void KURLPrivate::setASCII(const CString& string)
{
m_utf8 = string;
m_utf8IsASCII = true;
m_stringIsValid = false;
initProtocolIsInHTTPFamily();
initInnerURL();
}
KURL::KURL(const AtomicString& canonicalString, const url::Parsed& parsed, bool isValid)
: m_isValid(isValid)
, m_protocolIsInHTTPFamily(false)
, m_parsed(parsed)
, m_string(canonicalString)
{
initProtocolIsInHTTPFamily();
initInnerURL();
}
void KURL::init(const KURL& base,
const String& relative,
const WTF::TextEncoding* queryEncoding) {
// As a performance optimization, we do not use the charset converter
// if encoding is UTF-8 or other Unicode encodings. Note that this is
// per HTML5 2.5.3 (resolving URL). The URL canonicalizer will be more
// efficient with no charset converter object because it can do UTF-8
// internally with no extra copies.
StringUTF8Adaptor baseUTF8(base.getString());
// We feel free to make the charset converter object every time since it's
// just a wrapper around a reference.
KURLCharsetConverter charsetConverterObject(queryEncoding);
KURLCharsetConverter* charsetConverter =
(!queryEncoding || isUnicodeEncoding(queryEncoding))
? 0
: &charsetConverterObject;
// Clamp to int max to avoid overflow.
url::RawCanonOutputT<char> output;
if (!relative.isNull() && relative.is8Bit()) {
StringUTF8Adaptor relativeUTF8(relative);
m_isValid = url::ResolveRelative(baseUTF8.data(), baseUTF8.length(),
base.m_parsed, relativeUTF8.data(),
clampTo<int>(relativeUTF8.length()),
charsetConverter, &output, &m_parsed);
} else {
m_isValid = url::ResolveRelative(baseUTF8.data(), baseUTF8.length(),
base.m_parsed, relative.characters16(),
clampTo<int>(relative.length()),
charsetConverter, &output, &m_parsed);
}
// AtomicString::fromUTF8 will re-hash the raw output and check the
// AtomicStringTable (addWithTranslator) for the string. This can be very
// expensive for large URLs. However, since many URLs are generated from
// existing AtomicStrings (which already have their hashes computed), this
// fast path is used if the input string is already canonicalized.
//
// Because this optimization does not apply to non-AtomicStrings, explicitly
// check that the input is Atomic before moving forward with it. If we mark
// non-Atomic input as Atomic here, we will render the (const) input string
// thread unsafe.
if (!relative.isNull() && relative.impl()->isAtomic() &&
StringView(output.data(), static_cast<unsigned>(output.length())) ==
relative) {
m_string = relative;
} else {
m_string = AtomicString::fromUTF8(output.data(), output.length());
}
initProtocolIsInHTTPFamily();
initInnerURL();
DCHECK_EQ(protocol(), protocol().lower());
}
void KURL::init(const KURL& base, const String& relative, const WTF::TextEncoding* queryEncoding)
{
if (!relative.isNull() && relative.is8Bit()) {
StringUTF8Adaptor relativeUTF8(relative);
init(base, relativeUTF8.data(), relativeUTF8.length(), queryEncoding);
} else
init(base, relative.characters16(), relative.length(), queryEncoding);
initProtocolIsInHTTPFamily();
initInnerURL();
}
// Setters for the data. Using the ASCII version when you know the
// data is ASCII will be slightly more efficient. The UTF-8 version
// will always be correct if the caller is unsure.
void KURLPrivate::setUTF8(const CString& string)
{
const char* data = string.data();
unsigned dataLength = string.length();
// The m_utf8IsASCII must always be correct since the DeprecatedString
// getter must create it with the proper constructor. This test can be
// removed when DeprecatedString is gone, but it still might be a
// performance win.
m_utf8IsASCII = true;
for (unsigned i = 0; i < dataLength; i++) {
if (static_cast<unsigned char>(data[i]) >= 0x80) {
m_utf8IsASCII = false;
break;
}
}
m_utf8 = string;
m_stringIsValid = false;
initProtocolIsInHTTPFamily();
initInnerURL();
}