Paul Eggert wrote: > For example, lib/parse-duration.h currently > doesn't have the <config.h> reminder, even though it uses time_t and > therefore is vulnerable to a bug if a program includes it without > including config.h (and defining _TIME_BITS) first.
Good point. parse-duration.h is not defining the time_t type; therefore we don't need the reminder there. In POSIX:2008, the application can get the time_t type defined by including one of the following files: <sched.h> <sys/msg.h> <sys/select.h> <sys/sem.h> <sys/shm.h> <sys/stat.h> <sys/time.h> <sys/types.h> <time.h> <utime.h> Gnulib should therefore override each of these include files and make sure <config.h> (with the possible _TIME_BITS definition) is included first. > Do we need to make > another pass through the .h files and comments to deal with time_t, > struct timespec, _TIME_BITS, etc.? In my opinion, only the 10 modules mentioned above need to be changed or created. > And similarly for off_t? off_t is defined by one of these files: <aio.h> <fcntl.h> <stdio.h> <sys/mman.h> <sys/stat.h> <sys/types.h> <unistd.h> The same reasoning applies. But for off_t, IMO, the urgency is low, since most packages have already been using AC_SYS_LARGEFILE for nearly 20 years. Therefore there are not many application bugs that we can hope to uncover in this area now. Bruno
