Re: [liblo] TCP

Hey Stephen,


On 30 sept. 2011, at 17:42, Stephen Sinclair wrote:

> The server definitely does keep a list of sockets for incoming TCP
> connections.  I added this functionality so that connections could be
> persistent, since previously they had to be closed and opened for
> every OSC packet.

That is good and works fine.


> The address has one socket, and it may keep the socket open if it
> doesn't notice yet that the other side is closed.

This is the problem.
What is weird though, is that this client socket prevents a server socket from being opened on the same machine.


> There are often weird issues with waiting for sockets to time out.  I don't know if
> this is such a case.  I don't know whether the address keeping the
> socket open would stop a server from being able to bind that port for
> receiving.  Will have to test.

OK, thank you very much Stephen.
If you need more detail from me, please feel free to ask.


>> Consider this scenario:
>> 
>> - Client sends to Server
>> - Server is closed
>> - Client tried to lo_send -> results in an error.
>> 
>> Now I can opened a new server on the same port.
> 
> So you are saying that the address's socket is only closed when it
> tries to send and notices that there is no longer a receiver-side
> socket connected?

Yes, this is what I see.
I don't think this behavior is terribly wrong, it's just that I can't bind a server socket to the "hanging" port.


>> Does it suggest that I should get rid of the lo_address every time a message has been sent?  My application keeps track of lo_addresses and reuse them.
> 
> That is the desired behaviour, hopefully you don't need to change your
> application. :)
> 
> In the above you describe that the lo_send fails when the server was
> closed, wouldn't that imply that you only need to re-create the
> lo_address only when lo_send fails, not every time you send a message?

I don't even need to recreate a lo_address, it seems the socket is re-opened automatically by liblo, which is great.


> In any case, I will try to review the retry logic in address.c.
> Should it retry to open the connection at least once before reporting
> failure, or is failure on connection closed a reasonable behaviour?

No, unless it is required to solve the former problem.


Best,
Cam