"""W17B floor-plan image smoke slice — minimal safety tests only. Codex specified: "테스트 너무 강하게 늘리지 말 것". This file keeps three safety-only checks (dry-run safety, caller monkeypatch seam, methodology grep). No coverage tests, no enum tests, no image-rendering tests. """ from __future__ import annotations import json import re import sys from pathlib import Path import pytest _REPO_ROOT = Path(__file__).resolve().parents[3] _SCRIPTS_DIR = _REPO_ROOT / "backend" / "scripts" if str(_SCRIPTS_DIR) not in sys.path: sys.path.insert(0, str(_SCRIPTS_DIR)) def _seed_w17a_run(run_dir: Path) -> Path: """Create a minimal W17A success run dir layout the W17B loader needs.""" src = run_dir / "w17a_fake" src.mkdir(parents=True, exist_ok=True) (src / "floor_plan_image_prompt.json").write_text(json.dumps({ "floor_plan_image_prompt_by_fp": { "fp_l05_01": { "t2i_prompt_text": ( "Schematic plan. Markers #1 #2 #3. Do not renumber, " "omit, or invent." ), "numbered_marker_legend": [ {"marker_number": 1, "source_candidate_number": 1, "element_kind": "spatial_unit", "visual_encoding": "filled_area", "label": "u1", "priority": "must_show", "source_refs": [], "must_be_legible": True}, {"marker_number": 2, "source_candidate_number": 2, "element_kind": "spatial_unit", "visual_encoding": "filled_area", "label": "u2", "priority": "must_show", "source_refs": [], "must_be_legible": True}, {"marker_number": 3, "source_candidate_number": 3, "element_kind": "spatial_unit", "visual_encoding": "filled_area", "label": "u3", "priority": "must_show", "source_refs": [], "must_be_legible": True}, ], }, }, })) (src / "run_meta.json").write_text(json.dumps({ "run_id": "w17a_fake", "stage": "w17a_floor_plan_image_prompt_slice", "run_status": "succeeded", "exit_code": 0, "stage_status": "generated", "model_used": "gpt-5.5", })) return src def test_w17b_dry_run_makes_no_api_call_and_writes_no_png(tmp_path): """Default (no --generate) must be a pure dry-run: no PNG, no openai import, image_api_call_count == 0.""" import experiment_floor_plan_image_smoke_slice as mod src = _seed_w17a_run(tmp_path) out_root = tmp_path / "out" exit_code = mod.main([ "--derive-image-smoke-from", str(src), "--target-fp-ids", "fp_l05_01", "--output-root", str(out_root), ]) assert exit_code == 0 runs = sorted(out_root.iterdir()) assert len(runs) == 1 run_dir = runs[0] meta = json.loads((run_dir / "run_meta.json").read_text()) assert meta["run_status"] == "succeeded" assert meta["stage_status"] == "dry_run" assert meta["image_api_call_count"] == 0 assert meta["image_generation_count"] == 0 assert not (run_dir / "png").exists() or not any((run_dir / "png").iterdir()) def test_w17b_generate_uses_monkeypatched_caller_once(tmp_path, monkeypatch): """--generate must (a) refuse to run when OPENAI_API_KEY is absent and the openai SDK import is skipped, (b) when patched with a fake caller, invoke it EXACTLY ONCE for fp_l05_01, (c) write the PNG bytes returned by the caller to run-local png/.""" import experiment_floor_plan_image_smoke_slice as mod # Fail-closed: no API key set AND `_load_backend_env` patched to a # no-op so the test does not silently pick up `.env`. --generate must # then exit with validation_failed. monkeypatch.delenv("OPENAI_API_KEY", raising=False) monkeypatch.setattr(mod, "_load_backend_env", lambda: None, raising=False) src = _seed_w17a_run(tmp_path) out_root = tmp_path / "out_nokey" exit_code = mod.main([ "--derive-image-smoke-from", str(src), "--target-fp-ids", "fp_l05_01", "--output-root", str(out_root), "--generate", ]) assert exit_code == 1 meta_nokey = json.loads( sorted(out_root.iterdir())[0].joinpath("run_meta.json").read_text() ) assert meta_nokey["run_status"] == "validation_failed" assert "openai_api_key" in str(meta_nokey.get("failed_invariants", [])).lower() \ or "OPENAI_API_KEY" in str(meta_nokey) # Now patch a fake caller. The script must call it exactly once. monkeypatch.setenv("OPENAI_API_KEY", "sk-test-fake-key") attempts = {"count": 0} def _fake_caller(*, fp_id, prompt, model, size, quality, target_path, client): attempts["count"] += 1 target_path.parent.mkdir(parents=True, exist_ok=True) target_path.write_bytes(b"\x89PNG\r\n\x1a\nFAKEW17B") return { "status": "success", "png_size_bytes": target_path.stat().st_size, "actual_api_response_meta": {"latency_ms": 5}, "error_meta": {}, "cost_meta": {}, } monkeypatch.setattr(mod, "_openai_caller", _fake_caller, raising=False) monkeypatch.setattr(mod, "_create_openai_client", lambda: object(), raising=False) out_root2 = tmp_path / "out_gen" exit_code2 = mod.main([ "--derive-image-smoke-from", str(src), "--target-fp-ids", "fp_l05_01", "--output-root", str(out_root2), "--generate", ]) assert exit_code2 == 0 assert attempts["count"] == 1, ( f"expected exactly 1 openai call, got {attempts['count']}" ) run_dir = sorted(out_root2.iterdir())[0] meta = json.loads((run_dir / "run_meta.json").read_text()) assert meta["run_status"] == "succeeded" assert meta["stage_status"] == "generated" assert meta["image_api_call_count"] == 1 assert meta["image_generation_count"] == 1 png_path = run_dir / "png" / "fp_l05_01.png" assert png_path.exists() and png_path.stat().st_size > 0 def test_w17b2_dry_run_artifact_carries_marker_mapping_block(tmp_path): """Dry-run artifact (generated_image_meta.json) must surface the assembled image prompt with a deterministic marker mapping block. W17B3 extension: each mapping line must also carry `unit=...` and `placement=...` columns sourced from the W15e candidate (chained via W17A run_meta.args.derive_image_prompt_from). image_api_call_count must remain 0.""" import experiment_floor_plan_image_smoke_slice as mod # Seed a fake W15e candidate dir AND a W17A run that points at it. w15e_dir = tmp_path / "w15e_fake" w15e_dir.mkdir() (w15e_dir / "floor_plan_prompt_candidate.json").write_text(json.dumps({ "candidate_floor_plans": { "fp_l05_01": { "fp_id": "fp_l05_01", "group_id_pointer": "Gx", "candidate_diagram_t2i_prompt": "schematic.", "candidate_key_elements": [], "candidate_numbered_elements": [ {"number": 1, "label": "u1 area", "category": "area", "position_hint": "central core", "unit_id_pointer": "U1"}, {"number": 2, "label": "u2 area", "category": "area", "position_hint": "northwest run", "unit_id_pointer": "U2"}, {"number": 3, "label": "u3 area", "category": "area", "position_hint": "east enclosed", "unit_id_pointer": "U3"}, ], "candidate_camera_recommendations": [], "reconciliation_notes_vs_production": "", }, }, })) src = tmp_path / "w17a_fake_chained" src.mkdir() (src / "floor_plan_image_prompt.json").write_text(json.dumps({ "floor_plan_image_prompt_by_fp": { "fp_l05_01": { "t2i_prompt_text": ( "Schematic plan. Markers #1 #2 #3. Do not renumber, " "omit, or invent." ), "numbered_marker_legend": [ {"marker_number": 1, "source_candidate_number": 1, "element_kind": "spatial_unit", "visual_encoding": "filled_area", "label": "u1 area", "priority": "must_show", "source_refs": [], "must_be_legible": True}, {"marker_number": 2, "source_candidate_number": 2, "element_kind": "spatial_unit", "visual_encoding": "filled_area", "label": "u2 area", "priority": "must_show", "source_refs": [], "must_be_legible": True}, {"marker_number": 3, "source_candidate_number": 3, "element_kind": "spatial_unit", "visual_encoding": "filled_area", "label": "u3 area", "priority": "must_show", "source_refs": [], "must_be_legible": True}, ], }, }, })) (src / "run_meta.json").write_text(json.dumps({ "run_id": "w17a_fake", "stage": "w17a_floor_plan_image_prompt_slice", "run_status": "succeeded", "exit_code": 0, "stage_status": "generated", "model_used": "gpt-5.5", "args": {"derive_image_prompt_from": str(w15e_dir)}, "derived_from": w15e_dir.name, })) out_root = tmp_path / "out_mapping" exit_code = mod.main([ "--derive-image-smoke-from", str(src), "--target-fp-ids", "fp_l05_01", "--output-root", str(out_root), ]) assert exit_code == 0 run_dir = sorted(out_root.iterdir())[0] meta = json.loads((run_dir / "run_meta.json").read_text()) assert meta["image_api_call_count"] == 0 art = json.loads((run_dir / "generated_image_meta.json").read_text()) fp = art["fp_smoke_result"] assembled = fp.get("assembled_prompt") or "" # Mapping block lines per W17A legend entry, with placement columns # from the W15e candidate join. for n in (1, 2, 3): assert f"#{n} =" in assembled, ( f"assembled prompt missing mapping line for #{n}" ) assert "kind=" in assembled and "visual=" in assembled # W17B3 additions: room-local unit + placement. assert "unit=" in assembled assert "placement=" in assembled assert "unit=U1" in assembled and "unit=U2" in assembled and "unit=U3" in assembled assert "placement=central core" in assembled # The high-priority marker contract banner must appear above the # original prompt body. assert "Marker mapping contract" in assembled # Side-legend discouragement clause must be present. assert "side legend" in assembled.lower() or "no independent" in assembled.lower() # Room-local placement guidance must be present (W17B3). assert ( "room-local placement" in assembled.lower() or "inside or on the boundary" in assembled.lower() ) def test_w17b_methodology_grep_no_scenario_specific_static_tokens(): """Static script body must not embed scenario-specific tokens. Per-char assembly so this assertion source does not match itself.""" script_path = _SCRIPTS_DIR / "experiment_floor_plan_image_smoke_slice.py" assert script_path.exists(), f"script missing: {script_path}" forbidden_tokens = [ "b" + "edroom", "ki" + "tchen", "blood" + "stain", "cur" + "tain", "coo" + "ktop", "tele" + "vision", "cri" + "me", "vi" + "lla", "roo" + "ftop", "foot" + "print", "pol" + "ice", "de" + "ck", "wheel" + "house", "ba" + "throom", "su" + "ri-young", ] pat = re.compile(r"(?i)\b(" + "|".join(forbidden_tokens) + r")\b") m = pat.search(script_path.read_text()) assert m is None, ( f"{script_path.name}: scenario-specific token leaked → " f"{m.group(0) if m else ''}" )