Improve support for :read-only and :read-write pseudoclasses.

Source/core/html/HTMLElement.cpp

Index: Source/core/html/HTMLElement.cpp
diff --git a/Source/core/html/HTMLElement.cpp b/Source/core/html/HTMLElement.cpp
index 51b0a8cf0bf2415e0a3bd0a0dcc0aff9b9e2e3f6..f617e6a623d1c7f23ee87ddc66d8cab5f77593f9 100644
--- a/Source/core/html/HTMLElement.cpp
+++ b/Source/core/html/HTMLElement.cpp
@@ -970,6 +970,26 @@ void HTMLElement::defaultEventHandler(Event* event)
     Element::defaultEventHandler(event);
 }
 
+bool HTMLElement::matchesReadOnlyPseudoClass() const
+{
+    return !matchesReadWritePseudoClass();
+}
+
+bool HTMLElement::matchesReadWritePseudoClass() const
+{
+    if (fastHasAttribute(contenteditableAttr)) {
+        const AtomicString& value = fastGetAttribute(contenteditableAttr);
+
+        if (value.isEmpty() || equalIgnoringCase(value, "true") || equalIgnoringCase(value, "plaintext-only"))
+            return true;
+        if (equalIgnoringCase(value, "false"))
+            return false;
+        // All other values should be treated as "inherit".
+    }
+
+    return rendererIsEditable();

[rune, 2014/01/27 12:36:44]

rendererIsEditable() says that it should have an assert about !needsStyleRecalc
internally. matchesReadWritePseudoClass() is called during matching. I think
that's going to cause a problem here. If you change the contenteditable
attribute of some ascendant, wouldn't rendererIsEditable here check the computed
-webkit-user-modify of the renderer that  inherited style from the parent before
it changed?

[andersr, 2014/01/30 12:27:59]

Yes, you are right, of course.

Checking the parent element instead should produce the correct behavior.

I added a test which verifies that the correct style is calculated.

+}
+
 void HTMLElement::handleKeypressEvent(KeyboardEvent* event)
 {
     if (!document().settings() || !document().settings()->spatialNavigationEnabled() || !supportsFocus())