Re: [gnugo-devel] olib & xlib definition

[Gunnar Farneback]

Trevor wrote:
> Ah, right, you haven't profiled the experimental pattern-based reading 
> patterns (and I wouldn't recommend it until my next patch!).  They
> use olib & xlib quite a lot (perhaps more than they should...)  That
> was my motivation for the change.  approxlib really is a lot faster
> than accurate_approxlib.

To begin with, all constraint expressions which are equivalent to
olib(*) <= 1, and these are quite a few, are really just asking the
question whether the move would be a self atari (including suicide).
This is most efficiently solved using the is_self_atari() function and
it should have its own autohelpers.

For the rest of the patterns outside read_attack.db I'm pretty sure
that the increased accuracy provided by accurate_approxlib() is worth
the cost. Compared to a reading constraint the speed difference
between approxlib() and accurate_approxlib() is rather trivial.

In read_attack.db it seems reasonable that the cost is too high
though, so there an approxlib() based autohelper would make more
sense. Regarding the naming, however, I think we should let xlib and
olib remain accurate and give different names to the approxlib() based
ones.

/Gunnar