"""Unit tests for the read-only backtest harness (issue #648).

Covers the PURE aggregation / metric helpers, factored out of the DB code so
they're testable without a live database:
  - _compute_metrics   — signed/abs error %, median bias, MAPE, per-rooms split
  - _errors_summary    — bias / MAPE / p25 / p75 of a signed-error list
  - _bucketize_rooms / _rooms_label — 4+ collapse, студия labelling

No DB / network / mocks: these operate on plain lists/tuples.

NOTE: importing scripts.backtest_estimator pulls app.services.estimator →
app.core.config.Settings, which REQUIRES DATABASE_URL. Set a dummy value
BEFORE importing app modules (same pattern as tests/test_estimator_pure_units.py
and tests/test_audit_address_mismatch.py).
"""

import os

os.environ.setdefault("DATABASE_URL", "postgresql+psycopg://test:test@localhost:5432/test")

import math

import pytest

from scripts import backtest_estimator as bt

# --------------------------------------------------------------------------- #
# _bucketize_rooms / _rooms_label
# --------------------------------------------------------------------------- #


def test_bucketize_studio_and_negative_clamp_to_zero() -> None:
    assert bt._bucketize_rooms(0) == 0
    assert bt._bucketize_rooms(-3) == 0


def test_bucketize_four_plus_collapses() -> None:
    assert bt._bucketize_rooms(4) == 4
    assert bt._bucketize_rooms(5) == 4
    assert bt._bucketize_rooms(9) == 4


def test_bucketize_passthrough_for_one_to_three() -> None:
    assert bt._bucketize_rooms(1) == 1
    assert bt._bucketize_rooms(2) == 2
    assert bt._bucketize_rooms(3) == 3


def test_rooms_label() -> None:
    assert bt._rooms_label(0) == "студия"
    assert bt._rooms_label(1) == "1к"
    assert bt._rooms_label(3) == "3к"
    assert bt._rooms_label(4) == "4+"
    assert bt._rooms_label(7) == "4+"


# --------------------------------------------------------------------------- #
# _errors_summary
# --------------------------------------------------------------------------- #


def test_errors_summary_empty_returns_all_none() -> None:
    s = bt._errors_summary([])
    assert s["n"] == 0
    assert s["median_bias_pct"] is None
    assert s["mape_pct"] is None
    assert s["p25_pct"] is None
    assert s["p75_pct"] is None


def test_errors_summary_uses_median_abs_for_mape_not_mean() -> None:
    # signed errors with an asymmetric outlier: median |err| (=10) differs
    # sharply from the MEAN |err| (=40). The brief defines MAPE as the MEDIAN
    # absolute error, so we assert the robust median is used.
    signed = [10.0, 10.0, 10.0, 130.0]
    s = bt._errors_summary(signed)
    assert s["mape_pct"] == 10.0  # median(|10,10,10,130|) = 10, not mean 40
    assert s["median_bias_pct"] == 10.0  # median([10,10,10,130]) = 10


def test_errors_summary_signed_bias_can_be_negative() -> None:
    # Under-prediction → negative bias.
    s = bt._errors_summary([-20.0, -10.0, -30.0])
    assert s["median_bias_pct"] == -20.0
    assert s["mape_pct"] == 20.0  # median of |[-20,-10,-30]| = median[10,20,30]


# --------------------------------------------------------------------------- #
# _compute_metrics — signed/abs error %, bias, MAPE, per-rooms
# --------------------------------------------------------------------------- #


def test_compute_metrics_empty_overall_is_none_per_rooms_all_present() -> None:
    m = bt._compute_metrics([])
    assert m["overall"]["n"] == 0
    assert m["overall"]["median_bias_pct"] is None
    assert m["overall"]["mape_pct"] is None
    assert m["overall"]["n_no_analogs"] == 0
    # Every room bucket must still appear (with n=0) so the report renders.
    assert set(m["per_rooms"].keys()) == set(bt.ROOM_BUCKETS)
    for bucket in bt.ROOM_BUCKETS:
        assert m["per_rooms"][bucket]["n"] == 0
        assert m["per_rooms"][bucket]["median_bias_pct"] is None
        assert m["per_rooms"][bucket]["label"] == bt._rooms_label(bucket)


def test_compute_metrics_known_plus_22_pct_overprediction() -> None:
    # The headline finding: asking median over-predicts SOLD by ~+22%.
    # pred = 1.22 * sold for every row → signed error must be exactly +22%,
    # MAPE +22%, p25 == p75 == +22% (no spread).
    rows = [
        (122_000.0, 100_000.0, 1),
        (244_000.0, 200_000.0, 2),
        (366_000.0, 300_000.0, 3),
    ]
    m = bt._compute_metrics(rows)
    assert m["overall"]["n"] == 3
    assert m["overall"]["median_bias_pct"] == pytest.approx(22.0)
    assert m["overall"]["mape_pct"] == pytest.approx(22.0)
    assert m["overall"]["p25_pct"] == pytest.approx(22.0)
    assert m["overall"]["p75_pct"] == pytest.approx(22.0)


def test_compute_metrics_signed_error_formula() -> None:
    # Single row, hand-computed: 100*(150k-120k)/120k = +25.0%.
    m = bt._compute_metrics([(150_000.0, 120_000.0, 2)])
    assert m["overall"]["median_bias_pct"] == pytest.approx(25.0)
    assert m["overall"]["mape_pct"] == pytest.approx(25.0)


def test_compute_metrics_abs_error_distinct_from_signed() -> None:
    # Mixed over/under: signed bias near 0 but MAPE (median |err|) is positive.
    # rows: +50%, -50%, +50%, -50% → median signed in {-50,+50} band,
    # median |err| = 50.
    rows = [
        (150_000.0, 100_000.0, 1),  # +50
        (50_000.0, 100_000.0, 1),  # -50
        (150_000.0, 100_000.0, 1),  # +50
        (50_000.0, 100_000.0, 1),  # -50
    ]
    m = bt._compute_metrics(rows)
    assert m["overall"]["mape_pct"] == pytest.approx(50.0)
    # signed median of [-50,-50,50,50] = 0.0 (mean of two middles)
    assert m["overall"]["median_bias_pct"] == pytest.approx(0.0)


def test_compute_metrics_per_rooms_split_and_four_plus_collapse() -> None:
    rows = [
        (110_000.0, 100_000.0, 0),  # студия: +10
        (130_000.0, 100_000.0, 0),  # студия: +30  → median bucket 0 = +20
        (90_000.0, 100_000.0, 2),  # 2к: -10
        (200_000.0, 100_000.0, 5),  # 4+ (5 collapses): +100
        (300_000.0, 100_000.0, 4),  # 4+ : +200    → median bucket 4 = +150
    ]
    m = bt._compute_metrics(rows)

    assert m["per_rooms"][0]["n"] == 2
    assert m["per_rooms"][0]["median_bias_pct"] == pytest.approx(20.0)
    assert m["per_rooms"][0]["label"] == "студия"

    assert m["per_rooms"][2]["n"] == 1
    assert m["per_rooms"][2]["median_bias_pct"] == pytest.approx(-10.0)

    # rooms=5 and rooms=4 both land in bucket 4.
    assert m["per_rooms"][4]["n"] == 2
    assert m["per_rooms"][4]["median_bias_pct"] == pytest.approx(150.0)
    assert m["per_rooms"][4]["label"] == "4+"

    # buckets 1 and 3 had no rows.
    assert m["per_rooms"][1]["n"] == 0
    assert m["per_rooms"][3]["n"] == 0

    # overall n counts every matched row.
    assert m["overall"]["n"] == 5


def test_compute_metrics_drops_nonpositive_sold() -> None:
    # sold_ppm2 <= 0 cannot be divided → row dropped, not counted, no crash.
    rows = [
        (120_000.0, 0.0, 1),  # dropped
        (120_000.0, -5.0, 2),  # dropped
        (122_000.0, 100_000.0, 1),  # kept → +22
    ]
    m = bt._compute_metrics(rows)
    assert m["overall"]["n"] == 1
    assert m["overall"]["median_bias_pct"] == pytest.approx(22.0)


def test_compute_metrics_carries_no_analog_counts() -> None:
    rows = [(122_000.0, 100_000.0, 1)]
    m = bt._compute_metrics(
        rows,
        n_no_analogs=7,
        per_rooms_no_analogs={1: 4, 2: 3},
    )
    assert m["overall"]["n_no_analogs"] == 7
    assert m["per_rooms"][1]["n_no_analogs"] == 4
    assert m["per_rooms"][2]["n_no_analogs"] == 3
    # bucket with no skipped deals defaults to 0.
    assert m["per_rooms"][0]["n_no_analogs"] == 0


# --------------------------------------------------------------------------- #
# Rendering smoke tests — table + empty render must not crash.
# --------------------------------------------------------------------------- #


def test_render_table_runs_on_real_metrics() -> None:
    m = bt._compute_metrics([(122_000.0, 100_000.0, 1)], n_no_analogs=2)
    headline = {
        "deal_median_ppm2": 100_000.0,
        "ask_median_ppm2": 122_000.0,
        "spread_pct": 22.0,
    }
    out = bt._render_table(m, headline)
    assert "BACKTEST" in out
    assert "OVERALL" in out
    assert "+22.0" in out  # bias rendered with sign
    assert "100 000" in out  # ppm2 formatted with space thousands separator


def test_render_table_handles_empty_sample() -> None:
    m = bt._compute_metrics([])
    headline = {"deal_median_ppm2": None, "ask_median_ppm2": None, "spread_pct": None}
    out = bt._render_table(m, headline)
    assert "n/a" in out  # None metrics render as n/a, no crash


def test_fmt_helpers_handle_none_and_nan_safely() -> None:
    assert bt._fmt_pct(None) == "  n/a"
    assert bt._fmt_ppm2(None) == "n/a"
    # sanity: finite values format
    assert "+22" in bt._fmt_pct(22.0)
    assert not math.isnan(22.0)


# --------------------------------------------------------------------------- #
# argparse — defaults match the brief.
# --------------------------------------------------------------------------- #


def test_argparse_defaults() -> None:
    ns = bt._parse_args([])
    assert ns.sample == 300
    assert ns.since == "2025-06-01"
    assert ns.radius == 1000
    assert ns.rooms_tolerance == 0
    assert ns.json is False


def test_argparse_overrides() -> None:
    ns = bt._parse_args(
        ["--sample", "50", "--since", "2024-01-01", "--radius", "2000",
         "--rooms-tolerance", "1", "--json"]
    )
    assert ns.sample == 50
    assert ns.since == "2024-01-01"
    assert ns.radius == 2000
    assert ns.rooms_tolerance == 1
    assert ns.json is True