Re: [boost] read_write_mutex fundamentally broken in 1.33?

From: John Maddock (john_at_[hidden])
Date: 2005-09-09 06:34:58

> You can't lock two locks willy-nilly like this.  So ideally what I'd
> like to do is create a generic lock function that will lock multiple
> locks without danger of deadlock:
>
> template <class TryLock1, class TryLock2>
> void
> lock(TryLock1& l1, TryLock2& l2);
>
> and maybe also:
>
> template <class TryLock1, class TryLock2, class TryLock3>
> void
> lock(TryLock1& l1, TryLock2& l2, TryLock3& l3);

Yep, those would be useful additions to Boost.Thread IMO.

> Problem:  With the current interface, my "generic" lock algorithm is no
> longer generic.  It has to know about read_lock() and write_lock(),
> even though these concepts have absolutely nothing to do with how you
> lock two locks while avoiding deadlock.

Ah, that's what happens from looking at my not so good test case, rather 
than the docs :-)

try_read_write_mutex does have member typedefs try_read_lock and 
try_write_lock, which do what you want them to, I just happened to use a 
different locking primitive because it was easier to change the test code 
around to test different combinations of readers and writers.

> Sorry, but:  Houston, we have an even bigger problem...

Not this time, still it was big enough to begin with :-(

John.