Extend tests and add github actions

.github/workflows/pytest.yaml

@@ -0,0 +1,51 @@
+
+name: Pytest
+
+env:
+  GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+
+on:
+  push:
+    branches:
+      - main
+    paths-ignore:
+      - 'docs/**'
+      - 'README.md'
+      - 'mkdocs.yml'
+  pull_request:
+    branches:
+      - main
+    paths-ignore:
+      - 'docs/**'
+      - 'README.md'
+      - 'mkdocs.yml'
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        shell: bash -l {0}
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v2
+      - uses: actions/setup-python@v5
+        with:
+          python-version: '3.11'
+      - name: Install uv
+        run: |
+          curl -LsSf https://astral.sh/uv/install.sh | sh
+      - name: Install dependencies
+        run: |
+            uv pip install --python ${Python_ROOT_DIR} '.[dev]'
+      - name: Run pytest
+        uses: sjvrijn/pytest-last-failed@v2
+        with:
+          pytest-args: '--cov'
+      # - name: Explicitly convert coverage to xml
+      #   run: coverage xml
+      # - name: Upload coverage reports to Codecov
+      #   uses: codecov/[email protected]
+      #   with:
+      #     token: ${{ secrets.CODECOV_TOKEN }}
+      #     slug: princeton-nlp/SWE-agent

pyproject.toml

@@ -27,6 +27,13 @@ classifiers = [
   "Programming Language :: Python :: 3 :: Only",
 ]
 
+dependencies = [
+    "fastapi",
+    "uvicorn",
+    "requests",
+    "pydantic",
+]
+
 [project.optional-dependencies]
 dev = [
     # "mkdocs-material",
@@ -212,4 +219,4 @@ ACI = "ACI"
 
 [tool.typos.default.extend-words]
 # Don't correct the surname "Teh"
-aci = "aci"
+aci = "aci"

src/swebridge/local.py

@@ -111,9 +111,6 @@ def write_file(self, request: WriteFileRequest) -> WriteFileResponse:
 
 if __name__ == "__main__":
     runtime = RemoteRuntime("localhost:8000")
-    print(runtime.read_file(ReadFileRequest(path="README.md")))
-    print(runtime.write_file(WriteFileRequest(path="_test.txt", content="test")))
-    print(runtime.read_file(ReadFileRequest(path="_test.txt")))
     # ----
     # print(runtime.execute(Command(command="ls", shell=True)))
     # ----

src/swebridge/models.py

@@ -2,7 +2,7 @@
 
 
 class CreateShellRequest(BaseModel):
-    name: str = "default"
+    session: str = "default"
 
 
 class CreateShellResponse(BaseModel):

src/swebridge/runtime.py

@@ -105,20 +105,22 @@ def __init__(self):
         self.sessions: dict[str, Session] = {}
 
     async def create_shell(self, request: CreateShellRequest) -> CreateShellResponse:
-        if request.name in self.sessions:
-            return CreateShellResponse(success=False, failure_reason="session already exists")
+        if request.session in self.sessions:
+            return CreateShellResponse(success=False, failure_reason=f"session {request.session} already exists")
         shell = Session()
-        self.sessions[request.name] = shell
+        self.sessions[request.session] = shell
         return await shell.start()
 
     async def run_in_shell(self, action: Action) -> Observation:
         if action.session not in self.sessions:
-            return Observation(output="", exit_code_raw="-312", failure_reason="session does not exist")
+            return Observation(
+                output="", exit_code_raw="-312", failure_reason=f"session {action.session!r} does not exist"
+            )
         return await self.sessions[action.session].run(action)
 
     async def close_shell(self, request: CloseRequest) -> CloseResponse:
         if request.session not in self.sessions:
-            return CloseResponse(success=False, failure_reason="session does not exist")
+            return CloseResponse(success=False, failure_reason=f"session {request.session!r} does not exist")
         out = await self.sessions[request.session].close()
         del self.sessions[request.session]
         return out
@@ -132,7 +134,9 @@ async def execute(self, command: Command) -> CommandResponse:
                 exit_code=result.returncode,
             )
         except subprocess.TimeoutExpired:
-            return CommandResponse(stdout="", stderr="", exit_code=-1)
+            return CommandResponse(
+                stdout="", stderr=f"Timeout ({command.timeout}s) exceeded while running command", exit_code=-1
+            )
         except Exception as e:
             return CommandResponse(stdout="", stderr=str(e), exit_code=-2)
 

tests/conftest.py

@@ -35,8 +35,6 @@ def run_server():
     time.sleep(0.1)
 
     return RemoteServer(port)
-    # The thread will be automatically terminated when the test session ends
-    # because it's a daemon thread
 
 
 @pytest.fixture

tests/test_dress_rehearsal.py

@@ -1,5 +1,107 @@
+from pathlib import Path
+
 from swebridge.local import RemoteRuntime
+from swebridge.models import Action, CloseRequest, Command, CreateShellRequest, ReadFileRequest, WriteFileRequest
 
 
 def test_server_alive(remote_runtime: RemoteRuntime):
     assert remote_runtime.is_alive()
+
+
+def test_server_dead():
+    r = RemoteRuntime("http://doesnotexistadsfasdfasdf234123qw34.com")
+    assert not r.is_alive()
+
+
+def test_read_write_file(remote_runtime: RemoteRuntime, tmp_path: Path):
+    path = tmp_path / "test.txt"
+    remote_runtime.write_file(WriteFileRequest(path=str(path), content="test"))
+    assert path.read_text() == "test"
+    assert remote_runtime.read_file(ReadFileRequest(path=str(path))).content == "test"
+
+
+def test_read_non_existent_file(remote_runtime: RemoteRuntime):
+    assert not remote_runtime.read_file(ReadFileRequest(path="non_existent.txt")).success
+
+
+def test_execute_command(remote_runtime: RemoteRuntime):
+    assert remote_runtime.execute(Command(command="echo 'hello world'", shell=True)).stdout == "hello world\n"
+
+
+def test_execute_command_shell_false(remote_runtime: RemoteRuntime):
+    assert remote_runtime.execute(Command(command=["echo", "hello world"], shell=False)).stdout == "hello world\n"
+
+
+def test_execute_command_timeout(remote_runtime: RemoteRuntime):
+    r = remote_runtime.execute(Command(command=["sleep", "10"], timeout=0.1))
+    assert not r.success
+    assert "timeout" in r.stderr.lower()
+    assert not r.stdout
+
+
+def test_create_close_shell(remote_runtime: RemoteRuntime):
+    r = remote_runtime.create_shell(CreateShellRequest())
+    assert r.success
+    r = remote_runtime.close_shell(CloseRequest())
+    assert r.success
+
+
+def test_run_in_shell(remote_runtime: RemoteRuntime):
+    name = "test_run_in_shell"
+    r = remote_runtime.create_shell(CreateShellRequest(session=name))
+    assert r.success
+    r = remote_runtime.run_in_shell(Action(command="echo 'hello world'", session=name))
+    assert r.success
+    r = remote_runtime.run_in_shell(Action(command="doesntexit", session=name))
+    assert not r.success
+    r = remote_runtime.close_shell(CloseRequest(session=name))
+    assert r.success
+
+
+def test_run_in_shell_non_existent_session(remote_runtime: RemoteRuntime):
+    r = remote_runtime.run_in_shell(Action(command="echo 'hello world'", session="non_existent"))
+    assert not r.success
+    assert "does not exist" in r.failure_reason
+
+
+def test_close_shell_non_existent_session(remote_runtime: RemoteRuntime):
+    r = remote_runtime.close_shell(CloseRequest(session="non_existent"))
+    assert not r.success
+    assert "does not exist" in r.failure_reason
+
+
+def test_close_shell_twice(remote_runtime: RemoteRuntime):
+    r = remote_runtime.create_shell(CreateShellRequest())
+    assert r.success
+    r = remote_runtime.close_shell(CloseRequest())
+    assert r.success
+    r = remote_runtime.close_shell(CloseRequest())
+    assert not r.success
+    assert "does not exist" in r.failure_reason
+
+
+def test_run_in_shell_timeout(remote_runtime: RemoteRuntime):
+    print("in test")
+    r = remote_runtime.create_shell(CreateShellRequest())
+    assert r.success
+    r = remote_runtime.run_in_shell(Action(command="sleep 10", timeout=0.1))
+    assert not r.success
+    assert "timeout" in r.failure_reason
+    assert not r.output
+    r = remote_runtime.close_shell(CloseRequest())
+    assert r.success
+
+
+def test_run_in_shell_interactive_command(remote_runtime: RemoteRuntime):
+    r = remote_runtime.create_shell(CreateShellRequest())
+    assert r.success
+    r = remote_runtime.run_in_shell(Action(command="python", is_interactive_command=True, expect=[">>> "]))
+    assert r.success
+    r = remote_runtime.run_in_shell(
+        Action(command="print('hello world')", is_interactive_command=True, expect=[">>> "])
+    )
+    assert r.success
+    r = remote_runtime.run_in_shell(Action(command="quit()\n", is_interactive_quit=True))
+    assert r.success
+    r = remote_runtime.close_shell(CloseRequest())
+    assert r.success