On the cusp...

PhilD · #1 October 18th, 2006, 03:38 PM

Quote:

curtadams said:
An interesting idea for SP would be to have the computer pre-compute the turns with background threads. First plan the AI turns. Then, once that's done, calculate all the battles as the player sets them up. Because the turn calc time is small compared to the playing time it should be pretty easy to stuff most of the work into the background from a time viewpoint (the programming, admittedly, isn't so easy).

Yeah, I think making this kind of change would be a major reworking of the code - and from what I've understood, the skills in the mammoth "Illwinter development team" may not be that specialized in multithreading an AI (time to get Brad Wardell on the team

)

curtadams · #2 October 18th, 2006, 04:07 PM

A couple of comments -

Yes, I know the programming is complex. A plan - which doesn't require changing the basic model - is for "turn end" procressing to go in the background. The catch would be allowing changes to turn orders - the background process would have to figure out what to back out and redo for each change. Because of "wego" AI choices could be precomputed as is but that seems a relatively trivial amount of time. A crude start would be to precompute and cache each possible battle and redo if anything would change the battle setup. When the background thread finishes, it jumps back to the beginning and reruns the whole turn (if the player has changed anything), consulting its cache of prerun battles. In the most complicated games I suspect there's a lot of AI-AI interaction which would not be affected by most player action - so the turn processing would shorten dramatically. If the player's last action is to pitch a global, oh well, that turn may last a while.

Obviously the change is only relevant and possible for SP.

Lasu · #3 October 18th, 2006, 04:59 PM

Exactly.

Taqwus · #4 October 18th, 2006, 06:01 PM

Er, "every possible battle" would be insanely numerous.

Think about 800 units neighboring 800 units. *Any combination* of those 1600 units could be involved... AND summons and fliers, AND the gems could be changed, AND commanders may or may not have been killed through earlier spells...

Gandalf Parker · #5 October 18th, 2006, 06:41 PM

I think that background processing is fantastic. Many games use it. But I dont think it will work here. Personally I feel it would be moving in the wrong direction.

In a way, we are background processing. The game allows you to do your turns and send them to a server. The server processes everyones turns and sends them back. This allows for larger maps, larger armies, more nations, more events. I would rather see it move farther in that direction. Make MORE use of basically unlimited processing time. Id be fine even it took turn processing up to an hour or more.

And thats with me bing a solo player. I easily play Dom games where doing one turn a day is fine with me. Taking months to finish a game is fine with me. It allows me to do other things and makes dominions the game that lasts forever on my machine. Changing it into something that plays faster might be ok but its not anything I will push too hard for.

curtadams · #6 October 18th, 2006, 07:41 PM

Quote:

Taqwus said:
Er, "every possible battle" would be insanely numerous.

Think about 800 units neighboring 800 units. *Any combination* of those 1600 units could be involved... AND summons and fliers, AND the gems could be changed, AND commanders may or may not have been killed through earlier spells...

Every possible battle that gets set up that turn by the AI orders and the current state of the player's orders. No, not the factorial abstract possibilities.