* requiring null fields (pat.t#187 and split.t#{13,14}
* (as of patchlevel 7877) will fail. Then again,
* this code seems to be necessary or otherwise
- * building DynaLoader will fail:
- * "Error: '*' not in typemap in DynaLoader.xs, line 164"
- * --jhi */
+ * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
+ * --jhi updated by dapm */
for (i = *PL_reglastparen + 1; i <= rex->nparens; i++) {
if (i > PL_regsize)
PL_regoffs[i].start = -1;
return NULL;
}
-
+#define DECL_TRIE_TYPE(scan) \
+ const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold } \
+ trie_type = (scan->flags != EXACT) \
+ ? (do_utf8 ? trie_utf8_fold : (UTF ? trie_latin_utf8_fold : trie_plain)) \
+ : (do_utf8 ? trie_utf8 : trie_plain)
#define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, \
uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \
uscan = foldbuf + UNISKIP( uvc ); \
} \
break; \
+ case trie_latin_utf8_fold: \
+ if ( foldlen>0 ) { \
+ uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \
+ foldlen -= len; \
+ uscan += len; \
+ len=0; \
+ } else { \
+ len = 1; \
+ uvc = to_uni_fold( *(U8*)uc, foldbuf, &foldlen ); \
+ foldlen -= UNISKIP( uvc ); \
+ uscan = foldbuf + UNISKIP( uvc ); \
+ } \
+ break; \
case trie_utf8: \
uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \
break; \
} \
} STMT_END
-#define REXEC_FBC_EXACTISH_CHECK(CoNd) \
+#define REXEC_FBC_EXACTISH_CHECK(CoNd) \
+{ \
+ char *my_strend= (char *)strend; \
if ( (CoNd) \
&& (ln == len || \
- ibcmp_utf8(s, NULL, 0, do_utf8, \
+ !ibcmp_utf8(s, &my_strend, 0, do_utf8, \
m, NULL, ln, (bool)UTF)) \
- && (!reginfo || regtry(reginfo, &s)) ) \
+ && (!reginfo || regtry(reginfo, &s)) ) \
goto got_it; \
else { \
U8 foldbuf[UTF8_MAXBYTES_CASE+1]; \
f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); \
if ( f != c \
&& (f == c1 || f == c2) \
- && (ln == foldlen || \
- !ibcmp_utf8((char *) foldbuf, \
- NULL, foldlen, do_utf8, \
- m, \
- NULL, ln, (bool)UTF)) \
- && (!reginfo || regtry(reginfo, &s)) ) \
+ && (ln == len || \
+ !ibcmp_utf8(s, &my_strend, 0, do_utf8,\
+ m, NULL, ln, (bool)UTF)) \
+ && (!reginfo || regtry(reginfo, &s)) ) \
goto got_it; \
} \
- s += len
+} \
+s += len
#define REXEC_FBC_EXACTISH_SCAN(CoNd) \
STMT_START { \
U8 *sm = (U8 *) m;
U8 tmpbuf1[UTF8_MAXBYTES_CASE+1];
U8 tmpbuf2[UTF8_MAXBYTES_CASE+1];
- const U32 uniflags = UTF8_ALLOW_DEFAULT;
-
- to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
- to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
-
+ /* used by commented-out code below */
+ /*const U32 uniflags = UTF8_ALLOW_DEFAULT;*/
+
+ /* XXX: Since the node will be case folded at compile
+ time this logic is a little odd, although im not
+ sure that its actually wrong. --dmq */
+
+ c1 = to_utf8_lower((U8*)m, tmpbuf1, &ulen1);
+ c2 = to_utf8_upper((U8*)m, tmpbuf2, &ulen2);
+
+ /* XXX: This is kinda strange. to_utf8_XYZ returns the
+ codepoint of the first character in the converted
+ form, yet originally we did the extra step.
+ No tests fail by commenting this code out however
+ so Ive left it out. -- dmq.
+
c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXBYTES_CASE,
0, uniflags);
c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXBYTES_CASE,
0, uniflags);
+ */
+
lnc = 0;
while (sm < ((U8 *) m + ln)) {
lnc++;
* matching (called "loose matching" in Unicode).
* ibcmp_utf8() will do just that. */
- if (do_utf8) {
+ if (do_utf8 || UTF) {
UV c, f;
U8 tmpbuf [UTF8_MAXBYTES+1];
- STRLEN len, foldlen;
+ STRLEN len = 1;
+ STRLEN foldlen;
const U32 uniflags = UTF8_ALLOW_DEFAULT;
if (c1 == c2) {
/* Upper and lower of 1st char are equal -
* probably not a "letter". */
while (s <= e) {
- c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
+ if (do_utf8) {
+ c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
uniflags);
+ } else {
+ c = *((U8*)s);
+ }
REXEC_FBC_EXACTISH_CHECK(c == c1);
}
}
else {
while (s <= e) {
- c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
+ if (do_utf8) {
+ c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len,
uniflags);
+ } else {
+ c = *((U8*)s);
+ }
/* Handle some of the three Greek sigmas cases.
* Note that not all the possible combinations
}
}
else {
+ /* Neither pattern nor string are UTF8 */
if (c1 == c2)
REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
else
case AHOCORASICKC:
case AHOCORASICK:
{
- const enum { trie_plain, trie_utf8, trie_utf8_fold }
- trie_type = do_utf8 ?
- (c->flags == EXACT ? trie_utf8 : trie_utf8_fold)
- : trie_plain;
+ DECL_TRIE_TYPE(c);
/* what trie are we using right now */
reg_ac_data *aho
= (reg_ac_data*)progi->data->data[ ARG( c ) ];
U8 **points; /* map of where we were in the input string
when reading a given char. For ASCII this
is unnecessary overhead as the relationship
- is always 1:1, but for unicode, especially
- case folded unicode this is not true. */
+ is always 1:1, but for Unicode, especially
+ case folded Unicode this is not true. */
U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
U8 *bitmap=NULL;
/* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
* Actually, the code in regcppop() (which Ilya may be meaning by
* PL_reglastparen), is not needed at all by the test suite
- * (op/regexp, op/pat, op/split), but that code is needed, oddly
- * enough, for building DynaLoader, or otherwise this
- * "Error: '*' not in typemap in DynaLoader.xs, line 164"
- * will happen. Meanwhile, this code *is* needed for the
+ * (op/regexp, op/pat, op/split), but that code is needed otherwise
+ * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
+ * Meanwhile, this code *is* needed for the
* above-mentioned test suite tests to succeed. The common theme
* on those tests seems to be returning null fields from matches.
- * --jhi */
+ * --jhi updated by dapm */
#if 1
if (prog->nparens) {
regexp_paren_pair *pp = PL_regoffs;
case TRIE:
{
/* what type of TRIE am I? (utf8 makes this contextual) */
- const enum { trie_plain, trie_utf8, trie_utf8_fold }
- trie_type = do_utf8 ?
- (scan->flags == EXACT ? trie_utf8 : trie_utf8_fold)
- : trie_plain;
+ DECL_TRIE_TYPE(scan);
/* what trie are we using right now */
reg_trie_data * const trie
if ( got_wordnum ) {
if ( ! ST.accepted ) {
ENTER;
- SAVETMPS;
+ /* SAVETMPS; */ /* XXX is this necessary? dmq */
bufflen = TRIE_INITAL_ACCEPT_BUFFLEN;
sv_accept_buff=newSV(bufflen *
sizeof(reg_trie_accepted) - 1);
PL_reginput = (char *)ST.accept_buff[ best ].endpos;
if ( !ST.jump || !ST.jump[ST.accept_buff[best].wordnum]) {
scan = ST.B;
- /* NOTREACHED */
} else {
scan = ST.me + ST.jump[ST.accept_buff[best].wordnum];
- /* NOTREACHED */
- }
- if (has_cutgroup) {
- PUSH_YES_STATE_GOTO(TRIE_next, scan);
- /* NOTREACHED */
- } else {
- PUSH_STATE_GOTO(TRIE_next, scan);
- /* NOTREACHED */
}
+ PUSH_YES_STATE_GOTO(TRIE_next, scan);
/* NOTREACHED */
}
/* NOTREACHED */
case BRANCH: /* /(...|A|...)/ */
scan = NEXTOPER(scan); /* scan now points to inner node */
- if ((!next || (OP(next) != BRANCH && OP(next) != BRANCHJ))
- && !has_cutgroup)
- {
- /* last branch; skip state push and jump direct to node */
- continue;
- }
ST.lastparen = *PL_reglastparen;
ST.next_branch = next;
REGCP_SET(ST.cp);
} else {
while (scan < loceol && !isSPACE(*scan))
scan++;
- break;
}
+ break;
case NSPACEL:
PL_reg_flags |= RF_tainted;
if (do_utf8) {
* documentation of these array elements. */
si = *ary;
- a = SvROK(ary[1]) ? &ary[1] : 0;
- b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : 0;
+ a = SvROK(ary[1]) ? &ary[1] : NULL;
+ b = SvTYPE(ary[2]) == SVt_PVAV ? &ary[2] : NULL;
if (a)
sw = *a;