from __future__ import annotations

from PIL import Image, ImageDraw

from api.services.docling import extract
from api.services.docling.regions import detect_response_regions_from_image


def test_detects_grouped_answer_lines() -> None:
    image = Image.new("RGB", (900, 1200), "white")
    draw = ImageDraw.Draw(image)
    for y in (420, 470, 520):
        draw.line((160, y, 760, y), fill="black", width=3)

    candidates = detect_response_regions_from_image(image, page_index=2)

    line_regions = [c.to_mapper_dict() for c in candidates if c.region_type == "answer_lines"]
    assert line_regions
    best = line_regions[0]
    assert best["kind"] == "response"
    assert best["source"] == "ai"
    assert best["confirmed"] is False
    assert best["page_index"] == 2
    assert best["line_count"] == 3
    assert best["bbox"]["coord_origin"] == "TOPLEFT"
    assert best["bbox"]["w"] > 550
    assert best["bbox"]["h"] > 80


def test_detects_answer_box() -> None:
    image = Image.new("RGB", (900, 1200), "white")
    draw = ImageDraw.Draw(image)
    draw.rectangle((140, 300, 780, 520), outline="black", width=3)

    candidates = detect_response_regions_from_image(image, page_index=0)

    boxes = [c.to_mapper_dict() for c in candidates if c.region_type == "answer_box"]
    assert boxes
    assert boxes[0]["bbox"]["w"] > 600
    assert boxes[0]["bbox"]["h"] > 200


def test_detect_response_region_taxonomy_for_lines_and_boxes():
    img = Image.new("RGB", (800, 1000), "white")
    draw = ImageDraw.Draw(img)
    for y in (220, 260, 300):
        draw.line((120, y, 680, y), fill="black", width=2)
    draw.rectangle((140, 520, 660, 640), outline="black", width=3)

    regions = detect_response_regions_from_image(img, min_confidence=0.1)
    types = {r.region_type for r in regions}

    assert "answer_lines" in types
    assert "answer_box" in types


def test_attach_detected_response_regions_normalizes_taxonomy_and_uses_pdf_y(monkeypatch, tmp_path):
    pdf = tmp_path / "paper.pdf"
    pdf.write_bytes(b"%PDF test placeholder")
    parts = {
        "01.1": {"q": "01", "page": 1, "bbox": {"l": 50, "r": 80, "t": 700, "b": 680}, "regions": []},
        "01.2": {"q": "01", "page": 1, "bbox": {"l": 50, "r": 80, "t": 500, "b": 480}, "regions": []},
    }

    def fake_detect(path, min_confidence=0.32):
        return [{
            "page_index": 0,
            "region_type": "answer-box",
            "confidence": 0.77,
            "bbox": {"x": 100, "y": 335, "w": 500, "h": 40},
            "detection_method": "test",
            "meta": {"page_height_px": 1000, "page_height_pdf": 800},
        }]

    monkeypatch.setattr(extract.region_mod, "detect_response_regions_from_pdf", fake_detect)

    attached, candidates = extract.attach_detected_response_regions(parts, str(pdf))

    assert attached == 1
    assert len(candidates) == 1
    assert parts["01.1"]["regions"] == []
    assert parts["01.2"]["regions"][0]["type"] == "answer_box"
    assert parts["01.2"]["regions"][0]["source"] == "opencv"