Re: [gnugo-devel] make komaster variables global?

[Gunnar Farneback]

Arend wrote:
> I think it would make sense to make komaster and kom_pos static privat
> variables in board.c (or just global variables, no big difference
> here).

Static is usually better, unless they need to be accessible elsewhere.

> They would then be accessed only by komaster_trymove; if they change,
> they can be stored on the stack with the PUSH_VALUE construct, so that
> they get automatically restored in undo_trymove() (called from popgo() )
> -- I think this PUSH_VALUE / POP_MOVE construct is quite elegant.

It is indeed. Credits go to Tanguy who introduced this idea.

> A side effect, if e.g. the connection reading already depends on a ko, say
> O is komaster, then the default would be that any attack() etc. that
> is called as a helper function in the reading code would already start
> with O being komaster. I think this side effect should be desirable
> (and if not, we can still provide clear_komaster(&save_komaster)...)

I agree that it's probably desirable.

> Opinions?

I think it would work, although I'm not completely certain there
aren't any complications. One thing to consider is whether a
non-komaster trymove() should reset the komaster variables.

> The only thing that makes me hesitate a little before starting to write
> a patch to do this is that 5 komaster_trymove routines would have to be
> changed. Are there some komaster schemes that are clearly redundant and
> could get removed?

In my opinion scheme 5 is superior, but I don't think there have been
any serious attempts to evaluate them accurately. Scheme 4 was
introduced by Trevor in an attempt to solve certain problems with the
pattern based tactical reading and those may or may not be addressed
by scheme 5.

/Gunnar