I want to have the number of milliseconds since epoch. A popular solution looks like follows (one of the solutions of this question asked here Get time since epoch in milliseconds, preferably using C++11 chrono )
#include <iostream>
#include <chrono>
int main() {
auto millitime = std::chrono::duration_cast<std::chrono::milliseconds>
(std::chrono::system_clock::now().time_since_epoch()).count();
std::cout << millitime << std::endl;
return 0;
}
compiling this with a call to g++
like g++ -std=c++11 main.cpp -o timetest
results in the output
1372686001
which is equal to the number of seconds since epoch!
Is this a bug in the glibc? in g++? my mistake?
g++ (Debian 4.7.3-4) 4.7.3
ldd (Debian EGLIBC 2.17-6) 2.17
Update: it works when using the g++ 4.8. So it is a gcc bug?!
g++-4.8 (Debian 4.8.1-2) 4.8.1
constexpr ToDuration duration_cast(const std::chrono::duration<Rep,Period>& d); (since C++11) Converts a std::chrono::duration to a duration of different type ToDuration . The function does not participate in overload resolution unless ToDuration is a specialization of std::chrono::duration.
Chronos time is how we measure our days and our lives quantitatively.
I think what's happening is that you are compiling with GCC 4.7 but the run-time linker is using the libstdc++.so
from a different GCC version, and they are configured with different precision for std::chrono:system_clock
. If you use LD_LIBRARY_PATH
or suitable linker options to ensure you compile with GCC 4.7 and use its libstdc++.so
then the results should be correct.
For example:
$ $HOME/gcc/4.7.1/bin/g++ -std=c++11 t.cc
$ ./a.out
1372693222
$ LD_LIBRARY_PATH=$HOME/gcc/4.7.1/lib64 ./a.out
1372693225128
The difference happens because the call to system_clock::now()
is in the libstdc++.so
library so the result depends on which library is used at run-time, but the duration_cast
conversion from that value to milliseconds
is done by inline templates which are instantiated at compile-time. If the compile-time conversion is not consistent with the run-time call then the results are inconsistent.
For GCC 4.8.1 the system_clock
implementation was improved to always use the clock_gettime
system call if it's available, which was not the case for 4.7, so it consistently uses the high-precision clock no matter how GCC was configured, which probably explains why you don't see the problem with 4.8.1.
You should always ensure the right version of libstdc++.so
is used at run-time.
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With