Spaces:
Configuration error
Configuration error
| """Persistence and the runtime lookup path, against the built database. | |
| These run on `data/catalog.db`; they are skipped if the index has not been built. | |
| """ | |
| from pathlib import Path | |
| import pytest | |
| from spareparts import config, index_db | |
| pytestmark = pytest.mark.skipif( | |
| not config.DB_PATH.exists(), reason="run `python -m spareparts.build_index` first" | |
| ) | |
| def conn(): | |
| connection = index_db.connect() | |
| yield connection | |
| connection.close() | |
| def _annotated(conn): | |
| return conn.execute("SELECT COUNT(*) FROM labels").fetchone()[0] > 0 | |
| def test_labels_and_components_are_paired(conn): | |
| if not _annotated(conn): | |
| pytest.skip("annotation stage has not been run") | |
| counts = index_db.annotation_counts(conn) | |
| assert counts["labels"] > 800 | |
| # Every label gets exactly one component row, even when unresolved. | |
| assert conn.execute( | |
| "SELECT COUNT(*) FROM labels l" | |
| " LEFT JOIN components c ON c.label_id = l.id WHERE c.label_id IS NULL" | |
| ).fetchone()[0] == 0 | |
| def test_every_label_references_a_known_part_and_page(conn): | |
| if not _annotated(conn): | |
| pytest.skip("annotation stage has not been run") | |
| assert conn.execute( | |
| "SELECT COUNT(*) FROM labels l" | |
| " LEFT JOIN parts p ON p.part_no = l.part_no WHERE p.part_no IS NULL" | |
| ).fetchone()[0] == 0 | |
| assert conn.execute( | |
| "SELECT COUNT(*) FROM labels l" | |
| " LEFT JOIN pages g ON g.pdf_page = l.pdf_page WHERE g.pdf_page IS NULL" | |
| ).fetchone()[0] == 0 | |
| def test_boxes_lie_inside_their_diagram(conn): | |
| if not _annotated(conn): | |
| pytest.skip("annotation stage has not been run") | |
| bad = conn.execute( | |
| """ | |
| SELECT COUNT(*) FROM components c | |
| JOIN labels l ON l.id = c.label_id | |
| JOIN pages p ON p.pdf_page = l.pdf_page | |
| WHERE c.cx0 IS NOT NULL AND ( | |
| c.cx0 < 0 OR c.cy0 < 0 OR c.cx1 > p.img_w OR c.cy1 > p.img_h | |
| OR c.cx1 <= c.cx0 OR c.cy1 <= c.cy0) | |
| """ | |
| ).fetchone()[0] | |
| assert bad == 0 | |
| def test_every_method_is_a_known_tier(conn): | |
| if not _annotated(conn): | |
| pytest.skip("annotation stage has not been run") | |
| methods = set(index_db.method_counts(conn)) | |
| assert methods <= {"leader", "assembly", "capped", "nearest", "override", "none"} | |
| def test_lookup_answers_part_to_page_box_method_confidence(conn): | |
| """The one query the runtime needs, with no OCR or image work repeated.""" | |
| if not _annotated(conn): | |
| pytest.skip("annotation stage has not been run") | |
| rows = index_db.lookup(conn, "30642") | |
| assert rows, "30642 is a real part with a callout on page 34" | |
| row = rows[0] | |
| assert row["pdf_page"] == 34 | |
| assert row["section_code"] == "07" | |
| assert row["method"] in {"leader", "assembly", "capped", "nearest", "override"} | |
| assert row["cx0"] is not None and row["cx1"] > row["cx0"] | |
| assert 0.0 <= row["confidence"] <= 1.0 | |
| assert Path(row["png_path"]).name == "p34.png" | |
| # The whole point of the phase: the box is the worm gear, not the digits. | |
| # A box may legitimately abut the callout it belongs to, so the test is that | |
| # it is a different, much larger region centred elsewhere -- not that the two | |
| # are disjoint. | |
| label_box = (row["lx0"], row["ly0"], row["lx1"], row["ly1"]) | |
| comp_box = (row["cx0"], row["cy0"], row["cx1"], row["cy1"]) | |
| label_area = (label_box[2] - label_box[0]) * (label_box[3] - label_box[1]) | |
| comp_area = (comp_box[2] - comp_box[0]) * (comp_box[3] - comp_box[1]) | |
| assert comp_area > 4 * label_area | |
| label_centre = ((label_box[0] + label_box[2]) / 2, (label_box[1] + label_box[3]) / 2) | |
| comp_centre = ((comp_box[0] + comp_box[2]) / 2, (comp_box[1] + comp_box[3]) / 2) | |
| assert not ( | |
| label_box[0] <= comp_centre[0] <= label_box[2] | |
| and label_box[1] <= comp_centre[1] <= label_box[3] | |
| ), "component box is centred on the printed number" | |
| assert abs(comp_centre[1] - label_centre[1]) > 50 | |
| def test_multi_page_part_returns_every_page(conn): | |
| if not _annotated(conn): | |
| pytest.skip("annotation stage has not been run") | |
| # 73290 is listed on nine pages. Callouts exist on a subset of those, and | |
| # also on page 44, where the part is drawn but absent from that page's table | |
| # (PSD.md 4.10 records this happening). Such a callout is still valid -- it | |
| # is snapped against the global part set -- but must be flagged. | |
| rows = index_db.lookup(conn, "73290") | |
| pages = {r["pdf_page"] for r in rows} | |
| listed = { | |
| r[0] for r in conn.execute( | |
| "SELECT pdf_page FROM occurrences WHERE part_no = '73290'" | |
| ) | |
| } | |
| assert pages | |
| for page in pages - listed: | |
| flag = conn.execute( | |
| "SELECT in_page_table FROM labels WHERE part_no = '73290' AND pdf_page = ?", | |
| (page,), | |
| ).fetchone()[0] | |
| assert flag == 0, f"page {page} is not in the table but was not flagged" | |
| # Every page it points at is a real catalogue page. | |
| catalogue = {r[0] for r in conn.execute("SELECT pdf_page FROM pages")} | |
| assert pages <= catalogue | |
| def test_normalized_view_is_resolution_independent(conn): | |
| if not _annotated(conn): | |
| pytest.skip("annotation stage has not been run") | |
| row = conn.execute( | |
| "SELECT * FROM normalized_boxes WHERE part_no = '30642' AND pdf_page = 34" | |
| ).fetchone() | |
| assert row is not None | |
| for field in ("label_x0", "label_y1", "comp_x0", "comp_y1"): | |
| assert 0.0 <= row[field] <= 1.0 | |
| page = conn.execute("SELECT img_w, img_h FROM pages WHERE pdf_page = 34").fetchone() | |
| native = conn.execute( | |
| "SELECT c.cx0 FROM labels l JOIN components c ON c.label_id = l.id" | |
| " WHERE l.part_no = '30642' AND l.pdf_page = 34" | |
| ).fetchone()[0] | |
| assert abs(row["comp_x0"] * page["img_w"] - native) < 1.0 | |
| def test_reannotation_does_not_duplicate(conn): | |
| if not _annotated(conn): | |
| pytest.skip("annotation stage has not been run") | |
| dupes = conn.execute( | |
| "SELECT COUNT(*) FROM (SELECT 1 FROM labels" | |
| " GROUP BY pdf_page, part_no, instance HAVING COUNT(*) > 1)" | |
| ).fetchone()[0] | |
| assert dupes == 0 | |