Merge pull request princeton-nlp#186 from yuxqiu/typings

fix: correct some typings and a incorrect function call

swebench/harness/docker_build.py

@@ -239,7 +239,7 @@ def get_env_configs_to_build(
 
             if env_image.attrs["Created"] < base_image.attrs["Created"]:
                 # Remove the environment image if it was built after the base_image
-                for dep in find_dependent_images(env_image):
+                for dep in find_dependent_images(client, test_spec.env_image_key):
                     # Remove instance images that depend on this environment image
                     remove_image(client, dep.image_id, "quiet")
                 remove_image(client, test_spec.env_image_key, "quiet")
@@ -339,7 +339,7 @@ def build_instance_images(
     ):
     """
     Builds the instance images required for the dataset if they do not already exist.
-    
+
     Args:
         dataset (list): List of test specs or dataset to build images for
         client (docker.DockerClient): Docker client to use for building the images

swebench/harness/test_spec.py

@@ -4,7 +4,7 @@
 import re
 
 from dataclasses import dataclass
-from typing import Any, Union
+from typing import Any, Union, cast
 
 from swebench.harness.constants import (
     SWEbenchInstance,
@@ -37,9 +37,9 @@ class TestSpec:
     instance_id: str
     repo: str
     version: str
-    repo_script_list: str
-    eval_script_list: str
-    env_script_list: str
+    repo_script_list: list[str]
+    eval_script_list: list[str]
+    env_script_list: list[str]
     arch: str
     FAIL_TO_PASS: list[str]
     PASS_TO_PASS: list[str]
@@ -104,15 +104,15 @@ def platform(self):
             return "linux/arm64/v8"
         else:
             raise ValueError(f"Invalid architecture: {self.arch}")
-        
+
 
 def get_test_specs_from_dataset(dataset: Union[list[SWEbenchInstance], list[TestSpec]]) -> list[TestSpec]:
     """
     Idempotent function that converts a list of SWEbenchInstance objects to a list of TestSpec objects.
     """
     if isinstance(dataset[0], TestSpec):
-        return dataset
-    return list(map(make_test_spec, dataset))
+        return cast(list[TestSpec], dataset)
+    return list(map(make_test_spec, cast(list[SWEbenchInstance], dataset)))
 
 
 def make_repo_script_list(specs, repo, repo_directory, base_commit, env_name):