[windows] boost::this_thread::sleep_for() sleeps longer than it should in Windows

[Ondrej Siler <ondrej.siler@…>]

Calling the this_thread::sleep_for() results in a sleep almost twice as long as desired.

code:

#include <iostream>
#include <boost/thread.hpp>

int main(int argc, char * argv[])
{
  boost::chrono::system_clock::time_point start = boost::chrono::system_clock::now();
  boost::this_thread::sleep_for(boost::chrono::seconds(1));
  boost::chrono::system_clock::time_point end = boost::chrono::system_clock::now();
  std::cout << (boost::chrono::duration_cast<boost::chrono::microseconds>(end - start)).count() << std::endl;
  return 0;
}

When built in MS Windows 7 x64, the program outputs 2000114

The same code in linux x64 outputs 1000298

[viboes]

I have no access to a windows machine now. Could you do the test for 5 seconds? and 100ms?

[Ondrej Siler <ondrej.siler@…>]

Same behaviour - for 5 seconds sleeps 10 secs, for 100ms sleeps 200ms.

[viboes]

Please, could you check if the following update works for you?

Committed in trunk revision [83282].

[Ondrej Siler <ondrej.siler@…>]

I cannot link without BOOST_USE_WINDOWS_H defined.

With BOOST_USE_WINDOWS_H defined, it works ok.

Shouldn't be the test in thread/win32/thread_primitives.hpp:21

#if _WIN32_WINNT >= 0x0600

instead of

#if WINNT >= 0x0600

MS Visual Studio (IDE) doesn't know WINNT, but is happy with _WIN32_WINNT

[viboes]

Replying to Ondrej Siler <ondrej.siler@…>:

I cannot link without BOOST_USE_WINDOWS_H defined.

With BOOST_USE_WINDOWS_H defined, it works ok.

Shouldn't be the test in thread/win32/thread_primitives.hpp:21

#if _WIN32_WINNT >= 0x0600

instead of

#if WINNT >= 0x0600

MS Visual Studio (IDE) doesn't know WINNT, but is happy with _WIN32_WINNT

I suspect that you are talking of #8070, isn't it?

[Ondrej Siler <ondrej.siler@…>]

Yes, you are right. According to ​MSDN: "To compile an application that uses this function, define _WIN32_WINNT as 0x0600 or later."

[viboes]

Committed revision [83294][83318].

[viboes]

Committed revision [83525].

[anonymous]

on boost 1.56 this bug is back. sleeping for 1000 microseconds sleeps for about 2,000 microseconds. sleeping for 1 millisecond sleeps for about 2 milliseconds. sleeping for 5 seconds sleeps for 5 seconds.

Micro and milliseconds are doubling, but seconds are not doubling.

[viboes]

Niall, do you think this could be related to you improvement with GetTickCount64 or the PR related to WinRT?

[Niall Douglas]

I ran this test on Win8.1 x86 using VS2013 release x86 build on Boost trunk (1.57):

1015707

I ran it a few more times, all were minor variations around 1015xxx.

I'd say this bug is stale.

Niall