Re: [gnugo-devel] gmp mode sgf output

[Gunnar Farneback]

Paul wrote:
>   for some reason it handles passes coming from gmp client incorrectly
>   (at least on 19x19 board). however, the same is true for 3.3.5. it
>   may be a bug in jago 4.42 which i used for testing.
> 
>   please, test with other clients.
> 
>   if mistake is not jago-specific, it is gmp.c which should be fixed.
> 
>   lines 414-418:
>     s = val & 0x1ff;
>     if (s == 0)  {
>       x = -1;
>       y = 0;
>     } else if (s == 0x1ff)  {
> 
>    when pass is coming from jago, val=342 (19x19 board). gmp.c expects
>    that val=0 is sent for passes.

I assume you mean val=362 since 342 would be T18. But it's wrong
anyway. The GMP specification is very clear that pass is represented
by 0. Excerpt from the specification:

|         5. Move command - 1101 0pii 1iii iiii
| Interpretation of the move command is game specific. For the game 
| of Go, top value bit (p) is 1 if player to move is white or 0 if black.   
| 9 bit value (i)  is a board intersection value or 0 for pass.  
                                                    ^^^^^^^^^^
| Intersection values run 1 at A1, 2 at B1 ... with value wrapping at 
| end of row (varies with board size) so next row is A2.  Range for 
| 19x19 board is thus 1...361 (A1 ...T19). A1 is in the lower left corner
| and the letter I is omitted. [...]

My conclusion is that this would indeed be a Jago bug.

/Gunnar