import unittest from tests.fakes import Harness, config_data class CoreTest(unittest.TestCase): def setUp(self): self.h = Harness() self.p = self.h.plane def tearDown(self): self.h.close() def codex(self): return self.h.execs[("win10", "codex")] def test_pick_starts_lowest_todo_and_marks_in_progress(self): a, b = self.p.add_issue(7), self.p.add_issue(3) self.p.add_issue(1, state="Backlog") self.p.add_issue(2, assignee="kappiee") # not routed self.h.tick() runs = self.codex().started self.assertEqual([r.run_id for r in runs], ["GW-3-1"]) # concurrency 1 on win10 self.assertEqual(self.p.state_of(b), "In Progress") self.assertEqual(self.p.state_of(a), "Todo") self.assertTrue(self.p.texts(b)[-1].startswith("[worker] 第 1 轮开始：codex / sol / medium @ win10，新线程")) self.assertIsNone(runs[0].thread_id) self.assertIn("do it", runs[0].prompt) def test_review_question_progress_transitions(self): for tag, state, who in (("review", "Review", "oregon"), ("question", "Waiting", "kappiee"), ("blocked", "Waiting", "kappiee"), ("progress", "Todo", "window10")): with self.subTest(tag=tag): h = Harness() iid = h.plane.add_issue(1) h.tick() ex = h.execs[("win10", "codex")] ex.finish("[%s] text" % tag) h.tick() if tag == "progress": # picked again in the same tick, resuming the thread self.assertEqual(len(ex.started), 2) self.assertEqual(ex.started[1].thread_id, "th-1") self.assertEqual(h.plane.state_of(iid), "In Progress") self.assertIn(("i1", "s-todo", ["u-w10"]), h.plane.patches) else: self.assertEqual((h.plane.state_of(iid), h.plane.assignee_of(iid)), (state, who)) texts = h.plane.texts(iid) result = [t for t in texts if t.startswith("[codex]")][0] self.assertIn("第 1 轮结束（exit=0，run=GW-1-1，终端 win10", result) self.assertIn("[%s] text" % tag, result) self.assertEqual(h.rec(iid)["tags"], [tag]) h.close() def test_untagged_message_becomes_review(self): iid = self.p.add_issue(1) self.h.tick() self.codex().finish("all done") self.h.tick() self.assertEqual(self.p.state_of(iid), "Review") self.assertIn("未按协议标注首行", self.p.texts(iid)[-1]) def test_resume_sends_only_new_comments(self): iid = self.p.add_issue(1) self.p.add_comment(iid, "

old context

") self.h.tick() self.codex().finish("[question] A or B?") self.h.tick() self.p.add_comment(iid, "

B please

") self.p.issues[iid].update(state="s-todo", assignees=["u-w10"]) self.h.tick() run = self.codex().started[-1] self.assertEqual(run.thread_id, "th-1") self.assertIn("B please", run.prompt) self.assertNotIn("old context", run.prompt) self.assertNotIn("A or B?", run.prompt) self.assertIn("续线程 th-1", self.p.texts(iid)[-1]) def test_transient_failure_retries_with_backoff_then_waits(self): iid = self.p.add_issue(1) self.h.tick() self.codex().finish("", exit_code=1, tail="Connection reset by peer") self.h.tick() note = self.p.texts(iid)[-1] self.assertIn("类型=crash（临时性）", note) self.assertIn("重试 1/2", note) self.assertEqual(self.p.state_of(iid), "In Progress") self.assertEqual(len(self.codex().started), 1) # backoff 2 min not elapsed self.h.clock[0] += 121 self.h.tick() self.assertEqual(len(self.codex().started), 2) self.assertEqual(self.codex().started[1].thread_id, "th-1") # resumes the original thread self.assertIn("临时性错误中断", self.codex().started[1].prompt) self.codex().finish("", exit_code=1, tail="HTTP 503 Service Unavailable") self.h.tick() self.h.clock[0] += 301 self.h.tick() self.assertEqual(len(self.codex().started), 3) self.codex().finish("", exit_code=1, tail="banner exchange failed") self.h.tick() self.assertEqual((self.p.state_of(iid), self.p.assignee_of(iid)), ("Waiting", "kappiee")) self.assertIn("不再自动重试", self.p.texts(iid)[-1]) self.assertEqual(self.h.rec(iid)["retries"], 0) def test_logic_failure_on_fresh_thread_goes_waiting(self): iid = self.p.add_issue(1) self.h.tick() self.codex().finish("", exit_code=2, thread=None, tail="panic: bad config") self.h.tick() self.assertEqual(self.p.state_of(iid), "Waiting") self.assertIn("类型=crash，run=GW-1-1，exit=2，重试 0/2", self.p.texts(iid)[-1]) self.assertIn("panic: bad config", self.p.texts(iid)[-1]) def test_resume_failure_rebuilds_thread_with_handoff(self): iid = self.p.add_issue(1) self.h.tick() self.codex().finish("[progress] half way") self.h.tick() # finishes round 1, starts round 2 resuming th-1 self.assertEqual(self.codex().started[1].thread_id, "th-1") self.codex().finish("", exit_code=1, thread=None, tail="thread not found") self.h.tick() # round 2 fails -> rebuild immediately run = self.codex().started[2] self.assertIsNone(run.thread_id) self.assertIn("## 交接摘要", run.prompt) self.assertIn("续线程失败", run.prompt) self.assertIn("[progress] half way", run.prompt) self.assertIn("新线程（重建：续线程失败（crash））", self.p.texts(iid)[-1]) def test_thread_rebuilt_after_max_rounds(self): iid = self.p.add_issue(1) self.h.tick() for _ in range(3): self.codex().finish("[progress] more") self.h.tick() runs = self.codex().started self.assertEqual([r.thread_id for r in runs], [None, "th-1", "th-1", None]) self.assertIn("线程已续 3 轮（上限 3）", runs[3].prompt) self.assertIn("progress, progress, progress", runs[3].prompt) self.assertEqual(self.h.rec(iid)["thread_rounds"], 1) def test_timeout_kills_and_waits(self): iid = self.p.add_issue(1, labels=["budget:30m"]) self.h.tick() self.h.clock[0] += 29 * 60 self.h.tick() self.assertEqual(self.codex().killed, []) self.h.clock[0] += 2 * 60 self.h.tick() self.assertEqual(self.codex().killed, ["GW-1-1"]) self.assertEqual(self.p.state_of(iid), "Waiting") self.assertIn("类型=timeout", self.p.texts(iid)[-1]) self.assertIn("0.5 h", self.p.texts(iid)[-1]) def test_serialize_group_and_concurrency(self): a = self.p.add_issue(1, assignee="oregon", labels=["needs-game"]) b = self.p.add_issue(2, assignee="oregon", labels=["needs-game"]) c = self.p.add_issue(3, assignee="oregon") d = self.p.add_issue(4, assignee="oregon") self.h.tick() started = [r.run_id for r in self.h.execs[("oregon", "claude-code")].started] self.assertEqual(started, ["GW-1-1", "GW-3-1"]) # 2 slots; GW-2 waits for the game group self.assertEqual([self.p.state_of(x) for x in (a, b, c, d)], ["In Progress", "Todo", "In Progress", "Todo"]) def test_label_routing(self): self.p.add_issue(1, labels=["model:astra"]) self.p.add_issue(2, assignee="oregon", labels=["cap:gpu"]) # oregon lacks the capability self.h.tick() run = self.codex().started[0] self.assertEqual((run.model, run.effort), ("astra", "high")) self.assertNotIn(("oregon", "claude-code"), self.h.execs) def test_human_moving_item_during_run_is_respected(self): iid = self.p.add_issue(1) self.h.tick() self.p.issues[iid]["state"] = "s-done" self.codex().finish("[review] ok") self.h.tick() self.assertEqual(self.p.state_of(iid), "Done") self.assertIn("[review] ok", self.p.texts(iid)[-1]) def test_post_failure_keeps_run_in_flight(self): iid = self.p.add_issue(1) self.h.tick() self.codex().finish("[review] ok") orig = self.p.post_comment self.p.post_comment = lambda *a: (_ for _ in ()).throw(RuntimeError("plane down")) self.h.tick() self.assertIn("inflight", self.h.rec(iid)) self.p.post_comment = orig self.h.tick() self.assertNotIn("inflight", self.h.rec(iid)) self.assertEqual(self.p.state_of(iid), "Review") def test_heartbeat_written(self): self.p.add_issue(1) self.h.tick() import json, pathlib hb = json.loads((pathlib.Path(self.h.tmp.name) / "heartbeat.json").read_text()) self.assertEqual(hb["inflight"][0]["run"], "GW-1-1") self.assertEqual(hb["errors"], []) class EnvFailureTest(unittest.TestCase): def test_env_failure_pauses_group_until_item_leaves_waiting(self): data = config_data() data["projects"]["GWENT"]["retry"]["env_patterns"] = ["foreground window lost"] h = Harness(data) a = h.plane.add_issue(1, labels=["needs-game"]) b = h.plane.add_issue(2, labels=["needs-game"]) h.tick() ex = h.execs[("win10", "codex")] ex.finish("", exit_code=1, tail="foreground window lost") h.tick() self.assertEqual(h.plane.state_of(a), "Waiting") self.assertEqual(h.plane.state_of(b), "Todo") # group paused self.assertIn("暂停派发", h.plane.texts(a)[-1]) h.plane.issues[a]["state"] = "s-done" h.tick() self.assertEqual(h.plane.state_of(b), "In Progress") h.close() if __name__ == "__main__": unittest.main()