Entry 167

The gold goblins: thinkGold, moveGold and actEscapeGold

2026-09-26world

Filescrates/piney-battle/src/enemy_ai.rscrates/piney-battle/src/enemy_motion.rscrates/piney-battle/examples/battle_probe/spawn.rstools/test_battle_spawn_rs.py

The goblin rows 131-138, 140-143, 147-150 and 154-157 are gold goblins (goldFlag). Their constructor (checkGold) was ported, but they thought with defaultThink and moved like any goblin. That was wrong in every field once the portals came (worklog 0164). Now ccEnemyG::think sends them to thinkGold, and ccEnemyG::action to moveGold. docs/engine/battle.md, "The gold goblins", has the rules.

What they do. They run from the party and wear sleep, confusion, charm and paralysis off by their hoard's size. A volume 2 or larger goblin that is hit while held, or dragged more than 500 from where the hold caught it, shakes the hold off. For one frame it sets game.inBattleDist to -2 and goes off the command lists, then puts the distance back to 2200 and comes back on them. That is a new enemy_ai::Out::InBattleDist, which the runtime writes into its ccGame battle state after the entry frame. actEscapeGold picks a heading away from the target and a turn rate from goldParam, both with a little randomness, and sets its speed from the zoom and crisisRate. The hold escape overwrites crisisRate.

Four things came up on the way:

• The dropped draw. One of actEscapeGold's branches draws ccRandF(pi/10) and drops the result; the generator still moves.
• The widened ranges. checkGold writes 50000 into a volume 3 or 4 row's area, territory and viewRange in the table itself. Out::GoldRanges said so, but nothing applied it. The first comparison found a goblin 8,000 from Kite that the game's goblin saw and the port's did not. The port's think() now reads the widened ranges for those rows; every goblin of a row has the row's volume.
• The dust. A walking gold goblin's dust goes through ccTransPosFW2LW first. The runtime did this already, but the probe printed the raw position, one unit in the last place off.
• Delay slots in my listings. My no-nop listings hide the delay slots. An instruction that follows a branch in them is not always in its slot. That was checked against the full listing where it mattered. The same check showed that worklog 0164's note about effDungeonEntrance was wrong: its pTexMod store runs only when the colour is not -1. The result is the same (the generator starts at -1), and the code comment and effects.md are corrected.

Checks. tools/test_battle_spawn_rs.py's frame_enemies now spawns gold goblins as well: rows of each volume and type, in 40% of the cases. Between frames it holds them, lets them go, puts them to sleep, paralyses, confuses or charms them (event kind 2, which the probe already had). game.inBattleDist is part of each frame's scene (left at 0x7f7f7f7f before, so a write shows). All 200 cases were equal. A 60-case sample saw every act but wander and home, and the hold escapes set the distance to -2 and 2200. The whole spawn suite, the enemy AI and motion harnesses and 605 workspace tests pass.