The problem was that libretroshare handles files in UTF-8 but Windows's ANSI/POSIX C functions automatically assume that the char* parameters are encoded with the system's code page. There is no way to set that code page as UTF-8. So now under Windows the code translates the file name to UTF-16 before feeding it to one of the Unicode functions (they are usually prefixed or suffixed by 'w'). Please note that it is not very efficient.
Furthermore, Windows does not provide a Unicode version of opendir/readdir/closedir, so it was necessary to use FindFirstFileW/FindNextFileW/FindClose which has a different behaviour as well as different structures. The FolderIterator class was created in order to mimic the Unix way of traversing folders contents. Hence the algorithm is unchanged and the systems differences masked.
As it was necessary to use some functions from the Windows API, <windows.h> had to be included in a few files were it didn't appear before, creating macros and #define conflicts. In order to solve them, util/rswin.h must be included first in a file (if necessary). Otherwise the preprocessor will fail on purpose to avoid the code being compiled with different _WIN32_WINNT values.
As another side-effect, rstlvutil.h and rstlvutil.cc have been removed from libretroshare.pro file. They are only used by testing units and include util/utest.h which defines a macro FAILED that already exists in <windows.h>. I don't know if unit tests are still in use and I don't plan on coding often on Windows, so I'll leave that as an exercise (hot potato?) to a motivated fellow Windows programmer.
git-svn-id: http://svn.code.sf.net/p/retroshare/code/trunk@2924 b45a01b8-16f6-495d-af2f-9b41ad6348cc
b30056e17f