely to change, so are hard-coded here. */
+ if (to_fold) {
+ if ( table_index == UNI_XPOSIXUPPER
+ || table_index == UNI_XPOSIXLOWER
+ || table_index == UNI_TITLE)
+ {
+ table_index = UNI_CASED;
+ }
+ else if ( table_index == UNI_UPPERCASELETTER
+ || table_index == UNI_LOWERCASELETTER
+# ifdef UNI_TITLECASELETTER /* Missing from early Unicodes */
+ || table_index == UNI_TITLECASELETTER
+# endif
+ ) {
+ table_index = UNI_CASEDLETTER;
+ }
+ else if ( table_index == UNI_POSIXUPPER
+ || table_index == UNI_POSIXLOWER)
+ {
+ table_index = UNI_POSIXALPHA;
+ }
+ }
+
+ /* Create and return the inversion list */
+ return _new_invlist_C_array(uni_prop_ptrs[table_index]);
+}
+
+#endif
+
/*
* ex: set ts=8 sts=4 sw=4 et:
*/
https://perl5.git.perl.org / perl5.git / blobdiff