tests/test_embedder_cache.py

import pytest
from mediastorm.vectorize.embedder import Embedder


@pytest.fixture(scope="module")
def embedder():
    return Embedder()


def test_embed_query_returns_384_dimensions(embedder):
    """embed_query should produce a 384-dimensional vector."""
    vec = embedder.embed_query("documentary about war")
    assert len(vec) == 384


def test_embed_query_is_normalized(embedder):
    """embed_query result should be L2-normalized."""
    vec = embedder.embed_query("human rights in Africa")
    norm = sum(x ** 2 for x in vec) ** 0.5
    assert norm == pytest.approx(1.0, abs=1e-4)


def test_embed_query_repeated_calls_return_identical_results(embedder):
    """Repeated calls with same text should return identical vectors."""
    text = "award winning photography"
    v1 = embedder.embed_query(text)
    v2 = embedder.embed_query(text)
    assert v1 == v2


def test_embed_query_cache_avoids_recomputation(embedder):
    """Second call with same text should not invoke embed_texts again."""
    # Clear the cache so we start fresh
    embedder._embed_query_cached.cache_clear()

    call_count = 0
    original_embed_texts = embedder.embed_texts

    def counting_embed_texts(texts):
        nonlocal call_count
        call_count += 1
        return original_embed_texts(texts)

    embedder.embed_texts = counting_embed_texts
    try:
        embedder.embed_query("cache test query")
        embedder.embed_query("cache test query")
    finally:
        embedder.embed_texts = original_embed_texts

    assert call_count == 1, f"embed_texts called {call_count} times, expected 1"


def test_embed_query_different_texts_produce_different_vectors(embedder):
    """Different queries should produce different embeddings."""
    v1 = embedder.embed_query("war documentary")
    v2 = embedder.embed_query("cooking show")
    assert v1 != v2


def test_embed_query_matches_embed_texts_output(embedder):
    """embed_query result should match embed_texts([text])[0]."""
    text = "journalism and press freedom"
    # Use embed_texts directly to bypass cache for comparison
    expected = embedder.embed_texts([text])[0]
    result = embedder.embed_query(text)
    assert result == pytest.approx(expected, abs=1e-6)