"""B04 등고선 추출 엔진.

5종 표현(regular_grid/triangular_mesh/bspline_surface/local_rbf_height_field/
meshfree_surfels)의 npz 모델에서 표고 격자를 환원하고, marching squares로
지정 간격 등고선 라인을 추출한다. DTM valid_mask를 footprint로 사용해
경계 누출을 차단한다.
"""

from pathlib import Path
from typing import Any

import numpy as np
from scipy.interpolate import RBFInterpolator
from skimage import measure

# 등고선 캐시 형식/추출 규칙이 바뀔 때 증가시킨다.
CONTOUR_EXTRACTOR_VERSION = 4


def extract_contours_from_grid(
    x_coords: np.ndarray,
    y_coords: np.ndarray,
    z_grid: np.ndarray,
    valid_mask: np.ndarray | None,
    interval: float,
    min_interval: float = 0.5,
    scene_center: tuple[float, float, float] | None = None,
) -> list[dict[str, Any]]:
    """정규 표고 격자로부터 등고선 라인을 추출한다."""
    interval = max(interval, min_interval)
    finite_mask = np.isfinite(z_grid)
    if valid_mask is not None:
        finite_mask &= valid_mask
    if not finite_mask.any():
        return []

    z_min = float(np.min(z_grid[finite_mask]))
    z_max = float(np.max(z_grid[finite_mask]))
    span = z_max - z_min
    if span <= 0:
        return []
    # 레벨 수 상한은 arange **앞에서** 건다. 뒤에 두면 표고 범위가 비정상적으로 넓을 때
    # np.arange가 배열을 만들다 "Maximum allowed size exceeded"로 먼저 터진다.
    if span / interval > 500:
        interval = span / 100.0
    start_level = np.ceil(z_min / interval) * interval
    levels = np.arange(start_level, z_max, interval)
    if len(levels) == 0:
        return []

    contours_geojson_list: list[dict[str, Any]] = []

    # marching squares의 NaN 문제 예방: 무효 영역을 sentinel(z_min-1000)로 채운다.
    z_grid_masked = z_grid.copy()
    if valid_mask is not None:
        z_grid_masked[~valid_mask] = z_min - 1000.0
    invalid_mask = ~np.isfinite(z_grid_masked)
    if invalid_mask.any():
        z_grid_masked[invalid_mask] = z_min - 1000.0

    cx, cy, cz = scene_center if scene_center is not None else (0.0, 0.0, 0.0)

    for level in levels:
        for contour in measure.find_contours(z_grid_masked, level):
            current_segment: list[list[float]] = []
            for y_idx, x_idx in contour:
                x_idx_c = np.clip(x_idx, 0, len(x_coords) - 1)
                y_idx_c = np.clip(y_idx, 0, len(y_coords) - 1)
                x_0, x_1 = int(np.floor(x_idx_c)), int(np.ceil(x_idx_c))
                y_0, y_1 = int(np.floor(y_idx_c)), int(np.ceil(y_idx_c))

                is_valid = True
                if valid_mask is not None and not (
                    valid_mask[y_0, x_0]
                    and valid_mask[y_0, x_1]
                    and valid_mask[y_1, x_0]
                    and valid_mask[y_1, x_1]
                ):
                    is_valid = False

                if not is_valid:
                    if len(current_segment) >= 2:
                        mid_idx = len(current_segment) // 2
                        contours_geojson_list.append(
                            {
                                "level": float(level),
                                "coordinates": current_segment,
                                "label_position": current_segment[mid_idx],
                            }
                        )
                    current_segment = []
                    continue

                tx = x_idx_c - x_0
                ty = y_idx_c - y_0
                x_val = (1.0 - tx) * x_coords[x_0] + tx * x_coords[x_1]
                y_val = (1.0 - ty) * y_coords[y_0] + ty * y_coords[y_1]

                if scene_center is not None:
                    current_segment.append(
                        [
                            round(float(x_val - cx), 3),
                            round(float(level - cz), 3),
                            round(float(-(y_val - cy)), 3),
                        ]
                    )
                else:
                    current_segment.append(
                        [round(float(x_val), 3), round(float(y_val), 3), round(float(level), 3)]
                    )

            if len(current_segment) >= 2:
                mid_idx = len(current_segment) // 2
                contours_geojson_list.append(
                    {
                        "level": float(level),
                        "coordinates": current_segment,
                        "label_position": current_segment[mid_idx],
                    }
                )

    return contours_geojson_list


def _load_footprint_mask(
    model_npz_path: Path, x_coords: np.ndarray, y_coords: np.ndarray
) -> np.ndarray | None:
    """같은 source filter의 DTM valid_mask를 현재 격자에 최근접 리샘플한다."""
    stem = Path(model_npz_path).stem
    if stem.endswith("_smooth"):
        stem = stem[:-7]
    parts = stem.split("_", 1)
    if len(parts) < 2:
        return None
    filter_key = parts[1]
    dtm_path = Path(model_npz_path).parent / f"dtm_{filter_key}.npz"
    if not dtm_path.exists():
        return None
    try:
        d = np.load(dtm_path)
        dtm_x = np.asarray(d["x"]).ravel()
        dtm_y = np.asarray(d["y"]).ravel()
        dtm_mask = np.asarray(d["valid_mask"], dtype=bool)
    except Exception:
        return None
    if len(dtm_x) < 2 or len(dtm_y) < 2:
        return None

    def _nearest_idx(axis: np.ndarray, coords: np.ndarray) -> np.ndarray:
        ascending = bool(axis[0] <= axis[-1])
        a = axis if ascending else axis[::-1]
        idx = np.clip(np.searchsorted(a, coords), 1, len(a) - 1)
        idx = np.where(np.abs(a[idx - 1] - coords) <= np.abs(a[idx] - coords), idx - 1, idx)
        return idx if ascending else (len(axis) - 1 - idx)

    xi = _nearest_idx(dtm_x, np.asarray(x_coords, dtype=np.float64))
    yi = _nearest_idx(dtm_y, np.asarray(y_coords, dtype=np.float64))
    return dtm_mask[np.ix_(yi, xi)]


def _apply_footprint(
    model_npz_path: Path, x_coords: np.ndarray, y_coords: np.ndarray, valid_mask: np.ndarray
) -> np.ndarray:
    """valid_mask에 DTM footprint를 교집합으로 적용한다 (형상 다르면 최근접 리샘플)."""
    fp = _load_footprint_mask(model_npz_path, x_coords, y_coords)
    if fp is not None:
        if fp.shape == valid_mask.shape:
            return valid_mask & fp
        from scipy.ndimage import zoom

        zoom_y = valid_mask.shape[0] / fp.shape[0]
        zoom_x = valid_mask.shape[1] / fp.shape[1]
        fp_resized = zoom(fp.astype(float), (zoom_y, zoom_x), order=0) > 0.5
        if fp_resized.shape == valid_mask.shape:
            return valid_mask & fp_resized
    return valid_mask


def _tin_face_coverage_mask(
    vertices: np.ndarray, faces: np.ndarray, xx: np.ndarray, yy: np.ndarray
) -> np.ndarray:
    """저장된 TIN 면이 실제로 덮는 XY 영역만 True로 반환한다."""
    vertices = np.asarray(vertices)
    faces = np.asarray(faces, dtype=np.int64)
    if vertices.ndim != 2 or vertices.shape[1] < 2 or not len(faces):
        return np.zeros(xx.shape, dtype=bool)

    edges = np.vstack((faces[:, [0, 1]], faces[:, [1, 2]], faces[:, [2, 0]]))
    edges = np.sort(edges, axis=1)
    unique_edges, counts = np.unique(edges, axis=0, return_counts=True)
    boundary_edges = unique_edges[counts == 1]
    if not len(boundary_edges):
        return np.zeros(xx.shape, dtype=bool)

    import affine
    import rasterio.features
    from shapely import get_parts, linestrings, polygonize

    boundary_lines = linestrings(vertices[boundary_edges, :2])
    polygons = list(get_parts(polygonize(boundary_lines)))
    if not polygons:
        return np.zeros(xx.shape, dtype=bool)

    x_coords = xx[0, :]
    y_coords = yy[:, 0]
    dx = float(x_coords[1] - x_coords[0]) if len(x_coords) > 1 else 1.0
    dy = float(y_coords[1] - y_coords[0]) if len(y_coords) > 1 else 1.0

    transform = affine.Affine(dx, 0.0, x_coords[0] - dx / 2.0, 0.0, dy, y_coords[0] - dy / 2.0)

    mask = rasterio.features.rasterize(
        polygons, out_shape=xx.shape, transform=transform, fill=0, default_value=1, dtype="uint8"
    )
    return mask.astype(bool)


def _grid_axes(x_min: float, x_max: float, y_min: float, y_max: float, target_grid_m: float):
    cols = max(2, int(np.ceil((x_max - x_min) / target_grid_m)) + 1)
    rows = max(2, int(np.ceil((y_max - y_min) / target_grid_m)) + 1)
    x_coords = np.linspace(x_min, x_max, cols, dtype=np.float32)
    y_coords = np.linspace(y_min, y_max, rows, dtype=np.float32)
    return x_coords, y_coords


# 후보 셀 배열이 한 번에 커지지 않도록 삼각형을 나눠 처리하는 단위.
_RASTER_CHUNK = 200_000
# 격자점이 변 위에 놓였을 때 양쪽 삼각형 모두에서 탈락하지 않도록 두는 여유.
_BARYCENTRIC_EPS = 1e-9


def rasterize_triangle_mesh(
    vertices: np.ndarray,
    triangles: np.ndarray,
    x_coords: np.ndarray,
    y_coords: np.ndarray,
) -> np.ndarray:
    """삼각망을 규칙격자에 직접 래스터화해 표고 격자를 만든다.

    삼각형마다 자기가 덮는 격자 셀만 무게중심 선형보간으로 채우므로, 질의점별
    위치탐색(scipy `find_simplex`)이 사라진다. 삼각망 밖 셀은 NaN으로 남아
    기존 `griddata` 결과와 같은 유효 판정을 유지한다.
    """
    rows, cols = len(y_coords), len(x_coords)
    z_grid = np.full((rows, cols), np.nan, dtype=np.float64)
    vertices = np.asarray(vertices, dtype=np.float64)
    triangles = np.asarray(triangles, dtype=np.int64)
    if not len(triangles) or vertices.ndim != 2 or vertices.shape[1] < 3:
        return z_grid

    axis_x = np.asarray(x_coords, dtype=np.float64)
    axis_y = np.asarray(y_coords, dtype=np.float64)

    def _index_range(axis: np.ndarray, low: np.ndarray, high: np.ndarray, size: int):
        """축 배열을 직접 탐색해 [low, high]를 덮는 인덱스 구간을 낸다.

        `(좌표 - 원점) / 간격` 식은 쓰지 않는다. `_grid_axes`가 float32 축을 만드는데
        UTM 좌표(18만대)에서 float32 해상도가 0.015625m라 격자 간격이 0.984~1.0으로
        흔들리고, 등간격으로 환산하면 최대 0.81셀까지 어긋난다. 삼각형이 평균 0.3m라
        그만큼 어긋나면 격자점을 통째로 놓친다 (2026-08-17 실측).
        """
        ascending = size < 2 or axis[-1] >= axis[0]
        sorted_axis = axis if ascending else axis[::-1]
        lo = np.searchsorted(sorted_axis, low, side="left")
        hi = np.searchsorted(sorted_axis, high, side="right") - 1
        if ascending:
            return lo, hi
        return size - 1 - hi, size - 1 - lo

    for begin in range(0, len(triangles), _RASTER_CHUNK):
        chunk = triangles[begin : begin + _RASTER_CHUNK]
        corners = vertices[chunk]
        ax, ay, az = corners[:, 0, 0], corners[:, 0, 1], corners[:, 0, 2]
        bx, by, bz = corners[:, 1, 0], corners[:, 1, 1], corners[:, 1, 2]
        cx, cy, cz = corners[:, 2, 0], corners[:, 2, 1], corners[:, 2, 2]

        denom = (by - cy) * (ax - cx) + (cx - bx) * (ay - cy)
        col_lo, col_hi = _index_range(
            axis_x, np.minimum(np.minimum(ax, bx), cx), np.maximum(np.maximum(ax, bx), cx), cols
        )
        row_lo, row_hi = _index_range(
            axis_y, np.minimum(np.minimum(ay, by), cy), np.maximum(np.maximum(ay, by), cy), rows
        )

        # 클립 전에 격자 밖 삼각형을 걸러야 경계에 눌러붙지 않는다.
        selected = (
            np.isfinite(denom)
            & (np.abs(denom) > 1e-12)
            & (col_hi >= col_lo)
            & (row_hi >= row_lo)
            & (col_hi >= 0)
            & (row_hi >= 0)
            & (col_lo <= cols - 1)
            & (row_lo <= rows - 1)
        )
        if not selected.any():
            continue

        col_lo = np.clip(col_lo[selected], 0, cols - 1).astype(np.int64)
        col_hi = np.clip(col_hi[selected], 0, cols - 1).astype(np.int64)
        row_lo = np.clip(row_lo[selected], 0, rows - 1).astype(np.int64)
        row_hi = np.clip(row_hi[selected], 0, rows - 1).astype(np.int64)
        width = col_hi - col_lo + 1
        counts = width * (row_hi - row_lo + 1)
        total = int(counts.sum())
        if total <= 0:
            continue

        # 삼각형별 후보 셀 수가 제각각이라 반복 인덱스를 펼쳐서 한 번에 계산한다.
        starts = np.concatenate(([0], np.cumsum(counts)[:-1]))
        local = np.arange(total, dtype=np.int64) - np.repeat(starts, counts)
        width_rep = np.repeat(width, counts)
        offset_row = local // width_rep
        cell_row = np.repeat(row_lo, counts) + offset_row
        cell_col = np.repeat(col_lo, counts) + (local - offset_row * width_rep)

        point_x = axis_x[cell_col]
        point_y = axis_y[cell_row]
        ax_r, ay_r, az_r = (
            np.repeat(ax[selected], counts),
            np.repeat(ay[selected], counts),
            np.repeat(az[selected], counts),
        )
        bx_r, by_r, bz_r = (
            np.repeat(bx[selected], counts),
            np.repeat(by[selected], counts),
            np.repeat(bz[selected], counts),
        )
        cx_r, cy_r, cz_r = (
            np.repeat(cx[selected], counts),
            np.repeat(cy[selected], counts),
            np.repeat(cz[selected], counts),
        )
        denom_r = np.repeat(denom[selected], counts)

        weight_a = ((by_r - cy_r) * (point_x - cx_r) + (cx_r - bx_r) * (point_y - cy_r)) / denom_r
        weight_b = ((cy_r - ay_r) * (point_x - cx_r) + (ax_r - cx_r) * (point_y - cy_r)) / denom_r
        weight_c = 1.0 - weight_a - weight_b
        hit = (
            (weight_a >= -_BARYCENTRIC_EPS)
            & (weight_b >= -_BARYCENTRIC_EPS)
            & (weight_c >= -_BARYCENTRIC_EPS)
        )
        if not hit.any():
            continue
        z_grid[cell_row[hit], cell_col[hit]] = (
            weight_a[hit] * az_r[hit] + weight_b[hit] * bz_r[hit] + weight_c[hit] * cz_r[hit]
        )

    return z_grid


def extract_contours(
    model_npz_path: Path,
    representation: str,
    interval: float,
    target_grid_m: float = 1.0,
    scene_center: tuple[float, float, float] | None = None,
) -> list[dict[str, Any]]:
    """표현별 npz 모델에서 표고 격자를 환원한 뒤 등고선 리스트를 추출한다."""
    model_npz_path = Path(model_npz_path)
    if not model_npz_path.exists():
        raise FileNotFoundError(f"모델 파일이 존재하지 않습니다: {model_npz_path}")
    data = np.load(model_npz_path)

    if representation == "regular_grid":
        x_coords, y_coords, z_grid, valid_mask = (
            data["x"],
            data["y"],
            data["z"],
            data["valid_mask"],
        )
        current_res = (x_coords[-1] - x_coords[0]) / (len(x_coords) - 1)
        step = max(1, int(round(target_grid_m / current_res)))
        if step > 1:
            return extract_contours_from_grid(
                x_coords[::step],
                y_coords[::step],
                z_grid[::step, ::step],
                valid_mask[::step, ::step],
                interval,
                scene_center=scene_center,
            )
        return extract_contours_from_grid(
            x_coords, y_coords, z_grid, valid_mask, interval, scene_center=scene_center
        )

    if representation == "triangular_mesh":
        vertices, faces = data["vertices"], data["faces"]
        x_min, x_max = float(np.min(vertices[:, 0])), float(np.max(vertices[:, 0]))
        y_min, y_max = float(np.min(vertices[:, 1])), float(np.max(vertices[:, 1]))
        x_coords, y_coords = _grid_axes(x_min, x_max, y_min, y_max, target_grid_m)
        xx, yy = np.meshgrid(x_coords, y_coords)
        # 저장된 faces를 그대로 쓴다 — griddata는 정점을 다시 삼각분할하느라 긴 변
        # 제거·외곽 클리핑이 반영되지 않은 면까지 되살렸다 (PLAN 2026-08-17).
        z_grid = rasterize_triangle_mesh(vertices, faces, x_coords, y_coords)
        face_mask = _tin_face_coverage_mask(vertices, faces, xx, yy)
        valid_mask = np.isfinite(z_grid) & face_mask
        valid_mask = _apply_footprint(model_npz_path, x_coords, y_coords, valid_mask)
        return extract_contours_from_grid(
            x_coords, y_coords, z_grid, valid_mask, interval, scene_center=scene_center
        )

    if representation == "bspline_surface":
        # 모델 빌더와 같은 곡면을 다시 만든다 — 평활 계수·발산 방어를 공유한다.
        from B04_PreProcess.B04_PreProcess_Engine_ModelContext import (
            evaluate_nurbs_spline,
            fit_nurbs_spline,
        )

        control_x, control_y, control_z = data["control_x"], data["control_y"], data["control_z"]
        degree = int(data["degree"][0])
        spline, z_range = fit_nurbs_spline(control_x, control_y, control_z, degree)
        x_coords, y_coords = _grid_axes(
            float(control_x[0]),
            float(control_x[-1]),
            float(control_y[0]),
            float(control_y[-1]),
            target_grid_m,
        )
        z_grid = evaluate_nurbs_spline(
            spline, y_coords, x_coords, z_range, Path(model_npz_path).stem
        ).astype(np.float32)
        valid_mask = _apply_footprint(
            model_npz_path, x_coords, y_coords, np.ones_like(z_grid, dtype=bool)
        )
        return extract_contours_from_grid(
            x_coords, y_coords, z_grid, valid_mask, interval, scene_center=scene_center
        )

    if representation == "local_rbf_height_field":
        centers_xy, center_z = data["centers_xy"], data["center_z"]
        smoothing = float(data["smoothing"][0])
        interpolator = RBFInterpolator(
            centers_xy.astype(np.float64),
            center_z.astype(np.float64),
            neighbors=min(64, len(centers_xy)),
            smoothing=smoothing,
            kernel="thin_plate_spline",
        )
        x_min, x_max = float(np.min(centers_xy[:, 0])), float(np.max(centers_xy[:, 0]))
        y_min, y_max = float(np.min(centers_xy[:, 1])), float(np.max(centers_xy[:, 1]))
        x_coords, y_coords = _grid_axes(x_min, x_max, y_min, y_max, target_grid_m)
        xx, yy = np.meshgrid(x_coords, y_coords)
        z_values = interpolator(np.column_stack([xx.ravel(), yy.ravel()])).astype(np.float32)
        z_grid = z_values.reshape(len(y_coords), len(x_coords))
        valid_mask = _apply_footprint(
            model_npz_path, x_coords, y_coords, np.ones_like(z_grid, dtype=bool)
        )
        return extract_contours_from_grid(
            x_coords, y_coords, z_grid, valid_mask, interval, scene_center=scene_center
        )

    if representation == "meshfree_surfels":
        from scipy.spatial import Delaunay

        points = data["points"]
        x_min, x_max = float(np.min(points[:, 0])), float(np.max(points[:, 0]))
        y_min, y_max = float(np.min(points[:, 1])), float(np.max(points[:, 1]))
        x_coords, y_coords = _grid_axes(x_min, x_max, y_min, y_max, target_grid_m)
        # Delaunay 1회로 삼각분할과 hull 판정을 함께 해결한다. 기존에는 griddata가
        # 내부에서 한 번, hull 판정용으로 또 한 번 만들었다 (PLAN 2026-08-17).
        triangulation = Delaunay(np.asarray(points[:, :2], dtype=np.float64))
        z_grid = rasterize_triangle_mesh(points, triangulation.simplices, x_coords, y_coords)
        # hull 밖은 어느 삼각형에도 안 닿아 NaN으로 남는다 = 기존 hull_inside와 같다.
        valid_mask = np.isfinite(z_grid)
        valid_mask = _apply_footprint(model_npz_path, x_coords, y_coords, valid_mask)
        return extract_contours_from_grid(
            x_coords, y_coords, z_grid, valid_mask, interval, scene_center=scene_center
        )

    raise ValueError(f"지원하지 않는 표현 방식입니다: {representation}")