Re: [boost] More Threads Review

From: David Abrahams (dave_at_[hidden])
Date: 2002-10-18 10:19:43

"Peter Dimov" <pdimov_at_[hidden]> writes:

> Of historical interest: from my review of the Threads library, 31.08.2001:
> 
> "The mutex is missing one bit of functionality that I consider critical.
> When
> two threads need to lock two (or more) mutexes, the lock order must be the
> same to avoid deadlocks. No support is provided for this operation. Possible
> alternatives are
> 
> template<class M1, class M2, ...> class scoped_multiple_lock
> {
>     scoped_multiple_lock(M1 &, M2 &, ...);
> };
> 
> or
> 
> template<class M1, class M2, ...> class mutex_tuple
> {
>     typedef ... scoped_lock;
>     mutex_tuple(M1 &, M2 &, ...);
> };"
> 
> "I'd like to see
> 
> template<class M, class P, class F> void wait(M & m, P p, F f)
> {
>     typename M::scoped_lock lock(m);
>     while(!p()) this->wait(lock);
>     f();
> }
> 
> added, for people that want to go the extra mile (when someone bothers to
> define a predicate to use the wait(l, p) form, (s)he would probably define
> an action object, too.)"

FWIW, I support all of Peter's points above.

-- 
                    David Abrahams
dave_at_[hidden] * http://www.boost-consulting.com
Building C/C++ Extensions for Python: Dec 9-11, Austin, TX
http://www.enthought.com/training/building_extensions.html