[freeciv-ai] Re: definitely last version of active diplomats patch

["Per I. Mathisen" <per@xxxxxxxxxxx>]

On Fri, 6 Sep 2002, Raimar Falke wrote:
> Can you please describe how the AI uses diplomats (which functions are
> called in which order and do which tasks)? Can you also write down the
> limitations and assumptions?

It calls ai_manage_diplomat(punit) for each diplomat each time the AI
activates (which is twice per turn). Then you just follow the logic in
that function. Note that it works differently if we are following a goto
or not, which was set previously by another call to ai_manage_diplomat,
and this goto is checked differently depending on our ai_role.

The ai_choose_diplomat_offensive and ai_choose_diplomat_defensive are
called from the build code once per turn per city.

> BTW: the patch is big. IMHO too big to really read is carefully.

That happens.

> > +  if (pplayer->player_no == acity->original) {
> > +    incite_cost /= 2;
> > +  }
>
> Why isn't this included in city_incite_cost? It should be.

It can't be, since city_incite_cost() just updates
pcity->incite_revolt_cost, and the incite bonus is dependent on who does
the inciting. IMHO, city_incite_cost() should be called once for every
city at turn start, so we don't need to everywhere else before
pcity->incite_revolt_cost is used.

> > +  if (pplayers_at_war(pplayer, city_owner(acity))
>
> > +      && (find_palace(city_owner(acity)) != acity)
>
> Compare incite_cost against INCITE_IMPOSSIBLE_COST.

I'd rather assert that. Since we should check all possible reason why we
can't incite here. If we missed one reason, we should fix it.

> > +    /* incite gain (FIXME: we should count wonders too but need to
> > +       cache that somehow to avoid CPU hog -- Per) */
>
> What is the problem? Updating the cached value every turn is enough.
>
> > +    gain_incite *= SHIELD_WEIGHTING; /* cost to take city otherwise */
>
> ???

The incite algorithm is ugly. I know. It is just a bunch of WAGs, but it
works, and Greg said he'd try to make a better one.

But it is easier to experiment with new algorithms with existing code in
cvs that has been tested and that one can compare with.

As to wonders, show me an easy to way to calculate the worth of wonders in
enemy cities, please...

> > +{
> > +  int dist, urgency;
> > +  int best_dist = 30; /* any city closer than this is better than none */
> > +  int best_urgency = 0;
> > +  struct city *ctarget = NULL;
> > +  int continent = map_get_continent(x, y);
> > +
> > +  city_list_iterate(pplayer->cities, acity) {
>
> Move the stuff into the loop.

I'm with Mike here. I don't care if the compiler can do tricks so that
stuff like map_get_continent() doesn't get called on each loop - this is
just wrong by design. If we want it to be called once, then we should call
it outside the loop.

Yours
Per