made docker client local to handlers and lazy

loaded so that we don't have to start docker if we don't want it

src/utils/handlers.py

@@ -1,7 +1,7 @@
 from abc import ABC, abstractmethod
-import os, re, docker, signal, javalang
+import os, re, docker, signal, javalang, time
 from bs4 import BeautifulSoup
-from typing import Iterable, Tuple, Iterator
+from typing import Iterable, Optional, Tuple, Iterator
 import xml.etree.ElementTree as ET
 from javalang.tree import PackageDeclaration
 import tarfile
@@ -16,17 +16,18 @@ GROUP_ID = os.getgid()
 
 
 class BuildHandler(ABC):
+    DOCKER_CLIENT: Optional[docker.DockerClient] = None
+
     def __init__(self, repo_path: str, build_file: str, updates: dict) -> None:
         super().__init__()
         self.path: str = repo_path
         self.build_file: str = build_file
         self.updates = updates
 
-    def set_client(self, client: docker.DockerClient):
-        self.client = client
-
     def __enter__(self):
-        self.container = self.client.containers.run(
+        if BuildHandler.DOCKER_CLIENT is None:
+            BuildHandler.DOCKER_CLIENT = docker.from_env()
+        self.container = BuildHandler.DOCKER_CLIENT.containers.run(
             image=self.container_name(),
             command="tail -f /dev/null",  # to keep the container alive
             volumes={os.path.abspath(self.path): {"bind": "/repo", "mode": "rw"}},

src/utils/process_data.py

@@ -1,5 +1,4 @@
-import os
-import sys, docker
+import sys
 from utils.handlers import get_build_handler
 from .paths import get_project_path
 from sacrebleu import sentence_bleu as bleu
@@ -11,8 +10,6 @@ REFERENCE_MAP = Dataset.from_json(
 
 ARCHIVES_ROOT = str(get_project_path('../data/archives'))
 
-DOCKER_CLIENT = docker.from_env()
-
 
 def evaluate_comments(answers: dict[str, str], percent_cb):
     total = len(answers)
@@ -57,7 +54,6 @@ def evaluate_refinement(answers: dict[str, dict[str, str]], percent_cb):
             build_handler = get_build_handler(
                 ARCHIVES_ROOT, entry.metadata.archive_name(ArchiveState.MERGED)
             )
-            build_handler.set_client(DOCKER_CLIENT)
             current_progress += 1
             percent_cb(current_progress / total_number_of_steps * 100)
         except Exception as e: