HTMLPreloadScanner should only use valid <base> urls

Source/core/html/parser/HTMLPreloadScanner.cpp

Index: Source/core/html/parser/HTMLPreloadScanner.cpp
diff --git a/Source/core/html/parser/HTMLPreloadScanner.cpp b/Source/core/html/parser/HTMLPreloadScanner.cpp
index a98df60e30f99459ef336dd255f28deb2d1f4c31..83b1e5f221ba2c31b6c0f796338d9d8164abd431 100644
--- a/Source/core/html/parser/HTMLPreloadScanner.cpp
+++ b/Source/core/html/parser/HTMLPreloadScanner.cpp
@@ -515,8 +515,11 @@ template<typename Token>
 void TokenPreloadScanner::updatePredictedBaseURL(const Token& token)
 {
     ASSERT(m_predictedBaseElementURL.isEmpty());
-    if (const typename Token::Attribute* hrefAttribute = token.getAttributeItem(hrefAttr))
-        m_predictedBaseElementURL = KURL(m_documentURL, stripLeadingAndTrailingHTMLSpaces(hrefAttribute->value)).copy();
+    if (const typename Token::Attribute* hrefAttribute = token.getAttributeItem(hrefAttr)) {
+        KURL url(m_documentURL, stripLeadingAndTrailingHTMLSpaces(hrefAttribute->value));
+        if (url.isValid())
+            m_predictedBaseElementURL = url.copy();
+    }

[Yoav Weiss, 2015/03/31 11:04:43]

What does the parser do in case of an invalid base href? Does it ignore it as
well?

[kouhei (in TOK), 2015/03/31 11:15:38]

Good point. Let me verify.

[kouhei (in TOK), 2015/04/01 03:49:05]

It seems that the parser would set empty url as the base url when encountered
invalid base href.
I updated the code here to follow the behaviour.

 }
 
 HTMLPreloadScanner::HTMLPreloadScanner(const HTMLParserOptions& options, const KURL& documentURL, PassRefPtr<MediaValues> mediaValues)