@willmmiles: no sure if this block should be within the if (_pcb) { block or not. If yes, then maybe the same is true for close also and the latter should be updated too ?

What do you think ?

#79 moved the _pcb check in AsyncClient::close() to be done just once in the LwIP context in _tcp_close_api, which then returns whether or not the dispose callback needs to be run after checking _pcb and the queue contents. The key reason is that _pcb could be nullptr while there's still a FIN or ERROR event pending, so even if it is, we need to purge the queue. I didn't want to replicate the purge code just to avoid the LwIP lock, though that would be a technically valid approach.