Re: [Arianne-general] Seg Fault on client disconnect (solved)

[<x51...@fe...>]

Ulrich Eckhardt wrote:

> The client can now disconnect without crashing the server.
>
> The solution is not beautiful at all, though. When the client sends a
> logout-msg, the server sends an acknowledgement and continues as normal=
. When
> the client disconnects, the connection breaks and is discarded.(note: t=
he
> client could as well go on and start a new session with a login-msg)
>
> The segfault was because the Socket was closed but still registered in =
the
> SocketSet. The next access by the SocketSet triggered the segfault. In =
other
> words, the SocketSet depends on the internal state of another object, b=
ad
> design.

Well, as it was done, it was stable, I told you that perhaps the socket
design
was awful but as it use is confined to a very limited part of the
program, it
didn't matter.

> I know we had a discussion some months back that revolved around this d=
esign
> decision. I don=B4t exactly remember the results but I wonder if they s=
till
> hold true ? If so, I will make the Socket deregister itself from the
> SocketSet it is in, else I will take the other approach and make every =
Socket
> a completely distinct entity and incorporate the SocketSet=B4s function=
s there.

Well, I make a good idea. Although is it really needed?