[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: Build on OSX failed - (pthreads) (ITS#1152)



The test is specifically designed to detect pthread systems
which do not allow detachment of a joinable thread, a behavior
OpenLDAP depends up.  Some implementations inappropriately do
not allow a joinable thread which has completed its work from
being detached.  This should be viewed a bug in their
implementation of Pthreads.

Kurt

At 09:02 AM 5/17/01, maillist@clemens-benden.de wrote:

>Am Mittwoch, 16. Mai 2001 um 19:53 schrieb Kurt D. Zeilenga:
>
>I posting this to the forum - I hope this is ok (normally I do not 
>forward personal mail to a public forum)
>
>> or provide additional details (such as which test should pass and why 
>> it's not -- see config.log)
>> so that a developer can make a stab at fixing it...
>
>After some investigation I discovered the folling:
>
>(please correct any mistake - I know my knowlegde is very basic and 
>darwin seems to be not so good documented at all, addtionally I'am new 
>to POSIX pthreads.)
>
>a) The library which must be used for linking is the default library 
>(tried that with a small app)
>b) The threads are created in the JOINABLE state per default
>c) A thread which is in the exit state cannot be detached (only joinable 
>can)
>
>So the current configure script must fail ("test for linking with the 
>default libraries") because the test spawn a thread (this is ok) then 
>the main routine sleeps a second. In the meantime the thread dies 
>because there are no statements to execute and the thread will be 
>terminated (implicit call to pthread_exit(). The thread is now in the 
>exit state).
>
>By changing the execution order (putting the sleep AFTER the detach 
>call) the test will function because  pthread_detach will see an living 
>thread which is JOINABLE and can be detached (I did that and the test is 
>running with exit code 0, without the change, the test will exit with 
>rc=22 and configure will fail)
>
>Any comments available? May be from a serious darwin developer? (I only 
>have the sources of darwin 1.0, maybe that changed to 1.2 - this is the 
>current version of MacOS X Retail).
>
>> fix it yourself and provide us with a description of the
>> problem and a patch.
>
>Well... I do not think, that a fix (even if I ever could apply one) 
>would be a great help...
>
>--
>
>Sincerly,
>    Clemens
>
>May