[zeromq-dev] "cython inline checkrc" change in PyZMQ 13.0.0 changes handling of EINTR?

[MinRK]

To further clarify, there are two cases:

1. no sigint handler registered - ^C absolutely should result in
KeyboardInterrupt.  This behavior is intended, and has not changed since
2.1.9.
2. sigint handler registered - in this case, KeyboardInterrupt should *not*
be raised, and the regular ZMQError(EINTR) should happen.  This was broken
in 13.0.0, but is fixed in master.

13.0 was a big release, and there are unsurprisingly a few bugs.  There
will be a bugfix release around 30 days after the original (about two weeks
from now).

-MinRK


On Mon, Mar 11, 2013 at 2:16 PM, Min RK <benjaminrk at gmail.com> wrote:

>
>
> On Mar 11, 2013, at 13:13, Jonathan Kamens <jkamens at quantopian.com> wrote:
>
>  Greetings,
>
> With PyZMQ versions prior to 13.0.0, we were running into problems with
> certain PyZMQ calls getting interrupted by restartable signals (e.g.,
> SIGALRM) used by our application. We fixed this problem like this:
>
>             while True:
>                 try:
>                     self.context.term()
>                 except zmq.ZMQError as exc:
>                     if exc.errno == EINTR:
>                         log.info('zmq_term interrupted by signal,
> restarting')
>                     else:
>                         log.exception('Error terminating ZMQ context')
>                         raise
>                 except BaseException as exc:
>                     log.exception('Error terminating ZMQ context')
>                     raise
>                 else:
>                     break
>
> Note that 0MQ allows term() to be restarted when it gets EINTR<http://api.zeromq.org/2-1:zmq-term#toc4>
> .
>
> This worked just fine, but has stopped working in PyZMQ 13.0.0. Now,
> instead of a ZMQError with errno set to EINTR, we are getting a
> KeyboardInterrupt exception. I think this commit to PyZMQ<https://github.com/zeromq/pyzmq/commit/3959e4515f86db11f660738df5d36f62478c39e2>is the cause.
>
> Questions:
>
>    1. Was this change in behavior intentional? If so, it probably should
>    be documented in the release notes.
>    2. Am I understanding correctly that what needs to be done to fix the
>    problem in my app is to do "except KeyboardInterrupt" instead of "except
>    zmq.ZMQError as exc" and not bother to check errno?
>
> Thanks in advance for any help you can provide.
>
>
> This was not intentional, and is fixed in master.  There should only be a
> difference in behavior if you have non-default signal handler registered.
>
>
> Regards,
>
> Jonathan Kamens
>
> _______________________________________________
> zeromq-dev mailing list
> zeromq-dev at lists.zeromq.org
> http://lists.zeromq.org/mailman/listinfo/zeromq-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.zeromq.org/pipermail/zeromq-dev/attachments/20130311/d2ba2cdb/attachment.htm>