[gnugo-devel] Re: Fwd: gtp time report ?

[alain Baeckeroot]

Le Vendredi 3 Mars 2006 04:16, Don Dailey a écrit :
> 
> The cgos server does not try to account for network time lag in any way.
> I considered "trivial" solutions like always giving a fraction of a
> second for free after every move (something I may do) but that doesn't
> solve the real problem - unpredictable network lag time.   But it does
> try to give the engine the benefit of the doubt.   
> 
> What do other servers such as KGS do about this?
I don't know. It reports the time also, and i see the same small drift. 
> 
> By the way, the client does know anything about time.  The server 
> sends the reports to the client and the client simply forwards these
> reports to the engine.
I suppose i must dig into gnugo to see what it does ! It seems that  
gnugo/twinbot does not use this info, but instead use an internal timing (i 
see no other possible explaination, but i can be wrong again)

> Any solution involving the client is wide 
> open to cheating of course - especially so with a simple tcl client
> which anyone can hack.
I agree. I think this is why gnugo seems to use internal timing. It might need 
a "trusted-gtp-time-report" if this does not exist.

- Alain

> 
> - Don
> 
> On Fri, 2006-03-03 at 00:57 +0100, alain Baeckeroot wrote:
> > Hi Don
> > 
> > I have forgotten to CC you.
> > I have seen your name in GNU Go authors :) but i don't know if you are 
still 
> > on the list.
> > 
> > Best regards
> > alain
> > 
> > ----------  Message transmis  ----------
> > 
> > Subject: gtp time report ?
> > Date: Vendredi 3 Mars 2006 00:52
> > From: alain Baeckeroot <address@hidden>
> > To: address@hidden
> > 
> > Hi
> > 
> > I have a question:
> > On cgos.boardspace.net Twinbot lose one game on time:
> > 
> > twinbot(11.1k)      ReadyToGo(16.1k)        B+Time  03/02 22:36
> > 
> > When i look at timing report, i have not the same time given by the 
cgosgtp 
> > client and by my twinbot messages. The difference is only few seconds 
> > (between 3 to 10 depending on the game), but that enough to lose :(
> > 
> > I suppose that gnugo and the client does not exchange time info, but i m 
not 
> > really sure, and don"t know if it s a bug or a lacking feature, and who's 
> > "fault" it is ...  
> > 
> > Alain
> > 
> > ------------------------
> > time report from cgosGTP.tcl client:
> > 
> > genmove w
> > time_left w 1.012
> > 
> > genmove w
> > time_left w 0.77
> > 
> > genmove w
> > time_left w 0.425
> > 
> > genmove w
> > time_left w 0.0860000000001
> > 
> > genmove w
> > time_left w -0.125
> > gameover B+Time
> >     -----
> > The corresponding time report given by gnugo/twinbot:
> > 
> > New level 14 (39 white  9.25 )
> > New level 14 (40 white  9.05 )
> > New level 14 (41 white  8.99 )
> > New level 14 (42 white  8.79 )
> > New level 14 (43 white  8.59 )
> > new game ...
> > ---------------------------
> > Another game where the difference is more than 11s:
> > 
> > genmove b
> > time_left b 260.787
> > 
> > genmove b
> > time_left b 260.52
> > gameover B+73.5
> > 
> > New level 14 (64 black 0.10 271.89 10)
> >   129 th_move, gstatus 1.00, Best is C5 (1.0)
> > New level 14 (65 black 0.08 271.81 10)
> >   131 th_move, gstatus 1.00, Best is Pass (0.0)
> > 
> > 
> > 
> > -------------------------------------------------------
> 
>