tatlock/tests/responses/test_history.py

"""
Tests for conversation history management.
"""

import pytest
from fastapi.testclient import TestClient

from src.responses import service
from src.responses.context import ContextWindow
from src.responses.history import ConversationHistory
from src.responses.schemas import ResponseRequest


@pytest.mark.unit
@pytest.mark.asyncio
async def test_conversation_id_from_metadata():
    """Test conversation ID extraction from metadata."""
    history = ConversationHistory()

    request = ResponseRequest(
        model="lorem-tester",
        input=[{"role": "user", "content": "Hello"}],
        metadata={"conversation_id": "conv_123"},
    )

    conv_id = await history.get_conversation_id(request)
    assert conv_id == "conv_123"


@pytest.mark.unit
@pytest.mark.asyncio
async def test_conversation_id_generation():
    """Test conversation ID generation from first message."""
    history = ConversationHistory()

    request = ResponseRequest(
        model="lorem-tester",
        input=[{"role": "user", "content": "Hello"}],
        # No metadata provided
    )

    conv_id = await history.get_conversation_id(request)
    assert isinstance(conv_id, str)
    assert len(conv_id) == 16  # 16 character hex

    # Same first message should generate same ID
    request2 = ResponseRequest(model="lorem-tester", input=[{"role": "user", "content": "Hello"}])
    conv_id2 = await history.get_conversation_id(request2)
    assert conv_id == conv_id2


@pytest.mark.unit
def test_conversation_history_tracking(client: TestClient):
    """Test that conversations are tracked server-side."""
    # First request with conversation ID
    request_data = {
        "model": "lorem-tester",
        "input": [{"role": "user", "content": "Hello"}],
        "metadata": {"conversation_id": "test_conv_001"},
        "stream": False,
    }

    response = client.post("/v1/responses", json=request_data)
    assert response.status_code == 200

    # Second request with same conversation ID
    request_data2 = {
        "model": "lorem-tester",
        "input": [
            {"role": "user", "content": "Hello"},
            {"role": "assistant", "content": "Hi there!"},
            {"role": "user", "content": "How are you?"},
        ],
        "metadata": {"conversation_id": "test_conv_001"},
        "stream": False,
    }

    response2 = client.post("/v1/responses", json=request_data2)
    assert response2.status_code == 200

    # Both responses should be successful
    assert response.json()["status"] == "completed"
    assert response2.json()["status"] == "completed"


@pytest.mark.unit
@pytest.mark.asyncio
async def test_conversation_history_retrieval():
    """Test retrieving conversation history."""
    history = ConversationHistory()

    request = ResponseRequest(
        model="lorem-tester",
        input=[{"role": "user", "content": "Test"}],
        metadata={"conversation_id": "test_retrieve"},
    )

    conv_id = await history.get_conversation_id(request)

    # Initially empty
    retrieved = await history.get_history(conv_id)
    assert len(retrieved) == 0

    # Add mock response
    from src.responses.schemas import MessageOutputItem, OutputTextContent, Response, ResponseUsage

    mock_response = Response(
        id="resp_123",
        created_at=1234567890,
        model="lorem-tester",
        status="completed",
        output=[MessageOutputItem(id="msg_1", content=[OutputTextContent(text="Test response")])],
        usage=ResponseUsage(input_tokens=10, output_tokens=5, reasoning_tokens=0, total_tokens=15),
    )

    await history.add_response(conv_id, mock_response)

    # Should now have 1 response
    retrieved = await history.get_history(conv_id)
    assert len(retrieved) == 1
    assert retrieved[0].id == "resp_123"


@pytest.mark.unit
@pytest.mark.asyncio
async def test_conversation_history_trimming():
    """Test that history is trimmed to max_turns."""
    history = ConversationHistory(max_turns=3)

    from src.responses.schemas import MessageOutputItem, OutputTextContent, Response, ResponseUsage

    conv_id = "test_trim"

    # Add 5 responses (more than max_turns)
    for i in range(5):
        response = Response(
            id=f"resp_{i}",
            created_at=1234567890 + i,
            model="lorem-tester",
            status="completed",
            output=[
                MessageOutputItem(id=f"msg_{i}", content=[OutputTextContent(text=f"Response {i}")])
            ],
            usage=ResponseUsage(
                input_tokens=10, output_tokens=5, reasoning_tokens=0, total_tokens=15
            ),
        )
        await history.add_response(conv_id, response)

    # Should only keep last 3
    retrieved = await history.get_history(conv_id)
    assert len(retrieved) == 3
    assert retrieved[0].id == "resp_2"  # Oldest kept
    assert retrieved[1].id == "resp_3"
    assert retrieved[2].id == "resp_4"  # Most recent


@pytest.mark.unit
@pytest.mark.asyncio
async def test_clear_conversation():
    """Test clearing conversation history."""
    history = ConversationHistory()

    from src.responses.schemas import MessageOutputItem, OutputTextContent, Response, ResponseUsage

    conv_id = "test_clear"

    # Add a response
    response = Response(
        id="resp_123",
        created_at=1234567890,
        model="lorem-tester",
        status="completed",
        output=[MessageOutputItem(id="msg_1", content=[OutputTextContent(text="Test")])],
        usage=ResponseUsage(input_tokens=10, output_tokens=5, reasoning_tokens=0, total_tokens=15),
    )
    await history.add_response(conv_id, response)

    # Verify it exists
    assert len(await history.get_history(conv_id)) == 1

    # Clear it
    cleared = await history.clear_conversation(conv_id)
    assert cleared is True

    # Should be empty now
    assert len(await history.get_history(conv_id)) == 0

    # Clearing non-existent conversation should return False
    cleared_again = await history.clear_conversation(conv_id)
    assert cleared_again is False


@pytest.mark.unit
@pytest.mark.asyncio
async def test_conversation_count():
    """Test conversation count tracking."""
    history = ConversationHistory()

    initial_count = await history.get_conversation_count()

    # Add responses to 3 different conversations
    from src.responses.schemas import MessageOutputItem, OutputTextContent, Response, ResponseUsage

    for i in range(3):
        response = Response(
            id=f"resp_{i}",
            created_at=1234567890,
            model="lorem-tester",
            status="completed",
            output=[MessageOutputItem(id=f"msg_{i}", content=[OutputTextContent(text="Test")])],
            usage=ResponseUsage(
                input_tokens=10, output_tokens=5, reasoning_tokens=0, total_tokens=15
            ),
        )
        await history.add_response(f"conv_{i}", response)

    new_count = await history.get_conversation_count()
    assert new_count == initial_count + 3


@pytest.mark.unit
@pytest.mark.asyncio
async def test_context_window_token_counting():
    """Test token counting in context window."""
    context = ContextWindow(max_tokens=4096)

    # Test string counting
    text = "Hello world! " * 100  # ~1200 characters
    tokens = await context.count_tokens([text])
    assert tokens > 0
    assert tokens == len(text) // 4  # Approximate

    # Test dict counting
    message = {"role": "user", "content": "Test message"}
    tokens = await context.count_tokens([message])
    assert tokens > 0


@pytest.mark.unit
@pytest.mark.asyncio
async def test_context_window_trimming():
    """Test context window trimming to fit."""
    context = ContextWindow(max_tokens=100)

    # Create items that exceed limit
    items = [
        "This is a long message " * 20,  # ~480 chars = ~120 tokens
        "Another message " * 10,  # ~160 chars = ~40 tokens
        "Short message",  # ~13 chars = ~3 tokens
    ]

    # Trim with 10 token reserve
    trimmed = await context.trim_to_fit(items, reserve_tokens=10)

    # Should keep only items that fit (90 tokens available)
    # Most recent first: "Short message" (3 tokens) + "Another message..." (40 tokens) = 43 tokens
    assert len(trimmed) >= 1  # At least the short message
    assert "Short message" in trimmed  # Most recent always kept if it fits


@pytest.mark.unit
@pytest.mark.asyncio
async def test_context_window_fits_check():
    """Test checking if items fit in context."""
    context = ContextWindow(max_tokens=100)

    small_items = ["Hello", "World"]
    large_items = ["Very long text " * 200]  # ~3200 chars = ~800 tokens

    # Small items should fit
    assert await context.fits_in_context(small_items, reserve_tokens=10) is True

    # Large items should not fit
    assert await context.fits_in_context(large_items, reserve_tokens=10) is False


@pytest.mark.unit
@pytest.mark.asyncio
async def test_context_window_usage_stats():
    """Test context window usage statistics."""
    context = ContextWindow(max_tokens=1000)

    items = ["Test message " * 50]  # ~650 chars = ~162 tokens

    stats = await context.get_usage_stats(items, reserve_tokens=100)

    assert "total_tokens" in stats
    assert "max_tokens" in stats
    assert "reserved_tokens" in stats
    assert "available_tokens" in stats
    assert "usage_percent" in stats
    assert "fits" in stats

    assert stats["max_tokens"] == 1000
    assert stats["reserved_tokens"] == 100
    assert stats["available_tokens"] == 900
    assert isinstance(stats["usage_percent"], (int, float))


@pytest.mark.unit
@pytest.mark.asyncio
async def test_service_conversation_helpers():
    """Test service helper functions for conversation history."""
    # Get stats
    stats = await service.get_conversation_stats()
    assert "total_conversations" in stats
    assert "max_turns_per_conversation" in stats

    # Get context window
    context = service.get_context_window()
    assert context.max_tokens == 4096

    # Test clearing (should handle non-existent gracefully)
    result = await service.clear_conversation("non_existent_conv")
    assert result is False


@pytest.mark.unit
@pytest.mark.asyncio
async def test_streaming_tracks_history(async_client):
    """Test that streaming responses also track conversation history."""

    request_data = {
        "model": "lorem-tester",
        "input": [{"role": "user", "content": "Hello"}],
        "metadata": {"conversation_id": "stream_test_001"},
        "stream": True,
    }

    async with async_client.stream(
        "POST",
        "/v1/responses",
        json=request_data,
        timeout=20.0,
    ) as response:
        assert response.status_code == 200

        # Consume the stream
        async for line in response.aiter_lines():
            if line.startswith("data: ") and "[DONE]" not in line:
                continue

    # History should be tracked
    # (We can't easily verify this without exposing a GET endpoint,
    # but the integration is tested in the non-streaming tests)