Reduce test time with a global LLM mock

setup.py

@@ -46,6 +46,7 @@ def run(self):
     "chromadb==0.4.14",
     "gradio==3.0.0",
     "grpcio-status==1.48.2",
+    "mock==5.1.0",
 ]
 
 extras_require["pyppeteer"] = [

tests/conftest.py

@@ -12,57 +12,17 @@
 import os
 import re
 import uuid
-from typing import Optional
 
 import pytest
 
 from metagpt.config import CONFIG, Config
 from metagpt.const import DEFAULT_WORKSPACE_ROOT, TEST_DATA_PATH
 from metagpt.llm import LLM
 from metagpt.logs import logger
-from metagpt.provider.openai_api import OpenAILLM
 from metagpt.utils.git_repository import GitRepository
+from tests.mock.mock_llm import MockLLM
 
-
-class MockLLM(OpenAILLM):
-    rsp_cache: dict = {}
-
-    async def original_aask(
-        self,
-        msg: str,
-        system_msgs: Optional[list[str]] = None,
-        format_msgs: Optional[list[dict[str, str]]] = None,
-        timeout=3,
-        stream=True,
-    ):
-        """A copy of metagpt.provider.base_llm.BaseLLM.aask, we can't use super().aask because it will be mocked"""
-        if system_msgs:
-            message = self._system_msgs(system_msgs)
-        else:
-            message = [self._default_system_msg()] if self.use_system_prompt else []
-        if format_msgs:
-            message.extend(format_msgs)
-        message.append(self._user_msg(msg))
-        rsp = await self.acompletion_text(message, stream=stream, timeout=timeout)
-        return rsp
-
-    async def aask(
-        self,
-        msg: str,
-        system_msgs: Optional[list[str]] = None,
-        format_msgs: Optional[list[dict[str, str]]] = None,
-        timeout=3,
-        stream=True,
-    ) -> str:
-        if msg not in self.rsp_cache:
-            # Call the original unmocked method
-            rsp = await self.original_aask(msg, system_msgs, format_msgs, timeout, stream)
-            logger.info(f"Added '{rsp[:20]}' ... to response cache")
-            self.rsp_cache[msg] = rsp
-            return rsp
-        else:
-            logger.info("Use response cache")
-            return self.rsp_cache[msg]
+RSP_CACHE_NEW = {}  # used globally for producing new and useful only response cache
 
 
 @pytest.fixture(scope="session")
@@ -76,16 +36,37 @@ def rsp_cache():
     else:
         rsp_cache_json = {}
     yield rsp_cache_json
-    with open(new_rsp_cache_file_path, "w") as f2:
+    with open(rsp_cache_file_path, "w") as f2:
         json.dump(rsp_cache_json, f2, indent=4, ensure_ascii=False)
+    with open(new_rsp_cache_file_path, "w") as f2:
+        json.dump(RSP_CACHE_NEW, f2, indent=4, ensure_ascii=False)
 
 
-@pytest.fixture(scope="function")
-def llm_mock(rsp_cache, mocker):
+# Hook to capture the test result
+@pytest.hookimpl(tryfirst=True, hookwrapper=True)
+def pytest_runtest_makereport(item, call):
+    outcome = yield
+    rep = outcome.get_result()
+    if rep.when == "call":
+        item.test_outcome = rep
+
+
+@pytest.fixture(scope="function", autouse=True)
+def llm_mock(rsp_cache, mocker, request):
     llm = MockLLM()
     llm.rsp_cache = rsp_cache
     mocker.patch("metagpt.provider.base_llm.BaseLLM.aask", llm.aask)
+    mocker.patch("metagpt.provider.base_llm.BaseLLM.aask_batch", llm.aask_batch)
     yield mocker
+    if hasattr(request.node, "test_outcome") and request.node.test_outcome.passed:
+        if llm.rsp_candidates:
+            for rsp_candidate in llm.rsp_candidates:
+                cand_key = list(rsp_candidate.keys())[0]
+                cand_value = list(rsp_candidate.values())[0]
+                if cand_key not in llm.rsp_cache:
+                    logger.info(f"Added '{cand_key[:100]} ... -> {cand_value[:20]} ...' to response cache")
+                    llm.rsp_cache.update(rsp_candidate)
+                RSP_CACHE_NEW.update(rsp_candidate)
 
 
 class Context:
@@ -170,6 +151,12 @@ def init_config():
     Config()
 
 
+@pytest.fixture(scope="function")
+def new_filename(mocker):
+    mocker.patch("metagpt.utils.file_repository.FileRepository.new_filename", lambda: "20240101")
+    yield mocker
+
+
 @pytest.fixture
 def aiohttp_mocker(mocker):
     class MockAioResponse:

tests/metagpt/actions/test_debug_error.py

@@ -117,7 +117,6 @@ def test_player_calculate_score_with_multiple_aces(self):
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
 async def test_debug_error():
     CONFIG.src_workspace = CONFIG.git_repo.workdir / uuid.uuid4().hex
     ctx = RunCodeContext(

tests/metagpt/actions/test_design_api.py

@@ -17,7 +17,6 @@
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
 async def test_design_api():
     inputs = ["我们需要一个音乐播放器，它应该有播放、暂停、上一曲、下一曲等功能。", PRD_SAMPLE]
     for prd in inputs:

tests/metagpt/actions/test_design_api_review.py

@@ -11,7 +11,6 @@
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
 async def test_design_api_review():
     prd = "我们需要一个音乐播放器，它应该有播放、暂停、上一曲、下一曲等功能。"
     api_design = """

tests/metagpt/actions/test_generate_questions.py

@@ -20,7 +20,6 @@
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
 async def test_generate_questions():
     action = GenerateQuestions()
     rsp = await action.run(context)

tests/metagpt/actions/test_invoice_ocr.py

@@ -54,7 +54,6 @@ async def test_generate_table(invoice_path: Path, expected_result: dict):
     ("invoice_path", "query", "expected_result"),
     [(Path("invoices/invoice-1.pdf"), "Invoicing date", "2023年02月03日")],
 )
-@pytest.mark.usefixtures("llm_mock")
 async def test_reply_question(invoice_path: Path, query: dict, expected_result: str):
     invoice_path = TEST_DATA_PATH / invoice_path
     ocr_result = await InvoiceOCR().run(file_path=Path(invoice_path))

tests/metagpt/actions/test_prepare_interview.py

@@ -12,7 +12,6 @@
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
 async def test_prepare_interview():
     action = PrepareInterview()
     rsp = await action.run("I just graduated and hope to find a job as a Python engineer")

tests/metagpt/actions/test_project_management.py

@@ -18,7 +18,6 @@
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
 async def test_design_api():
     await FileRepository.save_file("1.txt", content=str(PRD), relative_path=PRDS_FILE_REPO)
     await FileRepository.save_file("1.txt", content=str(DESIGN), relative_path=SYSTEM_DESIGN_FILE_REPO)

tests/metagpt/actions/test_research.py

@@ -8,14 +8,7 @@
 
 import pytest
 
-from metagpt.actions import CollectLinks, research
-
-
-@pytest.mark.asyncio
-async def test_action():
-    action = CollectLinks()
-    result = await action.run(topic="baidu")
-    assert result
+from metagpt.actions import research
 
 
 @pytest.mark.asyncio

tests/metagpt/actions/test_summarize_code.py

@@ -177,7 +177,6 @@ def get_body(self):
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
 async def test_summarize_code():
     CONFIG.src_workspace = CONFIG.git_repo.workdir / "src"
     await FileRepository.save_file(filename="1.json", relative_path=SYSTEM_DESIGN_FILE_REPO, content=DESIGN_CONTENT)

tests/metagpt/actions/test_talk_action.py

@@ -33,7 +33,6 @@
         ),
     ],
 )
-@pytest.mark.usefixtures("llm_mock")
 async def test_prompt(agent_description, language, context, knowledge, history_summary):
     # Prerequisites
     CONFIG.agent_description = agent_description

tests/metagpt/actions/test_write_code.py

@@ -28,7 +28,6 @@
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
 async def test_write_code():
     context = CodingContext(
         filename="task_filename.py", design_doc=Document(content="设计一个名为'add'的函数，该函数接受两个整数作为输入，并返回它们的和。")
@@ -45,7 +44,6 @@ async def test_write_code():
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
 async def test_write_code_directly():
     prompt = WRITE_CODE_PROMPT_SAMPLE + "\n" + TASKS_2[0]
     llm = LLM()
@@ -54,7 +52,6 @@ async def test_write_code_directly():
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
 async def test_write_code_deps():
     # Prerequisites
     CONFIG.src_workspace = CONFIG.git_repo.workdir / "snake1/snake1"

tests/metagpt/actions/test_write_code_review.py

@@ -12,7 +12,6 @@
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
 async def test_write_code_review(capfd):
     code = """
 def add(a, b):

tests/metagpt/actions/test_write_docstring.py

@@ -27,14 +27,12 @@ def greet(self):
     ],
     ids=["google", "numpy", "sphinx"],
 )
-@pytest.mark.usefixtures("llm_mock")
 async def test_write_docstring(style: str, part: str):
     ret = await WriteDocstring().run(code, style=style)
     assert part in ret
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
 async def test_write():
     code = await WriteDocstring.write_docstring(__file__)
     assert code

tests/metagpt/actions/test_write_prd.py

@@ -18,8 +18,7 @@
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
-async def test_write_prd():
+async def test_write_prd(new_filename):
     product_manager = ProductManager()
     requirements = "开发一个基于大语言模型与私有知识库的搜索引擎，希望可以基于大语言模型进行搜索总结"
     await FileRepository.save_file(filename=REQUIREMENT_FILENAME, content=requirements, relative_path=DOCS_FILE_REPO)

tests/metagpt/actions/test_write_prd_review.py

@@ -11,7 +11,6 @@
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
 async def test_write_prd_review():
     prd = """
     Introduction: This is a new feature for our product.

tests/metagpt/actions/test_write_review.py

@@ -46,7 +46,6 @@
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
 async def test_write_review():
     write_review = WriteReview()
     review = await write_review.run(CONTEXT)

tests/metagpt/actions/test_write_teaching_plan.py

@@ -16,7 +16,6 @@
     ("topic", "context"),
     [("Title", "Lesson 1: Learn to draw an apple."), ("Teaching Content", "Lesson 1: Learn to draw an apple.")],
 )
-@pytest.mark.usefixtures("llm_mock")
 async def test_write_teaching_plan_part(topic, context):
     action = WriteTeachingPlanPart(topic=topic, context=context)
     rsp = await action.run()

tests/metagpt/actions/test_write_test.py

@@ -13,7 +13,6 @@
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
 async def test_write_test():
     code = """
     import random
@@ -40,7 +39,6 @@ def generate(self, max_y: int, max_x: int):
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
 async def test_write_code_invalid_code(mocker):
     # Mock the _aask method to return an invalid code string
     mocker.patch.object(WriteTest, "_aask", return_value="Invalid Code String")

tests/metagpt/actions/test_write_tutorial.py

@@ -14,7 +14,6 @@
 
 @pytest.mark.asyncio
 @pytest.mark.parametrize(("language", "topic"), [("English", "Write a tutorial about Python")])
-@pytest.mark.usefixtures("llm_mock")
 async def test_write_directory(language: str, topic: str):
     ret = await WriteDirectory(language=language).run(topic=topic)
     assert isinstance(ret, dict)
@@ -30,7 +29,6 @@ async def test_write_directory(language: str, topic: str):
     ("language", "topic", "directory"),
     [("English", "Write a tutorial about Python", {"Introduction": ["What is Python?", "Why learn Python?"]})],
 )
-@pytest.mark.usefixtures("llm_mock")
 async def test_write_content(language: str, topic: str, directory: Dict):
     ret = await WriteContent(language=language, directory=directory).run(topic=topic)
     assert isinstance(ret, str)

tests/metagpt/provider/conftest.py

@@ -0,0 +1,8 @@
+import pytest
+
+
+@pytest.fixture(autouse=True)
+def llm_mock(rsp_cache, mocker, request):
+    # An empty fixture to overwrite the global llm_mock fixture
+    # because in provider folder, we want to test the aask and aask functions for the specific models
+    pass

tests/metagpt/roles/test_architect.py

@@ -22,7 +22,6 @@
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
 async def test_architect():
     # Prerequisites
     filename = uuid.uuid4().hex + ".json"

tests/metagpt/roles/test_assistant.py

@@ -13,15 +13,13 @@
 from metagpt.actions.skill_action import SkillAction
 from metagpt.actions.talk_action import TalkAction
 from metagpt.config import CONFIG
-from metagpt.logs import logger
 from metagpt.memory.brain_memory import BrainMemory
 from metagpt.roles.assistant import Assistant
 from metagpt.schema import Message
 from metagpt.utils.common import any_to_str
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
 async def test_run():
     CONFIG.language = "Chinese"
 
@@ -88,7 +86,7 @@ class Input(BaseModel):
             if not has_action:
                 break
             msg: Message = await role.act()
-            logger.info(msg)
+            # logger.info(msg)
             assert msg
             assert msg.cause_by == seed.cause_by
             assert msg.content

tests/metagpt/roles/test_engineer.py

@@ -30,7 +30,6 @@
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
 async def test_engineer():
     # Prerequisites
     rqno = "20231221155954.json"
@@ -114,7 +113,6 @@ def test_todo():
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
 async def test_new_coding_context():
     # Prerequisites
     demo_path = Path(__file__).parent / "../../data/demo_project"

tests/metagpt/roles/test_invoice_ocr_assistant.py

@@ -41,7 +41,6 @@
         ),
     ],
 )
-@pytest.mark.usefixtures("llm_mock")
 async def test_invoice_ocr_assistant(query: str, invoice_path: Path, invoice_table_path: Path, expected_result: dict):
     invoice_path = TEST_DATA_PATH / invoice_path
     role = InvoiceOCRAssistant()

tests/metagpt/roles/test_product_manager.py

@@ -13,8 +13,7 @@
 
 
 @pytest.mark.asyncio
-@pytest.mark.usefixtures("llm_mock")
-async def test_product_manager():
+async def test_product_manager(new_filename):
     product_manager = ProductManager()
     rsp = await product_manager.run(MockMessages.req)
     logger.info(rsp)