Fix double-rounding in strtod for MinGW.

src/strtod.cc

-// 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:
 //
 //     * Redistributions of source code must retain the above copyright
 //       notice, this list of conditions and the following disclaimer.
 //     * Redistributions in binary form must reproduce the above
 //       copyright notice, this list of conditions and the following
 //       disclaimer in the documentation and/or other materials provided
 //       with the distribution.
@@ skipping 156 matching lines @@
     int exponent = 0;
     *result = DiyFp(significand, exponent);
     *remaining_decimals = buffer.length() - read_digits;
   }
 }
 
 
 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) {
     int read_digits;
     // The trimmed input fits into a double.
     // If the 10^exponent (resp. 10^-exponent) fits into a double too then we
     // can compute the result-double simply by multiplying (resp. dividing) the
     // two numbers.
     // This is possible because IEEE guarantees that floating-point operations
     // return the best possible approximation.
@@ skipping 236 matching lines @@
 
   double guess;
   if (DoubleStrtod(trimmed, exponent, &guess) ||
       DiyFpStrtod(trimmed, exponent, &guess)) {
     return guess;
   }
   return BignumStrtod(trimmed, exponent, guess);
 }
 
 } }  // namespace v8::internal