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))
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) ) \
(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_XPosix_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);
+ _invlist_intersection(PL_XPosix_ptrs[FLAGS(scan)],
+ PL_XPosix_ptrs[_CC_ASCII],
+ &my_invlist);
+ 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);
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;
/* 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);
}
c = SvUV(*c_p);
/* /aa doesn't allow folds between ASCII and non- */
- if ((OP(node) == EXACTFA || OP(node) == EXACTFA_NO_TRIE)
+ if ((OP(node) == EXACTFAA || OP(node) == EXACTFAA_NO_TRIE)
&& isASCII(c) != isASCII(uc))
{
continue;
* 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 ( *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, "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"))
+ {
+ paren = 's';
+
+ /* This indicates Unicode rules. */
+ REQUIRE_UNI_RULES(flagp, NULL);
+
+ if (! start_arg) {
+ goto no_colon;
+ }
+
+ RExC_parse = start_arg;
+
+ 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 (RExC_in_script_run) {
+ paren = ':';
+ ret = NULL;
+ goto parse_rest;
+ }
+ RExC_in_script_run = 1;
+
+ ret = reg_node(pRExC_state, SROPEN);
+
+ is_open = 1;
+ goto parse_rest;
+ }
+
+ 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++;
{
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] == '<' ? '>' : '\'';
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;
SvREFCNT_dec_NN(substitute_parse);
if (! *node_p) {
- if (flags & (RESTART_PASS1|NEED_UTF8)) {
- *flagp = flags & (RESTART_PASS1|NEED_UTF8);
- return FALSE;
- }
+ RETURN_X_ON_RESTART(FALSE, flags,flagp);
FAIL2("panic: reg returned NULL to grok_bslash_N, flags=%#" UVxf,
(UV) flags);
}
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;
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);
-
- /* 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;
+ bool maybe_exactfu = PASS2;
+ /* 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)));
/* 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
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;
+ * 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
* 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 {
+ 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;
}
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
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;
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{...}");
{
/* 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
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;
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)) {
#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;
https://perl5.git.perl.org / perl5.git / blobdiff