Rework float value/percentage scanning for VTTRegion

Source/core/html/track/vtt/VTTScanner.cpp

Index: Source/core/html/track/vtt/VTTScanner.cpp
diff --git a/Source/core/html/track/vtt/VTTScanner.cpp b/Source/core/html/track/vtt/VTTScanner.cpp
index fd6a662d8f43930ae4781b5d0b6698ec7e2574e3..590178f886450149786f234b05195c4932c82f4d 100644
--- a/Source/core/html/track/vtt/VTTScanner.cpp
+++ b/Source/core/html/track/vtt/VTTScanner.cpp
@@ -139,4 +139,33 @@ unsigned VTTScanner::scanDigits(int& number)
     return numDigits;
 }
 
+bool VTTScanner::scanFloat(float& number)
+{
+    Run integerRun = collectWhile<isASCIIDigit>();
+    seekTo(integerRun.end());
+    Run decimalRun(position(), position(), m_is8Bit);
+    if (scan('.')) {
+        decimalRun = collectWhile<isASCIIDigit>();
+        seekTo(decimalRun.end());
+    }
+
+    // At least one digit required.
+    if (integerRun.isEmpty() && decimalRun.isEmpty()) {
+        // Restore to starting position.
+        seekTo(integerRun.start());
+        return false;

[Mike West, 2014/01/21 07:59:24]

Here you return false without setting `number`. At the bottom, you set invalid
numbers to float::max() and return true. Is that intentional?

[fs, 2014/01/21 08:19:57]

Intentional, yes. At this point, we have not managed to scan a valid float, so
we leave |number| as is and return false to signal failure (and don't consume
input).

Below however, we've verified that the input is wellformed - charactersToFloat
does that too, but has a differing definition of wellformed. Hence we interpret
|validNumber| - or rather, it's negation - as meaning that the number was too
large. We still return true though because the input was wellformed and we did
consume it (and |number| is set to the "best approximation" of the value). (If
we needed to detect the overflow as well, then we'd probably return some other
value for !validNumber.)

+    }
+
+    size_t lengthOfFloat = Run(integerRun.start(), position(), m_is8Bit).length();
+    bool validNumber;
+    if (m_is8Bit)
+        number = charactersToFloat(integerRun.start(), lengthOfFloat, &validNumber);
+    else
+        number = charactersToFloat(reinterpret_cast<const UChar*>(integerRun.start()), lengthOfFloat, &validNumber);
+
+    if (!validNumber)
+        number = std::numeric_limits<float>::max();
+    return true;
+}
+
 }