Re: Cygwin port of Guile 2.2

[Andy Wingo <wingo@pobox.com>]

On Wed 03 May 2017 05:18, Derek Upham <sand@blarg.net> writes:

> Andy Wingo <wingo@pobox.com> writes:
>
>> On Mon 01 May 2017 22:48, Derek Upham <sand@blarg.net> writes:
>>
>>> Running pthread_join() on a thread only guarantees that the thread has
>>> returned an exit value.
>>
>> Would you mind providing a reference please?  It is not that I don't
>> believe you but I think it's important to know whether this is a bug in
>> Guile or in the pthreads implementation.
>
> It’s not explicit, but it’s heavily implied by the pthread_exit(3) man
> page:

A specific implementation's man page is interesting but not the
specification.  The specification is "The Open Group Base Specifications
Issue 7 IEEE Std 1003.1-2008, 2016 Edition" and can be found here:

  http://pubs.opengroup.org/onlinepubs/9699919799/

Note that pthread_exit() is not the normal way for a Guile thread to
exit.  I thought we removed uses of this but I see that there are a
couple more that need to be removed.  Are you using it?  It's certainly
not one that we suggest.  We don't use pthread_cleanup_push either --
the lifetime of the thread from Guile's POV isn't limited to a dynamic
extent of scm_with_guile, as the thread could be created by the user, as

  https://lists.gnu.org/archive/html/guile-devel/2011-04/msg00133.html

Anyway, I digress.  After looking at the specification of pthread_join
(http://pubs.opengroup.org/onlinepubs/9699919799/functions/pthread_join.html),
things do not seem clear to me.  Does the thread "terminate" before or
after the pthread_key destructors run?  Given that:

  * pthread_key destructors must run in the thread AFAIU; to be able to
    call pthread_setspecific in a destructor you need a fresh
    or from a fresh new thread, and spawning a new thread to run cleanup
    seems unlikely

  * pthread_key destructors can call any function

I think there's an argument that a thread doesn't "terminate" until its
thread-local key destructors have finished running, and therefore
pthread_join doesn't return until after the key destructors have run.
This is my understanding of what happens from reading NPTL.  Do I
understand correctly that you are on Cygwin?  Could it be a cygwin
pthreads incompatibility?

Andy