ensuring more land

[bvanevery]

Details 240x120 still dropping

240x120 still dropping

I changed the definition of a Giant map to 240x120.  This is not sufficient to prevent dropped factions, but it happens less frequently.  At 256x128 it was all but guaranteed, now it takes ~5 map generations to see factions dropped.  1x1 island starts are still exceedingly common and the generated maps are not basically usable.  I will try dropping to 120x120.

[bvanevery]

Details 80x160 is stable

80x160 is stable

80x160 seems to place factions just fine from run to run.  120x120 also works, but the aspect ratio in the minimap looks terrible.  Everything is squished horizontally.  In the 1:2 format, everything higher than 80x160 seems to result in dropped factions.  In particular, 100x200 will not work consistently.  So I guess I'll go with 80x160, which is only a little bit better than a Huge map, but I'll take what I can get.  Not certain if I have come up with the ideal settings, but the settings I used above for Huge maps work "ok" at 80x160.  Those are: Continent base=30, Continent modifier=60, Islands=0.

What will I call this size of map?  I don't want to use the word "Giant", because I've already been using that to describe 128x256 maps.  A word that's a bit huger than huge, but still isn't Giant.  Ogre?  Nah.  Really Huge?  Nah.  Trumpian?  No.  Titanic, Mega?  Getting bored.

For now I'm gonna call it "4X Standard" because that's exactly what it is.

[gwillybj]

For now I'm gonna call it "4X Standard" because that's exactly what it is.

Giant
"Bigger"?
Huge


I wrote a post in December 2017 addressing a different topic, but mentioning #WORLDBUILDER.

In 2004 smacksim wrote a series of posts at Apolyton: "Improving the AI: Part I: The Worldbuilder section of alphax.txt". His work has been invaluable to me.

I extracted the entire discussion, pasted it into Word, spell-checked it, and reformatted the paragraphs (in all of this changing none of the context). I saved the results both as a nice Word 2016 document with color and outlines and such, and as a simple text file. I've attached both files [t]here. Someone else might get as much out of smacksim's work as I have.

http://alphacentauri2.info/index.php?topic=20260.msg109203#msg109203

[bvanevery]

I can't take "Bigger" seriously.  I snigger and think of Valley Talk like "ginormous".  I did think of "Huger" but dismissed it.  1) Sounds like a stupid person saying "Huge", 2) very close in spelling to "Huge" and could be easily typoed.

New dumb idea: Bonus Huge.

The smacksim analysis, has some of the concerns I looked into, and others I wasn't even aware of.  "Won't settle on highland plains" is an interesting phenomenon or quandry, one that I can't say I've directly noticed.  I have noticed that the AI spreads itself out way too much, if it's got land to do so.  I wonder why it doesn't try to pack itself in more tightly?

I have indeed theorized that there's an optimal size of continent for a strong AI.  Problem is, the AI is crap at attacking over water.  That's part of why I've made worlds where enemies are reachable by land.  So far it seems like 4 times out of 5, the distribution of 7 factions in the world is reasonably spread out.  1 time in 5, there will be a substantial clump of factions all in the same area.  I think that's the "big continent" phenomenon that smacksim mentioned, or what I've called the placement algorithm being "cute".

I notice that smacksim did not change the continent size ratio settings.  I haven't either.  In the past 24 hours I've pretty much worn myself out rerolling worlds.  I'm ready to proceed to actual game playing for now.  But I think adjusting the continent size ratios, might be an area where better results could be obtained.  (Enough weasel words in that?)

[Buster's Uncle]

Bonus Huge would work.

[bvanevery]

I've got it!  I will call it an "Enormous" map.  Even though Huge and Enormous could be considered synonyms, Enormous is a longer word.

[Buster's Uncle]

-A 'bigger' word...

[bvanevery]

Not necessarily.

[Buster's Uncle]

Yessir.

[bvanevery]

I just tried a lot of different continent ratio settings.  The default settings are "3,6,12,18,24 ; Ratio    (Continent size ratios)".  Changing those didn't seem to have any impact on map generation at all.  I tried making them lower, higher, fewer than 5 settings, more than 5 settings.  No differences that I could discern.  I even tried completely quitting the game and loading "cold" rather than just starting new games, in case alphax.txt was only being read once.  No luck.  So until I run into someone with alternate experiences, I'll leave them at their defaults.

[bvanevery]

I changed the definition of a Giant map to 240x120.  This is not sufficient to prevent dropped factions, but it happens less frequently.  At 256x128 it was all but guaranteed, now it takes ~5 map generations to see factions dropped.  1x1 island starts are still exceedingly common and the generated maps are not basically usable.

In another thread, someone brought up that allowing a map coordinate to go all the way to an 8-bit integer boundary, might expose more bugs in the game.  They theorized that pathfinding, and the pesky "Interceptor crashing the game" bug, might be driven by this sort of thing.  I wonder if behavior improves simply by changing it to 252x126 ?  I can't really test the postulated midgame bugs, but I can test the faction dropping bug fairly readily.

My SMACX AI Growth mod is still using 256x128 for Giant maps.  I have no idea if anyone actually uses that, and Huge is my recommended map size nowadays.  It would still be good to find the threshold and record this wisdom in the mod, for posterity.

[Petek]

I looked into the problems with enormous maps around 10 years ago. The specific problems are that factions sometimes spawn on small islands, or don't spawn at all. I found that:

The above problems occur only in SMAX, not SMAC. I speculate that the "faction spawning algorithm" in SMAX had to be modified from the one in SMAC to account for several factions starting off later than the others. That is, suitable starting locations had to be reserved for the late-starting factions, especially on smaller maps. The unintended consequence was that starting locations were borked on larger maps. The workaround is to use pre-generated maps. For some reason these maps don't have the above problems. I posted some saved files of largish maps here and here.

[bvanevery]

The other workaround is to reduce map size to a threshold where the bad starts problems don't occur.  I did that in my mod, determining a threshold by sheer trial and error.  I defined the so-called Enormous map at 80x160.  160x160 would also work, but the pixel aspect ratio is horrible.  1:2 is necessary for the map to look nice in the minimap display.

I am currently playing a Giant game with the new 126x252 dimension.  No bugs so far.  That doesn't mean much.  Dropped factions are definitely possible at this size, and an earlier game was missing someone.  I wasn't expecting that problem to be solved though.  I just don't want any other game crashing bugs.  In my current game, 6 factions including myself are healthy and accounted for.  The Usurpers are extant, but I don't know their land extent yet.  I think they have a normal appearance on the power graph, so I don't think they were stillborn.

[scient]

In another thread, someone brought up that allowing a map coordinate to go all the way to an 8-bit integer boundary, might expose more bugs in the game.  They theorized that pathfinding, and the pesky "Interceptor crashing the game" bug, might be driven by this sort of thing.  I wonder if behavior improves simply by changing it to 252x126 ?  I can't really test the postulated midgame bugs, but I can test the faction dropping bug fairly readily.

Most coordinates are stored in game structures as 'signed short' (__int16). There could be instances where they are treated as signed 8-bit causing issues.

The interceptor crash might be a combination of things. The heart of it has to do with two globals not being set correctly (VehLift_xCoord/VehLift_yCoord), causing game to potentially attempt to read invalid memory address inside map struct. The crash occurs inside veh_scoot. The globals get set in veh_lift. My unofficial patch basically sets these globals directly further up in the code. The interceptor function is on my short list to poke around next and implement a more elegant bug fix.

https://github.com/b-casey/OpenSMACX/blob/master/src/veh.cpp#L1281

[bvanevery]

I feel like we're talking about a baby we're not even allowed to breathe on.   

How to raise a nerve stapled child.  Just rewrite its base code.