tests/conftest.py

"""Shared pytest fixtures.

Generalizes the in-memory-Qdrant + local-hashing-embedder trick (previously
hand-wired in the root offline scripts) into reusable fixtures, and provides
temp-SQLite account fixtures plus HF-sync mocks — so the whole default suite
runs with NO credentials, network, or Docker.
"""

from __future__ import annotations

import hashlib
import os
import re
import sqlite3
import sys
from pathlib import Path

import pytest

# Repo root on sys.path + as CWD-independent anchor.
ROOT = Path(__file__).resolve().parents[1]
if str(ROOT) not in sys.path:
    sys.path.insert(0, str(ROOT))

CDMS_DB = ROOT / "data" / "cdms_metadata.db"
_DIM = 1536
_TOKEN = re.compile(r"[a-z0-9]+")


# --------------------------------------------------------------------------- #
# Deterministic local embedder (stand-in for OpenAIEmbeddingService)
# --------------------------------------------------------------------------- #
def _embed(text: str):
    import numpy as np

    v = np.zeros(_DIM, dtype=np.float32)
    for tok in _TOKEN.findall((text or "").lower()):
        idx = int(hashlib.md5(tok.encode()).hexdigest(), 16) % _DIM
        v[idx] += 1.0
    n = np.linalg.norm(v)
    if n > 0:
        v /= n
    return v.tolist()


class LocalEmbedder:
    """Drop-in replacement for the OpenAI embedding service (offline, deterministic)."""

    def generate_embedding(self, text: str):
        return _embed(text)


@pytest.fixture(scope="session")
def local_embedder():
    return LocalEmbedder()


# --------------------------------------------------------------------------- #
# Accounts: temp SQLite store + service with HF sync disabled
# --------------------------------------------------------------------------- #
@pytest.fixture()
def account_store(tmp_path):
    from src.accounts.store import AccountStore

    return AccountStore(tmp_path / "accounts.db")


@pytest.fixture()
def accounts_service(tmp_path, monkeypatch):
    """Real AccountsService on a temp DB with sync OFF (no HF token) and a
    known SESSION_SECRET so token tests are deterministic."""
    monkeypatch.setenv("SESSION_SECRET", "unit-test-secret")
    monkeypatch.delenv("HF_DATA_REPO", raising=False)
    monkeypatch.delenv("HF_DATA_TOKEN", raising=False)
    from src.accounts.service import AccountsService

    svc = AccountsService(db_path=str(tmp_path / "accounts.db"), daily_quota=5)
    assert svc.sync.enabled is False  # local-dev guard: no sync without a token
    return svc


# --------------------------------------------------------------------------- #
# Integration: real CDMS pipeline over the committed index (in-memory Qdrant)
# --------------------------------------------------------------------------- #
@pytest.fixture(scope="session")
def in_memory_rag():
    """Build an in-memory Qdrant from the committed chunk DB and wire the REAL
    CDMSRAGSearch to it with the local embedder. Skips if the index or
    qdrant-client is unavailable."""
    if not CDMS_DB.exists():
        pytest.skip(f"CDMS index not present ({CDMS_DB}); skipping integration test")
    try:
        from src.rag.vector_store import QdrantVectorStore
        from src.cdms.rag_search import CDMSRAGSearch
    except Exception as e:  # missing qdrant-client etc.
        pytest.skip(f"RAG deps unavailable: {e}")

    conn = sqlite3.connect(str(CDMS_DB))
    try:
        rows = conn.execute(
            "SELECT dc.id, dc.content, dc.page_number, dc.document_id, d.filename "
            "FROM document_chunks dc JOIN documents d ON dc.document_id = d.id"
        ).fetchall()
    finally:
        conn.close()
    if not rows:
        pytest.skip("CDMS index has no chunks")

    store = QdrantVectorStore()  # Docker unavailable -> in-memory mode
    for cid, content, page, docid, filename in rows:
        if not content:
            continue
        store.add_document_chunk(
            str(cid),
            _embed(content),
            {
                "content": content,
                "source_file": filename,
                "page_number": page or 0,
                "document_id": docid,
            },
        )

    searcher = CDMSRAGSearch()
    searcher.vector_store = store
    searcher.embedding_service = LocalEmbedder()
    return searcher