Bomberland: make bomb blasts visible and dangerous; document detonate

- Blast (x) cells now persist BLAST_TTL=2 ticks so they actually appear in the
  observation agents receive (previously set to ttl=1 and cleared before any
  agent saw them), and store their owner for damage attribution.
- A unit moving onto an active blast cell now takes 1 HP of damage, credited to
  the blast owner (kills/damage_dealt), on top of the existing instant damage at
  detonation. Fire is now a real, avoidable hazard.
- Document the detonate action (already supported by the runtime) in the arena
  description and runtime README, plus the blast/x hazard behavior.

Author: john-b-yang

codeclash/arenas/bomberland/bomberland.py

@@ -22,9 +22,12 @@ class BomberlandArena(CodeArena):
     def next_actions(game_state):
         return {"unit_0": "up"}
 
-Valid actions are `up`, `down`, `left`, `right`, `bomb`, and `stay`. Each round runs several
-deterministic seeded games. Your units move on a destructible grid, place bombs, destroy blocks,
-damage opposing units, and score by survival, damage, kills, and block destruction.
+Valid actions are `up`, `down`, `left`, `right`, `bomb`, `stay`, and `detonate` (to blow up one of
+your own bombs early, e.g. the string `"detonate:x,y"` or `{"type": "detonate", "coordinates": [x, y]}`;
+bombs also explode automatically after their timer). Each round runs several deterministic seeded
+games. Your units move on a destructible grid, place bombs, destroy blocks, damage opposing units,
+and score by survival, damage, kills, and block destruction. Bomb blasts (`x` entities) stay active
+briefly and damage any unit standing on or moving into them.
 """
     default_args: dict = {
         "sims_per_round": 4,

codeclash/arenas/bomberland/runtime/README.md

@@ -12,11 +12,15 @@ def next_actions(game_state):
     return {"unit_0": "up"}
 ```
 
-Valid string actions are `up`, `down`, `left`, `right`, `bomb`, and `stay`.
+Valid string actions are `up`, `down`, `left`, `right`, `bomb`, `stay`, and
+`detonate` (blow up one of your own bombs early, e.g. `"detonate:x,y"` or
+`{"type": "detonate", "coordinates": [x, y]}`; bombs also explode on their timer).
 The game-state dictionary follows the upstream starter-kit shape where possible:
 `connection.agent_id` identifies the player, `agents[player].unit_ids` lists the
 controlled units, `unit_state` contains unit coordinates and health, and
-`entities` contains walls, destructible blocks, bombs, and blast tiles.
+`entities` contains walls, destructible blocks, bombs, and blast tiles (`x`).
+Blast tiles stay active briefly and damage any unit that stands on or moves onto
+them, so avoid walking into fire.
 
 Round simulation counts must be even so each player receives both starting sides.
 

codeclash/arenas/bomberland/runtime/run_bomberland.py

@@ -20,6 +20,11 @@
 START_BOMBS = 1
 BLAST_RADIUS = 3
 BOMB_TIMER = 6
+# How many ticks a blast (fire) cell stays active. It must be >= 2 so the cell
+# survives the start-of-tick decrement and is visible in the observation agents
+# receive on the following tick. While active, a blast damages any unit that
+# moves onto it.
+BLAST_TTL = 2
 
 
 def load_agent(name, path):
@@ -165,8 +170,8 @@ def entities_for_state(metal, wood, bombs, blasts):
                 "blast_diameter": bomb["radius"] * 2 - 1,
             }
         )
-    for (x, y), ttl in sorted(blasts.items()):
-        entities.append({"type": "x", "coordinates": [x, y], "ttl": ttl})
+    for (x, y), info in sorted(blasts.items()):
+        entities.append({"type": "x", "coordinates": [x, y], "ttl": info["ttl"]})
     return entities
 
 
@@ -241,7 +246,7 @@ def explode_bomb(index, bombs, metal, wood, units, stats, blasts):
         units[unit_id]["inventory"]["bombs"] += 1
 
     for cell in cells:
-        blasts[cell] = 1
+        blasts[cell] = {"ttl": BLAST_TTL, "owner": owner}
         if cell in wood:
             wood.remove(cell)
             stats[owner]["wood_destroyed"] += 1
@@ -345,6 +350,15 @@ def apply_actions(players, callbacks, agents, units, metal, wood, bombs, blasts,
             stats[player]["invalid_actions"] += 1
             continue
         units[unit_id]["coordinates"] = list(target)
+        blast = blasts.get(target)
+        if blast is not None:
+            unit = units[unit_id]
+            unit["hp"] -= 1
+            owner = blast["owner"]
+            if unit["agent_id"] != owner:
+                stats[owner]["damage_dealt"] += 1
+                if unit["hp"] <= 0:
+                    stats[owner]["kills"] += 1
 
 
 def tick_bombs(bombs, metal, wood, units, stats, blasts):
@@ -392,7 +406,7 @@ def run_game(players, callbacks, seed, ticks, width, height, unit_count, agent_t
     final_tick = 0
     for tick in range(ticks):
         final_tick = tick
-        blasts = {pos: ttl - 1 for pos, ttl in blasts.items() if ttl > 1}
+        blasts = {pos: {**info, "ttl": info["ttl"] - 1} for pos, info in blasts.items() if info["ttl"] > 1}
         apply_actions(
             players, callbacks, agents, units, metal, wood, bombs, blasts, stats, width, height, tick, agent_timeout
         )