Fix float parsing in srcset

Source/core/html/parser/HTMLSrcsetParser.cpp

Index: Source/core/html/parser/HTMLSrcsetParser.cpp
diff --git a/Source/core/html/parser/HTMLSrcsetParser.cpp b/Source/core/html/parser/HTMLSrcsetParser.cpp
index 8283ecd41536a733ce2d23e426c427da4eeffa4c..e14c4e4953ceb6fb67df34d2a843a0d5d279c798 100644
--- a/Source/core/html/parser/HTMLSrcsetParser.cpp
+++ b/Source/core/html/parser/HTMLSrcsetParser.cpp
@@ -97,7 +97,11 @@ struct DescriptorToken {
             isValid = false;
             return 0;
         }
-        return charactersToFloat(attribute + start, lengthExcludingDescriptor, &isValid);
+        Decimal result = parseToDecimalForNumberType(String(attribute + start, lengthExcludingDescriptor));
+        isValid = result.isFinite();
+        if (!isValid)
+            return 0;
+        return clampTo<float>(result.toDouble());

[tkent, 2015/08/25 10:02:46]

This looks to have a behavior change for numbers greater than the float limit.
Do we have a test?

[Yoav Weiss, 2015/08/25 10:07:58]

I didn't mean for that to be a behavior change. Is there a difference between
casting a double to a float (which was the previous behavior) and clampTo?
Should this be a static_case instead?

[tkent, 2015/08/25 10:14:23]

AFAIK, static_cast<float>(greater-than-float-max) returns Infinity, and
clampTo<float>() returns the maximum value of float.

     }
 };