[Computer-go] CGOS Dropped Connections - Two automatic losses on time.

Don Dailey dailey.don at gmail.com
Thu Nov 11 07:52:59 PST 2010


On Thu, Nov 11, 2010 at 10:42 AM, Christoph Birk <
birk at obs.carnegiescience.edu> wrote:

>
> On Nov 11, 2010, at 5:16 AM, <robertfinkng555 at o2.co.uk> <
> robertfinkng555 at o2.co.uk> wrote:
>
>> Eventually the current log
>> in times out at the server and you can log in again... but way too late
>> for you to reconnect and finish off the current game. Don's suggestion
>> would allow engines to reconnect after a temporary network glitch and
>> finish the game, avoiding bad results - especially important for
>> standard engines where several will be running simultaneously on the
>> same machine.
>>
>
> A new instance of the program can only finish the current game if it
> does not keep any state between moves. I don't know about your's but my
> program keeps a state, so that would not work anyway.
>

That does not make any sense.   Any bot that implements the standard
protocol GTP can reconnect and the server will send it the new state.    If
your program keeps state it will of course lose the state,  but the go
program will be restarted and have to rethink the move (from scratch) that
it was disconnected on.

I would like to implement ping with a "pong" required response,  but I don't
want to break existing clients.   Even though I can modify my own client
accordingly,  it will break the 3rd party clients that people have built.

Years ago I implemented ping but it had bugs and I removed it.   Because GTP
is not asynchronous there is a race condition issue that can happen if I
don't take care to implement it properly.   It's solvable but it's easy to
get wrong.


Don




>
> Christoph
>
>
> _______________________________________________
> Computer-go mailing list
> Computer-go at dvandva.org
> http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://computer-go.org/pipermail/computer-go/attachments/20101111/e37ce468/attachment.html>


More information about the Computer-go mailing list