diff --git a/app/application/classification/migration.py b/app/application/classification/migration.py index 5f8d3b825..b29aa81b0 100644 --- a/app/application/classification/migration.py +++ b/app/application/classification/migration.py @@ -121,7 +121,7 @@ class LegacyClassificationMigrationResult: @dataclass(frozen=True, slots=True) class _LegacyToken: - """保留一个旧逗号项展开后的值集合及其排除语义。""" + """保留旧字段展开后的同向值集合及其排除语义。""" negative: bool values: tuple[str, ...] @@ -530,26 +530,20 @@ def _legacy_field_definition( def _parse_legacy_tokens(value: str) -> tuple[tuple[_LegacyToken, ...], bool]: - """逐项复现旧逗号、排除前缀和连字符范围展开算法。""" + """沿用旧范围展开语义，并合并同向枚举，避免值数量膨胀为叶子数量。""" raw_tokens = [item for item in value.split(",") if item] - parsed: list[_LegacyToken] = [] + values_by_sign: dict[bool, list[str]] = {} requires_exists = not raw_tokens for raw_token in raw_tokens: expanded = _expand_legacy_token(raw_token) if not expanded: requires_exists = True continue - grouped: list[_LegacyToken] = [] for expanded_value in expanded: negative = expanded_value.startswith("!") plain_value = expanded_value[1:] if negative else expanded_value - if grouped and grouped[-1].negative == negative: - previous = grouped[-1] - grouped[-1] = _LegacyToken(negative, (*previous.values, plain_value)) - else: - grouped.append(_LegacyToken(negative, (plain_value,))) - parsed.extend(grouped) - return tuple(parsed), requires_exists + values_by_sign.setdefault(negative, []).append(plain_value) + return tuple(_LegacyToken(negative, tuple(values)) for negative, values in values_by_sign.items()), requires_exists def _expand_legacy_token(value: str) -> tuple[str, ...]: diff --git a/app/application/classification/projection.py b/app/application/classification/projection.py index e9639f9cf..212ed92ca 100644 --- a/app/application/classification/projection.py +++ b/app/application/classification/projection.py @@ -219,11 +219,11 @@ def _project_rule( continue if not values: continue - token = _render_legacy_token(values) - rendered = f"!{token}" if negative else token field_tokens = tokens_by_field.setdefault(field_name, []) - if rendered not in field_tokens: - field_tokens.append(rendered) + for value in values: + rendered = f"!{value}" if negative else value + if rendered not in field_tokens: + field_tokens.append(rendered) return ( {field_name: ",".join(tokens) for field_name, tokens in tokens_by_field.items()}, diagnostics, @@ -300,17 +300,6 @@ def _original_alias_value(value: str, aliases: Mapping[str, str]) -> str: return value -def _render_legacy_token(values: Sequence[str]) -> str: - """把一个迁移时保留边界的值集合恢复为单个旧逗号项。""" - if len(values) == 1: - return values[0] - if values and all(value.isdigit() for value in values): - numbers = [int(value) for value in values] - if numbers == list(range(numbers[0], numbers[-1] + 1)): - return f"{numbers[0]}-{numbers[-1]}" - return "-".join(values) - - def _projection_warning( code: str, message: str, diff --git a/app/startup/composition/classification.py b/app/startup/composition/classification.py index b0e2ebdd5..c0144d5b7 100644 --- a/app/startup/composition/classification.py +++ b/app/startup/composition/classification.py @@ -13,6 +13,7 @@ from pydantic import ValidationError from app.application.classification.configuration import ( ClassificationPolicyConfigurationService, + ClassificationPolicyValidationError, ) from app.application.classification.contract import ( ClassificationPolicyStateCorruptError, @@ -196,7 +197,18 @@ async def compose_classification( ) service.register_extra_fields(migration.extra_fields) - await service.async_initialize(migration.policy) + try: + await service.async_initialize(migration.policy) + except ClassificationPolicyValidationError as error: + logger.error("旧分类策略未通过发布校验，继续保留 legacy 只读兼容行为") + return finish( + ClassificationRuntime( + service, + legacy_config=legacy_config, + diagnostics=(*_migration_issues(migration.issues), *error.result.issues), + ), + migrated=False, + ) _log_migration_issues(migration) logger.info("已将 category.yaml 无损迁移为 MediaClassificationPolicy revision 1") return finish( diff --git a/docs/architecture/media-classification-design.md b/docs/architecture/media-classification-design.md index e9af92a44..666ef2a7e 100644 --- a/docs/architecture/media-classification-design.md +++ b/docs/architecture/media-classification-design.md @@ -934,12 +934,17 @@ API 常规读取只返回 `active`，历史接口按需读取 `history`。选择 8. 首个空规则分类转换为该媒体类型的全局 `fallbacks`，其后的 legacy 项在旧实现中本就不可达，迁移时保持禁用并向管理员报告。 9. 新策略只按媒体类型配置通用兜底；数据源只能作为规则的筛选条件，不能单独决定默认分类。 10. 保存新策略 revision 1，并保留原 YAML 文件只读备份，不再继续写入。 +11. 同一字段的正向枚举合并为 `contains_any`，排除枚举合并为 `contains_none`；国家、语言等值数量 + 不应展开为叶子数量。已知和未知 Genre ID 仍分别使用标准字段与受控扩展字段，保持原有 OR 语义。 +12. 迁移后的策略仍需通过完整发布校验；超出真实条件复杂度或引用约束时不写入新策略，保留旧分类 + 只读兼容行为和结构化错误诊断，不得让发布校验异常中断宿主启动。 ### 13.2 行为兼容 - 迁移测试必须证明同一批 TMDB fixture 在旧分类器和新分类器下得到完全相同的目录分类。 - 旧配置中的 `!值` 转换为明确的排除操作符。 - 旧逗号字符串转换为数组，后续 API 不再使用逗号编码多值。 +- 旧只读 API 将集合逐值投影为逗号项，每个排除值独立添加 `!`；范围允许展开为等价枚举，不改写原 YAML。 - 迁移后 TMDB 模块不再实例化 `CategoryHelper`，也不再写 `MediaInfo.category`。 ### 13.3 字段迁移 diff --git a/tests/test_media_classification_legacy.py b/tests/test_media_classification_legacy.py index 149018851..24feb3bca 100644 --- a/tests/test_media_classification_legacy.py +++ b/tests/test_media_classification_legacy.py @@ -273,6 +273,37 @@ def test_mixed_genre_ids_keep_positive_or_and_negative_and_semantics() -> None: assert _category_name(result, _tmdb_facts(result, {}, "电影")) == "兜底" +@pytest.mark.parametrize("negative", [False, True]) +@pytest.mark.parametrize("field_name", ["origin_country", "genre_ids"]) +def test_large_value_sets_remain_compact_and_preserve_legacy_semantics( + field_name: str, + negative: bool, +) -> None: + """大枚举只产生集合条件，旧 API 投影后每个正向或排除值仍保持原语义。""" + values = [str(index) for index in range(1, 61)] + prefix = "!" if negative else "" + config = { + "movie": { + "目标": {field_name: ",".join(f"{prefix}{value}" for value in values)}, + "兜底": None, + }, + "tv": {}, + } + result = migrate_legacy_category_config(config) + projection = project_policy_to_legacy_category_projection(result.policy) + remigrated = migrate_legacy_category_config(projection.config) + + assert result.valid + assert ClassificationPolicyValidator.validate(result.policy, result.extra_fields).valid + assert len(_leaf_fields(result.policy.rules[0].when)) <= 2 + assert projection.exact + for actual in [*([value] for value in values), ["999"], ["1", "999"], [], None]: + tmdb_info = {field_name: actual, "id": 1} + expected = _legacy_tmdb_category(config["movie"], tmdb_info) + assert _category_name(result, _tmdb_facts(result, tmdb_info, "电影")) == expected + assert _category_name(remigrated, _tmdb_facts(remigrated, tmdb_info, "电影")) == expected + + def test_safe_unknown_field_is_declared_but_unsafe_field_blocks_publish() -> None: """合法未知 TMDB 一级字段可迁移，非法字段段必须产生阻断错误。""" safe = migrate_legacy_category_config( @@ -387,7 +418,15 @@ def test_release_year_supports_values_ranges_and_non_numeric_hyphen_endpoints() assert _category_name(result, _tmdb_facts(result, {"release_date": "2023-01-01"}, "电影")) == "兜底" projection = project_policy_to_legacy_category_projection(result.policy) assert projection.exact - assert projection.config.movie == CategoryConfig.model_validate(config).movie + assert projection.config.movie == CategoryConfig.model_validate( + { + "movie": { + "近年": {"release_year": "2020,2021,2022,2024"}, + "字母年": {"release_year": "ABCD,EFGH"}, + "兜底": None, + } + } + ).movie def test_stable_ids_are_repeatable_ascii_and_media_type_scoped() -> None: @@ -485,5 +524,9 @@ def test_migrated_policy_round_trips_to_category_config() -> None: projected = project_policy_to_legacy_category_projection(migrated.policy) assert projected.exact - assert projected.config == CategoryConfig.model_validate(config) + expected = CategoryConfig.model_validate(config) + assert expected.movie is not None + assert expected.movie["组合"] is not None + expected.movie["组合"].release_year = "2020,2021,2022,2024" + assert projected.config == expected assert project_policy_to_legacy_category_config(migrated.policy) == projected.config diff --git a/tests/test_media_classification_startup.py b/tests/test_media_classification_startup.py index e97a81cf0..a1ce12f83 100644 --- a/tests/test_media_classification_startup.py +++ b/tests/test_media_classification_startup.py @@ -17,6 +17,7 @@ from app.application.classification.configuration import ( from app.application.classification.contract import ClassificationPolicyConflictError from app.application.classification.runtime import ClassificationRuntime from app.schemas.category import ( + ClassificationConditionGroup, ClassificationPolicyState, ) from app.schemas.types import SystemConfigKey @@ -152,6 +153,83 @@ async def test_absent_policy_migrates_yaml_once_without_rewriting_file( assert legacy_path.read_text(encoding="utf-8") == content +@pytest.mark.asyncio +async def test_large_legacy_enumerations_migrate_and_reload_without_losing_values( + monkeypatch: pytest.MonkeyPatch, + tmp_path: Path, +) -> None: + """复现 56/31 个旧枚举值的升级场景，验证首次发布、重启和旧接口均保留全部值。""" + languages = ",".join(f"language{index}" for index in range(26)) + countries = ",".join(f"country{index}" for index in range(30)) + content = ( + f"movie:\n 欧美电影:\n original_language: '{languages}'\n" + f" production_countries: '{countries}'\n 未分类:\n" + f"tv:\n 欧美漫:\n genre_ids: '16'\n origin_country: '{countries}'\n 未分类:\n" + ) + legacy_path = tmp_path / "category.yaml" + legacy_path.write_text(content, encoding="utf-8") + store = _MemoryPolicyStore() + monkeypatch.setattr(classification_composition, "SystemConfigClassificationPolicyStore", lambda *_args: store) + system_config = _SystemConfig({}) + + composition = await classification_composition.compose_classification( + executor=cast(Any, _InlineExecutor()), + settings=cast(Any, SimpleNamespace(CONFIG_PATH=tmp_path)), + system_config=cast(Any, system_config), + ) + + assert composition.migrated is True + policy = composition.runtime.require_policy() + assert policy.revision == 1 + assert len(policy.rules) == 2 + for rule in policy.rules: + assert isinstance(rule.when, ClassificationConditionGroup) + assert rule.when.all is not None and len(rule.when.all) == 2 + assert store.state is not None + system_config.publish_many({SystemConfigKey.MediaClassificationPolicy: store.state.model_dump(mode="json")}) + reloaded = await classification_composition.compose_classification( + executor=cast(Any, _InlineExecutor()), + settings=cast(Any, SimpleNamespace(CONFIG_PATH=tmp_path)), + system_config=cast(Any, system_config), + ) + + assert reloaded.migrated is False + assert reloaded.runtime.require_policy() == policy + projected = reloaded.runtime.legacy_config().model_dump(exclude_none=True) + assert projected["movie"]["欧美电影"]["original_language"] == languages + assert projected["movie"]["欧美电影"]["production_countries"] == countries + assert projected["tv"]["欧美漫"]["origin_country"] == countries + assert store.write_count == 1 + assert legacy_path.read_text(encoding="utf-8") == content + + +@pytest.mark.asyncio +async def test_publish_validation_failure_preserves_legacy_runtime_without_writes( + monkeypatch: pytest.MonkeyPatch, + tmp_path: Path, +) -> None: + """真正超出叶子上限的不同字段保留错误诊断和旧配置，不得中断启动或写入新策略。""" + content = "movie:\n 复杂规则:\n" + "".join(f" field{index}: 'X'\n" for index in range(31)) + legacy_path = tmp_path / "category.yaml" + legacy_path.write_text(content, encoding="utf-8") + store = _MemoryPolicyStore() + monkeypatch.setattr(classification_composition, "SystemConfigClassificationPolicyStore", lambda *_args: store) + + composition = await classification_composition.compose_classification( + executor=cast(Any, _InlineExecutor()), + settings=cast(Any, SimpleNamespace(CONFIG_PATH=tmp_path)), + system_config=cast(Any, _SystemConfig({})), + ) + + assert composition.migrated is False + assert composition.runtime.active_policy() is None + assert any(issue.code == "max_conditions_exceeded" for issue in composition.runtime.diagnostics()) + assert "复杂规则" in (composition.runtime.legacy_config().movie or {}) + assert store.write_count == 0 + assert store.state is None + assert legacy_path.read_text(encoding="utf-8") == content + + @pytest.mark.asyncio # type: ignore[misc] async def test_invalid_legacy_config_keeps_runtime_uninitialized_and_writes_nothing( monkeypatch: pytest.MonkeyPatch,