Re: [Boost-users] timed waits

Subject: Re: [Boost-users] timed waits
From: Peter Dimov (pdimov_at_[hidden])
Date: 2010-03-26 08:23:23

"Just C" wrote:
> I was looking at this article about conditon_variables.
> http://www.justsoftwaresolutions.co.uk/threading/condition-variable-spurious-wakes.html
>
> For a timeout, the article has the following sample program with condition
> variables.

...

> However, shouldn't this program instead start with
>
>        boost::mutex::scoped_lock lock(the_mutex, timeout);
>        if(!lock.owns())
>              return false;
>
>
> i.e. shouldn't it also take care of getting stuck while locking the mutex.
>
> i.e. if the timeout is 5 seconds & the getting the lock itself takes 10
> seconds, then, the earlier the program will return is 10+ seconds.

In principle, yes. But in practice this should only matter in realtime code 
that needs to guarantee a response within the timeout period (which is 
typically measured in milliseconds or microseconds). Mutexes should never 
stay locked for 10 seconds in a correctly designed program - they should 
protect short critical sections. (Unless there is a deadlock, of course, in 
which case a timed mutex might help you detect it.)

There is also the practical consideration that not all platforms support 
timed mutexes (in POSIX they are part of the realtime option, IIRC). When 
this is the case, a timed mutex is emulated with an ordinary mutex and a 
condition variable, so you will not be gaining much from it.