tests/test_llm.py

import pytest

from tutor.config import Settings
from tutor.services.llm.base import ChatMessage, LLMClient
from tutor.services.llm.factory import GEMINI_OPENAI_BASE_URL, create_llm_client
from tutor.services.llm.fake import FakeLLMClient
from tutor.services.llm.openai_compat import OpenAICompatClient


def _settings(**overrides: object) -> Settings:
    return Settings(_env_file=None, **overrides)  # type: ignore[arg-type]


async def test_fake_client_is_deterministic_and_records_calls() -> None:
    client = FakeLLMClient(responses=["first", "second"])
    messages = [ChatMessage(role="user", content="hello")]

    first = await client.complete(messages)
    second = await client.complete(messages)
    third = await client.complete(messages)  # queue exhausted -> default

    assert (first.text, second.text) == ("first", "second")
    assert third.text == "(fake) pong"
    assert len(client.calls) == 3
    assert client.calls[0][0].content == "hello"


def test_fake_satisfies_protocol() -> None:
    assert isinstance(FakeLLMClient(), LLMClient)


def test_factory_returns_fake_by_default() -> None:
    client = create_llm_client(_settings())
    assert isinstance(client, FakeLLMClient)


def test_factory_requires_api_key_for_remote_providers() -> None:
    with pytest.raises(ValueError, match="LLM_API_KEY"):
        create_llm_client(_settings(llm_provider="gemini"))


def test_factory_builds_openai_compat_client_with_gemini_preset() -> None:
    settings = _settings(llm_provider="gemini", llm_api_key="sk-test", llm_model="gemini-2.5-flash")
    client = create_llm_client(settings)
    assert isinstance(client, OpenAICompatClient)
    assert client.model == "gemini-2.5-flash"
    assert str(client._client.base_url) == GEMINI_OPENAI_BASE_URL


def test_factory_allows_keyless_ollama() -> None:
    client = create_llm_client(_settings(llm_provider="ollama", llm_model="llama3"))
    assert isinstance(client, OpenAICompatClient)