"""HF 선호학습 데이터 파이프라인 단위테스트 (계획서 2단계 고도화)."""
from __future__ import annotations

from app.engine.preference import (
    PreferenceCandidate,
    build_candidate,
    to_dpo_record,
    to_dpo_dataset,
    validate_labeled,
    dataset_stats,
)


def test_build_candidate_is_pending():
    c = build_candidate("p1", "원문", "글A", "글B", suggested_rejected="b")
    assert c.label_status == "pending"
    assert "원문" in c.prompt
    assert to_dpo_record(c) is None  # 미라벨은 학습 레코드 없음


def test_labeled_becomes_dpo_record():
    c = build_candidate("p1", "원문", "정상글", "표절글", suggested_rejected="b")
    c.label_status = "labeled"
    c.chosen = "정상글"
    c.rejected = "표절글"
    rec = to_dpo_record(c)
    assert rec is not None
    assert rec["chosen"] == "정상글" and rec["rejected"] == "표절글"


def test_validate_catches_same_chosen_rejected():
    c = PreferenceCandidate("p1", "p", "a", "b", label_status="labeled", chosen="x", rejected="x")
    errors = validate_labeled(c)
    assert any("chosen == rejected" in e for e in errors)


def test_validate_catches_missing():
    c = PreferenceCandidate("p1", "p", "a", "b", label_status="labeled", chosen="x", rejected=None)
    assert any("rejected 누락" in e for e in validate_labeled(c))


def test_dataset_stats():
    pending = build_candidate("p1", "o", "a", "b")
    labeled = build_candidate("p2", "o", "a", "b")
    labeled.label_status = "labeled"
    labeled.chosen, labeled.rejected = "a", "b"
    stats = dataset_stats([pending, labeled])
    assert stats["total"] == 2
    assert stats["labeled"] == 1
    assert stats["pending"] == 1
    assert stats["trainable"] == 1
    assert stats["errors"] == []


def test_to_dpo_dataset_filters_pending():
    cs = [build_candidate(f"p{i}", "o", "a", "b") for i in range(3)]
    cs[0].label_status = "labeled"
    cs[0].chosen, cs[0].rejected = "a", "b"
    assert len(to_dpo_dataset(cs)) == 1