pastor-online/backend/tests/test_main.py

"""Testes da API FastAPI."""

from __future__ import annotations

import json
import sys
import threading
from pathlib import Path

import pytest
from fastapi.testclient import TestClient

sys.path.insert(0, str(Path(__file__).resolve().parents[1]))

from app.config import Settings  # noqa: E402
from app.embedding import HashEmbedder  # noqa: E402
from app.main import create_app  # noqa: E402
from app.vector_store import InMemoryVectorStore  # noqa: E402


class StubProvider:
    async def stream(self, messages, system):
        yield "Resposta ancorada em Gênesis 1:1."

    async def aclose(self):
        return None


class FakeIndexer:
    """Indexador controlado para os testes de API.

    `release` permite segurar a reindexação em background, de forma que o teste
    do 409 seja determinístico.
    """

    def __init__(self, block: bool = False):
        self.calls: list[bool] = []
        self.release = threading.Event()
        self.release.set()
        if block:
            self.release.clear()

    def reindex(self, force: bool = False):
        self.calls.append(force)
        self.release.wait(timeout=10)
        from app.indexer import IndexStatus

        return IndexStatus(state="ready", sources_total=1, sources_indexed=1, chunks=42)

    def status(self):
        from app.indexer import IndexStatus

        return IndexStatus(state="ready", sources_total=1, sources_indexed=1, chunks=42)


@pytest.fixture
def settings(tmp_path):
    return Settings(
        corpus_dir=tmp_path / "corpus",
        vector_db_dir=tmp_path / "vector_db",
        embedding_model="hash-test",
        llm_provider="stub",
        cors_origins="http://localhost:8080",
    )


@pytest.fixture
def client(settings):
    embedder = HashEmbedder(dimensions=64)
    store = InMemoryVectorStore(embedder=embedder)
    app = create_app(
        settings=settings,
        store=store,
        embedder=embedder,
        provider=StubProvider(),
        indexer=FakeIndexer(),
    )
    return TestClient(app)


def test_health_returns_ok(client):
    response = client.get("/api/health")

    assert response.status_code == 200
    body = response.json()
    assert body["status"] == "ok"
    assert body["llm_provider"] == "stub"


def test_status_reports_indexation(client):
    response = client.get("/api/status")

    assert response.status_code == 200
    body = response.json()
    assert body["state"] == "ready"
    assert body["chunks"] == 42


def test_reindex_returns_202(client):
    response = client.post("/api/reindex")

    assert response.status_code == 202
    assert response.json()["state"] == "indexing"


def test_reindex_requires_token_when_configured(settings):
    embedder = HashEmbedder(dimensions=64)
    store = InMemoryVectorStore(embedder=embedder)
    settings.reindex_token = "segredo"
    app = create_app(settings=settings, store=store, embedder=embedder, provider=StubProvider())
    client = TestClient(app)

    assert client.post("/api/reindex").status_code == 401

    authorized = client.post("/api/reindex", headers={"X-Reindex-Token": "segredo"})
    assert authorized.status_code == 202


def test_concurrent_reindex_is_rejected_with_409(settings):
    settings.auto_index = False
    embedder = HashEmbedder(dimensions=64)
    store = InMemoryVectorStore(embedder=embedder)
    indexer = FakeIndexer(block=True)
    app = create_app(
        settings=settings,
        store=store,
        embedder=embedder,
        provider=StubProvider(),
        indexer=indexer,
    )

    # O context manager mantém um único event loop entre as requisições; sem ele,
    # o TestClient fecha o loop a cada chamada e a tarefa de background seria
    # cancelada, liberando o lock.
    with TestClient(app) as client:
        try:
            first = client.post("/api/reindex")
            assert first.status_code == 202

            second = client.post("/api/reindex")
            assert second.status_code == 409
        finally:
            indexer.release.set()


def test_chat_rejects_empty_message(client):
    response = client.post("/api/chat", json={"message": ""})

    assert response.status_code == 422


def test_chat_rejects_message_above_the_limit(client):
    response = client.post("/api/chat", json={"message": "a" * 5000})

    assert response.status_code == 422


def test_chat_streams_sources_token_and_done(client):
    from app.corpus import VerseChunk

    embedder = HashEmbedder(dimensions=64)
    store = client.app.state.services.store
    chunk = VerseChunk(
        book="Gênesis",
        chapter=1,
        verse=1,
        file="biblia.usfx.xml",
        text="No princípio criou Deus os céus e a terra.",
    )
    store.upsert([chunk], embedder.encode([chunk.text]))

    response = client.post(
        "/api/chat",
        json={"message": "Quem criou o mundo?", "history": []},
    )

    assert response.status_code == 200
    assert response.headers["content-type"].startswith("text/event-stream")

    body = response.text
    assert "event: sources" in body
    assert "event: token" in body
    assert "event: done" in body

    sources_block = body.split("event: sources")[1].split("event:")[0]
    payload = json.loads(sources_block.split("data:")[1].strip())
    assert payload["sources"][0]["reference"] == "Gênesis 1:1"


def test_chat_without_evidence_returns_the_fixed_message(client):
    response = client.post("/api/chat", json={"message": "fale sobre ações", "history": []})

    assert response.status_code == 200
    assert "event: token" in response.text
    assert "não aborda diretamente" in response.text


def test_cors_is_restricted_to_configured_origin(client):
    response = client.get("/api/health", headers={"Origin": "http://localhost:8080"})

    assert response.headers.get("access-control-allow-origin") == "http://localhost:8080"


def test_cors_rejects_unknown_origin(client):
    response = client.get("/api/health", headers={"Origin": "http://malvado.example"})

    assert "access-control-allow-origin" not in response.headers


def test_sources_endpoint_returns_a_list(client):
    response = client.get("/api/sources")

    assert response.status_code == 200
    assert "sources" in response.json()


def test_create_app_builds_the_configured_embedder(monkeypatch, tmp_path):
    """Produção: o modelo configurado precisa ser construído de fato.

    O `VectorStore` default para `HashEmbedder` quando nasce sem embedder. Se a
    aplicação criar o store primeiro e depois herdar o embedder dele, o modelo
    configurado nunca é carregado e a busca roda sem semântica nenhuma — com
    `/api/status` ainda anunciando o nome do modelo.
    """
    import app.main as main_module

    built: list[str] = []

    class FakeSentenceTransformer:
        def __init__(self, model_name, batch_size=32, device="cpu", module=None):
            built.append(model_name)
            self.name = model_name
            self.dimensions = 8

        def encode(self, texts):
            return [[0.0] * 8 for _ in texts]

    monkeypatch.setattr(main_module, "SentenceTransformerEmbedder", FakeSentenceTransformer)

    settings = Settings(
        corpus_dir=tmp_path / "corpus",
        vector_db_dir=tmp_path / "vector_db",
        embedding_model="modelo-configurado",
        llm_provider="stub",
        auto_index=False,
    )

    app = main_module.create_app(settings=settings)
    services = app.state.services

    assert built == ["modelo-configurado"]
    assert services.embedder.name == "modelo-configurado"
    assert services.store.embedder.name == "modelo-configurado"
    assert services.embedder_warning is None


def test_embedder_fallback_is_surfaced_in_status(monkeypatch, tmp_path):
    """Se o modelo não carrega, o status precisa dizer que a busca está sem semântica."""
    import app.main as main_module

    def explode(*args, **kwargs):
        raise RuntimeError("PyTorch >= 2.6 is required but found 2.5.1")

    monkeypatch.setattr(main_module, "SentenceTransformerEmbedder", explode)

    settings = Settings(
        corpus_dir=tmp_path / "corpus",
        vector_db_dir=tmp_path / "vector_db",
        embedding_model="modelo-configurado",
        llm_provider="stub",
        auto_index=False,
    )

    app = main_module.create_app(
        settings=settings, indexer=FakeIndexer(), provider=StubProvider()
    )
    body = TestClient(app).get("/api/status").json()

    assert any("semântica" in error for error in body["errors"])
    assert app.state.services.embedder.name == HashEmbedder.name


def test_health_reports_the_llm_endpoint_in_use(client):
    """/api/health precisa mostrar a URL efetiva, não a que o .env parece dizer.

    Variável de shell vence o `.env` na interpolação do compose: um
    `OLLAMA_BASE_URL` herdado do ambiente apontando para `localhost` dentro do
    container derrubava o chat sem nenhum indício do endereço tentado.
    """
    body = client.get("/api/health").json()

    assert body["llm_provider"] == "stub"
    assert body["llm_endpoint"] is None, "o stub não tem endpoint"


def test_health_exposes_ollama_endpoint_without_credentials(tmp_path):
    """A endpoint vai para o health, mas credenciais na URL não podem vazar."""
    from app.llm import OllamaProvider

    settings = Settings(
        corpus_dir=tmp_path / "corpus",
        vector_db_dir=tmp_path / "vector_db",
        embedding_model="hash-test",
        llm_provider="ollama",
        auto_index=False,
    )
    provider = OllamaProvider(
        base_url="http://user:segredo@ollama:11434", model="m"
    )
    app = create_app(
        settings=settings,
        store=InMemoryVectorStore(),
        embedder=HashEmbedder(dimensions=64),
        provider=provider,
        indexer=FakeIndexer(),
    )

    body = TestClient(app).get("/api/health").json()

    assert "segredo" not in body["llm_endpoint"]
    assert body["llm_endpoint"].endswith("ollama:11434")