[gnugo-devel] --score option

[Gunnar Farneback]

The --score option can be run in a number of different modes. The help
text (gnugo --help) says

   --score last            estimate score at last move in SGF file\n\
   --score until <n>       score at move 'n' in SGF file\n\
   --score end             generate moves to finish game, then score\n\
   --score aftermath       generate moves to finish, use best algorithm\n\

The second variant is bogus because there's no way for getopt to deal
with the syntax, but looking at the comment for
load_and_score_sgf_file() in play_solo.c we see

 * Load SGF file and score the game
 * untilstr:
 * end  - finish the game by selfplaying from the end of the file until 
 *        two passes
 * last - estimate territorial balance at the end of the of the file
 * move - load file until move is reached and estimate territorial balance
 *
 * aftermath - like 'end' but also plays out the aftermath

Thus "--score 143" or "--score L16" should work for that scoring mode.
However, this is really stupid since it overrides the -L option and in
fact we can see in main.c that --score and -L write their argument
into the same buffer. A consequence of that is that we can't do e.g.

--score aftermath -L 233

in order to see how the aftermath code would finish and score the game
at move 233.

Needless to say I will change the behaviour, but the question is
exactly how it should work. Clearly -L should decide how far to load
the game record and as usual it should be loaded to the very end if -L
is omitted. But what arguments do we want for --score?

I think we want to be able to
1. Estimate the score as is.
2. Finish the game and count the score quickly.
3. Finish the game and count the score accurately by playing out the
   aftermath.

Is there some more interesting variation? What should we call the
modes? I really don't like the "end" and "last" names. "aftermath" is
okay.

/Gunnar