segfault when zmq_poll() with closed socket

[potatogim]

Hi.

I met a segfault with my program.

this situation like below.

1. call zmq_poll() with a monitor socket gained by zmq_monitor_socket() and then some callback is triggered.
2. this callback checks connection and then if it is disconnected, close this socket.
3. after that, call zmq_poll() with same pollitems.
4. Bang!

I have rebuilt ZMQ::LibZMQ3 with ZMQ_TRACE.

1887                 mg = P5ZMQ3_Socket_mg_find( aTHX_ SvRV(*svr), &P5ZMQ3_Socket_vtbl );
1888
1889                 /* debugging logs added for tracing */
1890                 P5ZMQ3_TRACE(" + mg %d: %d", i, mg);
1891                 P5ZMQ3_TRACE(" + mg->mg_ptr %d: %d", i, mg->mg_ptr);
1892 
1893                 pollitems[i].socket = ((P5ZMQ3_Socket *) mg->mg_ptr)->socket;
1894                 P5ZMQ3_TRACE( " + via pollitem[%d].socket = %p", i, pollitems[i].socket );

and this is output from that.

[perlzmq (29306)] START mg_free (Message) [perlzmq (29306)] + zmq message 2e41f50 [perlzmq (29306)] END mg_free (Message) [perlzmq (29306)] START zmq_poll [perlzmq (29306)] + processing element 0 [perlzmq (29306)] + mg 0: 48468640 [perlzmq (29306)] + mg->mg_ptr 0: 0 Segmentation fault (core dumped)

Actually... it may be my fault, too.

but I think that we can handle this situation more gracefully :)

[lestrrat]

I haven't really been keeping up with Perl/XS for a while, would be cool if you can provide me with aPR

[potatogim]

Yep, I will perform PR soon :) Thank you for your answer!

[potatogim]

Oh....

I think this issue is duplicated with #63.

in master branch, the PR resolving this issue is included but 1.19 does not contains this PR :)