[1.6.6] Alchemist AI used Vargh Redemption in Angolwen

[WonderingMonster]

My alchemist had just bought some runes for their golem while in Angolwen. As I switched from the alchemist to the golem to use the runes, the alchemist's AI decided to use Vargh Redemption, hitting a nearby friendly pyromancer and causing it to go hostile. As far as I know there were no enemies nearby, with it being a friendly town. Why did the ring get triggered? It seems like a bug.

Will I be able to revisit the town later, or will the pyromancer remain permanently hostile?

[WonderingMonster]

I was able to reproduce this even if there were no other characters around. Whenever I switch to the golem, Vargh Redemption's Tidal Wave ability gets cast. Here's a save where this can be done: https://www.dropbox.com/s/rk9qnf8j5973z ... i.zip?dl=0

[Targuil]

I saw this happen once with my Writhing One as well when I switched to allocate talents to my Worm but I didn't think much about it back then. Definitely seems like a bug.

[WonderingMonster]

I've dug into this and turned on cheat mode and set log_detail_ai to 4. Sure enough, the tactics AI assigned a high score to activating the ring:

Code: Select all

[use_tactical AI] ===Want Buff parameters:
 ___ buff_best_attack_value =	2	best attack[nil]: 1, lvl:nil, ai_tactic: nil	
	___ best attackarea[T_ACTIVATE_OBJECT_2]: 2.7028125, lvl:2, ai_tactic: nil
 ___ buff_range_factor: 0, target dist: nil, desired range: 1	buff_duration_factor:	0.1
[use_tactical AI] ### foes_near_strength	0	### allies_near_strength	0	### want.life	0	want.escape	0
[use_tactical AI] ### ai_tactic:
		{}
[use_tactical AI] ### Final Wants (ai_tactic applied):
		{air=0, attack=2, attackall=2, attackarea=2, buff=0.1, defend=0.1, escape=0, heal=0, life=0, mana=0, special=1}
[use_tactical AI] === Tactical Action Summary === attempt:	1	2955	Daddar
  #:type:tid/ai name                          score  [xmLVL xmSPD  xmRNG] (tact=want*value, ...)
  1:  tid:T_ACTIVATE_OBJECT_2                 = +7.96[x1.40 x 1.00 x1.05] (escape=+0.00*+1.50,attackarea=+2.00*+2.70,)
  2:  tid:T_RUNE:_MANASURGE_3                 = +0.00[x1.20 x 1.00 x1.37] (mana=+0.00*+1.00,)
[use_tactical AI] === Tactical Action Summary (SORTED) === attempt:	1	2955	Daddar
  #:type:tid/ai name                          score  [xmLVL xmSPD  xmRNG] (tact=want*value, ...)
  1:  tid:T_ACTIVATE_OBJECT_2                 = +7.96[x1.40 x 1.00 x1.05] (escape=+0.00*+1.50,attackarea=+2.00*+2.70,)
  2:  tid:T_RUNE:_MANASURGE_3                 = +0.00[x1.20 x 1.00 x1.37] (mana=+0.00*+1.00,)
[LOG]	#GREY#  1:  tid:T_ACTIVATE_OBJECT_2                 score=+7.96[Lx1.40  Sx 1.00 Mx1.05] (escape=+0.00*+1.50,attackarea=+2.00*+2.70,)
[LOG]	#GREY#  2:  tid:T_RUNE:_MANASURGE_3                 score=+0.00[Lx1.20  Sx 1.00 Mx1.37] (mana=+0.00*+1.00,)
[use_tactical AI]### Daddar[2955] tactical AI picked action (attackarea)tid:T_ACTIVATE_OBJECT_2 [att:1, turn 5423: {score:+7.96 [escape=+0.00*+1.50,attackarea=+2.00*+2.70,]}
[LOG]	#ORCHID#__Daddar[2955] tactical AI picked action[att:1, turn 5423]: (attackarea)tid:T_ACTIVATE_OBJECT_2 {+7.96 [escape=+0.00*+1.50,attackarea=+2.00*+2.70,]}
[use_tactical AI] === Action Selected:	2955	Daddar	attackarea	 talent:T_ACTIVATE_OBJECT_2
[LOG]	Daddar brandishes #FFD700##UID:2866:0#Vargh Redemption#LAST#, calling forth the might of the oceans!
[use_tactical AI] turn	5423	2955	Daddar	### SUCCESSFUL ACTION returned:	T_ACTIVATE_OBJECT_2	true

Something that caught my attention earlier on in the log is that the AI concluded the ability would only affect the user ("may affect (SELF ONLY)"). That seemed wrong.

Code: Select all

[aiTalentTactics] COMPUTING TACTICs [2955]Daddar(OHash:nil) (all, wt_mod:nil, wt cache_turns:100) for talent: T_ACTIVATE_OBJECT_2 targeted on nil[nil], tg=nil
[aiTalentTactics]__ using talent tactical table for	T_ACTIVATE_OBJECT_2
	___[escape] = 1.5
	___[attackarea] = {
	___  [COLD] = 1.5
	___  [PHYSICAL] = 1.5
	___}
[aiTalentTactics] *** creating new turn_procs CACHE for talent	T_ACTIVATE_OBJECT_2
[aiTalentTactics] *** creating new target WEIGHT CACHE for talent	T_ACTIVATE_OBJECT_2	cache validity(turns):	100	aiOHash:	nil	vs (cache):	nil
[aiTalentTactics] extracting targets, selffire, friendlyfire from tg:	(from talent)
[aiTalentTargets] 	T_ACTIVATE_OBJECT_2	may affect (SELF ONLY)	Daddar	2955	at	20	12
[aiTalentTactics] *** updating turn_procs CACHE for talent	T_ACTIVATE_OBJECT_2
[aiTalentTactics]	targets for T_ACTIVATE_OBJECT_2: RT:nil(hostile:nil) TG:nil SF(x5)=1, FF(x1)=1
	____	T_ACTIVATE_OBJECT_2	may hit	2955	Daddar	at	20	12

When I compared the ring's use_power definition with Tidal Wave, I noticed that for the ring the radius was hardcoded to 1 instead of the full range the wave would grow to and also that required_target / target were missing. I added those based on Tidal Wave's values and this appears to have fixed the issue. Vargh Redemption no longer gets used when no enemies are around. I also checked that it does get used when enemies are near by luring some enemies to the alchemist with the golem.

I put my patch on https://github.com/rgeens/t-engine4/pull/1/files and I'll see if I can get it merged on the Netcore Gitlab next.

[WonderingMonster]

Patch was accepted, will be fixed in the next release.