https://bugs.webkit.org/show_bug.cgi?id=118768

Reviewed by Ryosuke Niwa.

The first thing done by createUninitialized() is check if the length passed
is zero. Internally, there are many cases for which we know the check will never succeed.

Clang is usually really smart for those kind of things, but there are a few cases where
the condition to avoid returning empty() is not simply a check for the length.
This patch adds an internal initializer to deal with that.

* wtf/text/StringImpl.cpp:
(WTF::StringImpl::createUninitializedInternal):
(WTF::StringImpl::createUninitializedInternalNonEmpty):
(WTF::StringImpl::createInternal): Create internal has a special case for null pointer
for the characters. The test also check length, the second check for length cannot fail.
(WTF::StringImpl::create8BitIfPossible): ditto.
(WTF::StringImpl::lower): 3 of the calls cannot be reached for empty length. On an empty length,
the test for (noUpper && !(ored & ~0x7F)) would have caused the function to return "this".

For the last createUninitialized(), there is no guarantee the realLength is not zero.

(WTF::StringImpl::replace): The first thing we do in replace(UChar,UChar) is check if there is anything
to replace. The check for length will never succeed as there must be a character to replace at that point.
* wtf/text/StringImpl.h:


git-svn-id: http://svn.webkit.org/repository/webkit/trunk@153686 268f45cc-cd09-0410-ab3c-d52691b4dbfc