Pull changes from #107 and #212

Author: matheper

debug_gym/gym/envs/aider.py

@@ -5,19 +5,37 @@
 from debug_gym.constants import DEBUG_GYM_CACHE_DIR
 from debug_gym.gym.entities import EvalOutput
 from debug_gym.gym.envs.env import RepoEnv
+from debug_gym.gym.terminal import DockerTerminal, Terminal
 
 
 class AiderBenchmarkEnv(RepoEnv):
     REPO_URL = "https://github.com/exercism/python"
     REPO_PATH = DEBUG_GYM_CACHE_DIR / "exercism"
 
+    def __init__(
+        self,
+        entrypoint: str = "python -m pytest -s .",
+        terminal: Terminal | None = None,
+        **kwargs,
+    ):
+        terminal = terminal or DockerTerminal(
+            base_image="python:3.12-slim",
+            setup_commands=[
+                "apt update",
+                "apt install -y git",
+                "pip install pytest",
+            ],
+            logger=kwargs.get("logger"),
+        )
+        if not isinstance(terminal, DockerTerminal):
+            raise ValueError("AiderBenchmarkEnv only supports DockerTerminal.")
+
+        super().__init__(entrypoint=entrypoint, terminal=terminal, **kwargs)
+
     @property
     def instructions(self) -> str:
         return self.current_sample["instructions"]
 
-    def __init__(self, entrypoint: str = "python -m pytest -s .", **kwargs):
-        super().__init__(entrypoint=entrypoint, **kwargs)
-
     def calculate_max_score(self, eval_output: EvalOutput) -> int:
         return utils.extract_max_score_from_pytest_output(eval_output.output)
 
@@ -30,11 +48,26 @@ def eval(self, **kwargs) -> EvalOutput:
         self.last_eval = EvalOutput(success, output)
         return self.last_eval
 
+    def setup_terminal(self):
+        self.logger.info(f"Configuring docker container: {self.terminal.container}")
+
+        self.terminal.run("git init")
+        self.terminal.run("git config user.name 'debug-gym'")
+        self.terminal.run("git config user.email '<>'")
+
+        self.terminal.run("git add *.py")
+        self.terminal.run("git commit -am 'Init'")
+
+        self.terminal.run("git add .debugignore")
+        self.terminal.run("git add .debugreadonly")
+        self.terminal.run("git commit -am 'Add debug-gym ignore and read-only files'")
+
     def reset(self, *, options: dict = None):
         options = options or {}
         self.current_sample = self.dataset[options["task_name"]]
         directory = self.current_sample["base_directory"]
         self.setup_workspace(directory, entrypoint=self.entrypoint)
+        self.setup_terminal()
         infos = super().reset(options=options)
         return infos
 

debug_gym/gym/envs/env.py

@@ -1,10 +1,7 @@
 import atexit
-import os
 import shutil
-import subprocess
 import tempfile
 from dataclasses import dataclass
-from glob import glob
 from pathlib import Path
 
 import numpy as np
@@ -300,20 +297,6 @@ def display_files(self):
         msg += self.directory_tree()
         return msg
 
-    def restore(self, *filepaths):
-        filepaths = filepaths or glob(
-            f"{self.path}/**",
-            root_dir=self.path,
-            recursive=True,
-        )
-        relative_filepaths = [os.path.relpath(f, self.path) for f in filepaths]
-        for filepath in relative_filepaths:
-            if os.path.isdir(self.path / filepath):
-                os.makedirs(self.working_dir / filepath, exist_ok=True)
-                continue
-
-            shutil.copy2(self.path / filepath, self.working_dir / filepath)
-
     def reset(self, *, options: dict = None):
         """Resets the environment and returns eval as the initial observation."""
         self.logger.info("Resetting environment")
@@ -507,10 +490,12 @@ def current_breakpoints(self):
 
     @property
     def patch(self):
-        command = ["git", "diff", "--no-index", self.path, self.working_dir]
-        result = subprocess.run(command, text=True, capture_output=True)
-        patch = result.stdout.replace(str(self.working_dir), str(self.path))
-        return patch
+        success, output = self.terminal.run("git diff")
+        if not success:
+            self.logger.error("Failed to get git diff. {output}")
+            return None
+
+        return output
 
     def apply_gold_patch(self):
         raise NotImplementedError(

debug_gym/gym/envs/mini_nightmare.py

@@ -4,6 +4,7 @@
 import debug_gym.gym.utils as utils
 from debug_gym.gym.entities import EvalOutput
 from debug_gym.gym.envs.env import RepoEnv
+from debug_gym.gym.terminal import DockerTerminal, Terminal
 
 
 class MiniNightmareEnv(RepoEnv):
@@ -21,13 +22,30 @@ class MiniNightmareEnv(RepoEnv):
         "tomorrow_date",
     ]
 
+    def __init__(
+        self,
+        entrypoint: str = "python -m pytest -s test.py",
+        terminal: Terminal | None = None,
+        **kwargs,
+    ):
+        terminal = terminal or DockerTerminal(
+            base_image="python:3.12-slim",
+            setup_commands=[
+                "apt update",
+                "apt install -y git",
+                "pip install pytest pandas",
+            ],
+            logger=kwargs.get("logger"),
+        )
+        if not isinstance(terminal, DockerTerminal):
+            raise ValueError("MiniNightmareEnv only supports DockerTerminal.")
+
+        super().__init__(entrypoint=entrypoint, terminal=terminal, **kwargs)
+
     @property
     def instructions(self) -> str:
         return self.current_sample["instructions"]
 
-    def __init__(self, entrypoint: str = "python -m pytest -s test.py", **kwargs):
-        super().__init__(entrypoint=entrypoint, **kwargs)
-
     def calculate_max_score(self, eval_output: EvalOutput) -> int:
         return utils.extract_max_score_from_pytest_output(eval_output.output)
 
@@ -40,11 +58,26 @@ def eval(self, **kwargs) -> EvalOutput:
         self.last_eval = EvalOutput(success, output)
         return self.last_eval
 
+    def setup_terminal(self):
+        self.logger.info(f"Configuring docker container: {self.terminal.container}")
+
+        self.terminal.run("git init")
+        self.terminal.run("git config user.name 'debug-gym'")
+        self.terminal.run("git config user.email '<>'")
+
+        self.terminal.run("git add *.py")
+        self.terminal.run("git commit -am 'Init'")
+
+        self.terminal.run("git add .debugignore")
+        self.terminal.run("git add .debugreadonly")
+        self.terminal.run("git commit -am 'Add debug-gym ignore and read-only files'")
+
     def reset(self, *, options: dict = None):
         options = options or {}
         self.current_sample = self.dataset[options["task_name"]]
         directory = self.current_sample["base_directory"]
         self.setup_workspace(directory, entrypoint=self.entrypoint)
+        self.setup_terminal()
         infos = super().reset(options=options)
         return infos
 

debug_gym/gym/terminal.py

@@ -457,15 +457,15 @@ def setup_container(self) -> docker.models.containers.Container:
         container.rename(container_name)
         container.reload()
         self._run_setup_commands(container)
-        self.logger.debug(f"Container {container_name} started successfully.")
+        self.logger.debug(f"{container} ({container_name}) started successfully.")
         atexit.register(self.clean_up)
         return container
 
     def _run_setup_commands(self, container):
         """Run setup commands if any. If commands fail, stop the container."""
         if self.setup_commands:
             setup_commands = " && ".join(self.setup_commands)
-            self.logger.debug(f"Running setup commands: {setup_commands}")
+            self.logger.debug(f"{container} Running setup commands: {setup_commands}")
             status, output = container.exec_run(
                 ["/bin/bash", "-c", setup_commands],
                 user="root",  # Run as root to allow installations

scripts/config_aider.yaml

@@ -20,7 +20,7 @@ base:
         # session_commands define commands that are always executed before starting a shell session or running a single command in the terminal.
         # session_commands:["conda activate aider"],
         # setup_commands define commands that are executed only once when the terminal is created. This is only supported for Docker terminal.
-        setup_commands: ["pip install pytest"],
+        setup_commands: ["apt update", "apt install -y git", "pip install pytest"],
     }
 
     # LLM configs

scripts/config_mini_nightmare.yaml

@@ -20,7 +20,7 @@ base:
         # session_commands define commands that are always executed before starting a shell session or running a single command in the terminal.
         # session_commands:["conda activate aider"],
         # setup_commands define commands that are executed only once when the terminal is created. This is only supported for Docker terminal.
-        setup_commands: ["pip install pytest pandas"],
+        setup_commands: ["apt update", "apt install -y git", "pip install pytest pandas"],
     }
 
     # LLM configs

tests/gym/envs/test_env.py

@@ -204,21 +204,6 @@ def env(tmp_path):
     return env
 
 
-def test_restore(env):
-    # Change the content of a file
-    file1 = env.working_dir / "file1.txt"
-    with open(file1, "w") as f:
-        f.write("Hello, World!")
-
-    def hash_file(file):
-        with open(file, "rb") as f:
-            return hash(f.read())
-
-    assert hash_file(env.path / "file1.txt") != hash_file(file1)
-    env.restore()
-    assert hash_file(env.path / "file1.txt") == hash_file(file1)
-
-
 def test_display_files(env):
     result = env.display_files()
     assert result == (