--- a/src/corelib/tools/qstring.cpp Tue Feb 02 00:43:10 2010 +0200
+++ b/src/corelib/tools/qstring.cpp Fri Apr 16 15:50:13 2010 +0300
@@ -1,6 +1,6 @@
/****************************************************************************
**
-** Copyright (C) 2009 Nokia Corporation and/or its subsidiary(-ies).
+** Copyright (C) 2010 Nokia Corporation and/or its subsidiary(-ies).
** All rights reserved.
** Contact: Nokia Corporation (qt-info@nokia.com)
**
@@ -885,7 +885,7 @@
QString QString::fromWCharArray(const wchar_t *string, int size)
{
if (sizeof(wchar_t) == sizeof(QChar)) {
- return fromUtf16((ushort *)string, size);
+ return fromUtf16((const ushort *)string, size);
} else {
return fromUcs4((uint *)string, size);
}
@@ -3857,6 +3857,12 @@
If \a size is -1 (default), \a unicode must be terminated
with a 0.
+ This function checks for a Byte Order Mark (BOM). If it is missing,
+ host byte order is assumed.
+
+ This function is comparatively slow.
+ Use QString(const ushort *, int) if possible.
+
QString makes a deep copy of the Unicode data.
\sa utf16(), setUtf16()
@@ -3923,6 +3929,9 @@
If \a unicode is 0, nothing is copied, but the string is still
resized to \a size.
+ Note that unlike fromUtf16(), this function does not consider BOMs and
+ possibly differing byte ordering.
+
\sa utf16(), setUnicode()
*/
@@ -4669,6 +4678,8 @@
Returns the QString as a '\\0\'-terminated array of unsigned
shorts. The result remains valid until the string is modified.
+ The returned string is in host byte order.
+
\sa unicode()
*/
@@ -7740,7 +7751,7 @@
return QString();
if (m_size && m_position == 0 && m_size == m_string->size())
return *m_string;
- return QString::fromUtf16(reinterpret_cast<const ushort*>(m_string->unicode() + m_position), m_size);
+ return QString(m_string->unicode() + m_position, m_size);
}