Re: [Bug-gnubg] Command line version

[Philippe Michel]

On Wed, 30 May 2007, Christian Anthon wrote:

> Mind you there is a bug in the rollout command in the cl interface,
> where gnubg doesn't double on the first roll even if it should.

This is http://savannah.gnu.org/bugs/index.php?16084

If I remember correctly, there was something of a mess with it :

- I opened the bug and sent a patch that worked for me.

- Whoever checked it in (I don't think it was you, Christian, even if you 
did close the case) made another change at the same time.

- This caused crashes, so everything was rolled back but the bug remained 
closed.

FWIW, since then I use a gnubg.c with the diff below against the CVS head. 
It fixes the problem, as far as I can tell.

*** gnubg.c     Wed May 30 23:07:15 2007
--- gnubg.c.perso       Wed May 30 23:03:57 2007
***************
*** 3991,3997 ****
        float aarNoOutput[ NUM_ROLLOUT_OUTPUTS ];
        float aarNoStdDev[ NUM_ROLLOUT_OUTPUTS ];
        evalsetup NoEs;
!       int false = FALSE;
        void *p;
        int         (** apBoard)[2][25];
        float       (** apOutput)[ NUM_ROLLOUT_OUTPUTS ];
--- 3991,3997 ----
        float aarNoOutput[ NUM_ROLLOUT_OUTPUTS ];
        float aarNoStdDev[ NUM_ROLLOUT_OUTPUTS ];
        evalsetup NoEs;
!       int fCubeDecTop = TRUE;
        void *p;
        int         (** apBoard)[2][25];
        float       (** apOutput)[ NUM_ROLLOUT_OUTPUTS ];
***************
*** 4026,4032 ****
          memcpy (&NoEs.rc, &rcRollout, sizeof (rolloutcontext));
        }
        apci[ i ] = &ci;
!       apCubeDecTop[ i ] = &false;
        }

        RolloutProgressStart (&ci, c, NULL, &rcRollout, asz, &p);
--- 4026,4032 ----
          memcpy (&NoEs.rc, &rcRollout, sizeof (rolloutcontext));
        }
        apci[ i ] = &ci;
!       apCubeDecTop[ i ] = &fCubeDecTop;
        }

        RolloutProgressStart (&ci, c, NULL, &rcRollout, asz, &p);