Code size records

I tried the top bots that came to mind and at first couldn't find any over 55k. But apparently you didn't quite set the record (but close). Only kept highest for each bot.

java -jar robocode_1.07/codesize.jar ./robocode.rumble_1730/robots/*.jar
...
45842	293199	74	kid.DeltaSquad.DeltaSquad_.1.jar
46143	289610	77	kid.Gladiator_.7.2.jar
47086	335204	209	Homer.Barney_1.0.jar
47269	458846	232	pedersen.Grishnakh_1.0.jar
50883	406124	107	voidious.Diamond_1.7.24.jar
52283	147604	107	positive.Portia_1.26c.jar
54049	307732	92	florent.XSeries.X2_0.17.jar
54352	249048	48	aw.Gilgalad_1.99.5c.jar
55297	509698	190	xander.cat.XanderCat_12.0.1.jar
58387	139768	35	jk.mega.DrussGT_2.8.8.jar
227423	474856	62	Krabb.sliNk.Garm_0.9y.jar

Not sure what the heck is going on with Garm! :-) Maybe he includes some big external library? I know I've thought about pulling in Guava, though there are ways to only pull in what you use.

Does he have the largest class size though? Speaking of codesize, I tried my hand at rewriting the tool.

Seems like it, but I don't really know what that means. And it (thankfully) sorts by code size, so it's non-trivial to check for greatest class size. Is that just your code + all classes you reference (including the Robocode and Java libraries)? Anyone know?

I think I have to throw in some extra bytes to the 12397 I use now, and pass Phoenix before you do. But hey, I am no Java programmer, for the current assignment I use OOPerl . . .

I am actually trying to aim towards the other end of the field lately.

I don't particularly care one way or another, but I also don't take pride in my large code size. It's worth noting that a laughable ~10k of Diamond's code size is the perceptual gun he uses for the first few shots. :-) Factoring that in, I don't think Diamond's too bloated for a full-featured 1v1 and Melee bot.

Heh, after Garm DrussGT comes in second =) I suspect it's because I have several Precise Prediction methods, my pixel-perfect Precise Intersection ones for first wave calculations and my high-speed ones for second wave calculations, and then I still have all of the code for the precise predictions which tracked when an enemy would fire and what segments they would see even though they aren't being called. Also, all of those buffers that are hard coded add ~17000 codesize.

I was looking into building an AS gun using spectral clustering, and when I packaged it, it pulled in the entire Colt library, which brought DrussGT to 217374 codesize bytes and 760KB of jar file! Needless to say, I started looking at other matrix libraries...