This is the mail archive of the cygwin mailing list for the Cygwin project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: pthread_join() problem

On 07/06/13 16:58, Christopher Faylor wrote:
On Mon, Jun 03, 2013 at 12:01:02PM +0200, Corinna Vinschen wrote:
>On Jun  2 23:01, David Stacey wrote:
>>I'm trying to get Poco[1] working under Cygwin, and have hit a
>>problem with the way it manages its threads. A short example
>>(attached) shows what's going on.
>>Poco creates a global object that is used for managing various
>>threads. In the destructor, the class calls pthread_join() to wait
>>for the threads to finish. This works in Fedora (and presumably
>>other Linux variants), but under Cygwin this call never returns.
>>Note that the object is global, and so pthread_join() is being
>>called after main() has returned and the global variables are being
>>mopped up.
>>In the attached example, the problem only exists if the
>>'thread_container' object is global. When run in this way,
>>pthread_join() never returns and the programme locks up. If
>>'thread_container' is moved local to main() then the programme works
>>correctly. Sadly, I can't make such a trivial fix to the Poco code,
>>which rather relies on this object being global.
>>The attached example (and indeed Poco) runs fine under Fedora 18
>>x64, but locks up on Cygwin (32-bit; haven't tried 64-bit). Problem
>>exists with a vanilla cygwin-1.7.18-1 and the most recent snapshot
>Thanks for the simple testcase!
>I tracked this down to a problem in the process exit handling which
>disallowed pthreads to exit when process exit was in progress (and
>running the global destructors is part of the process exit).  I fixed
>that in CVS.
>I'm just uploading a 2013-06-03 snapshot to
>as well as generating a 1.7.9-10 cygwin package for the 64 bit test
>distro.  Please give either one of them a try, especially with poco.
I've reworked this change since Corinna's fix was to what should have
been obsolete code.  A new snapshot should be uploaded soon.

Could you confirm that things still work as expected?

Thank you for looking at this problem and for providing a new snapshot. I have tested the 2013-06-07 snapshot, and can confirm that both my testcase and Poco threads function correctly.

Many thanks once again,


Problem reports:
Unsubscribe info:

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]