return default 1.0 for scaleFactor(), since android doesn't record at the same res...

diff --git a/WebKit/android/WebCoreSupport/ChromeClientAndroid.cpp b/WebKit/android/WebCoreSupport/ChromeClientAndroid.cpp
index dd608b6..015f9a0 100644 (file)
--- a/WebKit/android/WebCoreSupport/ChromeClientAndroid.cpp
+++ b/WebKit/android/WebCoreSupport/ChromeClientAndroid.cpp
@@ -81,8 +81,28 @@ FloatRect ChromeClientAndroid::pageRect() { notImplemented(); return FloatRect()
 
 float ChromeClientAndroid::scaleFactor()
 {
-    ASSERT(m_webFrame);
-    return m_webFrame->density();
+    /*  Conceptually, we should return the density of the device's screen,
+        except that we don't really model webkit's notion of drawing. Webkit
+        expects the drawing context's matrix (plus scale-factor) to reflect the
+        actually transformation to the pixels, and thus it uses scaleFactor to
+        perform some pre-rounding in a few places (e.g. HTMLCanvasElement).
+
+        ASSERT(m_webFrame);
+        return m_webFrame->density();
+
+        However, in Android we capture the drawing into a displayList, and then
+        replay that list at various scale factors (sometimes zoomed out, other
+        times zoomed in for "normal" reading, yet other times at arbitrary
+        zoom values based on the user's choice). In all of these cases, we do
+        not re-record the displayList, hence it is usually harmful to perform
+        any pre-rounding, since we just don't know the actual drawing resolution
+        at record time.
+
+        Given Android's device-independent drawing model, we always return 1.0
+        for this function, and we implement roundToDevicePixels() to be a no-op
+        so that we don't introduce pre-rounding artifacts during record.
+     */
+    return 1.0f;
 }
 
 void ChromeClientAndroid::focus() {