rpa_vision_v3/tests/unit/test_chat_interface.py

# tests/unit/test_chat_interface.py
"""
Tests unitaires du module chat_interface (Léa conversationnelle).

Vérifie :
1. Création de session (état initial, message d'accueil)
2. Envoi de message → appel TaskPlanner mocké
3. Historique (get_history)
4. Transitions d'états idle → planning → awaiting_confirmation → executing → done
5. Abandon (utilisateur répond "non")
6. Fallback gracieux quand gemma4/TaskPlanner indisponible
7. ChatManager (création, listing, cleanup)
"""

from __future__ import annotations

import sys
import time
from pathlib import Path
from unittest.mock import MagicMock

import pytest

_ROOT = str(Path(__file__).resolve().parents[2])
if _ROOT not in sys.path:
    sys.path.insert(0, _ROOT)

from agent_v0.server_v1.chat_interface import (
    ChatSession,
    ChatManager,
    STATE_IDLE,
    STATE_PLANNING,
    STATE_AWAITING_CONFIRMATION,
    STATE_EXECUTING,
    STATE_DONE,
    STATE_ERROR,
    ROLE_USER,
    ROLE_LEA,
)
from agent_v0.server_v1.task_planner import TaskPlan


# =============================================================================
# Fixtures
# =============================================================================

@pytest.fixture
def sample_workflows():
    return [
        {
            "session_id": "sess_bloc_notes",
            "name": "Bloc-notes",
            "description": "Ouvrir Bloc-notes via Exécuter (Win+R) et écrire du texte",
        },
    ]


@pytest.fixture
def understood_plan():
    """TaskPlan qui a compris l'ordre et matche un workflow."""
    return TaskPlan(
        instruction="ouvre le bloc-notes et écris bonjour",
        understood=True,
        workflow_match="sess_bloc_notes",
        workflow_name="Bloc-notes",
        match_confidence=0.9,
        parameters={"texte": "bonjour"},
        is_loop=False,
        mode="replay",
    )


@pytest.fixture
def unknown_plan():
    """TaskPlan qui n'a pas compris."""
    return TaskPlan(
        instruction="fais le café",
        understood=False,
        error="aucun workflow ne correspond",
    )


@pytest.fixture
def mock_task_planner(understood_plan):
    planner = MagicMock()
    planner.understand.return_value = understood_plan
    return planner


@pytest.fixture
def mock_replay_callback():
    return MagicMock(return_value="replay_abc123")


@pytest.fixture
def mock_status_provider():
    """Retourne un dict par défaut 'running' — peut être modifié dans les tests."""
    return MagicMock(return_value={
        "status": "running",
        "completed_actions": 1,
        "total_actions": 5,
    })


@pytest.fixture
def session(mock_task_planner, sample_workflows, mock_replay_callback, mock_status_provider):
    return ChatSession(
        task_planner=mock_task_planner,
        workflows_provider=lambda: sample_workflows,
        replay_callback=mock_replay_callback,
        status_provider=mock_status_provider,
    )


# =============================================================================
# Tests création session
# =============================================================================

class TestSessionCreation:
    def test_session_id_generated(self):
        s = ChatSession()
        assert s.session_id.startswith("chat_")

    def test_initial_state_is_idle(self):
        s = ChatSession()
        assert s.state == STATE_IDLE

    def test_welcome_message_present(self):
        s = ChatSession()
        history = s.get_history()
        assert len(history) == 1
        assert history[0]["role"] == ROLE_LEA
        assert "Bonjour" in history[0]["content"] or "Léa" in history[0]["content"]

    def test_session_id_custom(self):
        s = ChatSession(session_id="custom_42")
        assert s.session_id == "custom_42"


# =============================================================================
# Tests send_message
# =============================================================================

class TestSendMessage:
    def test_empty_message_rejected(self, session):
        result = session.send_message("")
        assert result["ok"] is False

    def test_send_message_calls_planner(self, session, mock_task_planner):
        session.send_message("ouvre le bloc-notes")
        mock_task_planner.understand.assert_called_once()
        call = mock_task_planner.understand.call_args
        assert call.kwargs["instruction"] == "ouvre le bloc-notes"
        # workflows_provider a été appelé et passé
        assert "available_workflows" in call.kwargs
        assert len(call.kwargs["available_workflows"]) == 1

    def test_send_message_transitions_to_awaiting_confirmation(self, session):
        result = session.send_message("ouvre le bloc-notes")
        assert result["ok"] is True
        assert session.state == STATE_AWAITING_CONFIRMATION
        assert result["state"] == STATE_AWAITING_CONFIRMATION

    def test_user_message_added_to_history(self, session):
        session.send_message("ouvre le bloc-notes")
        history = session.get_history()
        user_msgs = [m for m in history if m["role"] == ROLE_USER]
        assert len(user_msgs) == 1
        assert user_msgs[0]["content"] == "ouvre le bloc-notes"

    def test_lea_proposal_added_to_history(self, session):
        session.send_message("ouvre le bloc-notes")
        history = session.get_history()
        lea_msgs = [m for m in history if m["role"] == ROLE_LEA]
        # Bienvenue + proposition
        assert len(lea_msgs) == 2
        proposal = lea_msgs[-1]["content"]
        assert "Bloc-notes" in proposal
        assert "oui" in proposal.lower() or "y aller" in proposal.lower()

    def test_proposal_contains_confidence(self, session):
        session.send_message("ouvre le bloc-notes")
        history = session.get_history()
        proposal = history[-1]["content"]
        # 0.9 → 90%
        assert "90" in proposal

    def test_proposal_contains_parameters(self, session):
        session.send_message("ouvre le bloc-notes")
        history = session.get_history()
        proposal = history[-1]["content"]
        assert "texte" in proposal
        assert "bonjour" in proposal


# =============================================================================
# Tests get_history
# =============================================================================

class TestGetHistory:
    def test_history_returns_list_of_dicts(self, session):
        history = session.get_history()
        assert isinstance(history, list)
        assert all(isinstance(m, dict) for m in history)

    def test_history_message_structure(self, session):
        history = session.get_history()
        msg = history[0]
        assert "role" in msg
        assert "content" in msg
        assert "timestamp" in msg
        assert "meta" in msg

    def test_history_grows_with_messages(self, session):
        initial = len(session.get_history())
        session.send_message("ouvre le bloc-notes")
        assert len(session.get_history()) > initial


# =============================================================================
# Tests transitions d'états
# =============================================================================

class TestStateTransitions:
    def test_full_happy_path(self, session, mock_task_planner, mock_replay_callback):
        """idle → planning → awaiting_confirmation → executing → done."""
        # Départ : idle
        assert session.state == STATE_IDLE

        # Envoi message → planning → awaiting_confirmation
        session.send_message("ouvre le bloc-notes")
        assert session.state == STATE_AWAITING_CONFIRMATION

        # Confirmation → executing
        result = session.confirm(confirmed=True)
        assert result["ok"] is True
        assert session.state == STATE_EXECUTING
        mock_replay_callback.assert_called_once()
        call = mock_replay_callback.call_args
        assert call.kwargs["session_id"] == "sess_bloc_notes"

        # Simulation : replay terminé → done
        session._status_provider.return_value = {
            "status": "done",
            "completed_actions": 5,
            "total_actions": 5,
        }
        session.refresh_progress()
        assert session.state == STATE_DONE

    def test_confirm_via_message_oui(self, session, mock_replay_callback):
        """Le TIM peut répondre 'oui' en message au lieu d'un bouton."""
        session.send_message("ouvre le bloc-notes")
        assert session.state == STATE_AWAITING_CONFIRMATION

        session.send_message("oui")
        assert session.state == STATE_EXECUTING
        mock_replay_callback.assert_called_once()

    def test_refusal_via_confirm_false(self, session, mock_replay_callback):
        """confirm(False) → retour à idle, pas d'exécution."""
        session.send_message("ouvre le bloc-notes")
        result = session.confirm(confirmed=False)
        assert result["ok"] is True
        assert result["confirmed"] is False
        assert session.state == STATE_IDLE
        mock_replay_callback.assert_not_called()

    def test_refusal_via_message_non(self, session, mock_replay_callback):
        """Le TIM répond 'non' → annulation."""
        session.send_message("ouvre le bloc-notes")
        session.send_message("non")
        assert session.state == STATE_IDLE
        mock_replay_callback.assert_not_called()
        # Le message d'annulation doit être dans l'historique
        history = session.get_history()
        assert any("annule" in m["content"].lower() for m in history)

    def test_ambiguous_confirmation_reply(self, session):
        """Réponse ambiguë pendant awaiting_confirmation → demande de clarification."""
        session.send_message("ouvre le bloc-notes")
        result = session.send_message("peut-être")
        assert session.state == STATE_AWAITING_CONFIRMATION
        assert result.get("needs_clarification") is True

    def test_failed_replay_transitions_to_error(self, session):
        """replay_callback lève une exception → état error."""
        session._replay_callback = MagicMock(side_effect=RuntimeError("boom"))
        session.send_message("ouvre le bloc-notes")
        result = session.confirm(confirmed=True)
        assert result["ok"] is False
        assert session.state == STATE_ERROR

    def test_replay_failure_from_status(self, session):
        """Le replay rapporte 'failed' → état error."""
        session.send_message("ouvre le bloc-notes")
        session.confirm(confirmed=True)
        assert session.state == STATE_EXECUTING

        session._status_provider.return_value = {
            "status": "failed",
            "error": "element introuvable",
        }
        session.refresh_progress()
        assert session.state == STATE_ERROR


# =============================================================================
# Tests fallback / résilience
# =============================================================================

class TestResilience:
    def test_no_task_planner_graceful(self):
        """Sans TaskPlanner, on reste gracieux."""
        s = ChatSession(task_planner=None)
        result = s.send_message("test")
        assert result["ok"] is False
        assert s.state == STATE_ERROR
        # Message d'erreur présent dans l'historique
        history = s.get_history()
        assert any("désolée" in m["content"].lower() or "indisponible" in m["content"].lower()
                   for m in history)

    def test_task_planner_exception_graceful(self, mock_replay_callback):
        """TaskPlanner lève une exception (gemma4 down) → état error propre."""
        planner = MagicMock()
        planner.understand.side_effect = RuntimeError("gemma4 offline")

        s = ChatSession(
            task_planner=planner,
            workflows_provider=lambda: [],
            replay_callback=mock_replay_callback,
        )
        result = s.send_message("test")
        assert result["ok"] is False
        assert s.state == STATE_ERROR

    def test_instruction_not_understood(self, unknown_plan, mock_replay_callback):
        """Plan.understood = False → message d'erreur explicite."""
        planner = MagicMock()
        planner.understand.return_value = unknown_plan

        s = ChatSession(
            task_planner=planner,
            workflows_provider=lambda: [],
            replay_callback=mock_replay_callback,
        )
        result = s.send_message("fais le café")
        assert result["ok"] is False
        assert s.state == STATE_ERROR
        history = s.get_history()
        assert any("reformuler" in m["content"].lower() for m in history)

    def test_no_replay_callback(self, mock_task_planner, sample_workflows):
        """Sans replay_callback, on refuse l'exécution proprement."""
        s = ChatSession(
            task_planner=mock_task_planner,
            workflows_provider=lambda: sample_workflows,
            replay_callback=None,
        )
        s.send_message("ouvre le bloc-notes")
        result = s.confirm(confirmed=True)
        assert result["ok"] is False
        assert s.state == STATE_ERROR


# =============================================================================
# Tests snapshot
# =============================================================================

class TestSnapshot:
    def test_snapshot_structure(self, session):
        snap = session.get_snapshot()
        assert "session_id" in snap
        assert "state" in snap
        assert "messages" in snap
        assert "pending_plan" in snap
        assert "active_replay_id" in snap
        assert "progress" in snap

    def test_snapshot_includes_pending_plan_when_awaiting(self, session):
        session.send_message("ouvre le bloc-notes")
        snap = session.get_snapshot()
        assert snap["state"] == STATE_AWAITING_CONFIRMATION
        assert snap["pending_plan"] is not None
        assert snap["pending_plan"]["workflow_name"] == "Bloc-notes"

    def test_snapshot_no_pending_plan_in_idle(self, session):
        snap = session.get_snapshot()
        assert snap["pending_plan"] is None


# =============================================================================
# Tests ChatManager
# =============================================================================

class TestChatManager:
    def test_create_session(self, mock_task_planner, sample_workflows):
        mgr = ChatManager(
            task_planner=mock_task_planner,
            workflows_provider=lambda: sample_workflows,
        )
        s = mgr.create_session()
        assert s is not None
        assert s.session_id in [x["session_id"] for x in mgr.list_sessions()]

    def test_get_session(self, mock_task_planner):
        mgr = ChatManager(task_planner=mock_task_planner)
        s = mgr.create_session()
        retrieved = mgr.get_session(s.session_id)
        assert retrieved is s

    def test_get_session_not_found(self):
        mgr = ChatManager()
        assert mgr.get_session("unknown") is None

    def test_delete_session(self, mock_task_planner):
        mgr = ChatManager(task_planner=mock_task_planner)
        s = mgr.create_session()
        assert mgr.delete_session(s.session_id) is True
        assert mgr.get_session(s.session_id) is None

    def test_cleanup_old_sessions(self, mock_task_planner):
        mgr = ChatManager(task_planner=mock_task_planner)
        s = mgr.create_session()
        # Simuler une session très ancienne
        s.updated_at = time.time() - 100000
        removed = mgr.cleanup_old(max_age_s=3600)
        assert removed == 1
        assert mgr.get_session(s.session_id) is None

    def test_list_sessions_structure(self, mock_task_planner):
        mgr = ChatManager(task_planner=mock_task_planner)
        mgr.create_session(machine_id="pc-01")
        sessions = mgr.list_sessions()
        assert len(sessions) == 1
        s = sessions[0]
        assert "session_id" in s
        assert "state" in s
        assert "machine_id" in s
        assert s["machine_id"] == "pc-01"