Changeset 63120 in webkit for trunk/JavaScriptCore/runtime/UString.cpp

Timestamp:

Jul 12, 2010, 2:32:34 PM (15 years ago)

Author:

Darin Adler

Message:

2010-07-09 Darin Adler <Darin Adler>

Reviewed by Geoffrey Garen.

String to number coercion is not spec compliant
​https://bugs.webkit.org/show_bug.cgi?id=31349

ToNumber should ignore NBSP (\u00a0)
​https://bugs.webkit.org/show_bug.cgi?id=25490

• runtime/JSGlobalObjectFunctions.cpp: (JSC::parseIntOverflow): Added a version that works on UChar.
• runtime/JSGlobalObjectFunctions.h: Ditto.

• runtime/UString.cpp: (JSC::isInfinity): Added helper functions. (JSC::UString::toDouble): Use isStrWhiteSpace instead of isSASCIISpace to define what we should skip. Got rid of the code that used CString and UTF8String, instead processing the UChar of the string directly, except for when we call strtod. For strtod, use our own home-grown conversion function that does not try to do any UTF-16 processing. Tidied up the logic a bit as well.

2010-07-09 Darin Adler <Darin Adler>

Reviewed by Geoffrey Garen.

String to number coercion is not spec compliant
​https://bugs.webkit.org/show_bug.cgi?id=31349

• fast/js/ToNumber-expected.txt: Updated to expect more tests to pass.
• fast/js/parseFloat-expected.txt: Ditto.
• fast/js/sputnik/Conformance/09_Type_Conversion/9.3_ToNumber/9.3.1_ToNumber_from_String/S9.3.1_A2-expected.txt: Ditto.
• fast/js/sputnik/Conformance/09_Type_Conversion/9.3_ToNumber/9.3.1_ToNumber_from_String/S9.3.1_A3_T1-expected.txt: Ditto.
• fast/js/sputnik/Conformance/09_Type_Conversion/9.3_ToNumber/9.3.1_ToNumber_from_String/S9.3.1_A3_T2-expected.txt: Ditto.
• fast/js/sputnik/Conformance/15_Native_Objects/15.1_The_Global_Object/15.1.2/15.1.2.3_parseFloat/S15.1.2.3_A2_T10-expected.txt: Ditto.
• fast/js/sputnik/Conformance/15_Native_Objects/15.1_The_Global_Object/15.1.2/15.1.2.3_parseFloat/S15.1.2.3_A2_T3-expected.txt: Ditto.
• fast/js/sputnik/Conformance/15_Native_Objects/15.1_The_Global_Object/15.1.2/15.1.2.3_parseFloat/S15.1.2.3_A2_T8-expected.txt: Ditto.
• fast/js/sputnik/Conformance/15_Native_Objects/15.1_The_Global_Object/15.1.2/15.1.2.3_parseFloat/S15.1.2.3_A2_T9-expected.txt: Ditto.
• fast/js/sputnik/Conformance/15_Native_Objects/15.1_The_Global_Object/15.1.2/15.1.2.3_parseFloat/S15.1.2.3_A6-expected.txt: Ditto.

File:

• trunk/JavaScriptCore/runtime/UString.cpp (modified) (4 diffs)

trunk/JavaScriptCore/runtime/UString.cpp

@@ -243 +243 @@
 }
 
+static inline bool isInfinity(double number)
+{
+    return number == Inf || number == -Inf;
+}
+
+static bool isInfinity(const UChar* data, const UChar* end)
+{
+    return data + 7 < end
+        && data[0] == 'I'
+        && data[1] == 'n'
+        && data[2] == 'f'
+        && data[3] == 'i'
+        && data[4] == 'n'
+        && data[5] == 'i'
+        && data[6] == 't'
+        && data[7] == 'y';
+}
+
 double UString::toDouble(bool tolerateTrailingJunk, bool tolerateEmptyString) const
 {
-    if (size() == 1) {
+    unsigned size = this->size();
+
+    if (size == 1) {
         UChar c = data()[0];
         if (isASCIIDigit(c))
             return c - '0';
-        if (isASCIISpace(c) && tolerateEmptyString)
+        if (isStrWhiteSpace(c) && tolerateEmptyString)
             return 0;
         return NaN;
@@ -265 +285 @@
     // right thing but requires UChar, not char, for its argument.
 
-    CString s = UTF8String();
-    if (s.isNull())
-        return NaN;
-    const char* c = s.data();
-
-    // skip leading white space
-    while (isASCIISpace(*c))
-        c++;
-
-    // empty string ?
-    if (*c == '\0')
+    const UChar* data = this->data();
+    const UChar* end = data + size;
+
+    // Skip leading white space.
+    for (; data < end; ++data) {
+        if (!isStrWhiteSpace(*data))
+            break;
+    }
+
+    // Empty string.
+    if (data == end)
         return tolerateEmptyString ? 0.0 : NaN;
 
-    double d;
-
-    // hex number ?
-    if (*c == '0' && (*(c + 1) == 'x' || *(c + 1) == 'X')) {
-        const char* firstDigitPosition = c + 2;
-        c++;
-        d = 0.0;
-        while (*(++c)) {
-            if (*c >= '0' && *c <= '9')
-                d = d * 16.0 + *c - '0';
-            else if ((*c >= 'A' && *c <= 'F') || (*c >= 'a' && *c <= 'f'))
-                d = d * 16.0 + (*c & 0xdf) - 'A' + 10.0;
-            else
+    double number;
+
+    if (data[0] == '0' && data + 2 < end && (data[1] | 0x20) == 'x' && isASCIIHexDigit(data[2])) {
+        // Hex number.
+        data += 2;
+        const UChar* firstDigitPosition = data;
+        number = 0;
+        while (true) {
+            number = number * 16 + toASCIIHexValue(*data);
+            ++data;
+            if (data == end)
                 break;
-        }
-
-        if (d >= mantissaOverflowLowerBound)
-            d = parseIntOverflow(firstDigitPosition, c - firstDigitPosition, 16);
+            if (!isASCIIHexDigit(*data))
+                break;
+        }
+        if (number >= mantissaOverflowLowerBound)
+            number = parseIntOverflow(firstDigitPosition, data - firstDigitPosition, 16);
     } else {
-        // regular number ?
-        char* end;
-        d = WTF::strtod(c, &end);
-        if ((d != 0.0 || end != c) && d != Inf && d != -Inf) {
-            c = end;
-        } else {
-            double sign = 1.0;
-
-            if (*c == '+')
-                c++;
-            else if (*c == '-') {
-                sign = -1.0;
-                c++;
-            }
-
+        // Decimal number.
+
+        // Put into a null-terminated byte buffer.
+        Vector<char, 32> byteBuffer;
+        for (const UChar* characters = data; characters < end; ++characters) {
+            UChar character = *characters;
+            byteBuffer.append(isASCII(character) ? character : 0);
+        }
+        byteBuffer.append(0);
+
+        char* byteBufferEnd;
+        number = WTF::strtod(byteBuffer.data(), &byteBufferEnd);
+        const UChar* pastNumber = data + (byteBufferEnd - byteBuffer.data());
+
+        if ((number || pastNumber != data) && !isInfinity(number))
+            data = pastNumber;
+        else {
             // We used strtod() to do the conversion. However, strtod() handles
             // infinite values slightly differently than JavaScript in that it
@@ -317 +338 @@
             // whereas the ECMA spec requires that it be converted to NaN.
 
-            if (c[0] == 'I' && c[1] == 'n' && c[2] == 'f' && c[3] == 'i' && c[4] == 'n' && c[5] == 'i' && c[6] == 't' && c[7] == 'y') {
-                d = sign * Inf;
-                c += 8;
-            } else if ((d == Inf || d == -Inf) && *c != 'I' && *c != 'i')
-                c = end;
+            double signedInfinity = Inf;
+            if (data < end) {
+                if (*data == '+')
+                    data++;
+                else if (*data == '-') {
+                    signedInfinity = -Inf;
+                    data++;
+                }
+            }
+            if (isInfinity(data, end)) {
+                number = signedInfinity;
+                data += 8;
+            } else if (isInfinity(number) && data < end && (*data | 0x20) != 'i')
+                data = pastNumber;
             else
                 return NaN;
@@ -327 +357 @@
     }
 
+    // Look for trailing junk.
     if (!tolerateTrailingJunk) {
-        // allow trailing white space
-        while (isASCIISpace(*c))
-            c++;
-        if (c != s.data() + s.length())
-            d = NaN;
-    }
-
-    return d;
+        // Allow trailing white space.
+        for (; data < end; ++data) {
+            if (!isStrWhiteSpace(*data))
+                break;
+        }
+        if (data != end)
+            return NaN;
+    }
+
+    return number;
 }