[freeciv-ai] Re: pre-patch to new settler code

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

On Wed, 30 Apr 2003, Juhani Heino wrote:
> As a whole, settlers work fine now. Here are some observations from
> my test. Most of them don't regard settlers at all, and some are
> probably taken care of already, but I include all in case somebody
> gets fresh ideas.

Thanks, but I am not sure exactly why you post these comments in this
thread, though, since clearly you haven't used my new settlers patch :-)

> Settlers were built "in vain",ie. the city was size 1. A couple
> of times they even bought settlers - in those cases it was only
> a matter of one turn, but in one city the settlers were ready
> (not bought) 15 turns before the size increase.

The AI will only buy a settler the turn before the city grows. While it
may not look optimal, I don't see a better alternative than building
settlers in size 1 cities. We do want to churn out those settlers as fast
as possible, and if we instead built something else while in size 1, we
might lose several turns on the settler while in size 2.

The AI doesn't lose shields anyway.

> Should settlers avoid huts? Once one was lucky, but the second
> case was killed by barbarians.

Hut = danger in new pf code, Greg?

> AI plants cities even denser than I do, but that's OK

Then just wait until you see my new settler code. It plants cities even
denser.

> because
> with AI strategy the cities don't get big anyway. But I suggest
> "oceanotropism": the settler would choose the place with max
> amount of ocean squares. That would give more room. I can clarify
> this, but sure you get the idea.

Room is not as important as getting production fast. Once you have
production, you can make room by making caravels and/or stomping other
players.

> Triremes were idling at the port, so they could have been used for
> exploring. Should we make an "appointment", ie. the ship must be
> THERE and THEN, and before that they can explore? They went also
> carelessly to outer sea, but I saw that discussed earlier.

Yes, the present ferry code absolutely sucks.

> Units should avoid stacking in a same tile. Especially diplomats!

Yes. Diplomats should try to stack with units that have higher defense,
though.

Greg, you said you would look at adding avoid-stacking code to pf? We need
that before we can get any useful unit coordination code - otherwise units
will just stack up and get eliminated by the dozens.

> This may be only a problem of the observation mode, but cities
> were oscillating wildly with their specs. Example: all workers
> were scientists, so food was getting lower. But the next turn
> they had produced food normally, and they were shown as scientists
> again.

I can't remember having seen this. I usually don't watch cities very
closely, though.

> This most importantly concerns govt model: many times they
> changed from republic to despotism and then back almost
> immediately.

Did citizens revolt? I can't imagine that the AI actually decided to
positively go for Despotism... never seen that myself.

> At least that caused loss of units, but it
> doesn't seem healthy in other respects either.

Why did it cause loss of units?

> A city was building barracks until the last moment and then
> changed that to a defensive unit (it had none). Only a matter
> of one turn, but loss of shields anyway.

AI doesn't lose shields (cheat).

> even buildings of which I didn't see many  8-)

Buildings are currently not very useful in freeciv. The AI "knows" this.

> Even more inexcusable was sending a task force and leaving
> the city undefended while the enemy was at the gates. The
> city even had walls, still rare at the time. And the suckers
> couldn't even invade the city they were targeting. It's an
> idea that they were avoiding being captured by the diplomat,
> but shooting him and his escort with cannon would have been
> more fruitful, I think.

Sounds like something we should have a look at. Can you supply savegames
for situations like this? That is the only way we can fix it.
Unfortunately it doesn't help much just describing it. We know the AI has
problems, but to fix them we need savegames.

> A trireme took a catapult to explore - quite expensive, I
> think. Generally the AI uses vulnerable attack machines;
> usually they get shot down by city defenders. I think escort
> would be good here, at least when they have musketeers.

Yes, known problems.

> Cities were still letting disorder continue.

Same.

> City defenders attacked a phalanx outside. I don't see why
> - they would be better off letting it attack. Not even
> counting the terrain bonus.

Not necessarily. If the phalanx got time to fortify, then it could be used
as a bridgehead for diplomats and cannons later.

> I got this message, but it didn't seem to affect functionality:
> 1: Encountered an inconsistency in add_adjust_workers() for city Dublin

Known problem. Unknown cause.

  - Per