[Bug libstdc++/92616] Inconsistency in time between system_clock::now() and time(nullptr)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92616 Jonathan Wakely changed: What|Removed |Added Status|UNCONFIRMED |RESOLVED Resolution|--- |MOVED --- Comment #8 from Jonathan Wakely --- OK, let's close this as MOVED then, thanks!
[Bug libstdc++/92616] Inconsistency in time between system_clock::now() and time(nullptr)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92616 --- Comment #7 from Anthony Williams --- Reported as ubuntu bug: https://bugs.launchpad.net/ubuntu/+source/linux-signed-oem/+bug/1853807
[Bug libstdc++/92616] Inconsistency in time between system_clock::now() and time(nullptr)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92616 --- Comment #6 from Jonathan Wakely --- https://bugzilla.kernel.org/ but I think they prefer bugs to be reported to distros unless you're actually using the upstream kernel. So report it to Ubuntu (but mention it's also seen in Fedora).
[Bug libstdc++/92616] Inconsistency in time between system_clock::now() and time(nullptr)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92616 --- Comment #5 from Anthony Williams --- Where can I file a bug in the vDSO?
[Bug libstdc++/92616] Inconsistency in time between system_clock::now() and time(nullptr)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92616
--- Comment #4 from Jonathan Wakely ---
gettimeofday agrees with clock_gettime(CLOCK_REALTIME):
void dumpNow() {
struct timespec ts;
clock_gettime(CLOCK_REALTIME, );
auto const now = time(nullptr);
struct timeval tv;
gettimeofday(, NULL);
tm nowTm;
localtime_r(_sec, );
std::cout << "Now (clock_gettime) time is " << ts.tv_sec << std::setw(9) <<
std::setfill('0') << ts.tv_nsec
<< "\n=>" << ts.tv_sec << "(time_t)\n=>" << nowTm.tm_hour
<< ":" << nowTm.tm_min << ":" << nowTm.tm_sec << std::endl;
localtime_r(, );
std::cout << "Now (time(nullptr)) time is " << now << "(time_t)\n=>"
<< nowTm.tm_hour << ":" << nowTm.tm_min << ":" << nowTm.tm_sec
<< std::endl;
localtime_r(_sec, );
std::cout << "Now (gettimeofday) time is " << tv.tv_sec << std::setw(6) <<
std::setfill('0') << tv.tv_usec << "\n=>" << tv.tv_sec << "(time_t)\n=>"
<< nowTm.tm_hour << ":" << nowTm.tm_min << ":" << nowTm.tm_sec
<< std::endl;
}
Now (clock_gettime) time is 1574344361836948044
=>1574344361(time_t)
=>13:52:41
Now (time(nullptr)) time is 1574344361(time_t)
=>13:52:41
Now (gettimeofday) time is 1574344361836949
=>1574344361(time_t)
=>13:52:41
Now (clock_gettime) time is 1574344363000108423
=>1574344363(time_t)
=>13:52:43
Now (time(nullptr)) time is 1574344362(time_t)
=>13:52:42
Now (gettimeofday) time is 1574344363000108
=>1574344363(time_t)
=>13:52:43
[Bug libstdc++/92616] Inconsistency in time between system_clock::now() and time(nullptr)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92616 --- Comment #3 from Jonathan Wakely --- (In reply to Jonathan Wakely from comment #1) > I can reproduce it, but I think this has to be a glibc problem. Libstdc++ > simply calls clock_gettime(3), and both that and time(3) come from glibc. Correction, they come from the linux kernel's vDSO, not glibc. Apparently there are different paths for computing the seconds for time(3) and the CLOCK_REALTIME result. Either way, there's nothing we can do about it in libstdc++.
[Bug libstdc++/92616] Inconsistency in time between system_clock::now() and time(nullptr)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92616
--- Comment #2 from Jonathan Wakely ---
I see the same result when using clock_gettime directly instead of
system_clock::now()
#include
#include
#include
#include
void dumpNow() {
struct timespec ts;
clock_gettime(CLOCK_REALTIME, );
auto const nowSc = ts.tv_sec;
auto const now = time(nullptr);
tm nowTm;
localtime_r(, );
std::cout << "Now (sc) time is " << ts.tv_sec << std::setw(9) <<
std::setfill('0') << ts.tv_nsec
<< " (system clock)\n=>" << nowSc << "(time_t)\n=>" <<
nowTm.tm_hour
<< ":" << nowTm.tm_min << ":" << nowTm.tm_sec << std::endl;
localtime_r(, );
std::cout << "Now (time(nullptr)) time is " << now << "(time_t)\n=>"
<< nowTm.tm_hour << ":" << nowTm.tm_min << ":" << nowTm.tm_sec
<< std::endl;
}
int main() {
dumpNow();
std::this_thread::sleep_until(
std::chrono::system_clock::from_time_t(time(nullptr) + 2));
dumpNow();
}
Now (sc) time is 1574343394343888130 (system clock)
=>1574343394(time_t)
=>13:36:34
Now (time(nullptr)) time is 1574343394(time_t)
=>13:36:34
Now (sc) time is 157434339683583 (system clock)
=>1574343396(time_t)
=>13:36:36
Now (time(nullptr)) time is 1574343395(time_t)
=>13:36:35
[Bug libstdc++/92616] Inconsistency in time between system_clock::now() and time(nullptr)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92616 --- Comment #1 from Jonathan Wakely --- I can reproduce it, but I think this has to be a glibc problem. Libstdc++ simply calls clock_gettime(3), and both that and time(3) come from glibc.
