|
Boost Users : |
Subject: Re: [Boost-users] Try-upgrade locking
From: Howard Hinnant (howard.hinnant_at_[hidden])
Date: 2011-12-13 17:53:31
On Dec 13, 2011, at 5:37 PM, Vicente J. Botet Escriba wrote:
> Le 13/12/11 18:57, Kelvin Chung a écrit :
>> Suppose I have a boost::upgrade_lock<boost::shared_mutex>. When I want to create the boost::upgrade_to_unique_lock<boost::shared_mutex> and upgrade to exclusive locking, this blocks until all the shared lock holders leave. However, it does not appear that there is a non-blocking version of this, unlike, say, making a boost::unique_lock<boost::shared_mutex> and passing in boost::try_to_lock.
>>
>> So what should I do in order to create my own "try-upgrade" lock? I'm trying to use it like so:
>>
>> boost::upgrade_lock<boost::shared_mutex> upgradeLock(mutex);
>> ...
>> if (...) {
>> // Substitute for boost::upgrade_to_unique_lock<boost::shared_mutex>, but also with aspects of
>> // boost::try_to_lock, basically
>> my_upgrade_to_unique_lock<boost::shared_mutex> deferredWriteLock(lock);
>>
>> // Do stuff where it doesn't matter whether I have exclusive or not, while the readers leave
>> if (!deferredWriteLock) deferredWriteLock.lock(); // Now we need exclusive
>> // We have exclusive, do more stuff
>> }
>> // Back to regular upgrade_lock here
>>
>>
> Hi,
>
> There is a draft proposal from Howard Hinnant "Shared locking in C++" http://home.roadrunner.com/~hinnant/bloomington/shared_mutex.html that doesn't have upgrade_to_unique_lock but that allows to try to lock an upgrade_lock. Even if the interface is a little bit different I guess that the lecture will help you a lot.
>
> If I understand you wan to lock a mutex with shared ownership until you need to have exclusive ownership and then retrieve the shard ownership again. Unfortunately I think that the current interface doesn't let you to try to move from shared to exclusive ownership without blocking, and I don't know how to move from a exclusive ownership to a shared on.
>
> If you can block when getting exclusive ownership, you will be able to implement it with the current interface as follows (Note pseudo-code)
>
> IUC, Howard's proposal contains all the ingredient to do it using a an initial shared_lock, trying to move to an upgrade_lock and if successful moving then to a unique_lock and last moving back to the initial shared_lock. See the example:
>
> upgrade_mutex mut;
>
> // ...
>
> void foo()
> {
> shared_lock<upgrade_mutex> sl(mut);
>
> // mut share locked here
>
>
> // ...
>
>
> // Try to convert mut from shared to upgrade ownership for 5ms
>
>
> // It can succeed if no other thread has upgrade ownership,
>
>
> // or is blocked on exclusive ownership.
>
> upgrade_lock<upgrade_mutex> ul(std::move(sl), chrono::milliseconds(5));
> if (ul.owns_lock())
> {
>
> // mut upgrade locked here
>
>
> // This thread will still share with other threads having shared ownership here.
>
>
> // ...
>
>
> // Convert mut from upgrade to exclusive, blocking as necessary
>
> unique_lock<upgrade_mutex> el(std::move(ul));
>
> // mut exclusive locked here
>
>
> // No other threads have shared, upgrade or exclusive ownership here.
>
>
> // ...
>
> }
> // mut.unlock()
>
> else
> {
>
> // mut still share locked here
>
>
> // ...
>
> }
> }
> // if mut is share locked, then mut.unlock_shared()
>
> I plan to adapt it to boost as a replacement of the current Boost.Thread/Boost.Interprocess implementation, but this will take some time.
>
> In the mean time you can create some feature requests, I will try to get them, but I really think the design must be changed globally.
If it helps anyone, here's my latest implementation:
http://home.roadrunner.com/~hinnant/mutexes/shared_mutex
http://home.roadrunner.com/~hinnant/mutexes/shared_mutex.cpp
There are probably bits in there that are libc++-specific. Haven't looked at it in awhile. But it is boost-open-sourced (not part of the boost library), and anyone who wants it is welcome to it.
Howard
Boost-users list run by williamkempf at hotmail.com, kalb at libertysoft.com, bjorn.karlsson at readsoft.com, gregod at cs.rpi.edu, wekempf at cox.net