Hi PJ peeps, Please apply this patch which removes the pj_thread_sleep() call from pjsip_endpt_handle_events2(). I think it makes more sense just to return the error to the user immediately and let them decide if they want to sleep (like pjsua-lib does). One concrete case when this sleep is problematic is when a signal occurs. In this case, the call to pj_ioqueue_poll() will return because of EINTR which will cause the main loop to additionally sleep for the full timeout. This means, if pjsip_endpt_handle_events2() is called with a large timeout, the signal/interrupt will block the app for the full timeout interval if a signal happens. This is a problem. Thanks, Kal

Hi Kal, We're thinking to reduce the delay to max 10 ms instead of removing it completely. This should allow for error recovery. Please let us know your opinion/feedback. Patch attached. Regards, Ming On Wed, Sep 27, 2017 at 4:00 AM, b17 c0de <b17c0de@gmail.com> wrote: > Hi PJ peeps, > Please apply this patch which removes the pj_thread_sleep() call from > pjsip_endpt_handle_events2(). I think it makes more sense just to > return the error to the user immediately and let them decide if they > want to sleep (like pjsua-lib does). > > One concrete case when this sleep is problematic is when a signal > occurs. In this case, the call to pj_ioqueue_poll() will return > because of EINTR which will cause the main loop to additionally sleep > for the full timeout. This means, if pjsip_endpt_handle_events2() is > called with a large timeout, the signal/interrupt will block the app > for the full timeout interval if a signal happens. This is a problem. > > Thanks, > Kal > > _______________________________________________ > Visit our blog: http://blog.pjsip.org > > pjsip mailing list > pjsip@lists.pjsip.org > http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org >

Hi Kal, We have committed the patch, with some modifications, in ticket #2048 (https://trac.pjsip.org/repos/ticket/2048). Thanks for the patch and the discussion that followed. Best regards, Ming On Wed, Sep 27, 2017 at 4:27 PM, Ming <ming@teluu.com> wrote: > Hi Kal, > > We're thinking to reduce the delay to max 10 ms instead of removing it > completely. This should allow for error recovery. Please let us know > your opinion/feedback. Patch attached. > > Regards, > Ming > > On Wed, Sep 27, 2017 at 4:00 AM, b17 c0de <b17c0de@gmail.com> wrote: >> Hi PJ peeps, >> Please apply this patch which removes the pj_thread_sleep() call from >> pjsip_endpt_handle_events2(). I think it makes more sense just to >> return the error to the user immediately and let them decide if they >> want to sleep (like pjsua-lib does). >> >> One concrete case when this sleep is problematic is when a signal >> occurs. In this case, the call to pj_ioqueue_poll() will return >> because of EINTR which will cause the main loop to additionally sleep >> for the full timeout. This means, if pjsip_endpt_handle_events2() is >> called with a large timeout, the signal/interrupt will block the app >> for the full timeout interval if a signal happens. This is a problem. >> >> Thanks, >> Kal >> >> _______________________________________________ >> Visit our blog: http://blog.pjsip.org >> >> pjsip mailing list >> pjsip@lists.pjsip.org >> http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org >>