There need never be a command to load an SFG file, because that can and should be implemented by interface software. As long as there is some way to set up the position or game, we are ok.
Don
As I understand it GTP as GnuGo use it currently contains a load SGF file command which means the engine has to understand SGF. So it is reasonable to consider setting things like boardsize by sending the SGF token SZ[] to do it rather than having an explicit GTP command.
But when I get to the point of needing to write GTP handling I'll be happy to run with whatever the consensus is providing there's enough functionality, and there is a way to connect GTP and GMP together.
Peter.
From: Peter.Smith@Smallworld.co.uk
I'm some way off doing anything with GTP yet so this is rather commenting from the sidelines.
GTP has a command to load a SGF file? Suppose GTP had a command to pass SGF elements to an engine like "KM[5.5]" ? Would that remove the need to have certain commands as you could pass them as the SGF equivalent?
I think having a GTP to GMP bridge will be important. Otherwise if I write a Go program having a GTP interface isn't going to be enough to play matches. Should GTP have equivalents for the main GMP commands?
Peter Smith. _______________________________________________ gtp mailing list gtp@lists.lysator.liu.se http://lists.lysator.liu.se/mailman/listinfo/gtp
Peter wrote:
As I understand it GTP as GnuGo use it currently contains a load SGF file command which means the engine has to understand SGF.
The GNU Go implementation of GTP contains tons of stuff that other programs shouldn't need or sometimes even want to implement. The loadsgf command is very useful for regression purposes so it should probably be standardized sooner or later but only as an optional command.
So it is reasonable to consider setting things like boardsize by sending the SGF token SZ[] to do it rather than having an explicit GTP command.
Inline SGF in GTP does not appeal to me, at all.
But when I get to the point of needing to write GTP handling I'll be happy to run with whatever the consensus is providing there's enough functionality, and there is a way to connect GTP and GMP together.
I don't think it will be possible to write a general GTP/GMP converter, but it should be possible to write a program speaking GMP in one end and connecting through GTP to an engine which provides the moves. For example one might start with Bill Shubert's goDummy program (see http://www.igoweb.org/~wms/comp/cgoban/index.html) and replace the (extremely simplistic) move generation with code that starts an engine with which it speaks GTP.
/Gunnar