Re: [boost] [winrt support] Adding support for Windows 8 store/phone to Boost libraries

Subject: Re: [boost] [winrt support] Adding support for Windows 8 store/phone to Boost libraries
From: Niall Douglas (s_sourceforge_at_[hidden])
Date: 2014-04-29 10:28:38

On 29 Apr 2014 at 16:49, Andrey Semashev wrote:

> > BTW, I'll shortly be coming to this list with a new sync object for
> > Boost.Thread called a "permit object" which wraps the failed POSIX
> > pthreads permit object proposal for C11 which solves some of the
> > problems of using condition variables directly.
> 
> <offtopic> I believe, the general target for low-level thread
> synchronization should be Boost.Sync. It's a little in a hanging state
> now, but in the long term it should supersede Boost.Thread
> synchronization primitives. That said, I'm not aware of "permit
> objects", so I don't know how well it would fit in Boost.Sync.
> </offtopic>

I should stress what I'm about to say is too early, because the Boost 
permit object is not yet ready, so please wait until next week for 
further discussion. However, I can tell you that the permit object 
implementation remains a completely standalone new POSIX pthreads 
object written in C. All the proposed boost::permit<> object does is 
to wrap that C object - and in fact, boost/thread/permit.hpp is quite 
literally a copy and paste of 
boost/thread/posix/condition_variable.hpp with every instance of 
pthread_cond_whatever replaced with pthread_permit_whatever. And of 
course a few small modifications to allow a pthread API to work 
correctly on Windows (the C permit object works on Windows) and to 
support thread interruptions etc.

In other words, if you wanted to add the permit to Boost.Sync, it 
ought to be absolutely trivial. Whatever refactor you're doing for 
condition_variable (POSIX implementation) you repeat on permit, and 
you're done.

Please do wait till next week though. I am currently writing a load 
soak test for Boost.Thread which tests mutexes, condvars and permits 
for perfection under heavy CPU load sustained over minutes or hours. 
As you may know, there have been some problems with Thread on Mac OS 
X, hopefully this test might illuminate the cause. I also don't want 
to bring boost::permit<> to here until I can prove it is perfect and 
has significant advantages over condvars for its particular use 
cases. As you know, it is extremely easy to accidentally have some 
extremely rare race condition in threading primitives, so on those I 
am extra special cautious.

Niall

-- 
Currently unemployed and looking for work in Ireland.
Work Portfolio: http://careers.stackoverflow.com/nialldouglas/