/* Certain characters are output as a sequence with the first being a
* backslash. */
-#define isBACKSLASHED_PUNCT(c) strchr("-[]\\^", c)
+#define isBACKSLASHED_PUNCT(c) memCHRs("-[]\\^", c)
struct RExC_state_t {
#define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
#define PBITVAL(paren) (1 << ((paren) & 7))
-#define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
-#define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
-#define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
+#define PAREN_OFFSET(depth) \
+ (RExC_study_chunk_recursed + (depth) * RExC_study_chunk_recursed_bytes)
+#define PAREN_TEST(depth, paren) \
+ (PBYTE(PAREN_OFFSET(depth), paren) & PBITVAL(paren))
+#define PAREN_SET(depth, paren) \
+ (PBYTE(PAREN_OFFSET(depth), paren) |= PBITVAL(paren))
+#define PAREN_UNSET(depth, paren) \
+ (PBYTE(PAREN_OFFSET(depth), paren) &= ~PBITVAL(paren))
#define REQUIRE_UTF8(flagp) STMT_START { \
if (!UTF) { \
/* Change from /d into /u rules, and restart the parse. RExC_uni_semantics is
* a flag that indicates we need to override /d with /u as a result of
* something in the pattern. It should only be used in regards to calling
- * set_regex_charset() or get_regex_charse() */
+ * set_regex_charset() or get_regex_charset() */
#define REQUIRE_UNI_RULES(flagp, restart_retval) \
STMT_START { \
if (DEPENDS_SEMANTICS) { \
#define _invlist_intersection_complement_2nd(a, b, output) \
_invlist_intersection_maybe_complement_2nd(a, b, TRUE, output)
+/* We add a marker if we are deferring expansion of a property that is both
+ * 1) potentiallly user-defined; and
+ * 2) could also be an official Unicode property.
+ *
+ * Without this marker, any deferred expansion can only be for a user-defined
+ * one. This marker shouldn't conflict with any that could be in a legal name,
+ * and is appended to its name to indicate this. There is a string and
+ * character form */
+#define DEFERRED_COULD_BE_OFFICIAL_MARKERs "~"
+#define DEFERRED_COULD_BE_OFFICIAL_MARKERc '~'
+
/* About scan_data_t.
During optimisation we recurse through the regexp program performing
} STMT_END
/* m is not necessarily a "literal string", in this macro */
-#define reg_warn_non_literal_string(loc, m) \
- _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
- Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
+#define warn_non_literal_string(loc, packed_warn, m) \
+ _WARN_HELPER(loc, packed_warn, \
+ Perl_warner(aTHX_ packed_warn, \
"%s" REPORT_LOCATION, \
m, REPORT_LOCATION_ARGS(loc)))
+#define reg_warn_non_literal_string(loc, m) \
+ warn_non_literal_string(loc, packWARN(WARN_REGEXP), m)
+
+#define ckWARN2_non_literal_string(loc, packwarn, m, a1) \
+ STMT_START { \
+ char * format; \
+ Size_t format_size = strlen(m) + strlen(REPORT_LOCATION)+ 1;\
+ Newx(format, format_size, char); \
+ my_strlcpy(format, m, format_size); \
+ my_strlcat(format, REPORT_LOCATION, format_size); \
+ SAVEFREEPV(format); \
+ _WARN_HELPER(loc, packwarn, \
+ Perl_ck_warner(aTHX_ packwarn, \
+ format, \
+ a1, REPORT_LOCATION_ARGS(loc))); \
+ } STMT_END
#define ckWARNreg(loc,m) \
_WARN_HELPER(loc, packWARN(WARN_REGEXP), \
unsigned int i;
const U32 n = ARG(node);
bool new_node_has_latin1 = FALSE;
- const U8 flags = (inRANGE(OP(node), ANYOFH, ANYOFHr))
+ const U8 flags = (inRANGE(OP(node), ANYOFH, ANYOFRb))
? 0
: ANYOF_FLAGS(node);
}
/* Add in the points from the bit map */
- if (! inRANGE(OP(node), ANYOFH, ANYOFHr)) {
+ if (! inRANGE(OP(node), ANYOFH, ANYOFRb)) {
for (i = 0; i < NUM_ANYOF_CODE_POINTS; i++) {
if (ANYOF_BITMAP_TEST(node, i)) {
unsigned int start = i++;
* another SSC or a regular ANYOF class. Can create false positives. */
SV* anded_cp_list;
- U8 and_with_flags = inRANGE(OP(and_with), ANYOFH, ANYOFHr)
+ U8 and_with_flags = inRANGE(OP(and_with), ANYOFH, ANYOFRb)
? 0
: ANYOF_FLAGS(and_with);
U8 anded_flags;
SV* ored_cp_list;
U8 ored_flags;
- U8 or_with_flags = inRANGE(OP(or_with), ANYOFH, ANYOFHr)
+ U8 or_with_flags = inRANGE(OP(or_with), ANYOFH, ANYOFRb)
? 0
: ANYOF_FLAGS(or_with);
populate_ANYOF_from_invlist( (regnode *) ssc, &invlist);
set_ANYOF_arg(pRExC_state, (regnode *) ssc, invlist, NULL, NULL);
+ SvREFCNT_dec(invlist);
/* Make sure is clone-safe */
ssc->invlist = NULL;
if (OP(noper) == NOTHING) {
/* skip past a NOTHING at the start of an alternation
* eg, /(?:)a|(?:b)/ should be the same as /a|b/
+ *
+ * If the next node is not something we are supposed to process
+ * we will just ignore it due to the condition guarding the
+ * next block.
*/
+
regnode *noper_next= regnext(noper);
if (noper_next < tail)
noper= noper_next;
regnode *noper_next= regnext(noper);
if (noper_next < tail)
noper= noper_next;
+ /* we will undo this assignment if noper does not
+ * point at a trieable type in the else clause of
+ * the following statement. */
}
if ( noper < tail
Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %" IVdf, uvc );
}
}
- }
+ } else {
+ /* If we end up here it is because we skipped past a NOTHING, but did not end up
+ * on a trieable type. So we need to reset noper back to point at the first regop
+ * in the branch before we call TRIE_HANDLE_WORD()
+ */
+ noper= NEXTOPER(cur);
+ }
TRIE_HANDLE_WORD(state);
} /* end second pass */
regnode *noper_next= regnext(noper);
if (noper_next < tail)
noper= noper_next;
+ /* we will undo this assignment if noper does not
+ * point at a trieable type in the else clause of
+ * the following statement. */
}
if ( noper < tail
/* charid is now 0 if we dont know the char read, or
* nonzero if we do */
}
+ } else {
+ /* If we end up here it is because we skipped past a NOTHING, but did not end up
+ * on a trieable type. So we need to reset noper back to point at the first regop
+ * in the branch before we call TRIE_HANDLE_WORD().
+ */
+ noper= NEXTOPER(cur);
}
accept_state = TRIE_NODENUM( state );
TRIE_HANDLE_WORD(accept_state);
* using /iaa matching will be doing so almost entirely with ASCII
* strings, so this should rarely be encountered in practice */
-#define JOIN_EXACT(scan,min_subtract,unfolded_multi_char, flags) \
- if (PL_regkind[OP(scan)] == EXACT && OP(scan) != LEXACT \
- && OP(scan) != LEXACT_REQ8) \
- join_exact(pRExC_state,(scan),(min_subtract),unfolded_multi_char, (flags), NULL, depth+1)
-
STATIC U32
S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan,
UV *min_subtract, bool *unfolded_multi_char,
}
#endif
}
-
- if ( STR_LEN(scan) == 1
- && isALPHA_A(* STRING(scan))
- && ( OP(scan) == EXACTFAA
- || ( OP(scan) == EXACTFU
- && ! HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(* STRING(scan)))))
- {
- U8 mask = ~ ('A' ^ 'a'); /* These differ in just one bit */
-
- /* Replace a length 1 ASCII fold pair node with an ANYOFM node,
- * with the mask set to the complement of the bit that differs
- * between upper and lower case, and the lowest code point of the
- * pair (which the '&' forces) */
- OP(scan) = ANYOFM;
- ARG_SET(scan, *STRING(scan) & mask);
- FLAGS(scan) = mask;
- }
}
#ifdef DEBUGGING
/* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
{
dVAR;
+ SSize_t final_minlen;
/* There must be at least this number of characters to match */
SSize_t min = 0;
I32 pars = 0, code;
U32 j;
for ( j = 0 ; j < recursed_depth ; j++ ) {
for ( i = 0 ; i < (U32)RExC_total_parens ; i++ ) {
- if (
- PAREN_TEST(RExC_study_chunk_recursed +
- ( j * RExC_study_chunk_recursed_bytes), i )
- && (
- !j ||
- !PAREN_TEST(RExC_study_chunk_recursed +
- (( j - 1 ) * RExC_study_chunk_recursed_bytes), i)
- )
- ) {
+ if (PAREN_TEST(j, i) && (!j || !PAREN_TEST(j - 1, i))) {
Perl_re_printf( aTHX_ " %d",(int)i);
break;
}
* parsing code, as each (?:..) is handled by a different invocation of
* reg() -- Yves
*/
- JOIN_EXACT(scan,&min_subtract, &unfolded_multi_char, 0);
+ if (PL_regkind[OP(scan)] == EXACT && OP(scan) != LEXACT
+ && OP(scan) != LEXACT_REQ8)
+ join_exact(pRExC_state, scan, &min_subtract, &unfolded_multi_char,
+ 0, NULL, depth + 1);
/* Follow the next-chain of the current node and optimize
away all the NOTHINGs from it. */
if ( SvIV(re_trie_maxbuff)>=0 ) {
regnode *cur;
regnode *first = (regnode *)NULL;
- regnode *last = (regnode *)NULL;
+ regnode *prev = (regnode *)NULL;
regnode *tail = scan;
U8 trietype = 0;
U32 count=0;
REG_NODE_NUM(noper_next), SvPV_nolen_const(RExC_mysv));
}
Perl_re_printf( aTHX_ "(First==%d,Last==%d,Cur==%d,tt==%s,ntt==%s,nntt==%s)\n",
- REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur),
+ REG_NODE_NUM(first), REG_NODE_NUM(prev), REG_NODE_NUM(cur),
PL_reg_name[trietype], PL_reg_name[noper_trietype], PL_reg_name[noper_next_trietype]
);
});
} else {
if ( trietype == NOTHING )
trietype = noper_trietype;
- last = cur;
+ prev = cur;
}
if (first)
count++;
* noper may either be a triable node which can
* not be tried together with the current trie,
* or a non triable node */
- if ( last ) {
+ if ( prev ) {
/* If last is set and trietype is not
* NOTHING then we have found at least two
* triable branch sequences in a row of a
make_trie( pRExC_state,
startbranch, first, cur, tail,
count, trietype, depth+1 );
- last = NULL; /* note: we clear/update
+ prev = NULL; /* note: we clear/update
first, trietype etc below,
so we dont do it here */
}
Perl_re_indentf( aTHX_ "- %s (%d) <SCAN FINISHED> ",
depth+1, SvPV_nolen_const( RExC_mysv ), REG_NODE_NUM(cur));
Perl_re_printf( aTHX_ "(First==%d, Last==%d, Cur==%d, tt==%s)\n",
- REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur),
+ REG_NODE_NUM(first), REG_NODE_NUM(prev), REG_NODE_NUM(cur),
PL_reg_name[trietype]
);
});
- if ( last && trietype ) {
+ if ( prev && trietype ) {
if ( trietype != NOTHING ) {
/* the last branch of the sequence was part of
* a trie, so we have to construct it here
OP(opt)= OPTIMIZED;
}
}
- } /* end if ( last) */
+ } /* end if ( prev) */
} /* TRIE_MAXBUF is non zero */
-
} /* do trie */
}
if (
!recursed_depth
- ||
- !PAREN_TEST(RExC_study_chunk_recursed + ((recursed_depth-1) * RExC_study_chunk_recursed_bytes), paren)
+ || !PAREN_TEST(recursed_depth - 1, paren)
) {
/* it is quite possible that there are more efficient ways
* to do this. We maintain a bitmap per level of recursion
if (!recursed_depth) {
Zero(RExC_study_chunk_recursed, RExC_study_chunk_recursed_bytes, U8);
} else {
- Copy(RExC_study_chunk_recursed + ((recursed_depth-1) * RExC_study_chunk_recursed_bytes),
- RExC_study_chunk_recursed + (recursed_depth * RExC_study_chunk_recursed_bytes),
+ Copy(PAREN_OFFSET(recursed_depth - 1),
+ PAREN_OFFSET(recursed_depth),
RExC_study_chunk_recursed_bytes, U8);
}
/* we havent recursed into this paren yet, so recurse into it */
DEBUG_STUDYDATA("gosub-set", data, depth, is_inf);
- PAREN_SET(RExC_study_chunk_recursed + (recursed_depth * RExC_study_chunk_recursed_bytes), paren);
+ PAREN_SET(recursed_depth, paren);
my_recursed_depth= recursed_depth + 1;
} else {
DEBUG_STUDYDATA("gosub-inf", data, depth, is_inf);
|| OP(scan) == LEXACT_REQ8
|| OP(scan) == EXACTL)
{
- SSize_t l = STR_LEN(scan);
+ SSize_t bytelen = STR_LEN(scan), charlen;
UV uc;
- assert(l);
+ assert(bytelen);
if (UTF) {
const U8 * const s = (U8*)STRING(scan);
- uc = utf8_to_uvchr_buf(s, s + l, NULL);
- l = utf8_length(s, s + l);
+ uc = utf8_to_uvchr_buf(s, s + bytelen, NULL);
+ charlen = utf8_length(s, s + bytelen);
} else {
uc = *((U8*)STRING(scan));
+ charlen = bytelen;
}
- min += l;
+ min += charlen;
if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
/* The code below prefers earlier match for fixed
offset, later match for variable offset. */
data->last_start_max = is_inf
? SSize_t_MAX : data->pos_min + data->pos_delta;
}
- sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
+ sv_catpvn(data->last_found, STRING(scan), bytelen);
if (UTF)
SvUTF8_on(data->last_found);
{
MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
mg_find(sv, PERL_MAGIC_utf8) : NULL;
if (mg && mg->mg_len >= 0)
- mg->mg_len += utf8_length((U8*)STRING(scan),
- (U8*)STRING(scan)+STR_LEN(scan));
+ mg->mg_len += charlen;
}
- data->last_end = data->pos_min + l;
- data->pos_min += l; /* As in the first entry. */
+ data->last_end = data->pos_min + charlen;
+ data->pos_min += charlen; /* As in the first entry. */
data->flags &= ~SF_BEFORE_EOL;
}
}
else if (PL_regkind[OP(scan)] == EXACT) {
/* But OP != EXACT!, so is EXACTFish */
- SSize_t l = STR_LEN(scan);
+ SSize_t bytelen = STR_LEN(scan), charlen;
const U8 * s = (U8*)STRING(scan);
+ /* Replace a length 1 ASCII fold pair node with an ANYOFM node,
+ * with the mask set to the complement of the bit that differs
+ * between upper and lower case, and the lowest code point of the
+ * pair (which the '&' forces) */
+ if ( bytelen == 1
+ && isALPHA_A(*s)
+ && ( OP(scan) == EXACTFAA
+ || ( OP(scan) == EXACTFU
+ && ! HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(*s))))
+ {
+ U8 mask = ~ ('A' ^ 'a'); /* These differ in just one bit */
+
+ OP(scan) = ANYOFM;
+ ARG_SET(scan, *s & mask);
+ FLAGS(scan) = mask;
+ /* we're not EXACTFish any more, so restudy */
+ continue;
+ }
+
/* Search for fixed substrings supports EXACT only. */
if (flags & SCF_DO_SUBSTR) {
assert(data);
scan_commit(pRExC_state, data, minlenp, is_inf);
}
- if (UTF) {
- l = utf8_length(s, s + l);
- }
+ charlen = UTF ? (SSize_t) utf8_length(s, s + bytelen) : bytelen;
if (unfolded_multi_char) {
RExC_seen |= REG_UNFOLDED_MULTI_SEEN;
}
- min += l - min_subtract;
+ min += charlen - min_subtract;
assert (min >= 0);
delta += min_subtract;
if (flags & SCF_DO_SUBSTR) {
- data->pos_min += l - min_subtract;
+ data->pos_min += charlen - min_subtract;
if (data->pos_min < 0) {
data->pos_min = 0;
}
continue;
default:
-#ifdef DEBUGGING
Perl_croak(aTHX_ "panic: unexpected varying REx opcode %d",
OP(scan));
-#endif
case REF:
case CLUMP:
if (flags & SCF_DO_SUBSTR) {
case ANYOFH:
case ANYOFHb:
case ANYOFHr:
+ case ANYOFHs:
case ANYOF:
if (flags & SCF_DO_STCLASS_AND)
ssc_and(pRExC_state, data->start_class,
break;
}
+ case ANYOFR:
+ case ANYOFRb:
+ {
+ SV* cp_list = NULL;
+
+ cp_list = _add_range_to_invlist(cp_list,
+ ANYOFRbase(scan),
+ ANYOFRbase(scan) + ANYOFRdelta(scan));
+
+ if (flags & SCF_DO_STCLASS_OR) {
+ ssc_union(data->start_class, cp_list, invert);
+ }
+ else if (flags & SCF_DO_STCLASS_AND) {
+ ssc_intersection(data->start_class, cp_list, invert);
+ }
+
+ SvREFCNT_dec_NN(cp_list);
+ break;
+ }
+
case NPOSIXL:
invert = 1;
/* FALLTHROUGH */
DEBUG_STUDYDATA("post-fin", data, depth, is_inf);
- {
- SSize_t final_minlen= min < stopmin ? min : stopmin;
-
- if (!(RExC_seen & REG_UNBOUNDED_QUANTIFIER_SEEN)) {
- if (final_minlen > SSize_t_MAX - delta)
- RExC_maxlen = SSize_t_MAX;
- else if (RExC_maxlen < final_minlen + delta)
- RExC_maxlen = final_minlen + delta;
- }
- return final_minlen;
+ final_minlen = min < stopmin
+ ? min : stopmin;
+ if (!(RExC_seen & REG_UNBOUNDED_QUANTIFIER_SEEN)) {
+ if (final_minlen > SSize_t_MAX - delta)
+ RExC_maxlen = SSize_t_MAX;
+ else if (RExC_maxlen < final_minlen + delta)
+ RExC_maxlen = final_minlen + delta;
}
- NOT_REACHED; /* NOTREACHED */
+ return final_minlen;
}
STATIC U32
DEBUG_r(if (!PL_colorset) reginitcolors());
- /* Initialize these here instead of as-needed, as is quick and avoids
- * having to test them each time otherwise */
- if (! PL_InBitmap) {
-#ifdef DEBUGGING
- char * dump_len_string;
-#endif
-
- /* This is calculated here, because the Perl program that generates the
- * static global ones doesn't currently have access to
- * NUM_ANYOF_CODE_POINTS */
- PL_InBitmap = _new_invlist(2);
- PL_InBitmap = _add_range_to_invlist(PL_InBitmap, 0,
- NUM_ANYOF_CODE_POINTS - 1);
-#ifdef DEBUGGING
- dump_len_string = PerlEnv_getenv("PERL_DUMP_RE_MAX_LEN");
- if ( ! dump_len_string
- || ! grok_atoUV(dump_len_string, (UV *)&PL_dump_re_max_len, NULL))
- {
- PL_dump_re_max_len = 60; /* A reasonable default */
- }
-#endif
- }
pRExC_state->warn_text = NULL;
pRExC_state->unlexed_names = NULL;
i = rx->sublen + rx->suboffset - rx->offs[0].end;
}
else
- if ( 0 <= n && n <= (I32)rx->nparens &&
- (s1 = rx->offs[n].start) != -1 &&
+ if (inRANGE(n, 0, (I32)rx->nparens) &&
+ (s1 = rx->offs[n].start) != -1 &&
(t1 = rx->offs[n].end) != -1)
{
/* $&, ${^MATCH}, $1 ... */
i = t1 - s1;
if (is_utf8_string_loclen((U8*)s, i, &ep, &el))
- i = el;
+ i = el;
}
return i;
}
#endif
+PERL_STATIC_INLINE UV
+S_invlist_lowest(SV* const invlist)
+{
+ /* Returns the lowest code point that matches an inversion list. This API
+ * has an ambiguity, as it returns 0 under either the lowest is actually
+ * 0, or if the list is empty. If this distinction matters to you, check
+ * for emptiness before calling this function */
+
+ UV len = _invlist_len(invlist);
+ UV *array;
+
+ PERL_ARGS_ASSERT_INVLIST_LOWEST;
+
+ if (len == 0) {
+ return 0;
+ }
+
+ array = invlist_array(invlist);
+
+ return array[0];
+}
+
STATIC SV *
S_invlist_contents(pTHX_ SV* const invlist, const bool traditional_style)
{
}
else { /* Single char fold */
unsigned int k;
- unsigned int first_fold;
- const unsigned int * remaining_folds;
+ U32 first_fold;
+ const U32 * remaining_folds;
Size_t folds_count;
/* It matches itself */
}
while (RExC_parse < RExC_end) {
- /* && strchr("iogcmsx", *RExC_parse) */
+ /* && memCHRs("iogcmsx", *RExC_parse) */
/* (?g), (?gc) and (?o) are useless here
and must be globally applied -- japhy */
switch (*RExC_parse) {
RExC_parse-seqstart, seqstart);
NOT_REACHED; /*NOTREACHED*/
case '<': /* (?<...) */
+ /* If you want to support (?<*...), first reconcile with GH #17363 */
if (*RExC_parse == '!')
paren = ',';
else if (*RExC_parse != '=')
ckWARNreg(RExC_parse, "Empty (?) without any modifiers");
}
/* FALLTHROUGH */
+ case '*': /* If you want to support (?*...), first reconcile with GH #17363 */
+ /* FALLTHROUGH */
default: /* e.g., (?i) */
RExC_parse = (char *) seqstart + 1;
parse_flags:
value = (U8 *) SvPV(value_sv, value_len);
/* See if the result is one code point vs 0 or multiple */
- if (value_len > 0 && value_len <= (UV) ((SvUTF8(value_sv))
- ? UTF8SKIP(value)
- : 1))
+ if (inRANGE(value_len, 1, ((UV) SvUTF8(value_sv)
+ ? UTF8SKIP(value)
+ : 1)))
{
/* Here, exactly one code point. If that isn't what is wanted,
* fail */
|| ! is_PATWS_safe((p), RExC_end, UTF));
switch ((U8)*p) {
+ const char* message;
+ U32 packed_warn;
+ U8 grok_c_char;
+
case '^':
case '$':
case '.':
p++;
break;
case 'o':
- {
- UV result;
- const char* error_msg;
-
- bool valid = grok_bslash_o(&p,
- RExC_end,
- &result,
- &error_msg,
- TO_OUTPUT_WARNINGS(p),
- (bool) RExC_strict,
- TRUE, /* Output warnings
- for non-
- portables */
- UTF);
- if (! valid) {
- RExC_parse = p; /* going to die anyway; point
- to exact spot of failure */
- vFAIL(error_msg);
- }
- UPDATE_WARNINGS_LOC(p - 1);
- ender = result;
- break;
- }
+ if (! grok_bslash_o(&p,
+ RExC_end,
+ &ender,
+ &message,
+ &packed_warn,
+ (bool) RExC_strict,
+ FALSE, /* No illegal cp's */
+ UTF))
+ {
+ RExC_parse = p; /* going to die anyway; point to
+ exact spot of failure */
+ vFAIL(message);
+ }
+
+ if (message && TO_OUTPUT_WARNINGS(p)) {
+ warn_non_literal_string(p, packed_warn, message);
+ }
+ break;
case 'x':
- {
- UV result = UV_MAX; /* initialize to erroneous
- value */
- const char* error_msg;
-
- bool valid = grok_bslash_x(&p,
- RExC_end,
- &result,
- &error_msg,
- TO_OUTPUT_WARNINGS(p),
- (bool) RExC_strict,
- TRUE, /* Silence warnings
- for non-
- portables */
- UTF);
- if (! valid) {
- RExC_parse = p; /* going to die anyway; point
- to exact spot of failure */
- vFAIL(error_msg);
- }
- UPDATE_WARNINGS_LOC(p - 1);
- ender = result;
+ if (! grok_bslash_x(&p,
+ RExC_end,
+ &ender,
+ &message,
+ &packed_warn,
+ (bool) RExC_strict,
+ FALSE, /* No illegal cp's */
+ UTF))
+ {
+ RExC_parse = p; /* going to die anyway; point
+ to exact spot of failure */
+ vFAIL(message);
+ }
+
+ if (message && TO_OUTPUT_WARNINGS(p)) {
+ warn_non_literal_string(p, packed_warn, message);
+ }
#ifdef EBCDIC
- if (ender < 0x100) {
- if (RExC_recode_x_to_native) {
- ender = LATIN1_TO_NATIVE(ender);
- }
- }
+ if (ender < 0x100) {
+ if (RExC_recode_x_to_native) {
+ ender = LATIN1_TO_NATIVE(ender);
+ }
+ }
#endif
- break;
- }
+ break;
case 'c':
- p++;
- ender = grok_bslash_c(*p, TO_OUTPUT_WARNINGS(p));
- UPDATE_WARNINGS_LOC(p);
p++;
+ if (! grok_bslash_c(*p, &grok_c_char,
+ &message, &packed_warn))
+ {
+ /* going to die anyway; point to exact spot of
+ * failure */
+ RExC_parse = p + ((UTF)
+ ? UTF8_SAFE_SKIP(p, RExC_end)
+ : 1);
+ vFAIL(message);
+ }
+
+ ender = grok_c_char;
+ p++;
+ if (message && TO_OUTPUT_WARNINGS(p)) {
+ warn_non_literal_string(p, packed_warn, message);
+ }
+
break;
case '8': case '9': /* must be a backreference */
--p;
/* FALLTHROUGH */
case '0':
{
- I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
+ I32 flags = PERL_SCAN_SILENT_ILLDIGIT
+ | PERL_SCAN_NOTIFY_ILLDIGIT;
STRLEN numlen = 3;
ender = grok_oct(p, &numlen, &flags, NULL);
p += numlen;
- if ( isDIGIT(*p) /* like \08, \178 */
- && ckWARN(WARN_REGEXP)
- && numlen < 3)
+ if ( (flags & PERL_SCAN_NOTIFY_ILLDIGIT)
+ && isDIGIT(*p) /* like \08, \178 */
+ && ckWARN(WARN_REGEXP))
{
reg_warn_non_literal_string(
- p + 1,
- form_short_octal_warning(p, numlen));
+ p + 1,
+ form_alien_digit_msg(8, numlen, p,
+ RExC_end, UTF, FALSE));
}
}
break;
if (ender > 255) {
REQUIRE_UTF8(flagp);
+ if ( UNICODE_IS_PERL_EXTENDED(ender)
+ && TO_OUTPUT_WARNINGS(p))
+ {
+ ckWARN2_non_literal_string(p,
+ packWARN(WARN_PORTABLE),
+ PL_extended_cp_format,
+ ender);
+ }
}
/* We need to check if the next non-ignored thing is a
goto continue_parse;
}
- else if (! LOC) { /* XXX shouldn't /l assume could be a UTF-8
- locale, and prepare for that? */
+ else if (FOLD) {
bool splittable = FALSE;
bool backed_up = FALSE;
- char * e = s;
-
- assert(FOLD);
+ char * e;
+ char * s_start;
/* Here is /i. Running out of room creates a problem if we are
* folding, and the split happens in the middle of a
* oldp points to the beginning byte in the input of
* 'ender'.
*
+ * In the case of /il, we haven't folded anything that could be
+ * affected by the locale. That means only above-Latin1
+ * characters that fold to other above-latin1 characters get
+ * folded at compile time. To check where a good place to
+ * split nodes is, everything in it will have to be folded.
+ * The boolean 'maybe_exactfu' keeps track in /il if there are
+ * any unfolded characters in the node. */
+ bool need_to_fold_loc = LOC && ! maybe_exactfu;
+
+ /* If we do need to fold the node, we need a place to store the
+ * folded copy, and a way to map back to the unfolded original
+ * */
+ char * locfold_buf = NULL;
+ Size_t * loc_correspondence = NULL;
+
+ if (! need_to_fold_loc) { /* The normal case. Just
+ initialize to the actual node */
+ e = s;
+ s_start = s0;
+ s = old_old_s; /* Point to the beginning of the final char
+ that fits in the node */
+ }
+ else {
+
+ /* Here, we have filled a /il node, and there are unfolded
+ * characters in it. If the runtime locale turns out to be
+ * UTF-8, there are possible multi-character folds, just
+ * like when not under /l. The node hence can't terminate
+ * in the middle of such a fold. To determine this, we
+ * have to create a folded copy of this node. That means
+ * reparsing the node, folding everything assuming a UTF-8
+ * locale. (If at runtime it isn't such a locale, the
+ * actions here wouldn't have been necessary, but we have
+ * to assume the worst case.) If we find we need to back
+ * off the folded string, we do so, and then map that
+ * position back to the original unfolded node, which then
+ * gets output, truncated at that spot */
+
+ char * redo_p = RExC_parse;
+ char * redo_e;
+ char * old_redo_e;
+
+ /* Allow enough space assuming a single byte input folds to
+ * a single byte output, plus assume that the two unparsed
+ * characters (that we may need) fold to the largest number
+ * of bytes possible, plus extra for one more worst case
+ * scenario. In the loop below, if we start eating into
+ * that final spare space, we enlarge this initial space */
+ Size_t size = max_string_len + (3 * UTF8_MAXBYTES_CASE) + 1;
+
+ Newxz(locfold_buf, size, char);
+ Newxz(loc_correspondence, size, Size_t);
+
+ /* Redo this node's parse, folding into 'locfold_buf' */
+ redo_p = RExC_parse;
+ old_redo_e = redo_e = locfold_buf;
+ while (redo_p <= oldp) {
+
+ old_redo_e = redo_e;
+ loc_correspondence[redo_e - locfold_buf]
+ = redo_p - RExC_parse;
+
+ if (UTF) {
+ Size_t added_len;
+
+ (void) _to_utf8_fold_flags((U8 *) redo_p,
+ (U8 *) RExC_end,
+ (U8 *) redo_e,
+ &added_len,
+ FOLD_FLAGS_FULL);
+ redo_e += added_len;
+ redo_p += UTF8SKIP(redo_p);
+ }
+ else {
+
+ /* Note that if this code is run on some ancient
+ * Unicode versions, SHARP S doesn't fold to 'ss',
+ * but rather than clutter the code with #ifdef's,
+ * as is done above, we ignore that possibility.
+ * This is ok because this code doesn't affect what
+ * gets matched, but merely where the node gets
+ * split */
+ if (UCHARAT(redo_p) != LATIN_SMALL_LETTER_SHARP_S) {
+ *redo_e++ = toLOWER_L1(UCHARAT(redo_p));
+ }
+ else {
+ *redo_e++ = 's';
+ *redo_e++ = 's';
+ }
+ redo_p++;
+ }
+
+
+ /* If we're getting so close to the end that a
+ * worst-case fold in the next character would cause us
+ * to overflow, increase, assuming one byte output byte
+ * per one byte input one, plus room for another worst
+ * case fold */
+ if ( redo_p <= oldp
+ && redo_e > locfold_buf + size
+ - (UTF8_MAXBYTES_CASE + 1))
+ {
+ Size_t new_size = size
+ + (oldp - redo_p)
+ + UTF8_MAXBYTES_CASE + 1;
+ Ptrdiff_t e_offset = redo_e - locfold_buf;
+
+ Renew(locfold_buf, new_size, char);
+ Renew(loc_correspondence, new_size, Size_t);
+ size = new_size;
+
+ redo_e = locfold_buf + e_offset;
+ }
+ }
+
+ /* Set so that things are in terms of the folded, temporary
+ * string */
+ s = old_redo_e;
+ s_start = locfold_buf;
+ e = redo_e;
+
+ }
+
+ /* Here, we have 's', 's_start' and 'e' set up to point to the
+ * input that goes into the node, folded.
+ *
* If the final character of the node and the fold of ender
* form the first two characters of a three character fold, we
* need to peek ahead at the next (unparsed) character in the
* and try again.
*
* Otherwise, the node can be split at the current position.
- */
- s = old_old_s; /* Point to the beginning of the final char
- that fits in the node */
-
- /* The same logic is used for UTF-8 patterns and not */
+ *
+ * The same logic is used for UTF-8 patterns and not */
if (UTF) {
Size_t added_len;
* drop down to try at that position */
if (isPUNCT(*p)) {
s = (char *) utf8_hop_back((U8 *) s, -1,
- (U8 *) s0);
+ (U8 *) s_start);
backed_up = TRUE;
}
else {
* either case would break apart a fold */
do {
char *prev_s = (char *) utf8_hop_back((U8 *) s, -1,
- (U8 *) s0);
+ (U8 *) s_start);
/* If is a multi-char fold, can't split here. Backup
* one char and try again */
* three character fold starting at the character
* before s, we can't split either before or after s.
* Backup two chars and try again */
- if ( LIKELY(s > s0)
+ if ( LIKELY(s > s_start)
&& UNLIKELY(is_THREE_CHAR_FOLD_utf8_safe(prev_s, e)))
{
s = prev_s;
- s = (char *) utf8_hop_back((U8 *) s, -1, (U8 *) s0);
+ s = (char *) utf8_hop_back((U8 *) s, -1, (U8 *) s_start);
backed_up = TRUE;
continue;
}
splittable = TRUE;
break;
- } while (s > s0); /* End of loops backing up through the node */
+ } while (s > s_start); /* End of loops backing up through the node */
/* Here we either couldn't find a place to split the node,
* or else we broke out of the loop setting 'splittable' to
continue;
}
- if ( LIKELY(s > s0)
+ if ( LIKELY(s > s_start)
&& UNLIKELY(is_THREE_CHAR_FOLD_latin1_safe(s - 1, e)))
{
s -= 2;
splittable = TRUE;
break;
- } while (s > s0);
+ } while (s > s_start);
if (splittable) {
s++;
/* If we did find a place to split, reparse the entire node
* stopping where we have calculated. */
if (splittable) {
- upper_fill = s - s0;
+
+ /* If we created a temporary folded string under /l, we
+ * have to map that back to the original */
+ if (need_to_fold_loc) {
+ upper_fill = loc_correspondence[s - s_start];
+ Safefree(locfold_buf);
+ Safefree(loc_correspondence);
+
+ if (upper_fill == 0) {
+ FAIL2("panic: loc_correspondence[%d] is 0",
+ (int) (s - s_start));
+ }
+ }
+ else {
+ upper_fill = s - s0;
+ }
goto reparse;
}
+ else if (need_to_fold_loc) {
+ Safefree(locfold_buf);
+ Safefree(loc_correspondence);
+ }
/* Here the node consists entirely of non-final multi-char
* folds. (Likely it is all 'f's or all 's's.) There's no
assert(PL_regkind[OP(node)] == ANYOF);
/* There is no bitmap for this node type */
- if (inRANGE(OP(node), ANYOFH, ANYOFHr)) {
+ if (inRANGE(OP(node), ANYOFH, ANYOFRb)) {
return;
}
current Unicode version */
{
Size_t folds_count;
- unsigned int first_fold;
- const unsigned int * remaining_folds;
+ U32 first_fold;
+ const U32 * remaining_folds;
UV folded_cp;
if (isASCII(cp)) {
PERL_ARGS_ASSERT_OUTPUT_POSIX_WARNINGS;
if (! TO_OUTPUT_WARNINGS(RExC_parse)) {
+ CLEAR_POSIX_WARNINGS();
return;
}
UPDATE_WARNINGS_LOC(RExC_parse);
}
+PERL_STATIC_INLINE Size_t
+S_find_first_differing_byte_pos(const U8 * s1, const U8 * s2, const Size_t max)
+{
+ const U8 * const start = s1;
+ const U8 * const send = start + max;
+
+ PERL_ARGS_ASSERT_FIND_FIRST_DIFFERING_BYTE_POS;
+
+ while (s1 < send && *s1 == *s2) {
+ s1++; s2++;
+ }
+
+ return s1 - start;
+}
+
+
STATIC AV *
S_add_multi_match(pTHX_ AV* multi_char_matches, SV* multi_string, const STRLEN cp_count)
{
* is already in 'value'. Otherwise, need to translate the escape
* into what it signifies. */
if (! skip_white || ! isBLANK_A(value)) switch ((I32)value) {
+ const char * message;
+ U32 packed_warn;
+ U8 grok_c_char;
case 'w': namedclass = ANYOF_WORDCHAR; break;
case 'W': namedclass = ANYOF_NWORDCHAR; break;
case 'a': value = '\a'; break;
case 'o':
RExC_parse--; /* function expects to be pointed at the 'o' */
- {
- const char* error_msg;
- bool valid = grok_bslash_o(&RExC_parse,
- RExC_end,
- &value,
- &error_msg,
- TO_OUTPUT_WARNINGS(RExC_parse),
- strict,
- silence_non_portable,
- UTF);
- if (! valid) {
- vFAIL(error_msg);
- }
- UPDATE_WARNINGS_LOC(RExC_parse - 1);
- }
- non_portable_endpoint++;
+ if (! grok_bslash_o(&RExC_parse,
+ RExC_end,
+ &value,
+ &message,
+ &packed_warn,
+ strict,
+ range, /* MAX_UV allowed for range
+ upper limit */
+ UTF))
+ {
+ vFAIL(message);
+ }
+ else if (message && TO_OUTPUT_WARNINGS(RExC_parse)) {
+ warn_non_literal_string(RExC_parse, packed_warn, message);
+ }
+
+ if (value < 256) {
+ non_portable_endpoint++;
+ }
break;
case 'x':
RExC_parse--; /* function expects to be pointed at the 'x' */
- {
- const char* error_msg;
- bool valid = grok_bslash_x(&RExC_parse,
- RExC_end,
- &value,
- &error_msg,
- TO_OUTPUT_WARNINGS(RExC_parse),
- strict,
- silence_non_portable,
- UTF);
- if (! valid) {
- vFAIL(error_msg);
- }
- UPDATE_WARNINGS_LOC(RExC_parse - 1);
- }
- non_portable_endpoint++;
+ if (! grok_bslash_x(&RExC_parse,
+ RExC_end,
+ &value,
+ &message,
+ &packed_warn,
+ strict,
+ range, /* MAX_UV allowed for range
+ upper limit */
+ UTF))
+ {
+ vFAIL(message);
+ }
+ else if (message && TO_OUTPUT_WARNINGS(RExC_parse)) {
+ warn_non_literal_string(RExC_parse, packed_warn, message);
+ }
+
+ if (value < 256) {
+ non_portable_endpoint++;
+ }
break;
case 'c':
- value = grok_bslash_c(*RExC_parse, TO_OUTPUT_WARNINGS(RExC_parse));
- UPDATE_WARNINGS_LOC(RExC_parse);
- RExC_parse++;
+ if (! grok_bslash_c(*RExC_parse, &grok_c_char, &message,
+ &packed_warn))
+ {
+ /* going to die anyway; point to exact spot of
+ * failure */
+ RExC_parse += (UTF)
+ ? UTF8_SAFE_SKIP(RExC_parse, RExC_end)
+ : 1;
+ vFAIL(message);
+ }
+
+ value = grok_c_char;
+ RExC_parse++;
+ if (message && TO_OUTPUT_WARNINGS(RExC_parse)) {
+ warn_non_literal_string(RExC_parse, packed_warn, message);
+ }
+
non_portable_endpoint++;
break;
case '0': case '1': case '2': case '3': case '4':
case '5': case '6': case '7':
{
/* Take 1-3 octal digits */
- I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
+ I32 flags = PERL_SCAN_SILENT_ILLDIGIT
+ | PERL_SCAN_NOTIFY_ILLDIGIT;
numlen = (strict) ? 4 : 3;
value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
RExC_parse += numlen;
: 1;
vFAIL("Need exactly 3 octal digits");
}
- else if ( numlen < 3 /* like \08, \178 */
+ else if ( (flags & PERL_SCAN_NOTIFY_ILLDIGIT)
&& RExC_parse < RExC_end
&& isDIGIT(*RExC_parse)
&& ckWARN(WARN_REGEXP))
{
reg_warn_non_literal_string(
RExC_parse + 1,
- form_short_octal_warning(RExC_parse, numlen));
+ form_alien_digit_msg(8, numlen, RExC_parse,
+ RExC_end, UTF, FALSE));
}
}
- non_portable_endpoint++;
+ if (value < 256) {
+ non_portable_endpoint++;
+ }
break;
}
default:
/* non-Latin1 code point implies unicode semantics. */
if (value > 255) {
+ if (value > MAX_LEGAL_CP && ( value != UV_MAX
+ || prevvalue > MAX_LEGAL_CP))
+ {
+ vFAIL(form_cp_too_large_msg(16, NULL, 0, value));
+ }
REQUIRE_UNI_RULES(flagp, 0);
+ if ( ! silence_non_portable
+ && UNICODE_IS_PERL_EXTENDED(value)
+ && TO_OUTPUT_WARNINGS(RExC_parse))
+ {
+ ckWARN2_non_literal_string(RExC_parse,
+ packWARN(WARN_PORTABLE),
+ PL_extended_cp_format,
+ value);
+ }
}
/* Ready to process either the single value, or the completed range.
STRLEN foldlen;
unsigned int k;
Size_t folds_count;
- unsigned int first_fold;
- const unsigned int * remaining_folds;
+ U32 first_fold;
+ const U32 * remaining_folds;
if (j < 256) {
|= ANYOFL_FOLD
| ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD;
}
- else if (cp_list) { /* Look to see if a 0-255 code point is in list */
- UV start, end;
- invlist_iterinit(cp_list);
- if (invlist_iternext(cp_list, &start, &end) && start < 256) {
- anyof_flags |= ANYOFL_FOLD;
- has_runtime_dependency |= HAS_L_RUNTIME_DEPENDENCY;
- }
- invlist_iterfinish(cp_list);
+ else if (cp_list && invlist_lowest(cp_list) < 256) {
+ /* If nothing is below 256, has no locale dependency; otherwise it
+ * does */
+ anyof_flags |= ANYOFL_FOLD;
+ has_runtime_dependency |= HAS_L_RUNTIME_DEPENDENCY;
}
}
else if ( DEPENDS_SEMANTICS
if (optimizable) {
PERL_UINT_FAST8_T i;
- Size_t partial_cp_count = 0;
+ UV partial_cp_count = 0;
UV start[MAX_FOLD_FROMS+1] = { 0 }; /* +1 for the folded-to char */
UV end[MAX_FOLD_FROMS+1] = { 0 };
+ bool single_range = FALSE;
if (cp_list) { /* Count the code points in enough ranges that we would
see all the ones possible in any fold in this version
partial_cp_count += end[i] - start[i] + 1;
}
+ if (i == 1) {
+ single_range = TRUE;
+ }
invlist_iterfinish(cp_list);
}
/* Only try if there are no more code points in the class than
* in the max possible fold */
- && partial_cp_count > 0 && partial_cp_count <= MAX_FOLD_FROMS + 1)
+ && inRANGE(partial_cp_count, 1, MAX_FOLD_FROMS + 1))
{
if (partial_cp_count == 1 && ! upper_latin1_only_utf8_matches)
{
U8 foldbuf[UTF8_MAXBYTES_CASE];
UV folded = _to_uni_fold_flags(start[0],
foldbuf, &foldlen, 0);
- unsigned int first_fold;
- const unsigned int * remaining_folds;
+ U32 first_fold;
+ const U32 * remaining_folds;
Size_t folds_to_this_cp_count = _inverse_folds(
folded,
&first_fold,
* convert to UTF-8 if not already there */
if (value > 255) {
if (! UTF) {
-
SvREFCNT_dec(cp_list);;
REQUIRE_UTF8(flagp);
}
len = (UTF) ? UVCHR_SKIP(value) : 1;
- ret = regnode_guts(pRExC_state, op, len, "exact");
- FILL_NODE(ret, op);
- RExC_emit += 1 + STR_SZ(len);
- setSTR_LEN(REGNODE_p(ret), len);
- if (len == 1) {
- *STRING(REGNODE_p(ret)) = (U8) value;
- }
- else {
- uvchr_to_utf8((U8 *) STRING(REGNODE_p(ret)), value);
- }
- goto not_anyof;
+ ret = regnode_guts(pRExC_state, op, len, "exact");
+ FILL_NODE(ret, op);
+ RExC_emit += 1 + STR_SZ(len);
+ setSTR_LEN(REGNODE_p(ret), len);
+ if (len == 1) {
+ *STRINGs(REGNODE_p(ret)) = (U8) value;
+ }
+ else {
+ uvchr_to_utf8((U8 *) STRINGs(REGNODE_p(ret)), value);
+ }
+ goto not_anyof;
}
}
SvREFCNT_dec(intersection);
}
+ /* If it is a single contiguous range, ANYOFR is an efficient regnode,
+ * both in size and speed. Currently, a 20 bit range base (smallest
+ * code point in the range), and a 12 bit maximum delta are packed into
+ * a 32 bit word. This allows for using it on all of the Unicode code
+ * points except for the highest plane, which is only for private use
+ * code points. khw doubts that a bigger delta is likely in real world
+ * applications */
+ if ( single_range
+ && ! has_runtime_dependency
+ && anyof_flags == 0
+ && start[0] < (1 << ANYOFR_BASE_BITS)
+ && end[0] - start[0]
+ < ((1U << (sizeof(((struct regnode_1 *)NULL)->arg1)
+ * CHARBITS - ANYOFR_BASE_BITS))))
+
+ {
+ U8 low_utf8[UTF8_MAXBYTES+1];
+ U8 high_utf8[UTF8_MAXBYTES+1];
+
+ ret = reganode(pRExC_state, ANYOFR,
+ (start[0] | (end[0] - start[0]) << ANYOFR_BASE_BITS));
+
+ /* Place the lowest UTF-8 start byte in the flags field, so as to
+ * allow efficient ruling out at run time of many possible inputs.
+ * */
+ (void) uvchr_to_utf8(low_utf8, start[0]);
+ (void) uvchr_to_utf8(high_utf8, end[0]);
+
+ /* If all code points share the same first byte, this can be an
+ * ANYOFRb. Otherwise store the lowest UTF-8 start byte which can
+ * quickly rule out many inputs at run-time without having to
+ * compute the code point from UTF-8. For EBCDIC, we use I8, as
+ * not doing that transformation would not rule out nearly so many
+ * things */
+ if (low_utf8[0] == high_utf8[0]) {
+ OP(REGNODE_p(ret)) = ANYOFRb;
+ ANYOF_FLAGS(REGNODE_p(ret)) = low_utf8[0];
+ }
+ else {
+ ANYOF_FLAGS(REGNODE_p(ret))
+ = NATIVE_UTF8_TO_I8(low_utf8[0]);
+ }
+
+ goto not_anyof;
+ }
+
/* If didn't find an optimization and there is no need for a bitmap,
* optimize to indicate that */
if ( start[0] >= NUM_ANYOF_CODE_POINTS
U8 low_utf8[UTF8_MAXBYTES+1];
UV highest_cp = invlist_highest(cp_list);
- op = ANYOFH;
-
/* Currently the maximum allowed code point by the system is
* IV_MAX. Higher ones are reserved for future internal use. This
* particular regnode can be used for higher ones, but we can't
* calculate the code point of those. IV_MAX suffices though, as
* it will be a large first byte */
- (void) uvchr_to_utf8(low_utf8, MIN(start[0], IV_MAX));
+ Size_t low_len = uvchr_to_utf8(low_utf8, MIN(start[0], IV_MAX))
+ - low_utf8;
/* We store the lowest possible first byte of the UTF-8
* representation, using the flags field. This allows for quick
* transformation would not rule out nearly so many things */
anyof_flags = NATIVE_UTF8_TO_I8(low_utf8[0]);
+ op = ANYOFH;
+
/* If the first UTF-8 start byte for the highest code point in the
* range is suitably small, we may be able to get an upper bound as
* well */
if (highest_cp <= IV_MAX) {
U8 high_utf8[UTF8_MAXBYTES+1];
-
- (void) uvchr_to_utf8(high_utf8, highest_cp);
+ Size_t high_len = uvchr_to_utf8(high_utf8, highest_cp)
+ - high_utf8;
/* If the lowest and highest are the same, we can get an exact
- * first byte instead of a just minimum. We signal this with a
- * different regnode */
+ * first byte instead of a just minimum or even a sequence of
+ * exact leading bytes. We signal these with different
+ * regnodes */
if (low_utf8[0] == high_utf8[0]) {
+ Size_t len = find_first_differing_byte_pos(low_utf8,
+ high_utf8,
+ MIN(low_len, high_len));
+
+ if (len == 1) {
- /* No need to convert to I8 for EBCDIC as this is an exact
- * match */
- anyof_flags = low_utf8[0];
- op = ANYOFHb;
+ /* No need to convert to I8 for EBCDIC as this is an
+ * exact match */
+ anyof_flags = low_utf8[0];
+ op = ANYOFHb;
+ }
+ else {
+ op = ANYOFHs;
+ ret = regnode_guts(pRExC_state, op,
+ regarglen[op] + STR_SZ(len),
+ "anyofhs");
+ FILL_NODE(ret, op);
+ ((struct regnode_anyofhs *) REGNODE_p(ret))->str_len
+ = len;
+ Copy(low_utf8, /* Add the common bytes */
+ ((struct regnode_anyofhs *) REGNODE_p(ret))->string,
+ len, U8);
+ RExC_emit += NODE_SZ_STR(REGNODE_p(ret));
+ set_ANYOF_arg(pRExC_state, REGNODE_p(ret), cp_list,
+ NULL, only_utf8_locale_list);
+ goto not_anyof;
+ }
}
else if (NATIVE_UTF8_TO_I8(high_utf8[0]) <= MAX_ANYOF_HRx_BYTE)
{
set_ANYOF_arg(pRExC_state, REGNODE_p(ret), cp_list,
(HAS_NONLOCALE_RUNTIME_PROPERTY_DEFINITION)
- ? listsv : NULL,
+ ? listsv
+ : NULL,
only_utf8_locale_list);
+ SvREFCNT_dec(cp_list);;
+ SvREFCNT_dec(only_utf8_locale_list);
return ret;
not_anyof:
Set_Node_Offset_Length(REGNODE_p(ret), orig_parse - RExC_start,
RExC_parse - orig_parse);;
SvREFCNT_dec(cp_list);;
+ SvREFCNT_dec(only_utf8_locale_list);
return ret;
}
SV *rv;
if (cp_list) {
- av_store(av, INVLIST_INDEX, cp_list);
+ av_store(av, INVLIST_INDEX, SvREFCNT_inc_NN(cp_list));
}
if (only_utf8_locale_list) {
- av_store(av, ONLY_LOCALE_MATCHES_INDEX, only_utf8_locale_list);
+ av_store(av, ONLY_LOCALE_MATCHES_INDEX,
+ SvREFCNT_inc_NN(only_utf8_locale_list));
}
if (runtime_defns) {
- av_store(av, DEFERRED_USER_DEFINED_INDEX, SvREFCNT_inc(runtime_defns));
+ av_store(av, DEFERRED_USER_DEFINED_INDEX,
+ SvREFCNT_inc_NN(runtime_defns));
}
rv = newRV_noinc(MUTABLE_SV(av));
STATIC_ASSERT_STMT(ONLY_LOCALE_MATCHES_INDEX == 1 + INVLIST_INDEX);
STATIC_ASSERT_STMT(DEFERRED_USER_DEFINED_INDEX == 1 + ONLY_LOCALE_MATCHES_INDEX);
- av_store(av, INVLIST_INDEX, invlist);
+ ary[INVLIST_INDEX] = invlist;
av_fill(av, (ary[ONLY_LOCALE_MATCHES_INDEX])
- ? ONLY_LOCALE_MATCHES_INDEX:
- INVLIST_INDEX);
+ ? ONLY_LOCALE_MATCHES_INDEX
+ : INVLIST_INDEX);
si = NULL;
}
}
UV prev_cp = 0;
U8 count = 0;
- /* Ignore everything before the first new-line */
- while (*si_string != '\n' && remaining > 0) {
- si_string++;
- remaining--;
- }
- assert(remaining > 0);
-
+ /* Ignore everything before and including the first new-line */
+ si_string = (const char *) memchr(si_string, '\n', SvCUR(si));
+ assert (si_string != NULL);
si_string++;
- remaining--;
+ remaining = SvPVX(si) + SvCUR(si) - si_string;
while (remaining > 0) {
continue;
}
- /* Here, didn't find a legal hex number. Just add it from
- * here to the next \n */
+ /* Here, didn't find a legal hex number. Just add the text
+ * from here up to the next \n, omitting any trailing
+ * markers. */
remaining -= len;
- while (*(si_string + len) != '\n' && remaining > 0) {
- remaining--;
- len++;
- }
- if (*(si_string + len) == '\n') {
- len++;
- remaining--;
- }
+ len = strcspn(si_string,
+ DEFERRED_COULD_BE_OFFICIAL_MARKERs "\n");
+ remaining -= len;
if (matches_string) {
- sv_catpvn(matches_string, si_string, len - 1);
+ sv_catpvn(matches_string, si_string, len);
}
else {
- matches_string = newSVpvn(si_string, len - 1);
+ matches_string = newSVpvn(si_string, len);
}
- si_string += len;
sv_catpvs(matches_string, " ");
+
+ si_string += len;
+ if ( remaining
+ && UCHARAT(si_string)
+ == DEFERRED_COULD_BE_OFFICIAL_MARKERc)
+ {
+ si_string++;
+ remaining--;
+ }
+ if (remaining && UCHARAT(si_string) == '\n') {
+ si_string++;
+ remaining--;
+ }
} /* end of loop through the text */
assert(matches_string);
S_regnode_guts(pTHX_ RExC_state_t *pRExC_state, const U8 op, const STRLEN extra_size, const char* const name)
{
/* Allocate a regnode for 'op', with 'extra_size' extra (smallest) regnode
- * equivalents space. It aligns and increments RExC_size and RExC_emit
+ * equivalents space. It aligns and increments RExC_size
*
* It returns the regnode's offset into the regex engine program */
scan = REGNODE_OFFSET(temp);
}
+ assert(val >= scan);
if (reg_off_by_arg[OP(REGNODE_p(scan))]) {
assert((UV) (val - scan) <= U32_MAX);
ARG_SET(REGNODE_p(scan), val - scan);
NULL,
NULL,
NULL,
+ 0,
FALSE
);
sv_catpvs(sv, "]");
else if (k == LOGICAL)
/* 2: embedded, otherwise 1 */
Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags);
- else if (k == ANYOF) {
- const U8 flags = inRANGE(OP(o), ANYOFH, ANYOFHr)
- ? 0
- : ANYOF_FLAGS(o);
+ else if (k == ANYOF || k == ANYOFR) {
+ U8 flags;
+ char * bitmap;
+ U32 arg;
bool do_sep = FALSE; /* Do we need to separate various components of
the output? */
/* Set if there is still an unresolved user-defined property */
/* And things that aren't in the bitmap, but are small enough to be */
SV* bitmap_range_not_in_bitmap = NULL;
- const bool inverted = flags & ANYOF_INVERT;
+ bool inverted;
+
+ if (inRANGE(OP(o), ANYOFH, ANYOFRb)) {
+ flags = 0;
+ bitmap = NULL;
+ arg = 0;
+ }
+ else {
+ flags = ANYOF_FLAGS(o);
+ bitmap = ANYOF_BITMAP(o);
+ arg = ARG(o);
+ }
if (OP(o) == ANYOFL || OP(o) == ANYOFPOSIXL) {
if (ANYOFL_UTF8_LOCALE_REQD(flags)) {
}
}
+ inverted = flags & ANYOF_INVERT;
+
/* If there is stuff outside the bitmap, get it */
- if (ARG(o) != ANYOF_ONLY_HAS_BITMAP) {
- (void) _get_regclass_nonbitmap_data(prog, o, FALSE,
+ if (arg != ANYOF_ONLY_HAS_BITMAP) {
+ if (inRANGE(OP(o), ANYOFR, ANYOFRb)) {
+ nonbitmap_invlist = _add_range_to_invlist(nonbitmap_invlist,
+ ANYOFRbase(o),
+ ANYOFRbase(o) + ANYOFRdelta(o));
+ }
+ else {
+ (void) _get_regclass_nonbitmap_data(prog, o, FALSE,
&unresolved,
&only_utf8_locale_invlist,
&nonbitmap_invlist);
+ }
+
/* The non-bitmap data may contain stuff that could fit in the
* bitmap. This could come from a user-defined property being
* finally resolved when this call was done; or much more likely
* because there are matches that require UTF-8 to be valid, and so
- * aren't in the bitmap. This is teased apart later */
+ * aren't in the bitmap (or ANYOFR). This is teased apart later */
_invlist_intersection(nonbitmap_invlist,
PL_InBitmap,
&bitmap_range_not_in_bitmap);
/* Ready to start outputting. First, the initial left bracket */
Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
- if (! inRANGE(OP(o), ANYOFH, ANYOFHr)) {
+ /* ANYOFH by definition doesn't have anything that will fit inside the
+ * bitmap; ANYOFR may or may not. */
+ if ( ! inRANGE(OP(o), ANYOFH, ANYOFHr)
+ && ( ! inRANGE(OP(o), ANYOFR, ANYOFRb)
+ || ANYOFRbase(o) < NUM_ANYOF_CODE_POINTS))
+ {
/* Then all the things that could fit in the bitmap */
do_sep = put_charclass_bitmap_innards(sv,
- ANYOF_BITMAP(o),
+ bitmap,
bitmap_range_not_in_bitmap,
only_utf8_locale_invlist,
o,
+ flags,
/* Can't try inverting for a
* better display if there
* are things that haven't
* been resolved */
- unresolved != NULL);
+ unresolved != NULL
+ || inRANGE(OP(o), ANYOFR, ANYOFRb));
SvREFCNT_dec(bitmap_range_not_in_bitmap);
/* If there are user-defined properties which haven't been defined
/* And finally the matching, closing ']' */
Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
- if (inRANGE(OP(o), ANYOFH, ANYOFHr)) {
+ if (OP(o) == ANYOFHs) {
+ Perl_sv_catpvf(aTHX_ sv, " (Leading UTF-8 bytes=%s", _byte_dump_string((U8 *) ((struct regnode_anyofhs *) o)->string, FLAGS(o), 1));
+ }
+ else if (inRANGE(OP(o), ANYOFH, ANYOFRb)) {
U8 lowest = (OP(o) != ANYOFHr)
? FLAGS(o)
: LOWEST_ANYOF_HRx_BYTE(FLAGS(o));
- U8 highest = (OP(o) == ANYOFHb)
- ? lowest
- : OP(o) == ANYOFH
+ U8 highest = (OP(o) == ANYOFHr)
+ ? HIGHEST_ANYOF_HRx_BYTE(FLAGS(o))
+ : (OP(o) == ANYOFH || OP(o) == ANYOFR)
? 0xFF
- : HIGHEST_ANYOF_HRx_BYTE(FLAGS(o));
+ : lowest;
Perl_sv_catpvf(aTHX_ sv, " (First UTF-8 byte=%02X", lowest);
if (lowest != highest) {
Perl_sv_catpvf(aTHX_ sv, "-%02X", highest);
_invlist_invert(cp_list);
}
- put_charclass_bitmap_innards(sv, NULL, cp_list, NULL, NULL, TRUE);
+ put_charclass_bitmap_innards(sv, NULL, cp_list, NULL, NULL, 0, TRUE);
Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
SvREFCNT_dec(cp_list);
SV *nonbitmap_invlist,
SV *only_utf8_locale_invlist,
const regnode * const node,
+ const U8 flags,
const bool force_as_is_display)
{
/* Appends to 'sv' a displayable version of the innards of the bracketed
* '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.
+ * 'flags' is the flags field of 'node'
* '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.
literally */
SV* inverted_display; /* The output string when we invert the inputs */
- U8 flags = (node) ? ANYOF_FLAGS(node) : 0;
-
bool invert = cBOOL(flags & ANYOF_INVERT); /* Is the input to be inverted
to match? */
/* We are biased in favor of displaying things without them being inverted,
else if ( op == PLUS || op == STAR) {
DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
}
- else if (PL_regkind[(U8)op] == EXACT) {
+ else if (PL_regkind[(U8)op] == EXACT || op == ANYOFHs) {
/* Literal string, where present. */
node += NODE_SZ_STR(node) - 1;
node = NEXTOPER(node);
{
dVAR;
+#ifdef DEBUGGING
+ char * dump_len_string;
+
+ dump_len_string = PerlEnv_getenv("PERL_DUMP_RE_MAX_LEN");
+ if ( ! dump_len_string
+ || ! grok_atoUV(dump_len_string, (UV *)&PL_dump_re_max_len, NULL))
+ {
+ PL_dump_re_max_len = 60; /* A reasonable default */
+ }
+#endif
+
PL_user_def_props = newHV();
#ifdef USE_ITHREADS
#endif
- /* Set up the inversion list global variables */
+ /* Set up the inversion list interpreter-level 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_LB_invlist = _new_invlist_C_array(_Perl_LB_invlist);
PL_SCX_invlist = _new_invlist_C_array(_Perl_SCX_invlist);
+ PL_InBitmap = _new_invlist_C_array(InBitmap_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);
* Other parameters will be set on return as described below */
const char * const name, /* The first non-blank in the \p{}, \P{} */
- const Size_t name_len, /* Its length in bytes, not including any
+ Size_t name_len, /* Its length in bytes, not including any
trailing space */
const bool is_utf8, /* ? Is 'name' encoded in UTF-8 */
const bool to_fold, /* ? Is this under /i */
user-defined property */
SV * msg, /* Any error or warning msg(s) are appended to
this */
- const STRLEN level) /* Recursion level of this call */
+ const STRLEN level) /* Recursion level of this call */
{
dVAR;
char* lookup_name; /* normalized name for lookup in our tables */
the normalized name in certain situations */
Size_t non_pkg_begin = 0; /* Offset of first byte in 'name' that isn't
part of a package name */
+ Size_t lun_non_pkg_begin = 0; /* Similarly for 'lookup_name' */
bool could_be_user_defined = TRUE; /* ? Could this be a user-defined
property rather than a Unicode
one. */
qualified name */
bool invert_return = FALSE; /* ? Do we need to complement the result before
returning it */
+ bool stripped_utf8_pkg = FALSE; /* Set TRUE if the input includes an
+ explicit utf8:: package that we strip
+ off */
+ /* The expansion of properties that could be either user-defined or
+ * official unicode ones is deferred until runtime, including a marker for
+ * those that might be in the latter category. This boolean indicates if
+ * we've seen that marker. If not, what we're parsing can't be such an
+ * official Unicode property whose expansion was deferred */
+ bool could_be_deferred_official = FALSE;
PERL_ARGS_ASSERT_PARSE_UNIPROP_STRING;
break;
}
+ /* If this looks like it is a marker we inserted at compile time,
+ * set a flag and otherwise ignore it. If it isn't in the final
+ * position, keep it as it would have been user input. */
+ if ( UNLIKELY(cur == DEFERRED_COULD_BE_OFFICIAL_MARKERc)
+ && ! deferrable
+ && could_be_user_defined
+ && i == name_len - 1)
+ {
+ name_len--;
+ could_be_deferred_official = TRUE;
+ continue;
+ }
+
/* Otherwise, this character is part of the name. */
lookup_name[j++] = cur;
i++;
non_pkg_begin = i + 1;
lookup_name[j++] = ':';
+ lun_non_pkg_begin = j;
}
else { /* Only word chars (and '::') can be in a user-defined name */
could_be_user_defined = FALSE;
lookup_name += STRLENs("utf8::");
j -= STRLENs("utf8::");
equals_pos -= STRLENs("utf8::");
+ stripped_utf8_pkg = TRUE;
}
/* Here, we are either done with the whole property name, if it was simple;
/* Most punctuation after the equals indicates a subpattern, like
* \p{foo=/bar/} */
if ( isPUNCT_A(name[i])
- && name[i] != '-'
- && name[i] != '+'
- && name[i] != '_'
- && name[i] != '{')
+ && name[i] != '-'
+ && name[i] != '+'
+ && name[i] != '_'
+ && name[i] != '{'
+ /* A backslash means the real delimitter is the next character,
+ * but it must be punctuation */
+ && (name[i] != '\\' || (i < name_len && isPUNCT_A(name[i+1]))))
{
/* Find the property. The table includes the equals sign, so we
* use 'j' as-is */
const char * pos_in_brackets;
bool escaped = 0;
- /* A backslash means the real delimitter is the next character.
- * */
+ /* Backslash => delimitter is the character following. We
+ * already checked that it is punctuation */
if (open == '\\') {
open = name[i++];
escaped = 1;
* set of closing is so that if the opening is something like
* ']', the closing will be that as well. Something similar is
* done in toke.c */
- pos_in_brackets = strchr("([<)]>)]>", open);
+ pos_in_brackets = memCHRs("([<)]>)]>", open);
close = (pos_in_brackets) ? pos_in_brackets[3] : open;
if ( i >= name_len
|| name[name_len-1] != close
- || (escaped && name[name_len-2] != '\\'))
+ || (escaped && name[name_len-2] != '\\')
+ /* Also make sure that there are enough characters.
+ * e.g., '\\\' would show up incorrectly as legal even
+ * though it is too short */
+ || (SSize_t) (name_len - i - 1 - escaped) < 0)
{
sv_catpvs(msg, "Unicode property wildcard not terminated");
goto append_name_to_msg;
/* Here, the name could be for a user defined property, which are
* implemented as subs. */
user_sub = get_cvn_flags(name, name_len, 0);
- if (user_sub) {
+ if (! user_sub) {
+
+ /* Here, the property name could be a user-defined one, but there
+ * is no subroutine to handle it (as of now). Defer handling it
+ * until runtime. Otherwise, a block defined by Unicode in a later
+ * release would get the synonym InFoo added for it, and existing
+ * code that used that name would suddenly break if it referred to
+ * the property before the sub was declared. See [perl #134146] */
+ if (deferrable) {
+ goto definition_deferred;
+ }
+
+ /* Here, we are at runtime, and didn't find the user property. It
+ * could be an official property, but only if no package was
+ * specified, or just the utf8:: package. */
+ if (could_be_deferred_official) {
+ lookup_name += lun_non_pkg_begin;
+ j -= lun_non_pkg_begin;
+ }
+ else if (! stripped_utf8_pkg) {
+ goto unknown_user_defined;
+ }
+
+ /* Drop down to look up in the official properties */
+ }
+ else {
const char insecure[] = "Insecure user-defined property";
/* Here, there is a sub by the correct name. Normally we call it
* property hasn't been encountered yet, but at runtime, it's
* an error to try to use an undefined one */
if (! deferrable) {
- if (SvCUR(msg) > 0) sv_catpvs(msg, "; ");
- sv_catpvs(msg, "Unknown user-defined property name");
- goto append_name_to_msg;
+ goto unknown_user_defined;;
}
goto definition_deferred;
}
return prop_definition;
+ unknown_user_defined:
+ if (SvCUR(msg) > 0) sv_catpvs(msg, "; ");
+ sv_catpvs(msg, "Unknown user-defined property name");
+ goto append_name_to_msg;
failed:
if (non_pkg_begin != 0) {
definition_deferred:
- /* Here it could yet to be defined, so defer evaluation of this
- * until its needed at runtime. We need the fully qualified property name
- * to avoid ambiguity, and a trailing newline */
- if (! fq_name) {
- fq_name = S_get_fq_name(aTHX_ name, name_len, is_utf8,
- non_pkg_begin != 0 /* If has "::" */
- );
- }
- sv_catpvs(fq_name, "\n");
+ {
+ bool is_qualified = non_pkg_begin != 0; /* If has "::" */
- *user_defined_ptr = TRUE;
- return fq_name;
+ /* Here it could yet to be defined, so defer evaluation of this until
+ * its needed at runtime. We need the fully qualified property name to
+ * avoid ambiguity */
+ if (! fq_name) {
+ fq_name = S_get_fq_name(aTHX_ name, name_len, is_utf8,
+ is_qualified);
+ }
+
+ /* If it didn't come with a package, or the package is utf8::, this
+ * actually could be an official Unicode property whose inclusion we
+ * are deferring until runtime to make sure that it isn't overridden by
+ * a user-defined property of the same name (which we haven't
+ * encountered yet). Add a marker to indicate this possibility, for
+ * use at such time when we first need the definition during pattern
+ * matching execution */
+ if (! is_qualified || memBEGINPs(name, non_pkg_begin, "utf8::")) {
+ sv_catpvs(fq_name, DEFERRED_COULD_BE_OFFICIAL_MARKERs);
+ }
+
+ /* We also need a trailing newline */
+ sv_catpvs(fq_name, "\n");
+
+ *user_defined_ptr = TRUE;
+ return fq_name;
+ }
}
#endif