30 replies to this topic

[ArkhanTheBlack]

Remove terrain_analyzer:GetPathingDistance() then.

I really ask myself why I do spend my time with endless test while you brush it aside with one remark.

I'm tired of this. Do whatever you like. Will not waste more time with this problem.

Your tests on Necrons, Tau and Orcs were essential because I can't reproduce those crashes so easily. I also wasted an insane amount of time for tests and my first solution wasn't working too.

The fact that terrain_analyzer:GetPathingDistance() is unstable is known for at least one year. You checked it even yourself that time. So please don't blame me if you've forgotten that the function is unstable.

Further the CompassMove() function in it's current state was written by me. And the only thing I can do is to deactivate it, because both functions GetPathingDistance() and CanJumpToPosition() are unstable now. I can't really say that it's funny for me.

[LarkinVB]

I offered a possible solution with just using the first three iterations of CompassMove(). This solution did work here on all maps/setups where the old code did crash. A bit disappointing that it won't get a beta only because I used a wrong sequence of functions.

Edited by LarkinVB, 05 April 2008 - 02:55 PM.

[ArkhanTheBlack]

The problem is that both functions are extremely unreliable. They are map, situation and maybe even machine dependent. It would require hundreds of test games on different maps with different races on different computers to be sure if it's working. The Necron Lord crash bug took me one time about 40 games in two days to reproduce it just one time. The Necron Warrior crash occured only once in probably 80+ test games. I'm so sick of it that I prefer to just deactivate it instead of starting endless test sessions again.
I know that GetPathingDistance() was unstable on Mountain Trail, but not on Kasyr Luthien. And even on Mountain Trail, it could take up to 10 games to reproduce it. Let's just leave it that way now. I think the AI is good enough even without the DeviationJumps.

[dreddnott]

What exactly will the AI be unable to do if you disable what you think you need to disable to prevent crashes, Arkhan?

[ArkhanTheBlack]

What exactly will the AI be unable to do if you disable what you think you need to disable to prevent crashes, Arkhan?

In some cases a unit with jump ability in a critical situation won't make an emergency jump out of trouble. The code searches for a save friendly unit or building in range. If it finds one, it will jump to that location and everything is fine. However, if it can't find a save unit or building we used some code that checked in several directions for a save place. Those position checks could hit impassable terrain. In DC, that wasn't a problem, the function just returned a failed value. In SS however, it can also cause a crash in certain situations.
If we remove this code the emergency jump won't be as reliable as before.

[thudo]

Relic is listening: post precisely what the issue is and my exec will listen for SS Patch2. Of course nothing is guaranteed but its something.

[ArkhanTheBlack]

Relic is listening: post precisely what the issue is and my exec will listen for SS Patch2. Of course nothing is guaranteed but its something.

Well, it's quite simple! The SquadAI function "bool CanJumpToPosition( Math::Vector3f pos )" can cause a CTD if it gets a position it doesn't like. It's possible that the function was broken in SS, but it is also possible that the maps use some very excentric terrain features. I'm pretty sure that Broken Lands is a new map in SS, but I'm not so sure about Eden. Though, I surely never used Eden as a test map before, so it might be a hidden bug in DC which just wasn't discovered until now.

[ArkhanTheBlack]

There's one thing that could work. The SquadAI function DoMoveToClosestSafePoint() accepts a range and can return a save position. We use it for dancing, so it's probably stable. Maybe I give it a try tomorrow. I've no motivation left today...

[dreddnott]

Eden is an old, old map. 1v1, 128x128, with lots of impassable areas, so the map proper is extremely small.

DC AI mod v2.6 of course does not crash with Necrons vs. IG like the betas of SS AI mod (I just tested it to be sure).

[ThetaOrion]

Actually, finding the cause of the crash and a couple of possible solutions went extremely fast here -- a couple of beta cycles. Arkhan and team really have the system down well, here.

I have been watching the development of another AI elsewhere, and so far it has been a solid two weeks and they still haven't isolated what's actually causing the crash, and there have been a dozen different betas put out, and a lot more people working on it than here. Still no success in finding the cause of the crash in that other game's user-made AI.

And, there is another game that has an enhancement mod for it that looks tempting, and I have been watching the development of that mod since November 07 or December 07 simply waiting for a stable non-crashing version of that mod to be made. Crash bugs are killed in a new version every month, but still no reports of a stable non-crashing version, and here it is April 08.

So, I think you guys did quite well here. You don't want to turn on each other, because this is the most efficient development that I have ever seen done.

[thudo]

Hooaarryy auto-email notifications work again!