Fix double-rounding in strtod for MinGW.

src/strtod.cc

Index: src/strtod.cc
diff --git a/src/strtod.cc b/src/strtod.cc
index be79c80085200dc8015e53687e6e3e9c80f38418..0dc618a399f9df8b0ecb9c8b3039bd4edb158727 100644
--- a/src/strtod.cc
+++ b/src/strtod.cc
@@ -1,4 +1,4 @@
-// Copyright 2011 the V8 project authors. All rights reserved.
+// Copyright 2012 the V8 project authors. All rights reserved.
 // Redistribution and use in source and binary forms, with or without
 // modification, are permitted provided that the following conditions are
 // met:
@@ -175,13 +175,15 @@ static void ReadDiyFp(Vector<const char> buffer,
 static bool DoubleStrtod(Vector<const char> trimmed,
                          int exponent,
                          double* result) {
-#if (defined(V8_TARGET_ARCH_IA32) || defined(USE_SIMULATOR)) && !defined(WIN32)
+#if (defined(V8_TARGET_ARCH_IA32) || defined(USE_SIMULATOR)) \
+    && !defined(_MSC_VER)
   // On x86 the floating-point stack can be 64 or 80 bits wide. If it is
   // 80 bits wide (as is the case on Linux) then double-rounding occurs and the
   // result is not accurate.
-  // We know that Windows32 uses 64 bits and is therefore accurate.
-  // Note that the ARM simulator is compiled for 32bits. It therefore exhibits
-  // the same problem.
+  // We know that Windows32 with MSVC, unlike with MinGW32, uses 64 bits and is
+  // therefore accurate.
+  // Note that the ARM and MIPS simulators are compiled for 32bits. They
+  // therefore exhibit the same problem.
   return false;
 #endif
   if (trimmed.length() <= kMaxExactDoubleIntegerDecimalDigits) {