boost thread's usage of CreateWaitableTimer wakes PC from sleep (doh)

[denger]

Hi,

after upgrading to boost 1.58 from 1.57 a lot of users complained that my app was prematurely aborting the PC's sleep state. The culprit can easily detected and it was indeed my process: "powercfg /lastwake"

It shows that the process had set a "wake timer". The changes in boost/libs/thread/src/win32/thread.cpp seem to be the cause.

Please remove usage of "CreateWaitableTimer" in function "interruptible_wait()" or at least provide some means (preprocessor variable, whatever) to allow compilation of boost::thread without "CreateWaitableTimer".

Thanks

[anonymous]

Test system: Windows 7 x64

[anonymous]

CreateWaitableTimer is used by two functions in boost/libs/thread/src/win32/thread.cpp:

bool interruptible_wait(detail::win32::handle handle_to_wait_for,detail::timeout target_time)
bool non_interruptible_wait(detail::win32::handle handle_to_wait_for,detail::timeout target_time)

[anonymous]

The problem isn't the use of waitable timers per se, but the use of SetWaitableTimerEx, which always wakes the system from sleep if possible. Boost.Thread should revert to using plain SetWaitableTimer, ensuring that false is passed for the bResume parameter so that the timer doesn't wake the system.

[anonymous]

but the use of SetWaitableTimerEx

You're right! I have reverted thread.cpp from version 1.58 to 1.57 and my users have confirmed that the unexpected wake up problem is solved! So this is indeed a boost regression after 1.57.

[viboes]

Niall, please could you take a look.

[Niall Douglas]

Replying to anonymous:

The problem isn't the use of waitable timers per se, but the use of SetWaitableTimerEx, which always wakes the system from sleep if possible. Boost.Thread should revert to using plain SetWaitableTimer, ensuring that false is passed for the bResume parameter so that the timer doesn't wake the system.

​https://social.msdn.microsoft.com/Forums/vstudio/en-US/8047b845-2a1b-4220-a4b7-b10234bce924/unable-to-resume-in-windows-7-through-set-waitable-timer?forum=vcgeneral claimed that resumption of power from SetWaitableTimerEx didn't work. I therefore made the untested assumption that the WAKE_CONTEXT I passed didn't wake the machine.

I'll test that over here and see what I can figure out. The Microsoft documentation on WAKE_CONTEXT approaches nil unfortunately.

Niall

[Niall Douglas]

Replying to ned14:

I'll test that over here and see what I can figure out. The Microsoft documentation on WAKE_CONTEXT approaches nil unfortunately.

I'll be honest: until the AFIO review and then CppCon is over, I won't have the time to investigate this. Anyone else willing to take it on is welcome.

Niall

[anonymous]

We have the same issue with 1.58. Any updates on a fix for this?

Thanks!

Markus

[viboes]

Hi, I believe that we will need to wait until Niall has some time to look at this. I'm unable to test on Windows and don't the Windows interface well.

[Niall Douglas]

Replying to viboes:

Hi, I believe that we will need to wait until Niall has some time to look at this. I'm unable to test on Windows and don't the Windows interface well.

I gave it a quick test on my Win8.1 machine here and found no wakes from sleep. I'd suggest it's something in the system settings maybe. I didn't have much time though, it was only a very quick test.

I'd suggest reverting the patch I committed. I am no longer employed by the client who wanted that patch, so I have no objections for it returning to the previous code as that is less hassle for everybody.

Niall

[viboes]

Niall please, be free to do it.

[Niall Douglas]

Replying to viboes:

Niall please, be free to do it.

I'm on vacation from Boost library development until January. Had enough of Boost for a while.

But if someone else raises a pull request on github reverting the commit, I will check it and approve it.

Niall

[Mikael Olenfalk]

This pull request fixes the issue, passing null as REASON_CONTEXT is not document but works according to tests we made.

​https://github.com/boostorg/thread/pull/67

[viboes]

Niall, what do you think of the patch? What others think?

[viboes]

​https://github.com/boostorg/thread/commit/9ac736a4c4214cdb01caaa84a1121b3846dc87b4

[anonymous]

what version is the issue fixed in ?

[anonymous]

Who can tell which version has the fix? is it 1.61.0?

[raad@…]

The fix is in 1.60, as you can tell from the tag of the commit on github.