Re: AI is not in a releasable state
Nils Kneuper <crazy-ivanovic <at> gmx.net>
2009-03-06 11:15:18 GMT
Okay, lets have a look at this list and add my personal view on the importance
of those for 1.6. I will explicitly try to compare to 1.4, that is if there is
no regression compared to this, we *can* release...
>> #13120 AI is not scouting
> This bug doesn't have any real instructions on how to reproduce. I have
> updated it asking for them.
Hmm, somehow a part of "make the AI better and easier to configure". Personally
I don't think that this was possible in 1.4 either. So this is not a real
>> #13119 The ai should only recruit level one units for scouting
> I think this is subjective and not really an engine bug. If you don't
> want the AI using level 2 scouts, don't make level 2 scouting units
> recruitable by the AI.
This is more a feature request about "make the AI recruits smarter for specific
purposes". Yes, it makes sense as request, but we know that the AI is not bright
when it comes to recruiting. This is why Dragonking worked on recruiting formulas.
>> #13105 AI is blocked by a certain game configuration.
> I have just fixed this bug, adding notes. Hopefully my fix won't cause
> regressions elsewhere.
This was a clear regression and it looks like Dave fixed it. Okay, the fix has
to be tested no.
>> #8410 AI makes big mistakes in custom maps!
> This is a very old report which contains numerous issues, most of which
> are subjective or not reasonably solvable. All issues are also very old,
> dating to 2007. I have closed this bug.
Ancient "problem" and a case of "the AI has to do *something* if all options are
equal". Yes, this makes the AI more predictable which is not good, but we know
that the AI is not too smart, don't we?
To sum the stuff up:
The only real blocker and regression was fixed. The rest is a case of "we should
improve our AI (in general) and make it easier scriptable".
* Making the AI in general better is a non trivial task. The most likely way to
go over time looks like it will be via formulas, like we talked about at FOSDEM.
Having the AI not be smart is an ancient problem and no "new release blocking
bug" since it is not worse at all than in 1.4.x.
* For the scripting part once upon the time PythonAI was introduced, but it was
not much used and now removed due to the security risk of the implementation.
Since last summer we also got FormulaAI which is meant to allow some scripting,
so in those regards we *do* have an improvement.
In general I think those "AI finetuning parameters" we have in WML for ages were
never working perfectly and thus it is hard to judge if anything improved or
broke there. So the AI is most likely not in a state worse than in 1.4.x, in
fact it is in a better state, since several crashes due to the AI were fixed. At
the moment I don't see the AI as a blocker for 1.4.x, unless you show some
"real" regressions that break gameplay.
Nils Kneuper aka Ivanovic
PS: But I think that bug #13118  should be fixed before rc2 is out!