This library implements a client to play on https://www.dragongoserver.net (DGS). The client knows how to log in, retrieve the list of games to be played, retrieve board positions, play moves and sometimes finish games. It uses the QuickSuite API, which is thoroughly documented in DGS sourcecode https://sourceforge.net/p/dragongoserver/dgs-main/ci/master/tree/specs/quick_suite.txt.
A Robot object is also provided, which makes it trivial to interface a GTP Go program to DGS.
GnuGo is used to evaluate the status of stones at the end. If GnuGo and DGS agree on dead stones, then the robot will terminate the game automatically. Otherwise, it will set the board aside, and the game will need to be finished manually.
The library isn't really documented yet. The easiest way to get acquinted with the code is probably to start with the example bots.
The library implements an object DGSBot, which does everything. A typical bot program is reduced to creating a bot object, configuring it, then running it.
Two examples are included: qrobot, which runs GnuGo, and
mogobot, which runs Mogo.
The do_everything method will connect, process all game
challenges, play one move on each pending game, try to
finish games if needed, then return.
Some notes about the various fields that can be set in the bot:
-
sgffile, sgfout: These are filenames used to temporarily store SGF files. If you are running several bots on the same machine (or on the same NFS shares), make sure these are unique. This caveat applies to all filenames.
-
finished_games: this is a file of games that the bot considers finished. This is used to skip games that are finished but where DGS and GnuGo disagree on the status of stones. This is for performance only, so that the bot doesn't fetch boards it doesn't need. If you erase that file, the bot will just fetch the boards, try to finish them, fail, and fill the file again.
-
error_games: a list of games that produced an error. The behaviour is the same as the previous setting.
-
gtp_engine: Specifies the executable to be run. It is expected to accept GTP commands on stdin, and reply GTP on stdout.
-
board_ok: This is a Perl reference. Whenever someone challenges the bot, it'll automatically accept the challenge if this callback returns true. The callback receives a DGSMessage as parameter, initialised with the challenge settings. This makes it easy to make the bot refuse all challenges ('return 0'), accept all challenges, or accept only certain kind of games. For example, GnuGo plays like crap at sizes below 13 and can't play above 19, and Mogo can only play specific sizes and handicap below 4.
-
badsize_msg: This is slightly mis-named. This is a string which will be included in messages when refusing a challenge. It should explain what restrictions apply.
-
pre_run: This is a Perl reference to a function that receives a GTP object initialised with the GTP agent, and the board on which we play. It is called after the GTP program has been started and the SGF loaded, and before the move is computed. This can be used to fine-tune the agent by sending additional GTP commands (see mogobot for komi hacking).
The only exotic module used is Games::Go::SGF, available from CPAN.
The bot needs to be run repeatedly. This can be done with a simple shell loop, or a crontab, for example:
* * * * * cd $HOME/dgs ; nice -n 17 ./robot
The bot code checks that it's not running already, so running it again when it hasn't finished the previous round is not a problem.
Happy go-ing!