server and client websocket closing #43
Conversation
There was a problem hiding this comment.
Apart from the race, seems to be working quite ok. The code itself is a bit messy, but it works - I think that a refactor would be in order, but that's a different task pretty far in the backlog. After resolving the race (and considering piodul's remarks) lgtm.
client/dfclient/src/menu_state.cpp
Outdated
| int ret = webSocketManager._ws->Connect(gameData.serverAddress); | ||
| if (ret < 0) { |
There was a problem hiding this comment.
The race is here - adding sleep(100ms) after Connect() seemed to solve the problem.
There was a problem hiding this comment.
Please use proper synchronization to fix it, though (e.g. mutexes, condition variables, semaphores).
There was a problem hiding this comment.
@piodul Don't get me wrong, it was just a note to pinpoint the problem, not a fix suggestion xD
There was a problem hiding this comment.
According to my research the race should not be present, even though I did nothing in the code to mitigate it. After the call to handshake finished successfully the socket should be immediately available to send data. The only thing that might be racing here is if there is something happening within ioc.run that makes the websocket ready to send messages and there is no way to synchronize anything within an external function call except for a sleep.
After some testing I have not observed the race happening on my machine. It is possible that the racing behaviour was introduced in a different PR.
this makes repeated connections to a game that is already in progress handle nicely
there still is a race somewhere but I've seen it only once so it cannot be that bad