Last from me.
Your example is essentially saying that gnubg should pick the interpretation that maximizes the user's chance of winning, as if that is surely what he intended. This is certainly one possible algorithm.
However, another algorithm is to never interpret a continuous slide as pick and pass.
I prefer the latter, but the important thing is that there be some method that program consistently follows.
----- Original Message -----
From: "Michael Petch" <address@hidden>
To: "Louis P Zulli" <address@hidden>
Cc: address@hidden, "Jonathan Kinsey" <address@hidden>, "Philippe Michel" <address@hidden>
Sent: Thursday, April 9, 2009 4:25:59 PM GMT -05:00 US/Canada Eastern
Subject: Re: [Bug-gnubg] Handling ambiguous checker moves
I know you consider this to be text book black and white that “if you drag directly to the end position” that pick and pass should never be selected. I disagree.
Lets assume the bot does what you say in all cases.
What if you are trying to save backgammon. You have a piece on the 24 point, no men taken off (But your house has a prime) and you opponent opened up and left his last checker on the 23 point. You roll 2-1 and you drag the mouse from 24 to 21.
Tell me, when the bot jumps over because it won’t hit, I think its reasonable for someone to complain that “Intent is clear, the bot should know I will save backgammon and hit”.
A persons expectations may not be the same given all circumstances. They might believe the bot should smart enough to recognize that the intent was to save backgammon. And I will quote from you “we should strive to have a parsing scheme that is as intuitive as possible”
So in my case I believe its intuitive to assume the bot should know full well that this is a hitting play. For it to choose any other play would be counter intuitive, at least from the perspective of the person who made the play.
I think it could be argued that there are likely more positions where the intuitive thing to do is to hit, rather than pass.
In the end, I’d still be happy with the bot just making the first play it found to be legal to get from position A to B.
On 09/04/09 2:05 PM, "Zulli, Louis P" <address@hidden> wrote:
There is currently some "parsing" of "user checker actions on the gtk-board" to yield "actual backgammon moves." For example, my continuous checker slide from 6 to 3 was parsed as 6/5*/3 rather than 6/4/3. Even though a user can readily undo and redo a move, we should strive to have a parsing scheme that is as intuitive as possible. Perhaps I am atypical, but parsing my action as 6/5*3 is counter-intuitive. Philippe's suggestion that a continuous slide like mine never be parsed as pick and pass makes sense to me. Ideally, parsing of actions should not simply default to "pick the first legal interpretation" of the action based on some arbitrarily ordered list of legal interpretations.
Of course this is a small issue, but it seems like the kind of issue that should be discussed before any "official" release.