Re: [gnugo-devel] tactical attack cleanup

[Gunnar Farneback]

Evan wrote:
> So anyway, after a bit of a hiatus, I thought I might go finish up that
> long overdue tactics cleanup.  I don't have timings, but attached is the
> regression delta and node counts.

I measure a node count increase of 0.13%, i.e. negligible.

> All the changes appear to be related to the change in the trymove
> setup, which now allows defender to pass,

The defender was allowed to pass before too (or strictly speaking the
attacker was and is required to demonstrate that it can capture),
except in find_cap2(). That was probably unintentional but I'm not
sure it was technically a bug since I can't figure a position where a
string with at most four liberties is capped (as defined by the
find_cap code) and then forced to fill an eye or break a seki if it's
not allowed to pass.

> and force the reading deeper.

This is slightly problematic. The reading in test case reading:178 now
hits a depth limit and is actually worse than before, in spite of the
PASS.

> -
> -    if (edge_closing_backfill(str, apos, &xpos))
> -      ADD_CANDIDATE_MOVE(xpos, 10, moves);
> -

This bonus of 10 for the move ordering is lost after the patch. I'm
not sure it's worth fixing though.

Overall I think the patch looks fine.

/Gunnar