I'm not even absolutely sure about why the ctypes logic is that way or the exact behavior that it implies. As far as I can tell the code here is doing the same thing, but not sure I can write relevant tests.

We can write a comment for the whatsnew and/or the documentation: when ffi.dlopen(filename, flags) is called on Windows with flags equal to zero or omitted, then:

1. cffi automatically provides some reasonable flags to LoadLibraryEx();
2. if filename has got a path, but a relative one, then it is replaced with an absolute path.

If needed, you can avoid all this by giving an explicit non-zero flags.

Is there still something to do here?

I guess we should recreate the reproducer in #139 on windows, and make sure it now passes.