from pathlib import Path import pytest pytest.importorskip("yaml") from utils.config import Config, apply_overrides, deep_merge, load_config, parse_override # noqa: E402 def test_deep_merge_is_recursive(): merged = deep_merge({"a": {"b": 1, "c": 2}}, {"a": {"c": 3}, "d": 4}) assert merged == {"a": {"b": 1, "c": 3}, "d": 4} def test_parse_override_literals(): assert parse_override("train.lr=1e-3") == ("train.lr", 1e-3) assert parse_override("train.amp=False") == ("train.amp", False) assert parse_override("model.backbone.name=tiny_conv") == ("model.backbone.name", "tiny_conv") assert parse_override("simclr.feature_stages=[5, 6]") == ("simclr.feature_stages", [5, 6]) def test_parse_override_accepts_yaml_style_scalars(): """配置文件里写 null/true/false，命令行上也得能这么写。 回归测试：只用 ast.literal_eval 时 `false` 会变成字符串 "false"， 而非空字符串是真值——`--set xxx.enabled=false` 反而把开关打开了。 """ assert parse_override("a=null") == ("a", None) assert parse_override("a=~") == ("a", None) assert parse_override("a=true") == ("a", True) assert parse_override("a=false") == ("a", False) assert parse_override("a=false")[1] is False # 不是字符串 # 路径、模型名这类仍应保持字符串 assert parse_override("a=resnet18") == ("a", "resnet18") assert parse_override("a=runs/x/best.pth") == ("a", "runs/x/best.pth") def test_apply_overrides_creates_missing_nodes(): """非严格模式下仍可新建嵌套节点；严格模式（load_config 默认）会拦下未知键。""" cfg = apply_overrides({}, ["a.b.c=1"], strict=False) assert cfg == {"a": {"b": {"c": 1}}} with pytest.raises(KeyError, match="不存在"): apply_overrides({}, ["a.b.c=1"]) def test_load_config_with_base(tmp_path): (tmp_path / "base.yaml").write_text("train:\n lr: 0.1\n epochs: 10\n", encoding="utf-8") (tmp_path / "child.yaml").write_text( "_base_: ./base.yaml\ntrain:\n lr: 0.01\nwork_dir: runs/x\n", encoding="utf-8" ) cfg = load_config(tmp_path / "child.yaml", ["train.epochs=2"]) assert isinstance(cfg, Config) assert cfg.train.lr == 0.01 assert cfg.train.epochs == 2 assert cfg.get_path("work_dir") == "runs/x" def test_load_config_detects_cycle(tmp_path): (tmp_path / "a.yaml").write_text("_base_: ./b.yaml\n", encoding="utf-8") (tmp_path / "b.yaml").write_text("_base_: ./a.yaml\n", encoding="utf-8") with pytest.raises(ValueError, match="循环引用"): load_config(tmp_path / "a.yaml") def test_project_configs_are_loadable(): cluster = load_config("configs/cluster/film.yaml") assert cluster["dataset"]["name"] == "unlabeled" assert cluster["num_clusters"] > 0 classify = load_config("configs/classify/film.yaml") assert classify["model"]["name"] == "classifier" assert classify["train"]["monitor"] == "macro_f1" def test_replace_key_swaps_a_whole_subdict(tmp_path): """换组件时要能整块替换，否则旧组件的专有字段会残留下来传给新组件。""" (tmp_path / "base.yaml").write_text( "backbone:\n name: tiny_conv\n widths: [32, 64]\n depths: [1, 2]\n block: dw\n", encoding="utf-8", ) (tmp_path / "child.yaml").write_text( "_base_: ./base.yaml\nbackbone:\n _replace_: true\n name: torchvision\n arch: resnet18\n", encoding="utf-8", ) cfg = load_config(tmp_path / "child.yaml") assert cfg["backbone"] == {"name": "torchvision", "arch": "resnet18"} assert "_replace_" not in cfg["backbone"] # 标记本身不能漏进构造参数 def test_without_replace_key_subdicts_still_merge(tmp_path): (tmp_path / "base.yaml").write_text("backbone:\n name: tiny_conv\n block: dw\n", encoding="utf-8") (tmp_path / "child.yaml").write_text( "_base_: ./base.yaml\nbackbone:\n block: res\n", encoding="utf-8" ) cfg = load_config(tmp_path / "child.yaml") assert cfg["backbone"] == {"name": "tiny_conv", "block": "res"} def test_registry_reports_incompatible_fields_actionably(): """--set 只改 name 时旧字段会残留，报错要说清楚是哪些字段、该怎么办。""" import pytest from models.registry import BACKBONES with pytest.raises(TypeError, match=r"不接受的字段.*widths"): BACKBONES.build({"name": "torchvision", "arch": "resnet18", "widths": [32]}) @pytest.mark.parametrize( "config_path", sorted(str(p) for p in Path("configs").rglob("*.yaml")), ) def test_every_shipped_config_builds_its_components(config_path): """仓库里每个配置都必须能真的构建出组件。 回归测试：派生配置曾经无法使用——它同时继承两份骨架配置， 合并后 backbone 混了两边的字段（tiny_conv 的 widths 混进 torchvision），构建时直接报错。 """ pytest.importorskip("torch") from models.registry import BACKBONES cfg = load_config(config_path) # 聚类配置：pretrained 用顶层 backbone，simclr 用 simclr.backbone for backbone_cfg in (cfg.get("backbone"), cfg.get_path("simclr.backbone")): if not backbone_cfg: continue backbone_cfg = dict(backbone_cfg) if backbone_cfg.get("name") == "torchvision": backbone_cfg["pretrained"] = False # 单测里不下载 ImageNet 权重 backbone = BACKBONES.build(backbone_cfg) assert backbone.out_channels > 0 # 分类配置 model_cfg = cfg.get("model") if model_cfg and "weights" in model_cfg: # 检测配置：model 是 ultralytics 的权重名（yolo26n.pt），不走本仓库的注册表， # 这里只确认它确实是个权重文件名。 assert model_cfg["weights"].endswith(".pt") elif model_cfg: import torch from models import build_model model_cfg = dict(model_cfg) model_cfg.setdefault("num_classes", 3) model_cfg["init_from"] = None if model_cfg.get("backbone", {}).get("name") == "torchvision": model_cfg["backbone"] = {**model_cfg["backbone"], "pretrained": False} model = build_model(model_cfg) with torch.no_grad(): assert model(torch.randn(1, 3, 64, 64)).shape[1] == model_cfg["num_classes"] @pytest.mark.parametrize( "override,expected_hint", [ ("simclr.optmizer.weight_decay=0.0", "optimizer"), # 少了个 i ("simclr.optimizer.lrr=1e-3", "lr"), # 多了个 r ("num_clusers=4", "num_clusters"), # 少了个 t ], ) def test_set_rejects_typos_with_suggestion(override, expected_hint): """--set 打错字母会凭空造出没人读的键，训练照跑但改动无效——必须拦下并给出建议。""" with pytest.raises(KeyError) as excinfo: load_config("configs/cluster/film_simclr_resnet18.yaml", [override]) assert expected_hint in str(excinfo.value) def test_set_accepts_existing_keys(): cfg = load_config( "configs/cluster/film_simclr_resnet18.yaml", ["simclr.optimizer.lr=0.001", "simclr.optimizer.weight_decay=0.0", "num_clusters=7"], ) assert cfg["simclr"]["optimizer"]["lr"] == 0.001 assert cfg["simclr"]["optimizer"]["weight_decay"] == 0.0 assert cfg["num_clusters"] == 7 def test_apply_overrides_can_add_keys_when_not_strict(): from utils.config import apply_overrides assert apply_overrides({"a": {"b": 1}}, ["a.c=2"], strict=False) == {"a": {"b": 1, "c": 2}}