Rename regex flag and field utf16 -> unicode

- rename is_utf16 structure member to is_unicode
- rename flag LRE_FLAG_UTF16 as LRE_FLAG_UNICODE
This commit is contained in:
Charlie Gordon 2024-02-19 16:30:08 +01:00
parent 97ae6f39e6
commit 8d932deb49
3 changed files with 65 additions and 64 deletions

View File

@ -66,7 +66,7 @@ typedef struct {
const uint8_t *buf_end; const uint8_t *buf_end;
const uint8_t *buf_start; const uint8_t *buf_start;
int re_flags; int re_flags;
BOOL is_utf16; BOOL is_unicode;
BOOL ignore_case; BOOL ignore_case;
BOOL dotall; BOOL dotall;
int capture_count; int capture_count;
@ -224,7 +224,7 @@ static __maybe_unused void lre_dump_bytecode(const uint8_t *buf,
assert(buf_len >= RE_HEADER_LEN); assert(buf_len >= RE_HEADER_LEN);
re_flags= buf[0]; re_flags = buf[0];
bc_len = get_u32(buf + 3); bc_len = get_u32(buf + 3);
assert(bc_len + RE_HEADER_LEN <= buf_len); assert(bc_len + RE_HEADER_LEN <= buf_len);
printf("flags: 0x%x capture_count=%d stack_size=%d\n", printf("flags: 0x%x capture_count=%d stack_size=%d\n",
@ -696,10 +696,10 @@ static int get_class_atom(REParseState *s, CharRange *cr,
if ((c >= 'a' && c <= 'z') || if ((c >= 'a' && c <= 'z') ||
(c >= 'A' && c <= 'Z') || (c >= 'A' && c <= 'Z') ||
(((c >= '0' && c <= '9') || c == '_') && (((c >= '0' && c <= '9') || c == '_') &&
inclass && !s->is_utf16)) { /* Annex B.1.4 */ inclass && !s->is_unicode)) { /* Annex B.1.4 */
c &= 0x1f; c &= 0x1f;
p++; p++;
} else if (s->is_utf16) { } else if (s->is_unicode) {
goto invalid_escape; goto invalid_escape;
} else { } else {
/* otherwise return '\' and 'c' */ /* otherwise return '\' and 'c' */
@ -710,7 +710,7 @@ static int get_class_atom(REParseState *s, CharRange *cr,
#ifdef CONFIG_ALL_UNICODE #ifdef CONFIG_ALL_UNICODE
case 'p': case 'p':
case 'P': case 'P':
if (s->is_utf16) { if (s->is_unicode) {
if (parse_unicode_property(s, cr, &p, (c == 'P'))) if (parse_unicode_property(s, cr, &p, (c == 'P')))
return -1; return -1;
c = CLASS_RANGE_BASE; c = CLASS_RANGE_BASE;
@ -720,14 +720,14 @@ static int get_class_atom(REParseState *s, CharRange *cr,
#endif #endif
default: default:
p--; p--;
ret = lre_parse_escape(&p, s->is_utf16 * 2); ret = lre_parse_escape(&p, s->is_unicode * 2);
if (ret >= 0) { if (ret >= 0) {
c = ret; c = ret;
} else { } else {
if (ret == -2 && *p != '\0' && strchr("^$\\.*+?()[]{}|/", *p)) { if (ret == -2 && *p != '\0' && strchr("^$\\.*+?()[]{}|/", *p)) {
/* always valid to escape these characters */ /* always valid to escape these characters */
goto normal_char; goto normal_char;
} else if (s->is_utf16) { } else if (s->is_unicode) {
invalid_escape: invalid_escape:
return re_parse_error(s, "invalid escape sequence in regular expression"); return re_parse_error(s, "invalid escape sequence in regular expression");
} else { } else {
@ -749,7 +749,7 @@ static int get_class_atom(REParseState *s, CharRange *cr,
/* normal char */ /* normal char */
if (c >= 128) { if (c >= 128) {
c = unicode_from_utf8(p, UTF8_CHAR_LEN_MAX, &p); c = unicode_from_utf8(p, UTF8_CHAR_LEN_MAX, &p);
if ((unsigned)c > 0xffff && !s->is_utf16) { if ((unsigned)c > 0xffff && !s->is_unicode) {
/* XXX: should handle non BMP-1 code points */ /* XXX: should handle non BMP-1 code points */
return re_parse_error(s, "malformed unicode char"); return re_parse_error(s, "malformed unicode char");
} }
@ -811,11 +811,13 @@ static int re_parse_char_class(REParseState *s, const uint8_t **pp)
cr_init(cr, s->opaque, lre_realloc); cr_init(cr, s->opaque, lre_realloc);
p = *pp; p = *pp;
p++; /* skip '[' */ p++; /* skip '[' */
invert = FALSE; invert = FALSE;
if (*p == '^') { if (*p == '^') {
p++; p++;
invert = TRUE; invert = TRUE;
} }
for(;;) { for(;;) {
if (*p == ']') if (*p == ']')
break; break;
@ -825,7 +827,7 @@ static int re_parse_char_class(REParseState *s, const uint8_t **pp)
if (*p == '-' && p[1] != ']') { if (*p == '-' && p[1] != ']') {
const uint8_t *p0 = p + 1; const uint8_t *p0 = p + 1;
if (c1 >= CLASS_RANGE_BASE) { if (c1 >= CLASS_RANGE_BASE) {
if (s->is_utf16) { if (s->is_unicode) {
cr_free(cr1); cr_free(cr1);
goto invalid_class_range; goto invalid_class_range;
} }
@ -837,7 +839,7 @@ static int re_parse_char_class(REParseState *s, const uint8_t **pp)
goto fail; goto fail;
if (c2 >= CLASS_RANGE_BASE) { if (c2 >= CLASS_RANGE_BASE) {
cr_free(cr1); cr_free(cr1);
if (s->is_utf16) { if (s->is_unicode) {
goto invalid_class_range; goto invalid_class_range;
} }
/* Annex B: match '-' character */ /* Annex B: match '-' character */
@ -866,7 +868,7 @@ static int re_parse_char_class(REParseState *s, const uint8_t **pp)
} }
} }
if (s->ignore_case) { if (s->ignore_case) {
if (cr_regexp_canonicalize(cr, s->is_utf16)) if (cr_regexp_canonicalize(cr, s->is_unicode))
goto memory_error; goto memory_error;
} }
if (invert) { if (invert) {
@ -1161,7 +1163,7 @@ static int re_parse_term(REParseState *s, BOOL is_backward_dir)
re_emit_op(s, REOP_prev); re_emit_op(s, REOP_prev);
break; break;
case '{': case '{':
if (s->is_utf16) { if (s->is_unicode) {
return re_parse_error(s, "syntax error"); return re_parse_error(s, "syntax error");
} else if (!is_digit(p[1])) { } else if (!is_digit(p[1])) {
/* Annex B: we accept '{' not followed by digits as a /* Annex B: we accept '{' not followed by digits as a
@ -1213,7 +1215,7 @@ static int re_parse_term(REParseState *s, BOOL is_backward_dir)
lookahead: lookahead:
/* Annex B allows lookahead to be used as an atom for /* Annex B allows lookahead to be used as an atom for
the quantifiers */ the quantifiers */
if (!s->is_utf16 && !is_backward_lookahead) { if (!s->is_unicode && !is_backward_lookahead) {
last_atom_start = s->byte_code.size; last_atom_start = s->byte_code.size;
last_capture_count = s->capture_count; last_capture_count = s->capture_count;
} }
@ -1289,7 +1291,7 @@ static int re_parse_term(REParseState *s, BOOL is_backward_dir)
/* annex B: we tolerate invalid group names in non /* annex B: we tolerate invalid group names in non
unicode mode if there is no named capture unicode mode if there is no named capture
definition */ definition */
if (s->is_utf16 || re_has_named_captures(s)) if (s->is_unicode || re_has_named_captures(s))
return re_parse_error(s, "expecting group name"); return re_parse_error(s, "expecting group name");
else else
goto parse_class_atom; goto parse_class_atom;
@ -1297,7 +1299,7 @@ static int re_parse_term(REParseState *s, BOOL is_backward_dir)
p1 += 3; p1 += 3;
if (re_parse_group_name(s->u.tmp_buf, sizeof(s->u.tmp_buf), if (re_parse_group_name(s->u.tmp_buf, sizeof(s->u.tmp_buf),
&p1)) { &p1)) {
if (s->is_utf16 || re_has_named_captures(s)) if (s->is_unicode || re_has_named_captures(s))
return re_parse_error(s, "invalid group name"); return re_parse_error(s, "invalid group name");
else else
goto parse_class_atom; goto parse_class_atom;
@ -1308,7 +1310,7 @@ static int re_parse_term(REParseState *s, BOOL is_backward_dir)
after (inefficient, but hopefully not common */ after (inefficient, but hopefully not common */
c = re_parse_captures(s, &dummy_res, s->u.tmp_buf); c = re_parse_captures(s, &dummy_res, s->u.tmp_buf);
if (c < 0) { if (c < 0) {
if (s->is_utf16 || re_has_named_captures(s)) if (s->is_unicode || re_has_named_captures(s))
return re_parse_error(s, "group name not defined"); return re_parse_error(s, "group name not defined");
else else
goto parse_class_atom; goto parse_class_atom;
@ -1320,7 +1322,7 @@ static int re_parse_term(REParseState *s, BOOL is_backward_dir)
case '0': case '0':
p += 2; p += 2;
c = 0; c = 0;
if (s->is_utf16) { if (s->is_unicode) {
if (is_digit(*p)) { if (is_digit(*p)) {
return re_parse_error(s, "invalid decimal escape in regular expression"); return re_parse_error(s, "invalid decimal escape in regular expression");
} }
@ -1342,7 +1344,7 @@ static int re_parse_term(REParseState *s, BOOL is_backward_dir)
c = parse_digits(&p, FALSE); c = parse_digits(&p, FALSE);
if (c < 0 || (c >= s->capture_count && c >= re_count_captures(s))) { if (c < 0 || (c >= s->capture_count && c >= re_count_captures(s))) {
if (!s->is_utf16) { if (!s->is_unicode) {
/* Annex B.1.4: accept legacy octal */ /* Annex B.1.4: accept legacy octal */
p = q; p = q;
if (*p <= '7') { if (*p <= '7') {
@ -1384,7 +1386,7 @@ static int re_parse_term(REParseState *s, BOOL is_backward_dir)
break; break;
case ']': case ']':
case '}': case '}':
if (s->is_utf16) if (s->is_unicode)
return re_parse_error(s, "syntax error"); return re_parse_error(s, "syntax error");
goto parse_class_atom; goto parse_class_atom;
default: default:
@ -1406,7 +1408,7 @@ static int re_parse_term(REParseState *s, BOOL is_backward_dir)
return -1; return -1;
} else { } else {
if (s->ignore_case) if (s->ignore_case)
c = lre_canonicalize(c, s->is_utf16); c = lre_canonicalize(c, s->is_unicode);
if (c <= 0xffff) if (c <= 0xffff)
re_emit_op_u16(s, REOP_char, c); re_emit_op_u16(s, REOP_char, c);
else else
@ -1442,7 +1444,7 @@ static int re_parse_term(REParseState *s, BOOL is_backward_dir)
/* As an extension (see ES6 annex B), we accept '{' not /* As an extension (see ES6 annex B), we accept '{' not
followed by digits as a normal atom */ followed by digits as a normal atom */
if (!is_digit(p[1])) { if (!is_digit(p[1])) {
if (s->is_utf16) if (s->is_unicode)
goto invalid_quant_count; goto invalid_quant_count;
break; break;
} }
@ -1461,7 +1463,7 @@ static int re_parse_term(REParseState *s, BOOL is_backward_dir)
quant_max = INT32_MAX; /* infinity */ quant_max = INT32_MAX; /* infinity */
} }
} }
if (*p != '}' && !s->is_utf16) { if (*p != '}' && !s->is_unicode) {
/* Annex B: normal atom if invalid '{' syntax */ /* Annex B: normal atom if invalid '{' syntax */
p = p1; p = p1;
break; break;
@ -1753,7 +1755,7 @@ uint8_t *lre_compile(int *plen, char *error_msg, int error_msg_size,
s->buf_end = s->buf_ptr + buf_len; s->buf_end = s->buf_ptr + buf_len;
s->buf_start = s->buf_ptr; s->buf_start = s->buf_ptr;
s->re_flags = re_flags; s->re_flags = re_flags;
s->is_utf16 = ((re_flags & LRE_FLAG_UTF16) != 0); s->is_unicode = ((re_flags & LRE_FLAG_UNICODE) != 0);
is_sticky = ((re_flags & LRE_FLAG_STICKY) != 0); is_sticky = ((re_flags & LRE_FLAG_STICKY) != 0);
s->ignore_case = ((re_flags & LRE_FLAG_IGNORECASE) != 0); s->ignore_case = ((re_flags & LRE_FLAG_IGNORECASE) != 0);
s->dotall = ((re_flags & LRE_FLAG_DOTALL) != 0); s->dotall = ((re_flags & LRE_FLAG_DOTALL) != 0);
@ -1861,11 +1863,11 @@ static BOOL is_word_char(uint32_t c)
} \ } \
} while (0) } while (0)
#define PEEK_CHAR(c, cptr, cbuf_end) \ #define PEEK_CHAR(c, cptr, cbuf_end) \
do { \ do { \
if (cbuf_type == 0) { \ if (cbuf_type == 0) { \
c = cptr[0]; \ c = cptr[0]; \
} else { \ } else { \
uint32_t __c1; \ uint32_t __c1; \
c = ((uint16_t *)cptr)[0]; \ c = ((uint16_t *)cptr)[0]; \
if (c >= 0xd800 && c < 0xdc00 && \ if (c >= 0xd800 && c < 0xdc00 && \
@ -1875,18 +1877,18 @@ static BOOL is_word_char(uint32_t c)
c = (((c & 0x3ff) << 10) | (__c1 & 0x3ff)) + 0x10000; \ c = (((c & 0x3ff) << 10) | (__c1 & 0x3ff)) + 0x10000; \
} \ } \
} \ } \
} \ } \
} while (0) } while (0)
#define PEEK_PREV_CHAR(c, cptr, cbuf_start) \ #define PEEK_PREV_CHAR(c, cptr, cbuf_start) \
do { \ do { \
if (cbuf_type == 0) { \ if (cbuf_type == 0) { \
c = cptr[-1]; \ c = cptr[-1]; \
} else { \ } else { \
uint32_t __c1; \ uint32_t __c1; \
c = ((uint16_t *)cptr)[-1]; \ c = ((uint16_t *)cptr)[-1]; \
if (c >= 0xdc00 && c < 0xe000 && \ if (c >= 0xdc00 && c < 0xe000 && \
cbuf_type == 2 && (cptr - 4) >= cbuf_start) { \ cbuf_type == 2 && (cptr - 4) >= cbuf_start) { \
__c1 = ((uint16_t *)cptr)[-2]; \ __c1 = ((uint16_t *)cptr)[-2]; \
if (__c1 >= 0xd800 && __c1 < 0xdc00 ) { \ if (__c1 >= 0xd800 && __c1 < 0xdc00 ) { \
c = (((__c1 & 0x3ff) << 10) | (c & 0x3ff)) + 0x10000; \ c = (((__c1 & 0x3ff) << 10) | (c & 0x3ff)) + 0x10000; \
@ -1895,15 +1897,15 @@ static BOOL is_word_char(uint32_t c)
} \ } \
} while (0) } while (0)
#define GET_PREV_CHAR(c, cptr, cbuf_start) \ #define GET_PREV_CHAR(c, cptr, cbuf_start) \
do { \ do { \
if (cbuf_type == 0) { \ if (cbuf_type == 0) { \
cptr--; \ cptr--; \
c = cptr[0]; \ c = cptr[0]; \
} else { \ } else { \
uint32_t __c1; \ uint32_t __c1; \
cptr -= 2; \ cptr -= 2; \
c = ((uint16_t *)cptr)[0]; \ c = ((uint16_t *)cptr)[0]; \
if (c >= 0xdc00 && c < 0xe000 && \ if (c >= 0xdc00 && c < 0xe000 && \
cbuf_type == 2 && cptr > cbuf_start) { \ cbuf_type == 2 && cptr > cbuf_start) { \
__c1 = ((uint16_t *)cptr)[-1]; \ __c1 = ((uint16_t *)cptr)[-1]; \
@ -1915,12 +1917,12 @@ static BOOL is_word_char(uint32_t c)
} \ } \
} while (0) } while (0)
#define PREV_CHAR(cptr, cbuf_start) \ #define PREV_CHAR(cptr, cbuf_start) \
do { \ do { \
if (cbuf_type == 0) { \ if (cbuf_type == 0) { \
cptr--; \ cptr--; \
} else { \ } else { \
cptr -= 2; \ cptr -= 2; \
if (cbuf_type == 2) { \ if (cbuf_type == 2) { \
c = ((uint16_t *)cptr)[0]; \ c = ((uint16_t *)cptr)[0]; \
if (c >= 0xdc00 && c < 0xe000 && cptr > cbuf_start) { \ if (c >= 0xdc00 && c < 0xe000 && cptr > cbuf_start) { \
@ -1959,7 +1961,7 @@ typedef struct {
int stack_size_max; int stack_size_max;
BOOL multi_line; BOOL multi_line;
BOOL ignore_case; BOOL ignore_case;
BOOL is_utf16; BOOL is_unicode;
void *opaque; /* used for stack overflow check */ void *opaque; /* used for stack overflow check */
size_t state_size; size_t state_size;
@ -2105,7 +2107,7 @@ static intptr_t lre_exec_backtrack(REExecContext *s, uint8_t **capture,
goto no_match; goto no_match;
GET_CHAR(c, cptr, cbuf_end); GET_CHAR(c, cptr, cbuf_end);
if (s->ignore_case) { if (s->ignore_case) {
c = lre_canonicalize(c, s->is_utf16); c = lre_canonicalize(c, s->is_unicode);
} }
if (val != c) if (val != c)
goto no_match; goto no_match;
@ -2260,8 +2262,8 @@ static intptr_t lre_exec_backtrack(REExecContext *s, uint8_t **capture,
GET_CHAR(c1, cptr1, cptr1_end); GET_CHAR(c1, cptr1, cptr1_end);
GET_CHAR(c2, cptr, cbuf_end); GET_CHAR(c2, cptr, cbuf_end);
if (s->ignore_case) { if (s->ignore_case) {
c1 = lre_canonicalize(c1, s->is_utf16); c1 = lre_canonicalize(c1, s->is_unicode);
c2 = lre_canonicalize(c2, s->is_utf16); c2 = lre_canonicalize(c2, s->is_unicode);
} }
if (c1 != c2) if (c1 != c2)
goto no_match; goto no_match;
@ -2274,8 +2276,8 @@ static intptr_t lre_exec_backtrack(REExecContext *s, uint8_t **capture,
GET_PREV_CHAR(c1, cptr1, cptr1_start); GET_PREV_CHAR(c1, cptr1, cptr1_start);
GET_PREV_CHAR(c2, cptr, s->cbuf); GET_PREV_CHAR(c2, cptr, s->cbuf);
if (s->ignore_case) { if (s->ignore_case) {
c1 = lre_canonicalize(c1, s->is_utf16); c1 = lre_canonicalize(c1, s->is_unicode);
c2 = lre_canonicalize(c2, s->is_utf16); c2 = lre_canonicalize(c2, s->is_unicode);
} }
if (c1 != c2) if (c1 != c2)
goto no_match; goto no_match;
@ -2294,7 +2296,7 @@ static intptr_t lre_exec_backtrack(REExecContext *s, uint8_t **capture,
goto no_match; goto no_match;
GET_CHAR(c, cptr, cbuf_end); GET_CHAR(c, cptr, cbuf_end);
if (s->ignore_case) { if (s->ignore_case) {
c = lre_canonicalize(c, s->is_utf16); c = lre_canonicalize(c, s->is_unicode);
} }
idx_min = 0; idx_min = 0;
low = get_u16(pc + 0 * 4); low = get_u16(pc + 0 * 4);
@ -2334,7 +2336,7 @@ static intptr_t lre_exec_backtrack(REExecContext *s, uint8_t **capture,
goto no_match; goto no_match;
GET_CHAR(c, cptr, cbuf_end); GET_CHAR(c, cptr, cbuf_end);
if (s->ignore_case) { if (s->ignore_case) {
c = lre_canonicalize(c, s->is_utf16); c = lre_canonicalize(c, s->is_unicode);
} }
idx_min = 0; idx_min = 0;
low = get_u32(pc + 0 * 8); low = get_u32(pc + 0 * 8);
@ -2426,13 +2428,13 @@ int lre_exec(uint8_t **capture,
re_flags = bc_buf[RE_HEADER_FLAGS]; re_flags = bc_buf[RE_HEADER_FLAGS];
s->multi_line = (re_flags & LRE_FLAG_MULTILINE) != 0; s->multi_line = (re_flags & LRE_FLAG_MULTILINE) != 0;
s->ignore_case = (re_flags & LRE_FLAG_IGNORECASE) != 0; s->ignore_case = (re_flags & LRE_FLAG_IGNORECASE) != 0;
s->is_utf16 = (re_flags & LRE_FLAG_UTF16) != 0; s->is_unicode = (re_flags & LRE_FLAG_UNICODE) != 0;
s->capture_count = bc_buf[RE_HEADER_CAPTURE_COUNT]; s->capture_count = bc_buf[RE_HEADER_CAPTURE_COUNT];
s->stack_size_max = bc_buf[RE_HEADER_STACK_SIZE]; s->stack_size_max = bc_buf[RE_HEADER_STACK_SIZE];
s->cbuf = cbuf; s->cbuf = cbuf;
s->cbuf_end = cbuf + (clen << cbuf_type); s->cbuf_end = cbuf + (clen << cbuf_type);
s->cbuf_type = cbuf_type; s->cbuf_type = cbuf_type;
if (s->cbuf_type == 1 && s->is_utf16) if (s->cbuf_type == 1 && s->is_unicode)
s->cbuf_type = 2; s->cbuf_type = 2;
s->opaque = opaque; s->opaque = opaque;

View File

@ -34,10 +34,9 @@
#define LRE_FLAG_IGNORECASE (1 << 1) #define LRE_FLAG_IGNORECASE (1 << 1)
#define LRE_FLAG_MULTILINE (1 << 2) #define LRE_FLAG_MULTILINE (1 << 2)
#define LRE_FLAG_DOTALL (1 << 3) #define LRE_FLAG_DOTALL (1 << 3)
#define LRE_FLAG_UTF16 (1 << 4) #define LRE_FLAG_UNICODE (1 << 4)
#define LRE_FLAG_STICKY (1 << 5) #define LRE_FLAG_STICKY (1 << 5)
#define LRE_FLAG_INDICES (1 << 6) /* Unused by libregexp, just recorded. */ #define LRE_FLAG_INDICES (1 << 6) /* Unused by libregexp, just recorded. */
#define LRE_FLAG_NAMED_GROUPS (1 << 7) /* named groups are present in the regexp */ #define LRE_FLAG_NAMED_GROUPS (1 << 7) /* named groups are present in the regexp */
uint8_t *lre_compile(int *plen, char *error_msg, int error_msg_size, uint8_t *lre_compile(int *plen, char *error_msg, int error_msg_size,

View File

@ -43355,7 +43355,7 @@ static JSValue js_compile_regexp(JSContext *ctx, JSValueConst pattern,
mask = LRE_FLAG_DOTALL; mask = LRE_FLAG_DOTALL;
break; break;
case 'u': case 'u':
mask = LRE_FLAG_UTF16; mask = LRE_FLAG_UNICODE;
break; break;
case 'y': case 'y':
mask = LRE_FLAG_STICKY; mask = LRE_FLAG_STICKY;
@ -43373,7 +43373,7 @@ static JSValue js_compile_regexp(JSContext *ctx, JSValueConst pattern,
JS_FreeCString(ctx, str); JS_FreeCString(ctx, str);
} }
str = JS_ToCStringLen2(ctx, &len, pattern, !(re_flags & LRE_FLAG_UTF16)); str = JS_ToCStringLen2(ctx, &len, pattern, !(re_flags & LRE_FLAG_UNICODE));
if (!str) if (!str)
return JS_EXCEPTION; return JS_EXCEPTION;
re_bytecode_buf = lre_compile(&re_bytecode_len, error_msg, re_bytecode_buf = lre_compile(&re_bytecode_len, error_msg,
@ -44040,7 +44040,7 @@ static JSValue JS_RegExpDelete(JSContext *ctx, JSValueConst this_val, JSValueCon
break; break;
} }
if (end == start) { if (end == start) {
if (!(re_flags & LRE_FLAG_UTF16) || (unsigned)end >= str->len || !str->is_wide_char) { if (!(re_flags & LRE_FLAG_UNICODE) || (unsigned)end >= str->len || !str->is_wide_char) {
end++; end++;
} else { } else {
string_getc(str, &end); string_getc(str, &end);
@ -44815,7 +44815,7 @@ static const JSCFunctionListEntry js_regexp_proto_funcs[] = {
JS_CGETSET_MAGIC_DEF("ignoreCase", js_regexp_get_flag, NULL, LRE_FLAG_IGNORECASE ), JS_CGETSET_MAGIC_DEF("ignoreCase", js_regexp_get_flag, NULL, LRE_FLAG_IGNORECASE ),
JS_CGETSET_MAGIC_DEF("multiline", js_regexp_get_flag, NULL, LRE_FLAG_MULTILINE ), JS_CGETSET_MAGIC_DEF("multiline", js_regexp_get_flag, NULL, LRE_FLAG_MULTILINE ),
JS_CGETSET_MAGIC_DEF("dotAll", js_regexp_get_flag, NULL, LRE_FLAG_DOTALL ), JS_CGETSET_MAGIC_DEF("dotAll", js_regexp_get_flag, NULL, LRE_FLAG_DOTALL ),
JS_CGETSET_MAGIC_DEF("unicode", js_regexp_get_flag, NULL, LRE_FLAG_UTF16 ), JS_CGETSET_MAGIC_DEF("unicode", js_regexp_get_flag, NULL, LRE_FLAG_UNICODE ),
JS_CGETSET_MAGIC_DEF("sticky", js_regexp_get_flag, NULL, LRE_FLAG_STICKY ), JS_CGETSET_MAGIC_DEF("sticky", js_regexp_get_flag, NULL, LRE_FLAG_STICKY ),
JS_CGETSET_MAGIC_DEF("hasIndices", js_regexp_get_flag, NULL, LRE_FLAG_INDICES ), JS_CGETSET_MAGIC_DEF("hasIndices", js_regexp_get_flag, NULL, LRE_FLAG_INDICES ),
JS_CFUNC_DEF("exec", 1, js_regexp_exec ), JS_CFUNC_DEF("exec", 1, js_regexp_exec ),