AI/ontology_platform/tests/unit/test_select_ontology.py

"""Regression tests for the OntoCast `select_ontology` agent.

Covers the None-selection index fix described in
`docs/통합설계서.md` §5 Phase 0 and OntoCast 분석 §13.1.

The original code checked ``answer_index == 0`` for "None", but the Pydantic
dynamic model constrains ``answer_index`` to ``[1, num_ontologies + 1]``.
The fix maps ``num_ontologies + 1`` to ``NULL_ONTOLOGY`` and removes the
unreachable ``answer_index == 0`` branch.
"""

from __future__ import annotations

import logging
import sys
from pathlib import Path
from types import SimpleNamespace
from unittest.mock import AsyncMock, MagicMock

import pytest

# Make the vendored OntoCast importable. The repo layout is:
#   ontology_platform/
#     vendored/ontocast/ontocast/...
# We point at `vendored/ontocast` so `import ontocast.<x>` resolves.
REPO_ROOT = Path(__file__).resolve().parents[2]
VENDORED_ONTOCAST = REPO_ROOT / "vendored" / "ontocast"
if str(VENDORED_ONTOCAST) not in sys.path:
    sys.path.insert(0, str(VENDORED_ONTOCAST))

# Imports must come after sys.path tweak.
from ontocast.agent import select_ontology as select_ontology_module  # noqa: E402
from ontocast.onto.enum import Status  # noqa: E402
from ontocast.onto.null import NULL_ONTOLOGY  # noqa: E402


def _make_state(num_content_units: int = 1) -> SimpleNamespace:
    """Build a minimal stand-in for `AgentState` used by `select_ontology`.

    Only attributes accessed inside the function are populated. Using
    `SimpleNamespace` avoids constructing the full Pydantic model and all
    its nested defaults, which would couple the test to OntoCast internals
    far beyond the scope of this regression.
    """
    units = [SimpleNamespace(text=f"chunk {i} text") for i in range(num_content_units)]
    return SimpleNamespace(
        current_ontology=NULL_ONTOLOGY,
        content_units=units,
        current_content_unit=units[0] if units else None,
        input_text="",
        status=None,
        # `get_content_unit_progress_string` is only used for logging
        get_content_unit_progress_string=lambda: f"{num_content_units} units",
    )


def _make_tools(ontologies: list) -> SimpleNamespace:
    om = MagicMock()
    om.has_ontologies = bool(ontologies)
    om.ontologies = ontologies
    return SimpleNamespace(llm=MagicMock(), ontology_manager=om)


def _patch_llm_call(monkeypatch: pytest.MonkeyPatch, answer_index: int) -> None:
    """Patch `call_llm_with_retry` to return a stub selector with the given index."""
    selector_stub = SimpleNamespace(answer_index=answer_index)
    monkeypatch.setattr(
        select_ontology_module,
        "call_llm_with_retry",
        AsyncMock(return_value=selector_stub),
    )


# ─── Case 1: no ontologies available ──────────────────────────────────────
@pytest.mark.asyncio
async def test_no_ontologies_returns_null(monkeypatch: pytest.MonkeyPatch) -> None:
    """When no ontologies are registered, the LLM is not called and the
    current ontology stays NULL."""
    state = _make_state()
    tools = _make_tools(ontologies=[])

    # Guard: ensure LLM is never invoked when there is nothing to pick from.
    sentinel = AsyncMock()
    monkeypatch.setattr(select_ontology_module, "call_llm_with_retry", sentinel)

    result = await select_ontology_module.select_ontology(state, tools)

    assert result.current_ontology is NULL_ONTOLOGY
    sentinel.assert_not_called()


# ─── Case 2: LLM picks a valid index ──────────────────────────────────────
@pytest.mark.asyncio
async def test_llm_picks_valid_index(monkeypatch: pytest.MonkeyPatch) -> None:
    """When the LLM returns an index within [1, num_ontologies], the
    corresponding ontology is selected (0-based after subtraction)."""
    onto_a = MagicMock(name="onto_a", ontology_id="onto-a", iri="urn:a")
    onto_a.is_null = MagicMock(return_value=False)
    onto_a.initial_version = None
    onto_a.version = "1.0"
    onto_a.describe = MagicMock(return_value="ontology A description")

    onto_b = MagicMock(name="onto_b", ontology_id="onto-b", iri="urn:b")
    onto_b.is_null = MagicMock(return_value=False)
    onto_b.initial_version = None
    onto_b.version = "1.0"
    onto_b.describe = MagicMock(return_value="ontology B description")

    state = _make_state()
    tools = _make_tools(ontologies=[onto_a, onto_b])

    # LLM picks the second ontology (1-based index 2)
    _patch_llm_call(monkeypatch, answer_index=2)

    result = await select_ontology_module.select_ontology(state, tools)

    assert result.current_ontology is onto_b
    assert result.status == Status.SUCCESS


# ─── Case 3: LLM picks None (num_ontologies + 1) — the bug fix case ──────
@pytest.mark.asyncio
async def test_llm_picks_none_index_returns_null_without_warning(
    monkeypatch: pytest.MonkeyPatch, caplog: pytest.LogCaptureFixture
) -> None:
    """When the LLM returns `num_ontologies + 1` (the encoded "None" choice),
    the ontology must be set to NULL_ONTOLOGY **and** no WARNING should be
    emitted. The pre-fix code logged a warning on every legitimate None
    selection because the path fell through to the defensive branch.
    """
    onto_a = MagicMock(ontology_id="onto-a", iri="urn:a")
    onto_a.is_null = MagicMock(return_value=False)
    onto_a.initial_version = None
    onto_a.version = "1.0"
    onto_a.describe = MagicMock(return_value="ontology A description")

    state = _make_state()
    tools = _make_tools(ontologies=[onto_a])

    # num_ontologies = 1, so None is encoded as 2.
    _patch_llm_call(monkeypatch, answer_index=2)

    with caplog.at_level(logging.WARNING, logger="ontocast.agent.select_ontology"):
        result = await select_ontology_module.select_ontology(state, tools)

    assert result.current_ontology is NULL_ONTOLOGY
    assert result.status == Status.SUCCESS
    warnings = [r for r in caplog.records if r.levelno >= logging.WARNING]
    assert warnings == [], f"Unexpected warnings: {[r.message for r in warnings]}"


# ─── Case 4: defensive fallback for out-of-range index ───────────────────
@pytest.mark.asyncio
async def test_out_of_range_index_falls_back_to_null(
    monkeypatch: pytest.MonkeyPatch, caplog: pytest.LogCaptureFixture
) -> None:
    """If the LLM somehow returns a value outside [1, num_ontologies + 1]
    (Pydantic should prevent this, but be defensive), the function must
    log a WARNING and default to NULL_ONTOLOGY."""
    onto_a = MagicMock(ontology_id="onto-a", iri="urn:a")
    onto_a.is_null = MagicMock(return_value=False)
    onto_a.initial_version = None
    onto_a.version = "1.0"
    onto_a.describe = MagicMock(return_value="ontology A description")

    state = _make_state()
    tools = _make_tools(ontologies=[onto_a])

    # Out of range: num_ontologies = 1, valid is {1, 2}; we pass 99.
    _patch_llm_call(monkeypatch, answer_index=99)

    with caplog.at_level(logging.WARNING, logger="ontocast.agent.select_ontology"):
        result = await select_ontology_module.select_ontology(state, tools)

    assert result.current_ontology is NULL_ONTOLOGY
    warnings = [r for r in caplog.records if r.levelno >= logging.WARNING]
    assert len(warnings) == 1
    assert "Out-of-range answer_index" in warnings[0].message