import numpy as np

from streamsearch import embed
from streamsearch.schema import EMBED_DIM


def test_embed_shape_and_norm():
    vecs = embed.embed_texts(["hello world", "a second document"])
    assert vecs.shape == (2, EMBED_DIM)
    assert vecs.dtype == np.float32
    # normalize_embeddings=True -> unit vectors
    norms = np.linalg.norm(vecs, axis=1)
    np.testing.assert_allclose(norms, 1.0, atol=1e-4)


def test_embed_is_deterministic():
    a = embed.embed_texts(["reproducible embedding"])
    b = embed.embed_texts(["reproducible embedding"])
    np.testing.assert_allclose(a, b, atol=1e-6)


def test_empty_input_returns_empty():
    vecs = embed.embed_texts([])
    assert vecs.shape == (0, EMBED_DIM)


def test_model_loads_only_once():
    # get_model is lru_cached: repeated calls return the same object.
    embed.get_model.cache_clear()
    m1 = embed.get_model()
    m2 = embed.get_model()
    assert m1 is m2
    assert embed.get_model.cache_info().hits >= 1