"""C1 perception_mode enum SOT v1 — integrated test (G1-G7). Spec ref: docs/superpowers/specs/2026-05-20-c1-perception-mode-enum-sot-v1-design.md Plan ref: docs/superpowers/plans/2026-05-20-c1-perception-mode-enum-sot-v1-implementation.md Codex APPROVED_FOR_EXECUTION. no live LLM + NO VLM + no regex semantic. """ import ast import json import pathlib import pytest from app.core.errors import AppError from app.core.perception_mode import ( PERCEPTION_MODES, REPRODUCTION_PERCEPTION_MODES, _PERCEPTION_GUIDES, _validate_perception_mode, is_reproduction_perception, get_perception_guide, ) from app.core.steps.render_prompt_card import build_id_policy REPO_ROOT = pathlib.Path(__file__).resolve().parents[3] def _resolve_latest_major_pack(stem: str, major: str) -> pathlib.Path: """N-3 fix: W1_UTC와 W3_UTC 분리 — glob resolve로 actual dir 찾기.""" candidates = sorted((REPO_ROOT / "prompts/_base" / stem).glob(f"{major}.*")) assert candidates, f"no prompt-pack found for {stem}/{major}.*" return candidates[-1] SHOT_STAGING_V14_DIR = _resolve_latest_major_pack("shot_staging", "14") SCENE_DETAIL_V29_DIR = _resolve_latest_major_pack("scene_detail", "29") # ── G1 — Producer schema enum 8-token strict (N-2: set equality) ── def test_g1_producer_schema_enum_8_token_strict(): schema = json.loads((SHOT_STAGING_V14_DIR / "schema.json").read_text()) shot_props = schema["properties"]["shots"]["items"]["properties"] enum_set = set(shot_props["perception_mode"]["enum"]) assert enum_set == PERCEPTION_MODES, f"enum mismatch: {enum_set} vs {PERCEPTION_MODES}" # ── G2 — mirror presence (N-5: helper-owned mapping + behavioral) ── def test_g2_mirror_presence_schema(): schema = json.loads((SHOT_STAGING_V14_DIR / "schema.json").read_text()) shot_props = schema["properties"]["shots"]["items"]["properties"] assert "mirror" in shot_props["perception_mode"]["enum"] def test_g2_mirror_presence_helper(): assert "mirror" in PERCEPTION_MODES assert "mirror" in REPRODUCTION_PERCEPTION_MODES assert "mirror" in _PERCEPTION_GUIDES def test_g2_mirror_behavioral(): guide = get_perception_guide("mirror") assert isinstance(guide, str) and len(guide) > 0 # ── G3 — Consumer code helper consume (no inline 4-tuple literal) ── def test_g3_render_prompt_card_no_inline_literal(): import re src = (REPO_ROOT / "backend/app/core/steps/render_prompt_card.py").read_text() forbidden = re.search( r'"reflection",\s*"mirror",\s*"through_device",\s*"projection"', src ) assert forbidden is None, "render_prompt_card.py still has 4-tuple inline literal" assert "is_reproduction_perception" in src # ── G4 — Cascade no generic fallback + AppError on unknown ── def test_g4_detail_steps_no_generic_fallback(): src = (REPO_ROOT / "backend/app/core/steps/detail_steps.py").read_text() assert "_perception_guides = {" not in src assert "_perception_guides.get(perception" not in src assert "get_perception_guide" in src # ── G5 — Consumer prompt cross-reference (no 4-token literal, both orders) ── def test_g5_scene_detail_v29_cross_reference(): sys_md = (SCENE_DETAIL_V29_DIR / "system.md").read_text() # N-13: 양쪽 order full-list residue 모두 fail assert "mirror / reflection / through_device / projection" not in sys_md, ( "old-order 4-token inline list residue" ) assert "reflection / mirror / through_device / projection" not in sys_md, ( "new-order 4-token inline list residue (N-13)" ) assert "REPRODUCTION_PERCEPTION_MODES" in sys_md or "perception_mode.py" in sys_md # ── G6 — Area C boundary preserve (scoped diff, N-3) ── def test_g6_area_c_boundary_anchors_preserve(): """Scoped marker/slice diff: directionality_class block + key_bg_elements required list.""" src = (REPO_ROOT / "backend/app/core/steps/render_prompt_card.py").read_text() assert "_ALLOWED_DIRECTIONALITY_CLASSES" in src assert "_REPRODUCTION_SURFACE_CLASSES" in src assert "directionality_class_missing" in src or "directionality_class_invalid" in src schema_v13 = json.loads( (REPO_ROOT / "prompts/_base/shot_staging/13.202605170422/schema.json").read_text() ) schema_v14 = json.loads((SHOT_STAGING_V14_DIR / "schema.json").read_text()) # N-2: schema nested under shots.items.properties try: v13_shot_props = schema_v13["properties"]["shots"]["items"]["properties"] v14_shot_props = schema_v14["properties"]["shots"]["items"]["properties"] dc_v13 = v13_shot_props["key_bg_elements"]["items"]["properties"]["directionality_class"] dc_v14 = v14_shot_props["key_bg_elements"]["items"]["properties"]["directionality_class"] except (KeyError, TypeError): pytest.fail("directionality_class block extraction failed; schema structure changed") assert dc_v13 == dc_v14, ( f"Area C directionality_class block diff non-empty: {dc_v13} vs {dc_v14}" ) req_v13 = set(v13_shot_props["key_bg_elements"]["items"]["required"]) req_v14 = set(v14_shot_props["key_bg_elements"]["items"]["required"]) assert "directionality_class" in req_v13 assert "directionality_class" in req_v14 assert req_v13 == req_v14 # ── G7 — NO VLM canary (AST-based, N-8) + N-1 invalid input + N-7 direct ── def test_g7_no_vlm_helper_module(): """AST-based import/call check (N-8: stricter than substring grep).""" tree = ast.parse((REPO_ROOT / "backend/app/core/perception_mode.py").read_text()) forbidden = {"vlm", "vision_lm", "vision_language", "lvm_judge", "lvm_validator"} for node in ast.walk(tree): if isinstance(node, (ast.Import, ast.ImportFrom)): for alias in node.names: name_lower = (alias.name or "").lower() for kw in forbidden: assert kw not in name_lower, f"VLM import leak in helper: {alias.name}" if isinstance(node, ast.ImportFrom): mod_lower = (node.module or "").lower() for kw in forbidden: assert kw not in mod_lower, f"VLM module leak: {node.module}" if isinstance(node, ast.Call): if isinstance(node.func, ast.Name): for kw in forbidden: assert kw not in node.func.id.lower(), f"VLM call leak: {node.func.id}" def test_g7_no_vlm_test_module(): """AST-based check on this test file itself (N-8: tautological substring 회피).""" tree = ast.parse(pathlib.Path(__file__).read_text()) forbidden = {"vlm", "vision_lm", "vision_language", "lvm_judge", "lvm_validator"} for node in ast.walk(tree): if isinstance(node, (ast.Import, ast.ImportFrom)): for alias in node.names: name_lower = (alias.name or "").lower() for kw in forbidden: assert kw not in name_lower, f"VLM import leak in test: {alias.name}" if isinstance(node, ast.ImportFrom): mod_lower = (node.module or "").lower() for kw in forbidden: assert kw not in mod_lower, f"VLM module leak: {node.module}" if isinstance(node, ast.Call): if isinstance(node.func, ast.Name): for kw in forbidden: assert kw not in node.func.id.lower(), f"VLM call leak: {node.func.id}" @pytest.mark.parametrize("invalid_pm", ["future_unknown_mode", "", "MIRROR"]) def test_g7_invalid_input_fail_fast(invalid_pm): """N-1: invalid input → AppError perception_mode.unknown (not silent false/fallback).""" with pytest.raises(AppError) as exc_info: is_reproduction_perception(invalid_pm) assert exc_info.value.code == "perception_mode.unknown" with pytest.raises(AppError) as exc_info: get_perception_guide(invalid_pm) assert exc_info.value.code == "perception_mode.unknown" def test_g7_direct_guide_not_applicable(): """N-7: get_perception_guide("direct") → AppError guide_not_applicable (not raw KeyError).""" with pytest.raises(AppError) as exc_info: get_perception_guide("direct") assert exc_info.value.code == "perception_mode.guide_not_applicable", ( f"expected guide_not_applicable, got {exc_info.value.code}" ) def test_g7_non_direct_all_have_guide(): """Behavioral: non-direct 7-key all return non-empty guide.""" for pm in PERCEPTION_MODES - {"direct"}: guide = get_perception_guide(pm) assert isinstance(guide, str) and len(guide) > 0, f"empty guide for {pm}" # ── Invariants (spec §4) ── def test_invariant_strict_subset(): assert REPRODUCTION_PERCEPTION_MODES < PERCEPTION_MODES def test_invariant_guide_coverage(): assert set(_PERCEPTION_GUIDES) == PERCEPTION_MODES - {"direct"} def test_validate_perception_mode_direct_ok(): """direct is enum-valid (no raise from _validate_perception_mode).""" _validate_perception_mode("direct") # no raise assert is_reproduction_perception("direct") is False # ── G8 — build_id_policy consumer-boundary (Wave4, e2e-review-fix-v1) ── # render_prompt_card.build_id_policy 가 perception_mode 를 helper 로 넘기는 consumer. # 이전 render_prompt_card.py 의 `(perception_mode or "direct").lower()` 가 비정규 # casing 을 silent 정규화해 helper fail-fast 를 무력화했음 — Wave4 가 `.lower()` 를 # 제거. 본 G8 은 그 consumer-boundary 동작을 회귀 가드 (helper-level G7 과 별개). def _call_build_id_policy(perception_mode): """build_id_policy minimal-valid 호출 — perception_mode 분기까지 도달.""" return build_id_policy( visible_entities=[], outlook_pairs=[], perception_mode=perception_mode, key_bg_elements=[], subject_reference_policies=[], ) @pytest.mark.parametrize("bad_casing", ["MIRROR", "Reflection", "DIRECT"]) def test_g8_build_id_policy_noncanonical_casing_fail_fast(bad_casing): """Wave4: build_id_policy 가 비정규 casing perception_mode 를 silent 정규화하지 않고 helper perception_mode.unknown 으로 fail-fast (이전 .lower() 무력화 회귀 가드).""" with pytest.raises(AppError) as exc_info: _call_build_id_policy(bad_casing) assert exc_info.value.code == "perception_mode.unknown" @pytest.mark.parametrize("legacy_empty", [None, ""]) def test_g8_build_id_policy_legacy_empty_coerced_to_direct(legacy_empty): """Wave4: None / empty perception_mode 는 "direct" 로 coerce (legacy caller responsibility 보존) — raise 없음 + reproduction-surface constraint 부재.""" policy = _call_build_id_policy(legacy_empty) joined = " ".join(policy["constraints"]) assert "reflective/projected surface" not in joined, ( "direct 모드인데 reproduction-surface constraint 가 잘못 부착됨" ) @pytest.mark.parametrize( "canonical_repro", ["mirror", "reflection", "through_device", "projection"] ) def test_g8_build_id_policy_canonical_reproduction_still_works(canonical_repro): """Wave4: canonical lowercase reproduction perception_mode 는 정상 동작 — reproduction-surface constraint 부착 (.lower() 제거가 정상 경로 무손상 회귀 가드).""" policy = _call_build_id_policy(canonical_repro) joined = " ".join(policy["constraints"]) assert "reflective/projected surface" in joined, ( f"{canonical_repro}: reproduction-surface constraint 누락" )