Re: [Boost-Users] bug in pthreads implementation of recursive_mut ex

From: John Van Horne (jvanhorne_at_[hidden])
Date: 2003-05-06 14:59:25

We have an application which uses boost threads and omnithreads. A 
developer encountered a problem where threads would deadlock when a 
condition variable was wait()-ed in a boost-created thread. Every once 
in a while the unlock was not seen by the CORBA created threads. The 
problem went away when we applied Yulik's patch.

Thanks, Yulik.

-John Van Horne

Feldman, Yulik wrote:

> Thanks Alexander for the pointer to this long and interesting
> discussion. I'll check my design whether I could get rid of the
> recursive mutexes in order to reduce the probability of not holding
> invariants when the wait() is called, especially since the code still
> doesn't behave correctly, even after the Boost bug is fixed :)
>
> William, unfortunately I don't have a small testcase for the bug and it
> seems that it will be too time-consuming for me to try to prepare a
> small "working" testcase. Thanks for confirming the bug report (it makes
> me feel better that the debugging time was not wasted).
>
> --Yulik.
>
> -----Original Message-----
> From: William E. Kempf [mailto:wekempf_at_[hidden]]
> Sent: Monday, May 05, 2003 6:44 PM
> To: Boost-Users_at_[hidden]
> Subject: Re: [Boost-Users] bug in pthreads implementation of
> recursive_mutex
>
>
> Feldman, Yulik said:
> > Hi,
> >
> >
> >
> > It looks that there is a bug in
> > .../boost_1_30_0/libs/thread/src/recursive_mutex.cpp. The bug shows in
> > the pthreads implementation of the recursive_mutex when the mutex is
> > used with the "condition" synchronization primitive. Here is the
> > description of the bug:
>
> You're evaluation of the bug is correct.  I'll fix this in CVS shortly.
> Do you maybe have a test case for this that I could add to the
> regression
> tests?
>
> BTW, Mr. Terekhov does provide a good resource in his posting on this
> subject.  I disagree with some points in the thread he posted, but it is
> quite true that you absolutely must gaurantee that invariants hold at
> the
> point cond.wait() is called, and that doing so is complicated by the use
> of recursive locks.  Either avoid recursive locks entirely, avoid
> condition variables with recursive mutexes, or be very sure of the
> design
> and the state of your invariants.
>
> -- 
> William E. Kempf
>
>
>
>
> Info: <http://www.boost.org>
> Wiki: <http://www.crystalclearsoftware.com/cgi-bin/boost_wiki/wiki.pl>
> Unsubscribe: <mailto:boost-users-unsubscribe_at_[hidden]>
>
>
> Your use of Yahoo! Groups is subject to
> http://docs.yahoo.com/info/terms/
>
>
>
>
> *Yahoo! Groups Sponsor*
> <http://rd.yahoo.com/M=249982.3179269.4495679.1261774/D=egroupweb/S=1705006788:HM/A=1524963/R=0/*http://hits.411web.com/cgi-bin/autoredir?camp=556&lineid=3179269&prop=egroupweb&pos=HM> 
>
>
>
> Info: <http://www.boost.org>
> Wiki: <http://www.crystalclearsoftware.com/cgi-bin/boost_wiki/wiki.pl>
> Unsubscribe: <mailto:boost-users-unsubscribe_at_[hidden]>
>
>
> Your use of Yahoo! Groups is subject to the Yahoo! Terms of Service 
> <http://docs.yahoo.com/info/terms/>.

[Non-text portions of this message have been removed]