games-development-ai/cheap-worker/tests/test_a11_m2_classify.py

"""test_a11_m2_classify.py — A11 切片三 M2 判意图(classify)纯函数单测。

判意图 = 用户原话 + base 源工程 → LLM 判 {category, mode, target, payload, riskLevel, needsConfirm, clarify}。
LLM 调用是集成(M5 e2e 真验),本测只守两个纯函数:
  · build_classify_prompt:把 base 源的可改面(资产清单 / 集中数值 / 文件列表)+ 用户原话喂进 prompt;
  · parse_classify_response:解析 LLM JSON + 落 mode×category 锁定映射 + 风险/确认规则 + 坏 JSON 安全兜底。

跑:cheap-worker/.venv/bin/python cheap-worker/tests/test_a11_m2_classify.py
"""

import json
import sys
from pathlib import Path

sys.path.insert(0, str(Path(__file__).resolve().parents[1]))  # → cheap-worker/
import cheap_classify as C  # noqa: E402


def _src_project() -> dict:
    """造一个便宜档 base 源工程(2.0):assets.js 资产清单 + core.js 集中数值 + game-logic.js 玩法。"""
    return {
        "schemaVersion": "2.0",
        "sourceHash": "deadbeef",
        "profile": {"tickModel": "realtime", "inputModel": "discrete-choice", "progressModel": "metric"},
        "files": {
            "src/assets.js": "export const IMAGE_FILES = { heroSprite: 'hero.png', bg: 'bg.jpg' };\n",
            "src/core.js": "export const ROUND_MS = 30000;\nexport const COLOR_PALETTE = ['red','blue'];\n",
            "src/game-logic.js": "// 彩点点击得分玩法\nimport { ROUND_MS } from './core.js';\n",
        },
        "entry": "entry.js",
        "globalName": "__GameBundle",
    }


# ── build_classify_prompt:可改面 + 原话进 prompt ──

def test_prompt_includes_instruction_and_changeable_surface():
    p = C.build_classify_prompt("把主角调成绿色", _src_project())
    assert "把主角调成绿色" in p, "用户原话必须进 prompt"
    assert "assets.js" in p and "heroSprite" in p, "资产清单(可改面)要进 prompt 供定位 asset 目标"
    assert "core.js" in p and "ROUND_MS" in p, "集中数值(可改面)要进 prompt 供定位 config 目标"
    assert "game-logic.js" in p, "玩法文件要进 prompt 供 behavior 判断"


def test_prompt_handles_string_source_project():
    """source_project 传 JSON 串也行(后端经 HTTP 传的是串)。"""
    p = C.build_classify_prompt("加快下落", json.dumps(_src_project()))
    assert "加快下落" in p and "core.js" in p


# ── parse_classify_response:映射 + 风险/确认 + 兜底 ──

def _llm(obj: dict) -> str:
    """模拟 LLM 返回(可能裹 ```json 围栏,parse 要能剥)。"""
    return "```json\n" + json.dumps(obj, ensure_ascii=False) + "\n```"


def test_parse_asset_change_low_risk_no_confirm():
    out = C.parse_classify_response(_llm({
        "category": "asset", "target": {"kind": "asset", "path": "/files/src/assets.js#heroSprite"},
        "payload": {"value": {"heroSprite": "hero-green.png"}}, "riskLevel": "low",
        "rationale": "换主角贴图", "clarify": None,
    }))
    assert out["category"] == "asset"
    assert out["mode"] == "deterministic", "asset/config/level → deterministic(锁定映射)"
    assert out["needsConfirm"] is False, "低风险确定性改 → 不强制确认"


def test_parse_behavior_change_is_regenerate_and_confirm():
    out = C.parse_classify_response(_llm({
        "category": "behavior", "target": {"kind": "behavior", "path": "/files/src/game-logic.js"},
        "payload": {"intent": "陨石改成左右摇摆"}, "riskLevel": "medium", "rationale": "改玩法逻辑",
    }))
    assert out["mode"] == "regenerate-module", "behavior → regenerate-module"
    assert out["needsConfirm"] is True, "改玩法(模块重生成)默认要确认"


def test_parse_big_change_high_risk_confirm_no_mode():
    out = C.parse_classify_response(_llm({
        "category": "big-change", "riskLevel": "high",
        "rationale": "把点击游戏改成 3D 赛车 = 换品类重写", "clarify": "这是大改动,需重新生成整个游戏,确认?",
    }))
    assert out["category"] == "big-change"
    assert out["mode"] is None, "大改不落 deterministic/regenerate-module(超局部调整)"
    assert out["riskLevel"] == "high" and out["needsConfirm"] is True
    assert out["clarify"]


def test_parse_unclear_asks_clarify():
    out = C.parse_classify_response(_llm({
        "category": "unclear", "rationale": "意图不明", "clarify": "你想改哪一处?颜色、速度还是玩法?",
    }))
    assert out["category"] == "unclear"
    assert out["needsConfirm"] is True and out["clarify"]


def test_parse_enforces_mode_category_lock():
    """LLM 给了矛盾的 category=asset + mode=regenerate-module → 按 category 纠正 mode(锁定映射可信边界)。"""
    out = C.parse_classify_response(_llm({
        "category": "asset", "mode": "regenerate-module",
        "target": {"kind": "asset", "path": "/x"}, "payload": {"value": {}}, "riskLevel": "low",
    }))
    assert out["mode"] == "deterministic", "category=asset 必落 deterministic,无视 LLM 给的矛盾 mode"


def test_parse_bad_json_safe_fallback():
    """坏 JSON / 非 JSON → 安全兜底为 unclear + 需确认(绝不抛、绝不静默判成可执行改动)。"""
    out = C.parse_classify_response("这不是 JSON,模型胡说一通")
    assert out["category"] == "unclear"
    assert out["needsConfirm"] is True
    assert out["mode"] is None


# ── worker /classify 端点路由(注入 classify_fn 桩,免真网络)──

def test_classify_endpoint_routes_and_returns_proposal():
    import json as _json
    import urllib.request

    import worker_service as WS

    stub = {"category": "config", "mode": "deterministic", "needsConfirm": False, "riskLevel": "low"}
    state = WS.WorkerState(classify_fn=lambda raw, sp: stub)
    server, port = WS.start_server(state, host="127.0.0.1", port=0)
    try:
        req = urllib.request.Request(
            f"http://127.0.0.1:{port}/classify",
            data=_json.dumps({"rawText": "加快下落", "sourceProject": "{}"}).encode("utf-8"),
            method="POST", headers={"Content-Type": "application/json"})
        with urllib.request.urlopen(req, timeout=5) as resp:
            assert resp.status == 200
            out = _json.loads(resp.read().decode("utf-8"))
        assert out["category"] == "config" and out["needsConfirm"] is False

        # 缺 rawText → 400
        bad = urllib.request.Request(
            f"http://127.0.0.1:{port}/classify",
            data=_json.dumps({"sourceProject": "{}"}).encode("utf-8"),
            method="POST", headers={"Content-Type": "application/json"})
        try:
            urllib.request.urlopen(bad, timeout=5)
            assert False, "缺 rawText 应 400"
        except urllib.error.HTTPError as e:
            assert e.code == 400
    finally:
        server.shutdown()


def _run_all():
    fns = [v for k, v in sorted(globals().items()) if k.startswith("test_") and callable(v)]
    for fn in fns:
        fn()
        print(f"  ✓ {fn.__name__}")
    print(f"\n[test_a11_m2_classify] {len(fns)}/{len(fns)} passed")


if __name__ == "__main__":
    _run_all()