@Cubbi, that is sad. Sad that MinGW is not fixing what they know is a bug.
It was several years ago that I started using TDM-GCC 4.9.2 and what at the time I believed was GCC 5.1. I ran across the std::random_device bug and started looking around for better alternatives. I was also less understanding of the differences between compilers.
I finally settled on MSVS, mostly because of the debugging. Any C++ features MSVC hasn't at this time implemented I haven't done any experimentation. Right now I'm delving into other aspects of C++11/C++14 I know MSVC has implemented.
To me as a user, though, unacceptable quality of implementation is a bug just the same.
My opinion as well.
@Sanboro, have you looked at the C++ working paper I linked to? There is a proposal for a very simple toolkit for using the C++ <random> library to replace rand()/srand(). All you need to do is include the toolkit source in your project and you get the robustness of C++ with a very simple and easy to use interface.