Re: [Bug-gnubg] Follow up to: Match statistics are wrong when match is loaded again

[Jim Segrave]

On Sun 23 Feb 2003 (21:27 +0100), Scott Steiner wrote:
> Hi,
> 
> as suggested by Jim Segrave, I am sending the files that display the
> error described in my earlier post.

I think there's a bug in the way gnubg handles a resignation, which is
a side effect of the Jellyfish format. The problem arises because
after your roll of 21, you resigned without moving any chequers.

The .sgf file you sent appears to be corrupted in the last line:

Line reformaated for clarity:

;B[21]A[19]
[jltu E 0.0000 0.0000 0.0000 0.1282 0.0000 -1.0000 2C 0 1 0.0000]
[jluv E 0.0000 0.0000 0.0000 0.1299 0.0000 -1.0000 2C 0 1 0.0000]
[uwjk E 0.0000 0.0000 0.0000 0.1699 0.0000 -1.0000 2C 0 1 0.0000]
[uwwx E 0.0000 0.0000 0.0000 0.1984 0.0000 -1.0000 2C 0 1 0.0000]
[jlwx E 0.0000 0.0000 0.0000 0.1299 0.0000 -1.0000 2C 0 1 0.0000]
[prjk E 0.0000 0.0000 0.0000 0.1466 0.0000 -1.0000 2C 0 1 0.0000]
[prrs E 0.0000 0.0000 0.0000 0.0498 0.0000 -1.0000 2C 0 1 0.0000]
[jlpq E 0.0000 0.0000 0.0000 0.1084 0.0000 -1.0000 2C 0 1 0.0000]
[uwuv E 0.0000 0.0000 0.0000 0.2061 0.0000 -1.0000 2C 0 1 0.0000]
[vxvw E 0.0000 0.0000 0.0000 0.2095 0.0000 -1.0000 2C 0 1 0.0000]
[vxpq E 0.0000 0.0000 0.0000 0.1891 0.0000 -1.0000 2C 0 1 0.0000]
[tvuv E 0.0000 0.0000 0.0000 0.1987 0.0000 -1.0000 2C 0 1 0.0000]
[jllm E 0.0000 0.0000 0.0000 0.0948 0.0000 -1.0000 2C 0 1 0.0000]
[tvwx E 0.0000 0.0000 0.0000 0.1982 0.0000 -1.0000 2C 0 1 0.0000]
[tvvw E 0.0000 0.0000 0.0000 0.1990 0.0000 -1.0000 2C 0 1 0.0000]
[jlvw E 0.0000 0.0000 0.0000 0.1298 0.0000 -1.0000 2C 0 1 0.0000]
[prtu E 0.0000 0.0000 0.0000 0.1033 0.0000 -1.0000 0C 0 1 0.0000]
[uwpq E 0.0000 0.0000 0.0000 0.1564 0.0000 -1.0000 0C 0 1 0.0000]
[pruv E 0.0000 0.0000 0.0000 0.1019 0.0000 -1.0000 0C 0 1 0.0000]
[\237\340^Lxxxxx ]
LU[+0.000])^M

The above is a list of the 18 legal moves you could have made, plus a
completely bogus entry for the move you "did make" (it appears gnubg
felt obliged to invent one for you, as there was no move listed and
you could have moved). The values immediately following the square
ackets should be letters in the range a-z (where a-x are points, y
is the bar and z is off) I think gnubg is writing a move record with
random memory contents, so it writes 8 pairs of characters, most of
which are not valid board points. 

When I load this file with gnubg, sometimes I get the strange match
statistics you posted, sometimes, depending on what else I've done
using gnubg, I get a crash and core dump, because of the invalid
move. I get this 100% of the time when trying to load this file
running the gdb debugger, which suggests it's a random contents of
memory related problem. Removing just the one bogus move from the
last line cures the problem.

The solution lies in preventing gnubg from generating an analysis
entry for a non-move.

-- 
Jim Segrave           address@hidden