In GCC and Clang, the double NAN (defined in math.h) is JSON-encoded like:

{"value": nan}

In Microsoft Visual Studio, it's:

{"value": -nan(ind)}

This is seems wrong and likely violates the Extended JSON Spec. Furthermore, Infinity is encoded as "inf" with GCC and Clang on Linux and Mac, but "Infinity" with GCC on Solaris.