>Sorry, I intended to answer that one, it just got buried under >a lot of other mails. No problem, i just feared it got lost because there was quite a lot of posts this week. >> I tried to move the Add call to the net_socket_private >>constructor, and >> then I felt in the problem of my previous email >>(see bellow). >This particular issue is fixed in r30188, thanks for your >investigation! nice. There is another issue i saw in create_socket in case of error: The mutex is destroyed before calling the socket destructor, but this socket is also destroyed in the destructor. This should be removed from create_socket. >> I can try to provide a patch, but I am afraid that I will do bad >> things >> without help since two points are not clear to me: >> - Is it ok to add all sockets in sSocketList as soon as they are >> created >> (this impacts socket_get_next_stats)? >This would not be the correct solution, as a socket can be in other >lists as well exclusively (ie. in its parent socket "pending" and >"connected" lists). ok. pending_children sockets are the ones waiting for an accept call, and connected_children sockets the ones that were accepted ? >> - Is it ok to load the net_socket module as soon as the stack is >> created >> (It seems this was supposed to be done automatically only when >> needed) ? >I will look into this problem now. You can't just load the module, as >you would then no longer be able to unload the stack. ok. I will watch the commits to see how it will be fixed :)