FYI, these patches and those I will send later supersede the previous patchsets I sent yesterday for fixing and making libvirt work on OpenBSD. On Tue, Sep 04, 2012 at 04:49:52PM +0200, Jasper Lievisse Adriaanse wrote: -- Cheers, Jasper "Stay Hungry. Stay Foolish"

On Tue, Sep 04, 2012 at 09:20:24AM -0600, Eric Blake wrote: OpenBSD's sys/time.h has this: /* * Structure returned by gettimeofday(2) system call, * and used in other calls. */ struct timeval { long tv_sec; /* seconds */ long tv_usec; /* and microseconds */ }; #ifndef _TIMESPEC_DECLARED #define _TIMESPEC_DECLARED /* * Structure defined by POSIX.1b to be like a timeval. */ struct timespec { time_t tv_sec; /* seconds */ long tv_nsec; /* and nanoseconds */ }; #endif Oversight probably.. -- Cheers, Jasper "Stay Hungry. Stay Foolish"

On 09/04/2012 08:20 AM, Eric Blake wrote: Most likely this problem arose because of the patch I pushed in gnulib commit e07d7c40f3ca5ec410cf5aa6fa03cfe51e712039. Previously, gnulib required timeval's tv_sec to be the same size as time_t. But now, it requires only that tv_sec be big enough to hold a time_t. This patch was needed for Emacs. Without the patch, gnulib replaced struct timeval on OpenBSD, and this messed up utimens.c, and Emacs wouldn't build. Alternatively, gnulib could substitute its own struct timeval for the system's, wrapping every struct timeval-using function (gettimeofday, futimesat, futimes, lutimes, etc. That'd be more work, though. And it would introduce some performance issues with gettimeofday, which is supposed to be fast. I've been trying to get away from using struct timeval, and to use the higher-resolution struct timespec instead, so messing with these obsolescent interfaces has been lower priority for me. But if someone wants to take the more-ambitious approach that'd be fine, I expect. For this particular case, though, how about if we avoid the problem entirely? libvirt doesn't need to use struct timeval here at all. I'd use the following (untested) patch: it makes libvirt smaller and probably faster, and it ports to OpenBSD without messing with gnulib: diff --git a/tools/virsh-domain.c b/tools/virsh-domain.c index f0ec742..3cef782 100644 --- a/tools/virsh-domain.c +++ b/tools/virsh-domain.c @@ -3707,7 +3707,7 @@ static char * vshGenFileName(vshControl *ctl, virDomainPtr dom, const char *mime) { char timestr[100]; - struct timeval cur_time; + time_t cur_time; struct tm time_info; const char *ext = NULL; char *ret = NULL; @@ -3723,8 +3723,8 @@ vshGenFileName(vshControl *ctl, virDomainPtr dom, const char *mime) ext = ".png"; /* add mime type here */ - gettimeofday(&cur_time, NULL); - localtime_r(&cur_time.tv_sec, &time_info); + time (&cur_time); + localtime_r(&cur_time, &time_info); strftime(timestr, sizeof(timestr), "%Y-%m-%d-%H:%M:%S", &time_info); if (virAsprintf(&ret, "%s-%s%s", virDomainGetName(dom), diff --git a/tools/virsh.c b/tools/virsh.c index 5cf3237..88da429 100644 --- a/tools/virsh.c +++ b/tools/virsh.c @@ -2187,7 +2187,7 @@ vshOutputLogFile(vshControl *ctl, int log_level, const char *msg_format, char *str; size_t len; const char *lvl = ""; - struct timeval stTimeval; + time_t stTime; struct tm *stTm; if (ctl->log_fd == -1) @@ -2198,8 +2198,8 @@ vshOutputLogFile(vshControl *ctl, int log_level, const char *msg_format, * * [YYYY.MM.DD HH:MM:SS SIGNATURE PID] LOG_LEVEL message */ - gettimeofday(&stTimeval, NULL); - stTm = localtime(&stTimeval.tv_sec); + time (&stTime); + stTm = localtime(&stTime); virBufferAsprintf(&buf, "[%d.%02d.%02d %02d:%02d:%02d %s %d] ", (1900 + stTm->tm_year), (1 + stTm->tm_mon),