Move viewport unit resolution to style recalc time

Source/platform/LengthFunctions.cpp

Index: Source/platform/LengthFunctions.cpp
diff --git a/Source/platform/LengthFunctions.cpp b/Source/platform/LengthFunctions.cpp
index b1b72fafb99abffe7a4244110a00e3306b9d53a5..d6af4604bc712e799d1f9370cdfc388dbd61bc44 100644
--- a/Source/platform/LengthFunctions.cpp
+++ b/Source/platform/LengthFunctions.cpp
@@ -43,17 +43,14 @@ float floatValueForLength(const Length& length, float maximumValue)
         return static_cast<float>(maximumValue);
     case Calculated:
         return length.nonNanCalculatedValue(maximumValue);
-    case ViewportPercentageWidth:

[kenneth.r.christiansen, 2013/12/16 10:08:19]

Ain't you breaking @viewport { width: 100%; } when removing this? If not, I
think we can solve the device-width in a similar way.

[Timothy Loh, 2013/12/18 07:18:27]

You probably mean 100vw here; this works because we re-resolve viewport
descriptions on resizes (and not yet for the meta tag).

The cases device-width and device-height get compared against in
PageScaleConstraintsSet.cpp and WebViewImpl.cpp in some cases. Also I suspect if
we resolve these and try to compare actual numbers, these might not be equal in
cases with zoom (since we take a zoomed value and divide by the zoom factor).

As for actually having new types in the enum, I have plans for reworking Length
so these will be restricted to only where it is relevant.

-    case ViewportPercentageHeight:
-    case ViewportPercentageMin:
-    case ViewportPercentageMax:
-        return 0;
     case Intrinsic:
     case MinIntrinsic:
     case MinContent:
     case MaxContent:
     case FitContent:
     case ExtendToZoom:
+    case DeviceWidth:
+    case DeviceHeight:
     case Undefined:
         ASSERT_NOT_REACHED();
         return 0;