"""Teste TDD pentru tools/mapare-llm/holdout.py.

Verifica logica de split + calcul hit-rate pe un fixture SINTETIC (nu pe date reale).
Fixture-ul nu testeaza numerele efective pe CSV-uri, ci CORECTITUDINEA functiilor.
"""
from __future__ import annotations

import sys
import os

# Adaugam tools/mapare-llm/ in path pentru import direct al holdout.py
sys.path.insert(0, os.path.join(os.path.dirname(__file__), '..', 'tools', 'mapare-llm'))

import pytest

# Fixture sintetic: 5 denumiri cu frecvente diferite
# Total volum = 100 + 80 + 50 + 30 + 10 + 1 + 1 = 272
FIXTURE = [
    ("Revizie motor", 100),
    ("Schimb ulei", 80),
    ("Reglat frane", 50),
    ("Diagnosticare", 30),
    ("Curatenie interior", 10),
    ("Altceva rar A", 1),
    ("Altceva rar B", 1),
]
FIXTURE_TOTAL_VOL = sum(n for _, n in FIXTURE)  # 272
FIXTURE_DISTINCT = len(FIXTURE)  # 7


# ---------------------------------------------------------------------------
# compute_volume_coverage
# ---------------------------------------------------------------------------

def test_compute_volume_coverage_sorted_descrescator():
    """Primul element trebuie sa fie cel cu NR cel mai mare."""
    from holdout import compute_volume_coverage

    rows = [("A", 10), ("B", 90), ("C", 0)]
    result = compute_volume_coverage([r for r in rows if r[1] > 0])
    assert result[0]["denumire"] == "B"
    assert result[0]["nr"] == 90


def test_compute_volume_coverage_cumul():
    """Acoperirea cumulativa e corecta."""
    from holdout import compute_volume_coverage

    rows = [("A", 90), ("B", 9), ("C", 1)]  # total=100
    result = compute_volume_coverage(rows)

    # Ordine: A(90), B(9), C(1) dupa sortare desc
    assert result[0]["denumire"] == "A"
    assert abs(result[0]["cumulative_volume_frac"] - 0.90) < 1e-9
    assert result[0]["cumulative_count"] == 1

    assert result[1]["denumire"] == "B"
    assert abs(result[1]["cumulative_volume_frac"] - 0.99) < 1e-9
    assert result[1]["cumulative_count"] == 2

    assert result[2]["denumire"] == "C"
    assert abs(result[2]["cumulative_volume_frac"] - 1.0) < 1e-9
    assert result[2]["cumulative_count"] == 3


def test_compute_volume_coverage_gol():
    """Lista goala -> lista goala (fara exceptii)."""
    from holdout import compute_volume_coverage

    assert compute_volume_coverage([]) == []


# ---------------------------------------------------------------------------
# corpus_size_for_threshold
# ---------------------------------------------------------------------------

def test_corpus_size_for_90pct():
    """Gaseste corect numarul de etichete pentru 90% acoperire."""
    from holdout import corpus_size_for_threshold

    rows = [("A", 90), ("B", 9), ("C", 1)]  # total=100
    # A singur = 90% -> 1 eticheta suficienta
    assert corpus_size_for_threshold(rows, threshold=0.90) == 1


def test_corpus_size_for_99pct():
    """Prag 99%: necesita 2 etichete (A+B = 99/100)."""
    from holdout import corpus_size_for_threshold

    rows = [("A", 90), ("B", 9), ("C", 1)]
    assert corpus_size_for_threshold(rows, threshold=0.99) == 2


def test_corpus_size_for_100pct():
    """Prag 100%: necesita toate etichetele."""
    from holdout import corpus_size_for_threshold

    rows = [("A", 90), ("B", 9), ("C", 1)]
    assert corpus_size_for_threshold(rows, threshold=1.0) == 3


# ---------------------------------------------------------------------------
# compute_hit_rate_at_k
# ---------------------------------------------------------------------------

def test_compute_hit_rate_at_k_1():
    """Top-1 eticheta (A=90): hit-rate = 90/100 = 0.90."""
    from holdout import compute_hit_rate_at_k

    rows = [("A", 90), ("B", 9), ("C", 1)]
    assert abs(compute_hit_rate_at_k(rows, k=1) - 0.90) < 1e-9


def test_compute_hit_rate_at_k_2():
    """Top-2 etichete (A+B=99): hit-rate = 0.99."""
    from holdout import compute_hit_rate_at_k

    rows = [("A", 90), ("B", 9), ("C", 1)]
    assert abs(compute_hit_rate_at_k(rows, k=2) - 0.99) < 1e-9


def test_compute_hit_rate_at_k_depasit():
    """k mai mare decat numarul de randuri: hit-rate = 1.0."""
    from holdout import compute_hit_rate_at_k

    rows = [("A", 90), ("B", 10)]
    assert abs(compute_hit_rate_at_k(rows, k=100) - 1.0) < 1e-9


def test_compute_hit_rate_at_k_gol():
    """Lista goala: hit-rate = 0.0 (fara ZeroDivisionError)."""
    from holdout import compute_hit_rate_at_k

    assert compute_hit_rate_at_k([], k=10) == 0.0


# ---------------------------------------------------------------------------
# leave_one_out_hit_rate
# ---------------------------------------------------------------------------

def test_leave_one_out_hit_rate_formula():
    """Hit-rate leave-first-out: (total_vol - total_distinct) / total_vol.

    Interpretare: pe oricare aparitie, dupa prima, e un hit (deja in corpus).
    Singletonii (NR=1) contribuie 0 hit-uri.
    """
    from holdout import leave_one_out_hit_rate

    rows = [("A", 10), ("B", 5), ("C", 1)]  # total=16, distinct=3
    # formula: (16 - 3) / 16 = 0.8125
    assert abs(leave_one_out_hit_rate(rows) - 13 / 16) < 1e-9


def test_leave_one_out_hit_rate_toate_singletons():
    """Toti singletons: hit-rate = 0 (fiecare aparitie e prima)."""
    from holdout import leave_one_out_hit_rate

    rows = [("A", 1), ("B", 1), ("C", 1)]
    assert leave_one_out_hit_rate(rows) == 0.0


def test_leave_one_out_hit_rate_gol():
    """Lista goala: returneaza 0.0 fara exceptie."""
    from holdout import leave_one_out_hit_rate

    assert leave_one_out_hit_rate([]) == 0.0


# ---------------------------------------------------------------------------
# singleton_stats
# ---------------------------------------------------------------------------

def test_singleton_stats_calcul():
    """Statistici singletons corecte."""
    from holdout import singleton_stats

    rows = [("A", 100), ("B", 1), ("C", 1)]  # total=102, 2 singletons
    stats = singleton_stats(rows)

    assert stats["singleton_count"] == 2
    assert stats["total_distinct"] == 3
    assert abs(stats["singleton_volume_frac"] - 2 / 102) < 1e-9
    assert abs(stats["singleton_distinct_frac"] - 2 / 3) < 1e-9


def test_singleton_stats_fara_singletons():
    """Fara singletons: toate fractiile singleton = 0."""
    from holdout import singleton_stats

    rows = [("A", 5), ("B", 10)]
    stats = singleton_stats(rows)
    assert stats["singleton_count"] == 0
    assert stats["singleton_volume_frac"] == 0.0


# ---------------------------------------------------------------------------
# normalize_for_match: cheia de potrivire refolosita din app/mapping.py
# ---------------------------------------------------------------------------

def test_normalize_for_match_diacritice():
    """normalize_for_match trateaza diacriticele identic (din app/mapping.py)."""
    from holdout import normalize_key

    # Variante cu si fara diacritice -> aceeasi cheie normalizata
    assert normalize_key("Reparație motor") == normalize_key("Reparatie motor")
    assert normalize_key("REPARATIE MOTOR") == normalize_key("Reparatie motor")


def test_normalize_for_match_spatii():
    """Spatiile multiple se colapseza."""
    from holdout import normalize_key

    assert normalize_key("revizie   periodica") == normalize_key("REVIZIE PERIODICA")


# ---------------------------------------------------------------------------
# run_holdout: structura si verdict
# ---------------------------------------------------------------------------

def test_run_holdout_campuri_obligatorii():
    """run_holdout returneaza toate campurile asteptate."""
    from holdout import run_holdout

    result = run_holdout(FIXTURE, client_name="test_client")

    campuri = [
        "client", "total_distinct", "total_volume",
        "coverage_at_100", "coverage_at_500", "coverage_at_1000",
        "labels_for_90pct", "frac_for_90pct",
        "leave_one_out_hit_rate",
        "singleton_count", "singleton_distinct_frac", "singleton_volume_frac",
        "verdict", "nota",
    ]
    for camp in campuri:
        assert camp in result, f"Camp lipsa: {camp}"


def test_run_holdout_client_name():
    """client_name se pastreaza corect in rezultat."""
    from holdout import run_holdout

    result = run_holdout(FIXTURE, client_name="test_client")
    assert result["client"] == "test_client"


def test_run_holdout_verdict_valid():
    """Verdict e unul din valorile definite."""
    from holdout import run_holdout

    result = run_holdout(FIXTURE, client_name="test_client")
    assert result["verdict"] in ("SUSTINUTA", "SLABA", "NEVALIDABILA")


def test_run_holdout_total_volum():
    """total_volume = suma NR din fixture."""
    from holdout import run_holdout

    result = run_holdout(FIXTURE, client_name="test_client")
    assert result["total_volume"] == FIXTURE_TOTAL_VOL


def test_run_holdout_distinct():
    """total_distinct = numarul de randuri din fixture."""
    from holdout import run_holdout

    result = run_holdout(FIXTURE, client_name="test_client")
    assert result["total_distinct"] == FIXTURE_DISTINCT


def test_run_holdout_verdict_sustinuta_pe_zipf_puternic():
    """Pe distributie Zipf puternica (o denumire = 95% din volum), verdict SUSTINUTA."""
    from holdout import run_holdout

    rows = [("REVIZIE", 9500)] + [(f"altceva_{i}", 1) for i in range(500)]
    result = run_holdout(rows, client_name="zipf")
    assert result["verdict"] == "SUSTINUTA"


def test_run_holdout_verdict_slaba_pe_distributie_plata():
    """Pe distributie uniforma (50 denumiri cu aceeasi frecventa), poate fi SLABA/NEVALIDABILA."""
    from holdout import run_holdout

    rows = [(f"op_{i}", 100) for i in range(100)]  # 100 denumiri cu NR egal
    result = run_holdout(rows, client_name="uniform")
    # 90% din 100*100=10000 = 9000; necesita 90 din 100 denumiri = 90% -> NEVALIDABILA
    assert result["verdict"] in ("SLABA", "NEVALIDABILA")