Use locale insensitive method to parse double literals.

runtime/vm/object.cc

Index: runtime/vm/object.cc
diff --git a/runtime/vm/object.cc b/runtime/vm/object.cc
index aa1aac6319890ba3e63a236a9cbd33827415942c..6841a3c3a96b7e818cab0c11678b48cc20b6fb81 100644
--- a/runtime/vm/object.cc
+++ b/runtime/vm/object.cc
@@ -10586,29 +10586,9 @@ RawDouble* Double::New(double d, Heap::Space space) {
 }
 
 
-static bool IsWhiteSpace(char ch) {
-  return ch == '\0' || ch == '\n' || ch == '\r' || ch == ' ' || ch == '\t';
-}
-
-
-static bool StringToDouble(const String& str, double* double_value) {
-  ASSERT(double_value != NULL);
-  // TODO(regis): For now, we use strtod to convert a string to double.
-  const char* nptr = str.ToCString();
-  char* endptr = NULL;
-  *double_value = strtod(nptr, &endptr);
-  // We do not treat overflow or underflow as an error and therefore do not
-  // check errno for ERANGE.
-  if (!IsWhiteSpace(*endptr)) {
-    return false;
-  }
-  return true;
-}
-
-
 RawDouble* Double::New(const String& str, Heap::Space space) {
   double double_value;
-  if (!StringToDouble(str, &double_value)) {
+  if (!StringToDouble(str.ToCString(), str.Length(), &double_value)) {

[Vyacheslav Egorov (Google), 2013/04/05 13:45:25]

I made it more strict than it was before. All tests still pass.

I am not sure what was the reason to allow white space after string
representation of a double.

     return Double::Handle().raw();
   }
   return New(double_value, space);
@@ -10645,7 +10625,7 @@ RawDouble* Double::NewCanonical(double value) {
 
 RawDouble* Double::NewCanonical(const String& str) {
   double double_value;
-  if (!StringToDouble(str, &double_value)) {
+  if (!StringToDouble(str.ToCString(), str.Length(), &double_value)) {
     return Double::Handle().raw();
   }
   return NewCanonical(double_value);