Re: [gnugo-devel] Good news

[Tanguy URVOY]

Hello,

Arend Bayer wrote:
> The biggest remaining challenge I see now is keeping track of what
> exactly changed on the board between two matchpat() calls (and doing
> this without making the code too ugly). Maybe s.th. like replacing
> all trymove()'s in owl.c with a dfa_trymove().

Yes this is the point.

The actual matcher is already brutal:
the function dfa_prepare_for_match(int color)
scans the whole board to setup the color at
each call of matchpat().
I am not sure that the cost of detecting changes
during this operation would be so big.

dfa_trymove() is a better solution
to maintain the board difference
but it may be more complex to handle.

Maybe we should first implement a pattern matcher
with the following interface :

void
matchpat(matchpat_callback_fn_ptr callback, ... , int last_change);

An check it with a
brutal detection of last_change.



We need the following function:

/* Compute the row of pos2 if scanned from pos1 with spiral path. */
static int spiral_row(int pos1, int pos2);


I think I already wrote it a year before.
It may kreep somewhere in Gnugo dead code...




We also need a dfa_cache_entry type
(LL is the number of transformations)


typedef struct dfa_cache_entry
{
   int last_raw[LL];
   int patno[LL];
   int pat_list[LL][DFA_MAX_MATCHED];
} dfa_cache_entry_t;


An a big array of dfa cache entries pointers.

dfa_cache_entry_t *dfa_cache[COLOR][DFA_MAX_BOARD * DFA_MAX_BOARD * 16];

With a cache entry allocated for each 'ON BOARD' position.



I have to go back to my phd writting.
I will have no time to work on it
this week. If someone is interested ...

Tanguy


--
---------------------------------------
Tanguy Urvoy http://www.irisa.fr/galion
---------------------------------------