168 lines
7.2 KiB
Python
168 lines
7.2 KiB
Python
import unittest
|
|
from unittest.mock import patch, MagicMock, ANY
|
|
import json
|
|
import subprocess
|
|
import io
|
|
import sys
|
|
import os
|
|
|
|
# Ensure the project root is in sys.path to resolve imports correctly
|
|
project_root = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
|
|
if project_root not in sys.path:
|
|
sys.path.append(project_root)
|
|
# Import the class to be tested
|
|
from gemini_cli_adapter import GeminiCliAdapter
|
|
|
|
class TestGeminiCliAdapterParity(unittest.TestCase):
|
|
|
|
def setUp(self) -> None:
|
|
"""Set up a fresh adapter instance and reset session state for each test."""
|
|
# Patch session_logger to prevent file operations during tests
|
|
self.session_logger_patcher = patch('gemini_cli_adapter.session_logger')
|
|
self.mock_session_logger = self.session_logger_patcher.start()
|
|
self.adapter = GeminiCliAdapter(binary_path="gemini")
|
|
self.adapter.session_id = None
|
|
self.adapter.last_usage = None
|
|
self.adapter.last_latency = 0.0
|
|
|
|
def tearDown(self) -> None:
|
|
self.session_logger_patcher.stop()
|
|
|
|
@patch('subprocess.Popen')
|
|
def test_count_tokens_uses_estimation(self, mock_popen: MagicMock) -> None:
|
|
"""
|
|
Test that count_tokens uses character-based estimation.
|
|
"""
|
|
contents_to_count = ["This is the first line.", "This is the second line."]
|
|
expected_chars = len("\n".join(contents_to_count))
|
|
expected_tokens = expected_chars // 4
|
|
token_count = self.adapter.count_tokens(contents=contents_to_count)
|
|
self.assertEqual(token_count, expected_tokens)
|
|
# Verify that NO subprocess was started for counting
|
|
mock_popen.assert_not_called()
|
|
|
|
@patch('subprocess.Popen')
|
|
def test_send_with_safety_settings_no_flags_added(self, mock_popen: MagicMock) -> None:
|
|
"""
|
|
Test that the send method does NOT add --safety flags when safety_settings are provided,
|
|
as this functionality is no longer supported via CLI flags.
|
|
"""
|
|
process_mock = MagicMock()
|
|
mock_stdout_content = [json.dumps({"type": "result", "usage": {}}) + "\n", ""]
|
|
process_mock.stdout.readline.side_effect = mock_stdout_content
|
|
process_mock.stderr.read.return_value = ""
|
|
process_mock.poll.return_value = 0
|
|
mock_popen.return_value = process_mock
|
|
message_content = "User's prompt here."
|
|
safety_settings = [
|
|
{"category": "HARM_CATEGORY_HARASSMENT", "threshold": "BLOCK_ONLY_HIGH"},
|
|
{"category": "HARM_CATEGORY_SEXUALLY_EXPLICIT", "threshold": "BLOCK_MEDIUM_AND_ABOVE"}
|
|
]
|
|
self.adapter.send(message=message_content, safety_settings=safety_settings)
|
|
args, kwargs = mock_popen.call_args
|
|
command = args[0]
|
|
# Verify that no --safety flags were added to the command
|
|
self.assertNotIn("--safety", command)
|
|
# Verify that the message was passed correctly via stdin
|
|
# We might need to wait a tiny bit for the thread, or just check if it was called
|
|
# In most cases it will be called by the time send() returns because of wait()
|
|
process_mock.stdin.write.assert_called_with(message_content)
|
|
|
|
@patch('subprocess.Popen')
|
|
def test_send_without_safety_settings_no_flags(self, mock_popen: MagicMock) -> None:
|
|
"""
|
|
Test that when safety_settings is None or an empty list, no --safety flags are added.
|
|
"""
|
|
process_mock = MagicMock()
|
|
mock_stdout_content = [json.dumps({"type": "result", "usage": {}}) + "\n", ""]
|
|
process_mock.stdout.readline.side_effect = mock_stdout_content
|
|
process_mock.stderr.read.return_value = ""
|
|
process_mock.poll.return_value = 0
|
|
mock_popen.return_value = process_mock
|
|
message_content = "Another prompt."
|
|
self.adapter.send(message=message_content, safety_settings=None)
|
|
args_none, _ = mock_popen.call_args
|
|
self.assertNotIn("--safety", args_none[0])
|
|
mock_popen.reset_mock()
|
|
|
|
# Reset side effects for the second call
|
|
process_mock.stdout.readline.side_effect = [json.dumps({"type": "result", "usage": {}}) + "\n", ""]
|
|
self.adapter.send(message=message_content, safety_settings=[])
|
|
args_empty, _ = mock_popen.call_args
|
|
self.assertNotIn("--safety", args_empty[0])
|
|
|
|
@patch('subprocess.Popen')
|
|
def test_send_with_system_instruction_prepended_to_stdin(self, mock_popen: MagicMock) -> None:
|
|
"""
|
|
Test that the send method prepends the system instruction to the prompt
|
|
sent via stdin, and does NOT add a --system flag to the command.
|
|
"""
|
|
process_mock = MagicMock()
|
|
mock_stdout_content = [json.dumps({"type": "result", "usage": {}}) + "\n", ""]
|
|
process_mock.stdout.readline.side_effect = mock_stdout_content
|
|
process_mock.stderr.read.return_value = ""
|
|
process_mock.poll.return_value = 0
|
|
mock_popen.return_value = process_mock
|
|
message_content = "User's prompt here."
|
|
system_instruction_text = "Some instruction"
|
|
expected_input = f"{system_instruction_text}\n\n{message_content}"
|
|
self.adapter.send(message=message_content, system_instruction=system_instruction_text)
|
|
args, kwargs = mock_popen.call_args
|
|
command = args[0]
|
|
# Verify that the system instruction was prepended to the input sent to write
|
|
process_mock.stdin.write.assert_called_with(expected_input)
|
|
# Verify that no --system flag was added to the command
|
|
self.assertNotIn("--system", command)
|
|
|
|
@patch('subprocess.Popen')
|
|
def test_send_with_model_parameter(self, mock_popen: MagicMock) -> None:
|
|
"""
|
|
Test that the send method correctly adds the -m <model> flag when a model is specified.
|
|
"""
|
|
process_mock = MagicMock()
|
|
mock_stdout_content = [json.dumps({"type": "result", "usage": {}}) + "\n", ""]
|
|
process_mock.stdout.readline.side_effect = mock_stdout_content
|
|
process_mock.stderr.read.return_value = ""
|
|
process_mock.poll.return_value = 0
|
|
mock_popen.return_value = process_mock
|
|
message_content = "User's prompt here."
|
|
model_name = "gemini-1.5-flash"
|
|
expected_command_part = f'-m "{model_name}"'
|
|
self.adapter.send(message=message_content, model=model_name)
|
|
args, kwargs = mock_popen.call_args
|
|
command = args[0]
|
|
# Verify that the -m <model> flag was added to the command
|
|
self.assertIn(expected_command_part, command)
|
|
# Verify that the message was passed correctly via stdin
|
|
process_mock.stdin.write.assert_called_with(message_content)
|
|
|
|
@patch('subprocess.Popen')
|
|
def test_send_parses_tool_calls_from_streaming_json(self, mock_popen: MagicMock) -> None:
|
|
"""
|
|
Test that tool_use messages in the streaming JSON are correctly parsed.
|
|
"""
|
|
process_mock = MagicMock()
|
|
mock_stdout_content = [
|
|
json.dumps({"type": "init", "session_id": "session-123"}) + "\n",
|
|
json.dumps({"type": "chunk", "text": "I will call a tool. "}) + "\n",
|
|
json.dumps({"type": "tool_use", "name": "get_weather", "args": {"location": "London"}, "id": "call-456"}) + "\n",
|
|
json.dumps({"type": "result", "usage": {"total_tokens": 100}}) + "\n",
|
|
""
|
|
]
|
|
process_mock.stdout.readline.side_effect = mock_stdout_content
|
|
process_mock.stderr.read.return_value = ""
|
|
process_mock.poll.return_value = 0
|
|
mock_popen.return_value = process_mock
|
|
|
|
result = self.adapter.send(message="What is the weather?")
|
|
|
|
self.assertEqual(result["text"], "I will call a tool. ")
|
|
self.assertEqual(len(result["tool_calls"]), 1)
|
|
self.assertEqual(result["tool_calls"][0]["name"], "get_weather")
|
|
self.assertEqual(result["tool_calls"][0]["args"], {"location": "London"})
|
|
self.assertEqual(self.adapter.session_id, "session-123")
|
|
self.assertEqual(self.adapter.last_usage, {"total_tokens": 100})
|
|
|
|
if __name__ == '__main__':
|
|
unittest.main()
|