On Sat, Nov 20, 2010 at 9:10 AM, Ted Unangst <ted.unan...@gmail.com> wrote: > On Sat, Nov 20, 2010 at 11:54 AM, Ted Unangst <ted.unan...@gmail.com> wrote: >> On Sat, Nov 20, 2010 at 9:18 AM, Mark Kettenis <mark.kette...@xs4all.nl> > wrote: >>>> openbsd is apparently among the last operating systems to require >>>> sys/types.h before sys/socket.h. posix doesn't require this and it runs >>>> contrary to current recommendations i think, so it's just one more weird >>>> thing to deal with when trying to get something to compile. >>>> >>>> i haven't really tested this, i'm just throwing it out there. >>> >>> Sorry, but blindly including <sys/types.h> isn't allowed by POSIX. >> >> Requiring everyone to include sys/types.h before socket.h isn't >> allowed by posix either. This way, at least the code compiles without >> changes. >> >> our socket.h header is broken. this makes it better.
IMO, this takes it from "not enough", right past "good", to "too much"... > By, the way, there's this comment in types.h: > > * XPG4.2 states that inclusion of <netinet/in.h> must pull these > * in and that inclusion of <sys/socket.h> must pull in sa_family_t. > * We put these here because there are other headers that require > * these types and <sys/socket.h> and <netinet/in.h> will indirectly > * include <sys/types.h>. > > So somebody already thinks socket.h includes types.h. The next line of that comment is interesting: > * XXX - now that we have protected versions these should move. Why not do what that suggests and move the sa_family_t and socklen_t types to socket.h? Copy the #ifndef...#endif blocks for ssize_t and size_t there at the same time, add includes of <sys/cdefs.h> and <sys/_types.h> and give compilation a whirl... Philip Guenther
