Single player mode initial version (#33)

* Single player mode initial version

* Ref: Create new Tournament class. Move main.py code

* WIP: Factored out tournament class from game class

* Working evaluation for single player mode

* Ref/simplify: Add empty patch for round 0

* Enh(viewer): Improve game session dropdown

Add warning sign next to probably failed; add # rounds; alphabetical sort

* Ref: Change get_agent and prompt rendering (needs to use tournament)

* Change: Move round setting to tournament section of config

Author: klieret

codeclash/agents/__init__.py

@@ -1,33 +1,18 @@
+from minisweagent.environments.docker import DockerEnvironment
+
 from codeclash.agents.abstract import Player
 from codeclash.agents.dummy import Dummy
 from codeclash.agents.minisweagent import MiniSWEAgent
 from codeclash.agents.utils import GameContext
-from codeclash.constants import DIR_WORK
-from codeclash.games.abstract import CodeGame
 
 
-def get_agent(config: dict, prompts: dict, game: CodeGame) -> Player:
+def get_agent(
+    config: dict, game_context: GameContext, environment: DockerEnvironment
+) -> Player:
     agents = {
         "dummy": Dummy,
         "mini": MiniSWEAgent,
     }.get(config["agent"])
     if agents is None:
         raise ValueError(f"Unknown agent type: {config['agent']}")
-    environment = game.get_environment(
-        f"{game.game_id}.{config['name']}"
-    )  # NOTE: MUST be branch_name (defined in agents/abstract.py)
-    return agents(
-        config,
-        environment,
-        GameContext(
-            id=game.game_id,
-            log_env=game.log_env,
-            log_local=game.log_local,
-            name=game.name,
-            player_id=config["name"],
-            prompts=prompts,
-            round=1,
-            rounds=game.rounds,
-            working_dir=str(DIR_WORK),
-        ),
-    )
+    return agents(config, environment, game_context)

codeclash/agents/abstract.py

@@ -1,12 +1,14 @@
 import os
+import uuid
 from abc import ABC, abstractmethod
 
 from dotenv import load_dotenv
-from minisweagent import Environment
+from minisweagent.environments.docker import DockerEnvironment
 
 from codeclash.agents.utils import GameContext
 from codeclash.constants import GH_ORG
-from codeclash.utils.environment import assert_zero_exit_code
+from codeclash.tournaments.utils.git_utils import filter_git_diff
+from codeclash.utils.environment import assert_zero_exit_code, create_file_on_container
 from codeclash.utils.log import get_logger
 
 load_dotenv()
@@ -16,41 +18,52 @@ class Player(ABC):
     def __init__(
         self,
         config: dict,
-        environment: Environment,
+        environment: DockerEnvironment,
         game_context: GameContext,
-    ):
+    ) -> None:
         self.config = config
         self.name = config["name"]
+        self._player_unique_id = uuid.uuid4()
+        """Unique ID that doesn't clash even accross multiple games. Used for git tags."""
         self.environment = environment
         self.game_context = game_context
-        self.game_context.render_and_set_prompts()
         self.logger = get_logger(
             self.name,
             log_path=self.game_context.log_local / f"{self.name}.log",
             emoji="👤",
         )
+        self._metadata = {
+            "name": self.name,
+            "player_unique_id": self._player_unique_id,
+            "diff": {0: ""},  # mapping round -> diff
+            "incremental_diff": {0: ""},  # mapping round -> diff
+        }
 
-    @property
-    def branch_name(self):
-        """Get the branch name for the agent's codebase."""
-        return f"{self.game_context.id}.{self.name}"
-
-    def commit(self):
-        """Commit changes to the agent's codebase."""
-        r, rounds = self.game_context.round, self.game_context.rounds
-        for cmd in [
-            "git add -A",
-            f"git commit --allow-empty -m 'Round {r}/{rounds} Update'",
-        ]:
-            assert_zero_exit_code(self.environment.execute(cmd), logger=self.logger)
-        self.logger.info(f"Committed changes for {self.name} for round {r}/{rounds}")
+    # --- Main methods ---
 
-    def on_round_update(self, new_round: int):
-        """Update the agent's round to match the game round."""
+    def pre_run_hook(self, *, new_round: int) -> None:
+        """Should be called before we call the run method."""
+        if new_round == 1:
+            self._tag_round(0)
         self.game_context.round = new_round
-        self.game_context.render_and_set_prompts()
 
-    def push(self):
+    def post_run_hook(self, *, round: int) -> None:
+        """Should be called after we called the run method."""
+        self._commit()
+        self._metadata["diff"][round] = self._get_round_diff(round)
+        self._metadata["incremental_diff"][round] = self._get_round_diff(
+            round, incremental=True
+        )
+
+    @abstractmethod
+    def run(self) -> None:
+        """Given the observation / recap, update the codebase"""
+
+    def get_metadata(self) -> dict:
+        """Get metadata for the agent."""
+        return self._metadata
+
+    def push(self) -> None:
         """Push codebase to a branch on the game's remote repository."""
         token = os.getenv("GITHUB_TOKEN")
         if not token:
@@ -59,13 +72,98 @@ def push(self):
         for cmd in [
             "git remote remove origin",
             f"git remote add origin https://x-access-token:{token}@github.com/{GH_ORG}/{self.game_context.name}.git",
-            f"git push origin {self.branch_name}",
+            f"git push origin {self._branch_name}",
+            "git push origin --tags",
         ]:
             assert_zero_exit_code(self.environment.execute(cmd), logger=self.logger)
         self.logger.info(
-            f"Pushed {self.name} commit history to remote repository (branch {self.branch_name})"
+            f"Pushed {self.name} commit history to remote repository (branch {self._branch_name})"
         )
 
-    @abstractmethod
-    def run(self):
-        """Given the observation / recap, update the codebase"""
+    def reset_and_apply_patch(
+        self, patch: str, *, base_commit: str = "", filter_patch: bool = True
+    ) -> None:
+        """Clean all uncommited changes. If base_commit is provided, reset to that commit.
+        Then apply the patch to the codebase.
+        """
+        # Need to clean before we copy over the patch (else it's gonna be removed by git clean)
+        self.logger.debug(
+            assert_zero_exit_code(
+                self.environment.execute(
+                    f"git reset --hard {base_commit} && git clean -fd"
+                )
+            )
+        )
+
+        patch = filter_git_diff(patch) if filter_patch else patch
+
+        if not patch.strip():
+            self.logger.debug("No patch to apply, skipping")
+            return
+
+        create_file_on_container(
+            container=self.environment,  # type: ignore
+            content=patch,
+            dest_path="tmp_patch.txt",
+        )
+
+        self.logger.debug(f"Applying patch to agent's codebase: {patch}")
+
+        commands = ["git status", "git apply tmp_patch.txt", "rm -f tmp_patch.txt"]
+        for cmd in commands:
+            self.logger.debug(f"Executing command: {cmd}")
+            out = assert_zero_exit_code(
+                self.environment.execute(cmd), logger=self.logger
+            )
+            self.logger.debug(out)
+
+    # --- Helper methods ---
+
+    def _tag_round(self, round: int) -> None:
+        """Git tag the codebase at the given round."""
+        assert_zero_exit_code(
+            self.environment.execute(
+                f"git tag -a {self._get_round_tag_name(round)} -m 'Round {round} Update'"
+            ),
+            logger=self.logger,
+        )
+
+    @property
+    def _branch_name(self) -> str:
+        """Get the branch name for the agent's codebase."""
+        return f"{self.game_context.id}.{self.name}"
+
+    def _get_round_tag_name(self, round: int) -> str:
+        """Get git tag name for the version of the codebase at the given round."""
+        return f"{self._player_unique_id}-round-{round}"
+
+    def _commit(self) -> None:
+        """Commit changes to the agent's codebase."""
+        r = self.game_context.round
+        for cmd in [
+            "git add -A",
+            f"git commit --allow-empty -m 'Round {r} Update'",
+        ]:
+            assert_zero_exit_code(self.environment.execute(cmd), logger=self.logger)
+        self._tag_round(r)
+        self.logger.info(f"Committed changes for {self.name} for round {r}")
+
+    def _get_round_diff(self, round: int, *, incremental: bool = False) -> str:
+        """Get the diff between the round and initial version (round 0).
+        If incremental is True, get the diff between the round and the previous round.
+        Returns empty string if round is 0.
+        """
+        if round == 0:
+            return ""
+        if incremental:
+            previous_round_tag = self._get_round_tag_name(round - 1)
+        else:
+            previous_round_tag = self._get_round_tag_name(0)
+        current_round_tag = self._get_round_tag_name(round)
+        out = assert_zero_exit_code(
+            self.environment.execute(
+                f"git diff {previous_round_tag}..{current_round_tag}"
+            ),
+            logger=self.logger,
+        )
+        return out["output"]

codeclash/agents/dummy.py

@@ -5,4 +5,5 @@ class Dummy(Player):
     """A dummy player that does nothing. Mainly for testing purposes."""
 
     def run(self):
-        self.commit()
+        pass
+        # self.commit()  # now called in post_round_hook

codeclash/agents/minisweagent.py

@@ -8,8 +8,9 @@
 
 import yaml
 from jinja2 import Template
-from minisweagent import Environment, Model
+from minisweagent import Model
 from minisweagent.agents.default import AgentConfig, DefaultAgent
+from minisweagent.environments.docker import DockerEnvironment
 from minisweagent.models.litellm_model import LitellmModel
 from minisweagent.run.utils.save import save_traj
 from rich.console import Console
@@ -28,7 +29,7 @@ class ClashAgent(DefaultAgent):
     def __init__(
         self,
         model: Model,
-        env: Environment,
+        env: DockerEnvironment,
         name: str,
         game_context: GameContext,
         *,
@@ -56,7 +57,7 @@ def render_template(self, template: str, **kwargs) -> str:
             | asdict(self.env.config)
             | asdict(self.model.config)
             | platform.uname()._asdict()
-            | self.game_context.to_dict()
+            | self.game_context.to_template_vars()
         )
         return Template(template).render(**kwargs, **cs, **os.environ)
 
@@ -69,7 +70,7 @@ class MiniSWEAgent(Player):
     """Player with agentic code editing capabilities"""
 
     def __init__(
-        self, config: dict, environment: Environment, game_context: GameContext
+        self, config: dict, environment: DockerEnvironment, game_context: GameContext
     ):
         super().__init__(config, environment=environment, game_context=game_context)
 
@@ -104,10 +105,11 @@ def run(self):
                 traj_path,
                 exit_status=exit_status,
                 result=result,
+                print_fct=self.logger.debug,
             )
             copy_file_to_container(
                 self.environment,
                 traj_path,
                 self.game_context.log_env / traj_path.name,
             )
-            self.commit()
+            # self.commit()  # now called in post_round_hook

codeclash/agents/utils.py

@@ -36,24 +36,15 @@ class GameContext:
     rounds: int
     working_dir: str
 
-    def render_and_set_prompts(self):
-        """Render and set prompts using the current game context."""
+    def _render_prompt_templates(self) -> dict:
         context = asdict(self)
-        del context["prompts"]
-        for key, template_str in self.prompts.items():
-            rendered = Template(template_str).render(**context)
-            setattr(self, key, rendered)
-
-    def to_dict(self):
-        """Convert the GameContext to a dictionary, including dynamically added attributes."""
-        result = asdict(self)
-        declared = set(self.__dataclass_fields__)
-        for attr in dir(self):
-            if (
-                not attr.startswith("_")
-                and attr not in declared
-                and not callable(getattr(self, attr))
-            ):
-                result[attr] = getattr(self, attr)
-        del result["prompts"]
-        return result
+        return {
+            key: Template(template_str).render(**context)
+            for key, template_str in self.prompts.items()
+        }
+
+    def to_template_vars(self) -> dict[str, str]:
+        """Convert the GameContext to a dictionary for rendering prompts in the agent"""
+        out = asdict(self) | self._render_prompt_templates()
+        out.pop("prompts")
+        return out

codeclash/games/__init__.py

@@ -1,3 +1,5 @@
+from pathlib import Path
+
 from codeclash.games.abstract import CodeGame
 from codeclash.games.battlecode.main import BattleCodeGame
 from codeclash.games.battlesnake.main import BattleSnakeGame
@@ -7,16 +9,17 @@
 
 
 # might consider postponing imports to avoid loading things we don't need
-def get_game(config: dict) -> CodeGame:
+def get_game(config: dict, *, tournament_id: str, local_output_dir: Path) -> CodeGame:
     game = {
-        x.name: x for x in [
+        x.name: x
+        for x in [
             BattleCodeGame,
             BattleSnakeGame,
             CoreWarGame,
             RoboCodeGame,
-            RobotRumbleGame
+            RobotRumbleGame,
         ]
     }.get(config["game"]["name"])
     if game is None:
         raise ValueError(f"Unknown game: {config['game']['name']}")
-    return game(config)
+    return game(config, tournament_id=tournament_id, local_output_dir=local_output_dir)