[freeciv-ai] Re: (PR#4026) Advanced rampage.

["Gregory Berkolaiko" <Gregory.Berkolaiko@xxxxxxxxxxxx>]

On Sat, 10 May 2003, Per I. Mathisen wrote:

> On Tue, 6 May 2003, Gregory Berkolaiko wrote:
> > Here is an up-to-date version of rampage.
> 
> +      /* If the victim is in the city, we increase the benefit and correct
> +       * it with our health because there may be more units in the city
> +       * stacked, and we won't destroy them all at once, so in the next
> +       * turn they may attack us. So we shouldn't send already injured
> +       * units to useless suicide. */
> +      if (map_get_city(x, y)) {
> +        /* A WAG for the city value */
> +        benefit += unit_value(get_role_unit(F_CITIES, 0));
> +        benefit = (benefit * punit->hp) / unit_type(punit)->hp;
> +      }
> 
> I don't like this. First, those who have commented that we should also
> check for fortress are right. Just because fortress is not commonly used
> in default rules, does not give you a license to ignore it. (If it is THAT
> useless, it should rather be removed from the game.)

Ok.

> Second, the benefit of taking a city is ridiculously small. The loss to
> the other player might be enormous. Our benefit of taking the city with
> punit depends on punit's ability to defend the city from attack. Taking a
> city with a Howitzer when an enemy Howitzer stands outside may not be a
> good idea, but checking this might be ardous. If we did assess_danger()
> also for enemy cities, we would know how big danger there was of a retake
> of this city, though. Also, if the enemy cannot steal any techs from us,
> defending from a position inside the city might well be much better than
> not. Finally, we may want to reduce want of taking size 1 cities that do
> not have city walls, since they will be destroyed.
> 
> I suggest we make a function for this right away, even if is presently not
> well populated with checks. Perhaps
>       int ai_city_attack_want(punit, pcity)
> which for now always returns
>       unit_value(get_role_unit(F_CITIES, 0)) * pcity->size
>         * unit_list_size(&pcity->units_supported)
> but is populated with TODO comments.

This is ridiculous in this situation.  This will cause a unit attack a 
city of size 4 at the win_chance of 20%.  Since we are talking about this, 
I hink we should remove the city bonus from rampage altogether.  Rampage 
is a hit-n-run operation, has no infrastructure (and no chance) to occupy 
cities.

> +      /* If we have non-zero attack rating... */
> +      if (att_rating > 0 && is_my_turn(punit, pdef)) {
> +        /* No need to amortize! We're doing it in one-turn horizon. */
> +        return kill_desire(benefit, attack, loss, vuln, 1);
> +      }
> +    }
> 
> What does this mean?

What is "this"?

>  /*************************************************************************
> +  Find and kill anything reachable within this turn and worth more than
> +  the relevant of the given thresholds until we have run out of juicy
> +  targets or movement.  The first threshold is for attacking which will
> +  leave us where we stand (attacking adjacent units), the second is for
> +  attacking distant (but within reach) targets.
> +
> +  For example, if unit is a bodyguard on duty, it should call
> +    ai_military_rampage(punit, 100, RAMPAGE_ONLY_FREE_CITY)
> +  meaning "we will move _only_ to pick up a free city but we are happy to
> +  attack adjacent squares as long as they are worthy of it".
> 
> Modify thres_adj by occupychance. For example:
>   assert(thres_adj <= thres_move);
>   thres_adj += ((tresh_move - tresh_adj) * game.occupychance / 100);
> 
> This teaches the AI about the dangers inherent in occupychance. Not that
> anyone uses it, but as above, if it is in the rules, we should support it.

Ok.

> +/**********************************************************************
> +  Sea attack is the same as overlap (consider bombardment) but we don't
> +  want to pass through enemy tiles.
> +**********************************************************************/
> +static int sea_attack_move(int x, int y, enum direction8 dir,
> +                         int x1, int y1, struct pf_parameter *param)
> +{
> +  struct tile *src_tile = map_get_tile(x, y);
> +
> +  if (src_tile->terrain == T_OCEAN) {
> 
> Use is_ocean(src_tile->terrain), applies everywhere.

Ok.

> +    if (is_non_allied_unit_tile(src_tile, param->owner)) {
> +      return PF_IMPOSSIBLE_MC;
> +    }
> 
> You need to check F_NO_LAND_ATTACK here. Which means that you have to pass
> another argument, I guess.

Mmmh.  Yes...

> (Don't we have code in common/ that does this kind of thing already?)

What thing?

> +    /* Sea-to-Land.  TODO: Marines support */
> 
> Please don't leave such important TODOs unimplemented.

Mmmmh.  Not sure about that one.  Will try...

> +    /* Land-to-Land */
> +    if (is_non_allied_unit_tile(src_tile, param->owner)) {
> +      /* Cannot pass through defended tiles */
> +      move_cost = PF_IMPOSSIBLE_MC;
> +    } else if(is_non_allied_unit_tile(tgt_tile, param->owner)) {
> +
> +      /* TODO: Intelligent checking */
> +      /* Attack! */
> 
> What would 'intelligent checking' mean here?

Like could_unit_attack_tile and checking whether all people areenemies 
there and such.

G.