Segmentation at sock.close() at loop.timer

[pavel-orekhov]

I tried to reopen socket at timer's handler. It runs in endless recursion at remove() because socketfd=0 after close() and commonfd=0 too.

`#include <iostream.>
#include <memory.>
#include <unistd.h>
#include "zmqpp.hpp"

int main(int argc, char* argv[]){
zmqpp::context zmqContext;
zmqpp::loop loop;
zmqpp::socket workerSock(zmqContext,zmqpp::socket_type::dealer);
workerSock.connect("tcp://localhost:5555");

loop.add(std::chrono::milliseconds(1000),0,[&](){
    loop.remove(workerSock);
    workerSock.close();
    //reopen sock in 5 sec
    return true;
 });

loop.start();
return 0;

}`

zmqpp::socket::socket(const zmqpp::context&, zmqpp::socket_type)0x55555577fd90

void zmqpp::loop::remove(const socket_t&)0x55555577fd90 1
void zmqpp::socket::close()0x55555577fd90
void zmqpp::loop::remove(const socket_t&)0 0

Thread 1 "closeAtTimer" received signal SIGSEGV, Segmentation fault.
0x000055555555d293 in zmqpp::poller::remove (this=0x7fffffffdd90, descriptor=0) at src/zmqpp/poller.cpp:108
108 remove(item);
(gdb) bt -10
#349301 0x000055555555d1d8 in zmqpp::poller::remove (this=, item=...) at src/zmqpp/poller.cpp:114
#349302 0x000055555555d298 in zmqpp::poller::remove (this=, descriptor=)
at src/zmqpp/poller.cpp:108
#349303 0x000055555555d1d8 in zmqpp::poller::remove (this=, item=...) at src/zmqpp/poller.cpp:114
#349304 0x000055555555d298 in zmqpp::poller::remove (this=, descriptor=)
at src/zmqpp/poller.cpp:108
#349305 0x000055555555d1d8 in zmqpp::poller::remove (this=this@entry=0x7fffffffdd90, item=...)
at src/zmqpp/poller.cpp:114
#349306 0x000055555555d247 in zmqpp::poller::remove (this=this@entry=0x7fffffffdd90, socket=...)
at src/zmqpp/poller.cpp:101
#349307 0x000055555555867d in zmqpp::loop::remove (this=0x7fffffffdd10, socket=...) at src/zmqpp/loop.cpp:124
#349308 0x0000555555558c6d in zmqpp::loop::flush_remove_later (this=this@entry=0x7fffffffdd10)
at src/zmqpp/loop.cpp:215
#349309 0x0000555555558e8f in zmqpp::loop::start (this=0x7fffffffdd10) at src/zmqpp/loop.cpp:151
#349310 0x00005555555567db in main (argc=, argv=) at closeAtTimer.cxx:20
(gdb)

[pavel-orekhov]

segv205.patch.txt
fd==0 is valid fd.
The patch suggested. Please test it.

[pavel-orekhov]

There are two problems after previous patch:

• after socket::close called from loop.timer's handler loop::start breaks with exception "it is not socket"
• after socket::close called from loop.timer's handler loop.remove cant remove socket from list

Both problems are because after socket::close() socket::_socket is null and it misscompares with list of sockets at loop and at poller level.

It is required to save socket::_socket at SOCKET::close() till LOOP::flush_remove_later().
Is it right to add field void* socket::_saved_socket for this problem?

[pavel-orekhov]

the 205_2n3.patch.txt
suggested as a fix of mentioned problems.