wine/gecko
0
0
Fork 0
mirror of https://gitlab.winehq.org/wine/wine-gecko.git synced 2024-09-13 09:24:08 -07:00
Code Issues Packages Projects Releases Wiki Activity

Bug 502789: correct regexp->native compiler of handling case-insensitive matching of characters with multiple lower-case forms, r=lw

Browse Source
This commit is contained in:
David Mandelin 2009-07-13 12:41:30 -07:00
parent 427ece476a
commit a10caff6bb
1 changed files with 98 additions and 32 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

130
js/src/jsregexp.cpp
View File

@ -2090,43 +2090,107 @@ class RegExpNativeCompiler {
LIns* compileFlatSingleChar(jschar ch, LIns* pos, LInsList& fails)
{
/*
* Fast case-insensitive test for ASCII letters: convert text
* char to lower case by bit-or-ing in 32 and compare.
*/
JSBool useFastCI = JS_FALSE;
jschar ch2 = ch; /* 2nd char to test for if ci */
if (cs->flags & JSREG_FOLD) {
if ((L'A' <= ch && ch <= L'Z') || (L'a' <= ch && ch <= L'z')) {
ch |= 32;
ch2 = ch;
useFastCI = JS_TRUE;
} else if (JS_TOLOWER(ch) != ch) {
ch2 = JS_TOLOWER(ch);
ch = JS_TOUPPER(ch);
}
}
LIns* to_fail = lir->insBranch(LIR_jf, lir->ins2(LIR_lt, pos, cpend), 0);
fails.add(to_fail);
LIns* text_ch = lir->insLoad(LIR_ldcs, pos, 0);
LIns* comp_ch = useFastCI ?
lir->ins2(LIR_or, text_ch, lir->insImm(32)) :
text_ch;
if (ch == ch2) {
fails.add(lir->insBranch(LIR_jf, lir->ins2(LIR_eq, comp_ch, lir->insImm(ch)), 0));
} else {
LIns* to_ok = lir->insBranch(LIR_jt, lir->ins2(LIR_eq, comp_ch, lir->insImm(ch)), 0);
fails.add(lir->insBranch(LIR_jf, lir->ins2(LIR_eq, comp_ch, lir->insImm(ch2)), 0));
if (!targetCurrentPoint(to_ok))
return NULL;
// Extra characters that need to be compared against when doing folding.
struct extra {
jschar ch;
LIns *match;
};
extra extras[5];
int nextras = 0;
if (cs->flags & JSREG_FOLD) {
ch = JS_TOUPPER(ch);
jschar lch = JS_TOLOWER(ch);
if (ch != lch) {
if (L'A' <= ch && ch <= L'Z') {
// Fast conversion of text character to lower case by OR-ing with 32.
text_ch = lir->ins2(LIR_or, text_ch, lir->insImm(32));
// These ASCII letters have 2 lower-case forms. We put the ASCII one in
// |extras| so it is tested first, because we expect that to be the common
// case. Note that the code points of the non-ASCII forms both have the
// 32 bit set, so it is OK to compare against the OR-32-converted text char.
ch = lch;
if (ch == L'i') {
extras[nextras++].ch = ch;
ch = 0x131;
} else if (ch == L's') {
extras[nextras++].ch = ch;
ch = 0x17f;
}
goto gen;
} else if (0x01c4 <= ch && ch <= 0x1e60) {
// The following group of conditionals handles characters that have 1 or 2
// lower-case forms in addition to JS_TOLOWER(ch).
if (ch <= 0x1f1) { // DZ,LJ,NJ
if (ch == 0x01c4) {
extras[nextras++].ch = 0x01c5;
} else if (ch == 0x01c7) {
extras[nextras++].ch = 0x01c8;
} else if (ch == 0x01ca) {
extras[nextras++].ch = 0x01cb;
} else if (ch == 0x01f1) {
extras[nextras++].ch = 0x01f2;
}
} else if (ch < 0x0392) { // no extra lower-case forms in this range
} else if (ch <= 0x03a6) { // Greek
if (ch == 0x0392) {
extras[nextras++].ch = 0x03d0;
} else if (ch == 0x0395) {
extras[nextras++].ch = 0x03f5;
} else if (ch == 0x0398) {
extras[nextras++].ch = 0x03d1;
} else if (ch == 0x0399) {
extras[nextras++].ch = 0x0345;
extras[nextras++].ch = 0x1fbe;
} else if (ch == 0x039a) {
extras[nextras++].ch = 0x03f0;
} else if (ch == 0x039c) {
extras[nextras++].ch = 0xb5;
} else if (ch == 0x03a0) {
extras[nextras++].ch = 0x03d6;
} else if (ch == 0x03a1) {
extras[nextras++].ch = 0x03f1;
} else if (ch == 0x03a3) {
extras[nextras++].ch = 0x03c2;
} else if (ch == 0x03a6) {
extras[nextras++].ch = 0x03d5;
}
} else if (ch == 0x1e60) { // S with dot above
extras[nextras++].ch = 0x1e9b;
}
}
extras[nextras++].ch = lch;
}
}
gen:
for (int i = 0; i < nextras; ++i) {
LIns *test = lir->ins2(LIR_eq, text_ch, lir->insImm(extras[i].ch));
LIns *branch = lir->insBranch(LIR_jt, test, 0);
extras[i].match = branch;
}
fails.add(lir->insBranch(LIR_jf, lir->ins2(LIR_eq, text_ch, lir->insImm(ch)), 0));
for (int i = 0; i < nextras; ++i) {
if (!targetCurrentPoint(extras[i].match))
return NULL;
}
return lir->ins2(LIR_piadd, pos, lir->insImm(2));
}
LIns* compileFlatDoubleChar(jschar ch1, jschar ch2, LIns* pos,
LInsList& fails)
JS_INLINE bool hasCases(jschar ch)
{
return JS_TOLOWER(ch) != JS_TOUPPER(ch);
}
LIns* compileFlatDoubleChar(jschar ch1, jschar ch2, LIns* pos, LInsList& fails)
{
#ifdef IS_BIG_ENDIAN
uint32 word = (ch1 << 16) | ch2;
@ -2140,9 +2204,11 @@ class RegExpNativeCompiler {
JSBool useFastCI = JS_FALSE;
union { jschar c[2]; uint32 i; } mask;
if (cs->flags & JSREG_FOLD) {
JSBool mask1 = (L'A' <= ch1 && ch1 <= L'Z') || (L'a' <= ch1 && ch1 <= L'z');
JSBool mask2 = (L'A' <= ch2 && ch2 <= L'Z') || (L'a' <= ch2 && ch2 <= L'z');
if ((!mask1 && JS_TOLOWER(ch1) != ch1) || (!mask2 && JS_TOLOWER(ch2) != ch2)) {
jschar uch1 = JS_TOUPPER(ch1);
jschar uch2 = JS_TOUPPER(ch2);
JSBool mask1 = (L'A' <= uch1 && uch1 <= L'Z' && uch1 != L'I' && uch1 != L'S');
JSBool mask2 = (L'A' <= uch2 && uch2 <= L'Z' && uch2 != L'I' && uch2 != L'S');
if ((!mask1 && hasCases(ch1)) || (!mask2 && hasCases(ch2))) {
pos = compileFlatSingleChar(ch1, pos, fails);
if (!pos) return NULL;
return compileFlatSingleChar(ch2, pos, fails);