I assumed that by using mode_t and preprocessor macros, I am working relatively portable so that I don't have to worry if mode_t is implemented as unsigned short or signed int.
But you do arithmetic operations on mode_t types. The ussual C language promotions apply.
Could you maybe just tell me how the code should look like to work on both Linux and BSD? Hardcoding a numeric literal really doesn't seem like the proper solution to me...
Just explicitely acknowledge the imlicit conversion:
__pyx_v_mode = (mode_t)((__pyx_v_mode & (~S_IFMT)) | S_IFDIR);
Petr