"""Committed tests for blackcat AW limit + compaction (stdlib, no network).""" # retoor import os import tempfile import unittest from unittest import mock from blackcat.agent import BlackCatBot, resolve_context_limit from blackcat.client import DeepSeekClient, DeepSeekError from blackcat.fstate import FileState, estimate_tokens from blackcat.tools import FileTools def _bot(**kwargs): kwargs.setdefault("provider", "echo") kwargs.setdefault("system", "t") return BlackCatBot(**kwargs) class TestEstimate(unittest.TestCase): def test_chars_per_token(self): self.assertEqual(estimate_tokens(""), 1) self.assertEqual(estimate_tokens("abcd"), 1) self.assertEqual(estimate_tokens("x" * 400), 100) class TestHistory(unittest.IsolatedAsyncioTestCase): async def test_prompts_accumulate_messages(self): async with _bot() as bot: self.assertEqual(await bot.prompt("hi"), "echo: hi") self.assertEqual(await bot.prompt("again"), "echo: again") self.assertEqual(len(bot.messages), 4) self.assertEqual(bot.messages[0]["role"], "user") async def test_report_carries_context(self): async with _bot() as bot: await bot.prompt("hi") status = bot.report()["context"] self.assertEqual(status["limit"], 1000000) self.assertFalse(status["summary"]) self.assertEqual(status["messages"], 2) async def test_wire_strips_meta(self): async with _bot() as bot: bot.messages.append({"role": "user", "tool": "read", "tool_ok": True, "content": "x"}) wire = bot._wire_messages() self.assertEqual(wire, [{"role": "user", "content": "x"}]) class TestRunTool(unittest.IsolatedAsyncioTestCase): def setUp(self): self.tmp = tempfile.TemporaryDirectory() self.addCleanup(self.tmp.cleanup) tools = FileTools(FileState(self.tmp.name)) self.bot = BlackCatBot(provider="echo", system="t", tools=tools) self.addAsyncCleanup(self.bot.aclose) async def test_unknown_tool_recorded(self): out = await self.bot.run_tool("nope") self.assertFalse(out["ok"]) self.assertEqual(out["code"], "unknown_tool") kinds = [f["kind"] for f in self.bot.transcript] self.assertIn("tool", kinds) async def test_read_flows_to_messages(self): with open(os.path.join(self.tmp.name, "f.txt"), "w") as fh: fh.write("hello\n") out = await self.bot.run_tool("read", path="f.txt") self.assertTrue(out["ok"]) last = self.bot.messages[-1] self.assertEqual(last["tool"], "read") self.assertTrue(last["tool_ok"]) self.assertIn("hello", last["content"]) async def test_bulky_args_abbreviated(self): with open(os.path.join(self.tmp.name, "f.txt"), "w") as fh: fh.write("a\n") await self.bot.run_tool("read", path="f.txt") await self.bot.run_tool("edit", path="f.txt", old_string="a" * 500, new_string="b") frame = self.bot.transcript[-1] self.assertEqual(frame["args"]["old_string"], "<500 chars>") self.assertEqual(frame["args"]["new_string"], "<1 chars>") class TestCompaction(unittest.IsolatedAsyncioTestCase): async def test_warn_gate_fires_once(self): async with _bot(context_limit=100) as bot: bot._usage_input = 76 await bot._maybe_compact() warns = [f for f in bot.transcript if f.get("event") == "warn"] self.assertEqual(len(warns), 1) await bot._maybe_compact() warns = [f for f in bot.transcript if f.get("event") == "warn"] self.assertEqual(len(warns), 1) async def test_manual_compact_clears_old_tools_only(self): async with _bot(keep_recent_tools=2) as bot: for i in range(4): bot.messages.append({"role": "user", "tool": "read", "tool_ok": True, "content": "result %d" % i}) info = await bot.compact(reason="manual") self.assertTrue(info["ok"]) self.assertEqual(info["tier"], "clear") self.assertEqual(info["cleared"], 2) self.assertEqual(len(bot.messages), 4) self.assertEqual(bot.messages[-1]["content"], "result 3") self.assertTrue(bot.messages[0]["content"]["cleared"]) async def test_auto_compact_summarizes(self): async with _bot(context_limit=200, keep_recent_tokens=40) as bot: bot.note("decision: use tabs") await bot.prompt("please summarize " + "x" * 800) compacts = [f for f in bot.transcript if f["kind"] == "compact"] self.assertTrue(compacts) self.assertTrue(bot.context_status()["summary"]) self.assertIn("Messages condensed", bot._summary) self.assertIn("decision: use tabs", bot._summary) self.assertEqual(bot.system, "t") self.assertTrue(bot.messages) self.assertEqual(bot.messages[-1]["role"], "assistant") async def test_memory_flush_writes_notes(self): with tempfile.TemporaryDirectory() as tmp: mem = os.path.join(tmp, "MEMORY.md") async with _bot(context_limit=200, keep_recent_tokens=40, memory_path=mem) as bot: bot.note("decision: use tabs") await bot.prompt("go " + "y" * 800) with open(mem, encoding="utf-8") as fh: body = fh.read() self.assertIn("decision: use tabs", body) async def test_headdrop_when_nothing_to_summarize(self): async with _bot(context_limit=50, keep_recent_tokens=100000) as bot: for i in range(10): bot.messages.append({"role": "user", "content": "m%d" % i}) bot._usage_input = 1000 info = await bot.compact(reason="manual") self.assertEqual(info["tier"], "headdrop") self.assertEqual(info["dropped_messages"], 2) self.assertEqual(len(bot.messages), 8) async def test_reentrant_compact_noops(self): async with _bot() as bot: bot._compacting = True try: info = await bot.compact() finally: bot._compacting = False self.assertFalse(info["ok"]) self.assertEqual(info["code"], "busy") async def test_llm_summary_path_mocked(self): bot = BlackCatBot(provider="deepseek", system="t", api_key="k", context_limit=100, keep_recent_tokens=10) for i in range(4): bot.messages.append({"role": "user", "content": "body %d " % i + "z" * 200}) bot._usage_input = 1000 payload = {"choices": [{"message": {"content": "CANNED"}}]} with mock.patch.object( DeepSeekClient, "chat", new=mock.AsyncMock(return_value=payload)) as chat: info = await bot.compact(reason="manual") self.assertEqual(info["tier"], "summary") self.assertEqual(bot._summary, "CANNED") sent = chat.call_args[0][0] self.assertIn("Files touched", sent[0]["content"]) async def test_failed_summarizer_falls_back(self): bot = BlackCatBot(provider="deepseek", system="t", api_key="k", context_limit=100, keep_recent_tokens=10) for i in range(5): bot.messages.append({"role": "user", "content": "body %d " % i + "z" * 200}) bot._usage_input = 1000 with mock.patch.object( DeepSeekClient, "chat", new=mock.AsyncMock( side_effect=DeepSeekError("down"))): info = await bot.compact(reason="manual") self.assertEqual(info["tier"], "headdrop") class TestContextLimitResolution(unittest.IsolatedAsyncioTestCase): async def test_model_table(self): self.assertEqual(resolve_context_limit("deepseek-flash"), 1000000) self.assertEqual(resolve_context_limit("deepseek-chat"), 128000) self.assertEqual(resolve_context_limit("openrouter/auto"), 128000) self.assertEqual(resolve_context_limit(None), 128000) async def test_default_uses_model(self): async with _bot() as bot: self.assertEqual(bot.context_limit, 1000000) async with _bot(model="deepseek-chat") as bot: self.assertEqual(bot.context_limit, 128000) async def test_explicit_limit_wins(self): async with _bot(context_limit=200) as bot: self.assertEqual(bot.context_limit, 200) async def test_bad_limit_rejected(self): with self.assertRaises(ValueError): _bot(context_limit=0) class TestTodoTool(unittest.IsolatedAsyncioTestCase): async def test_set_update_add_list(self): async with _bot() as bot: out = await bot.run_tool("todo", action="set", items=["plan", "build", "verify"]) self.assertTrue(out["ok"]) self.assertEqual(out["counts"], {"pending": 3, "in_progress": 0, "done": 0}) out = await bot.run_tool("todo", action="update", index=1, status="in_progress") self.assertTrue(out["ok"]) out = await bot.run_tool("todo", action="update", index=2, status="in_progress") states = [t["status"] for t in bot.todo_list()] self.assertEqual(states, ["pending", "in_progress", "pending"]) await bot.run_tool("todo", action="update", index=2, status="done") await bot.run_tool("todo", action="add", items=["ship"]) out = await bot.run_tool("todo", action="list") self.assertEqual(out["counts"], {"pending": 3, "in_progress": 0, "done": 1}) self.assertIn("[x] 2. build", out["rendered"]) async def test_bad_calls(self): async with _bot() as bot: self.assertEqual((await bot.run_tool("todo", action="set", items="nope"))["code"], "bad_args") self.assertEqual((await bot.run_tool("todo", action="update", index=9, status="done"))["code"], "bad_args") self.assertEqual((await bot.run_tool("todo", action="bogus"))["code"], "bad_args") async def test_todos_on_wire_and_survive_compaction(self): async with _bot(context_limit=100, keep_recent_tokens=10) as bot: await bot.run_tool("todo", action="set", items=["big task"]) for i in range(5): bot.messages.append({"role": "user", "content": "body %d " % i + "z" * 200}) bot._usage_input = 1000 await bot.compact(reason="manual") wire = bot._build_wire() joined = "\n".join(str(m.get("content", "")) for m in wire) self.assertIn("Current task list", joined) self.assertIn("big task", joined) class TestSessionPersistence(unittest.IsolatedAsyncioTestCase): async def test_roundtrip_restores_history_todos_summary(self): with tempfile.TemporaryDirectory() as tmp: path = os.path.join(tmp, "s.json") async with _bot() as bot: await bot.prompt("hello") await bot.run_tool("todo", action="set", items=["step one"]) await bot.save_session(path) async with _bot() as bot: info = await bot.load_session(path) self.assertTrue(info["ok"]) self.assertEqual(info["messages"], 3) self.assertEqual(bot.messages[0]["content"], "hello") self.assertEqual(bot.todo_list()[0]["text"], "step one") self.assertEqual(await bot.prompt("again"), "echo: again") self.assertEqual(len(bot.messages), 5) async def test_model_mismatch_rejected(self): with tempfile.TemporaryDirectory() as tmp: path = os.path.join(tmp, "s.json") async with _bot() as bot: await bot.save_session(path) async with _bot(model="deepseek-chat") as bot: with self.assertRaises(ValueError): await bot.load_session(path) async def test_malformed_rejected(self): with tempfile.TemporaryDirectory() as tmp: path = os.path.join(tmp, "s.json") with open(path, "w") as fh: fh.write('{"version": 99}') async with _bot() as bot: with self.assertRaises(ValueError): await bot.load_session(path) if __name__ == "__main__": unittest.main()