auto-commit-service/tests/test_multi_model_client.py

"""Tests for MultiModelLlamaClient — retry logic and pipeline methods."""

import asyncio
from unittest.mock import AsyncMock, MagicMock, call, patch

import pytest

from auto_commit_service.llm.multi_model_client import (
    MultiModelLlamaClient,
    _TRANSIENT_EXCEPTIONS,
    _transient_exc_types,
)


class TestTransientExcTypes:
    """_transient_exc_types builds the correct tuple of retryable exceptions."""

    def test_always_includes_builtins(self) -> None:
        types = _transient_exc_types()
        assert ConnectionError in types
        assert TimeoutError in types

    def test_no_asyncio_timeout_error_duplicate(self) -> None:
        # asyncio.TimeoutError is TimeoutError on Python 3.11+ — must not be
        # included separately (it would be an identical duplicate).
        types = _transient_exc_types()
        assert asyncio.TimeoutError not in types or asyncio.TimeoutError is TimeoutError

    def test_returns_tuple(self) -> None:
        assert isinstance(_transient_exc_types(), tuple)

    def test_module_constant_matches_function(self) -> None:
        assert set(_TRANSIENT_EXCEPTIONS) == set(_transient_exc_types())

    def test_httpx_included_when_available(self) -> None:
        try:
            import httpx
            assert httpx.RemoteProtocolError in _TRANSIENT_EXCEPTIONS
            assert httpx.ReadError in _TRANSIENT_EXCEPTIONS
            assert httpx.ConnectError in _TRANSIENT_EXCEPTIONS
        except ImportError:
            pytest.skip("httpx not installed")


class TestRetryHelper:
    """_retry retries transient transport errors with exponential backoff."""

    @pytest.fixture
    def client(self) -> MultiModelLlamaClient:
        with patch("auto_commit_service.llm.multi_model_client.InferenceClient"):
            return MultiModelLlamaClient(retry_attempts=2)

    @pytest.mark.asyncio
    async def test_success_on_first_attempt(self, client: MultiModelLlamaClient) -> None:
        coro_fn = AsyncMock(return_value="result")
        result = await client._retry("test", coro_fn)
        assert result == "result"
        coro_fn.assert_awaited_once()

    @pytest.mark.asyncio
    async def test_retries_on_transient_error(self, client: MultiModelLlamaClient) -> None:
        exc = ConnectionError("disconnected")
        coro_fn = AsyncMock(side_effect=[exc, "result"])

        with patch("asyncio.sleep", new_callable=AsyncMock) as mock_sleep:
            result = await client._retry("test", coro_fn)

        assert result == "result"
        assert coro_fn.await_count == 2
        mock_sleep.assert_awaited_once_with(1)  # 2**0 = 1s backoff

    @pytest.mark.asyncio
    async def test_exponential_backoff_sequence(self, client: MultiModelLlamaClient) -> None:
        exc = ConnectionError("disconnected")
        # 3 total attempts (retry_attempts=2): fail, fail, succeed
        coro_fn = AsyncMock(side_effect=[exc, exc, "result"])

        with patch("asyncio.sleep", new_callable=AsyncMock) as mock_sleep:
            result = await client._retry("test", coro_fn)

        assert result == "result"
        assert mock_sleep.await_args_list == [call(1), call(2)]  # 2**0, 2**1

    @pytest.mark.asyncio
    async def test_raises_after_all_attempts_exhausted(self, client: MultiModelLlamaClient) -> None:
        exc = ConnectionError("disconnected")
        coro_fn = AsyncMock(side_effect=exc)

        with patch("asyncio.sleep", new_callable=AsyncMock):
            with pytest.raises(ConnectionError):
                await client._retry("test", coro_fn)

        assert coro_fn.await_count == 3  # 1 initial + 2 retries

    @pytest.mark.asyncio
    async def test_does_not_retry_non_transient_errors(self, client: MultiModelLlamaClient) -> None:
        exc = ValueError("bad input")
        coro_fn = AsyncMock(side_effect=exc)

        with pytest.raises(ValueError):
            await client._retry("test", coro_fn)

        coro_fn.assert_awaited_once()

    @pytest.mark.asyncio
    async def test_zero_retries_no_sleep(self, client: MultiModelLlamaClient) -> None:
        with patch("auto_commit_service.llm.multi_model_client.InferenceClient"):
            no_retry_client = MultiModelLlamaClient(retry_attempts=0)

        exc = ConnectionError("disconnected")
        coro_fn = AsyncMock(side_effect=exc)

        with patch("asyncio.sleep", new_callable=AsyncMock) as mock_sleep:
            with pytest.raises(ConnectionError):
                await no_retry_client._retry("test", coro_fn)

        mock_sleep.assert_not_awaited()
        coro_fn.assert_awaited_once()


class TestAnalyzeCommitViaPipeline:
    """analyze_commit_via_pipeline: fill_step is called once; only wait_step is retried."""

    @pytest.fixture
    def client(self) -> MultiModelLlamaClient:
        with patch("auto_commit_service.llm.multi_model_client.InferenceClient"):
            c = MultiModelLlamaClient(retry_attempts=2)
            c._client = AsyncMock()
            return c

    @pytest.fixture
    def handle(self) -> MagicMock:
        h = MagicMock()
        h.pipeline_id = "pipe-123"
        h.wait_step = AsyncMock(return_value={
            "result": {"choices": [{"message": {"content": "feat(auth): ✨ Add login"}}]}
        })
        return h

    @pytest.mark.asyncio
    async def test_fill_step_called_exactly_once_on_success(
        self, client: MultiModelLlamaClient, handle: MagicMock
    ) -> None:
        await client.analyze_commit_via_pipeline(handle, "prompt", "system")
        client._client.fill_step.assert_awaited_once()

    @pytest.mark.asyncio
    async def test_wait_step_retried_on_transient_error(
        self, client: MultiModelLlamaClient, handle: MagicMock
    ) -> None:
        """fill_step must NOT be called again when wait_step fails transiently."""
        exc = ConnectionError("read error")
        good = {"result": {"choices": [{"message": {"content": "feat: ✨ x"}}]}}
        handle.wait_step = AsyncMock(side_effect=[exc, good])

        with patch("asyncio.sleep", new_callable=AsyncMock):
            result = await client.analyze_commit_via_pipeline(handle, "prompt", "system")

        # fill_step called exactly once — not retried
        client._client.fill_step.assert_awaited_once()
        # wait_step called twice (initial + 1 retry)
        assert handle.wait_step.await_count == 2
        assert result == "feat: ✨ x"

    @pytest.mark.asyncio
    async def test_fill_step_failure_propagates_without_retry(
        self, client: MultiModelLlamaClient, handle: MagicMock
    ) -> None:
        """fill_step is outside the retry boundary — errors propagate immediately."""
        client._client.fill_step = AsyncMock(side_effect=ConnectionError("fill failed"))

        with pytest.raises(ConnectionError):
            await client.analyze_commit_via_pipeline(handle, "prompt", "system")

        # fill_step called once, wait_step never reached
        client._client.fill_step.assert_awaited_once()
        handle.wait_step.assert_not_awaited()


class TestBuildLlmClientFactory:
    """_build_llm_client in app.py produces a correctly configured client."""

    def test_factory_passes_all_settings(self) -> None:
        from auto_commit_service.app import _build_llm_client
        from auto_commit_service.config import AutoCommitSettings

        settings = AutoCommitSettings(
            service_name="test",
            reasoning_model_id="ministral-14b-reasoning",
            instruct_model_id="ministral-3b-instruct",
            llm_timeout=45.0,
            llm_retry_attempts=3,
        )

        with patch("auto_commit_service.llm.multi_model_client.InferenceClient"):
            client = _build_llm_client(settings)

        assert client._reasoning_model_id == "ministral-14b-reasoning"
        assert client._instruct_model_id == "ministral-3b-instruct"
        assert client._timeout == 45.0
        assert client._retry_attempts == 3