"""DeterministicEmbedder 与工厂：纯数学，不连库不联网（T-P2-03 接口位）."""

from __future__ import annotations

import math

import pytest

from packages.retrieval.embedding import (
    DETERMINISTIC_MODEL_NAME,
    DeterministicEmbedder,
    build_embedding_provider,
)
from packages.shared.config import get_settings
from packages.shared.errors import AppError, ErrorCode


def _dot(a: list[float], b: list[float]) -> float:
    return sum(x * y for x, y in zip(a, b, strict=True))


async def test_same_text_embeds_to_same_vector() -> None:
    embedder = DeterministicEmbedder(dimensions=1024)
    first, second = await embedder.embed(["白衬衫怎么搭配", "白衬衫怎么搭配"])
    assert first == second


async def test_dimensions_match_declaration() -> None:
    vec = (await DeterministicEmbedder(dimensions=1024).embed(["hello"]))[0]
    assert len(vec) == 1024
    small = (await DeterministicEmbedder(dimensions=8).embed(["hello"]))[0]
    assert len(small) == 8


async def test_vectors_are_l2_normalized() -> None:
    for text in ("hello", "中文知识片段", ""):
        vec = (await DeterministicEmbedder(dimensions=1024).embed([text]))[0]
        assert math.isclose(_dot(vec, vec), 1.0, rel_tol=1e-6, abs_tol=1e-9)


async def test_different_texts_are_near_orthogonal() -> None:
    """不同文本近似正交（|dot| ≪ 1）：向量召回管线在无真模型时仍可测."""

    left, right = await DeterministicEmbedder(dimensions=1024).embed(
        ["interview outfit advice", "winter skincare routine"]
    )
    assert left != right
    assert abs(_dot(left, right)) < 0.1


async def test_batch_preserves_order_and_length() -> None:
    texts = ["alpha", "beta", "gamma"]
    vectors = await DeterministicEmbedder(dimensions=16).embed(texts)
    assert len(vectors) == len(texts)
    singles = [(await DeterministicEmbedder(dimensions=16).embed([t]))[0] for t in texts]
    assert vectors == singles


def test_metadata_is_stable() -> None:
    embedder = DeterministicEmbedder(dimensions=8)
    assert embedder.model_name == DETERMINISTIC_MODEL_NAME
    assert embedder.version
    assert embedder.dimensions == 8


def test_invalid_dimensions_rejected() -> None:
    with pytest.raises(ValueError, match="dimensions"):
        DeterministicEmbedder(dimensions=0)


def test_factory_builds_deterministic_embedder(monkeypatch: pytest.MonkeyPatch) -> None:
    # 带 alias 的 Settings 字段按字段名构造会被静默忽略 —— 统一走
    # setenv + cache_clear（上批经验，见任务文档 §2）。
    monkeypatch.setenv("KB_EMBEDDING_MODEL", "deterministic-v1")
    monkeypatch.setenv("KB_VECTOR_DIMENSIONS", "1024")
    get_settings.cache_clear()
    try:
        settings = get_settings()
        provider = build_embedding_provider(settings)
        assert isinstance(provider, DeterministicEmbedder)
        assert provider.model_name == "deterministic-v1"
        assert provider.dimensions == 1024
    finally:
        monkeypatch.undo()
        get_settings.cache_clear()


def test_factory_rejects_unknown_model(monkeypatch: pytest.MonkeyPatch) -> None:
    monkeypatch.setenv("KB_EMBEDDING_MODEL", "bge-m3-onnx")
    get_settings.cache_clear()
    try:
        settings = get_settings()
        assert settings.kb_embedding_model == "bge-m3-onnx"
        with pytest.raises(AppError) as excinfo:
            build_embedding_provider(settings)
        assert excinfo.value.code == ErrorCode.CONFIG_ERROR
    finally:
        monkeypatch.undo()
        get_settings.cache_clear()