The agent can end in a few cases where the process crashes.

[mkm85]

The agent can end in a few cases where the process crashes, since it uses indices to candidates which has been removed.

Issue 1

https://github.com/webrtc-rs/rtc/blob/master/rtc-ice/src/agent/mod.rs#L736

            if new_state == ConnectionState::Failed {
                self.set_selected_pair(None);
                self.delete_all_candidates(false);
            }

this code does not reset the selected pairs, which means they now can point to invalid indexes in the candidate arrays.

Issue 2

https://github.com/webrtc-rs/rtc/blob/master/rtc-ice/src/agent/mod.rs#L1330

if let Ok(added) = self.add_remote_candidate(prflx_candidate)
                            && added
                        {
                            remote_candidate_index = Some(self.remote_candidates.len() - 1);
                        }

add_remote_candidate can trigger a state change to failed, which clears the remote candidates (now the index becomes -1), because add_remote_candidate runs the contact() function which checks for timeouts.

There is a fix for these problems here: https://github.com/nabto/webrtc-rs-rtc/tree/ice_state_fail it defers the contact calls to the timeout handler and implements proper candidate pair cleanup.

Is it the preferred way to fix the issue?