pion: That's the interesting case I saw as well. It's a known possible race condition and it is handled correctly now. It also shouldn't affect anything as the 20 second connection timer (or the idle_timeout) will close the listening/connecting socket and everything should continue normally. Still, when did that appear? It sounds like just after things went bad which sorta makes sense...
It's looking like winsock stops delivering these async socket events notifications completely... I'll have to explore this some more... perhaps a build for you with extra debugging around this since you can obviously reproduce the issue.
I can also look into using the data connection async notification method to build another feature into the deadlock detector so it can catch that as well for you.... Or perhaps change the way this is done entirely.
|