Merge pull request #141 from openqj/feat-harness2-gaps

harness2: upstream gap fill — trivial fix, session budget, tag validation, preflight

cloud/tests/test_harness2_tasks.py

@@ -24,8 +24,8 @@ try:
     )
     from cloud.harness2_tasks import (
         CodexEditor, CodexProbeAdapter, HttpTaskClient, TaskState,
-        build_multipart, run_improvement_via_tasks, run_task,
-        unwrap_result_text,
+        build_multipart, preflight_checks, run_improvement_via_tasks, run_task,
+        unwrap_result_text, validate_run_tag,
     )
     from cloud.uploads import parse_multipart, parse_paths_field
 except ImportError:  # discover -s cloud/tests 时把项目根目录补进 sys.path
@@ -37,8 +37,8 @@ except ImportError:  # discover -s cloud/tests 时把项目根目录补进 sys.p
     )
     from cloud.harness2_tasks import (
         CodexEditor, CodexProbeAdapter, HttpTaskClient, TaskState,
-        build_multipart, run_improvement_via_tasks, run_task,
-        unwrap_result_text,
+        build_multipart, preflight_checks, run_improvement_via_tasks, run_task,
+        unwrap_result_text, validate_run_tag,
     )
     from cloud.uploads import parse_multipart, parse_paths_field
 
@@ -211,7 +211,7 @@ class TestParseHelpers(unittest.TestCase):
 
 
 class TestFullRunViaTasks(unittest.TestCase):
-    def _run(self, client, run_tag="run1", work_sub="work"):
+    def _run(self, client, run_tag="run1", work_sub="work", k=1, max_sessions=None):
         tmp = tempfile.TemporaryDirectory()
         self.addCleanup(tmp.cleanup)
         root = Path(tmp.name)
@@ -221,7 +221,8 @@ class TestFullRunViaTasks(unittest.TestCase):
             base_harness=dict(BASE_TREE), run_tag=run_tag,
             work_dir=root / work_sub, domain="accounting",
             source_files={"ledger.csv": LEDGER},
-            k=1, mode="parallel", library_root=root / "library")
+            k=k, mode="parallel", library_root=root / "library",
+            max_sessions=max_sessions)
         return result, client, root
 
     def test_parallel_k1_via_real_task_shape(self):
@@ -229,10 +230,11 @@ class TestFullRunViaTasks(unittest.TestCase):
         result, _, root = self._run(client)
         self.assertTrue(result.ok)
         self.assertEqual(result.selection.chosen_label, "agg1")
+        self.assertTrue(result.selection.trivial)  # k=1 单终版不开选择会话
         self.assertEqual(result.mode_result.rollouts["agg1"].deliverable_text,
                          "Total: 85.15")
-        # 4 探针 + 3 编辑 + 1 选择 + 1 反思 = 9 个任务
-        self.assertEqual(client.created, 9)
+        # 4 探针 + 3 编辑 + 1 反思 = 8 个任务（选择 trivial，不建任务）
+        self.assertEqual(client.created, 8)
         self.assertIsNotNone(result.library_entry)
         self.assertTrue((root / "library" / "accounting" / "INDEX.md").is_file())
 
@@ -253,8 +255,9 @@ class TestFullRunViaTasks(unittest.TestCase):
         self.assertIsNone(result.library_entry)
 
     def test_selector_gibberish_falls_back(self):
+        # k=1 时选择 trivial，不走选择器；用 k=2 让选择器真正跑起来。
         client = FakeTaskClient(selector_mode="gibberish")
-        result, _, _ = self._run(client)
+        result, _, _ = self._run(client, k=2)
         self.assertTrue(result.ok)
         self.assertTrue(result.selection.fallback)
 
@@ -338,6 +341,79 @@ class TestEnvelopeFullRound(unittest.TestCase):
         self.assertIsNotNone(result.library_entry)
 
 
+class TestSessionBudget(unittest.TestCase):
+    """会话数门：0 挡全部、2 挡在中途；fmt2 与选择/反思不计入。"""
+
+    def _run(self, max_sessions):
+        client = FakeTaskClient()
+        tmp = tempfile.TemporaryDirectory()
+        self.addCleanup(tmp.cleanup)
+        root = Path(tmp.name)
+        result = run_improvement_via_tasks(
+            client, project_id="p1", task_id="ledger-1",
+            prompt="Sum every signed amount in ledger.csv. Return Total: with two decimals.",
+            base_harness=dict(BASE_TREE), run_tag="sess",
+            work_dir=root / "work", domain="accounting",
+            source_files={"ledger.csv": LEDGER},
+            k=1, mode="parallel", library_root=root / "library",
+            max_sessions=max_sessions)
+        return result, client
+
+    def test_zero_blocks_everything(self):
+        result, client = self._run(0)
+        self.assertEqual(client.created, 0)
+        self.assertEqual(set(result.mode_result.fell_back), {"inv1", "dom1", "agg1"})
+
+    def test_mid_cap_stops_after_two_sessions(self):
+        # base 探针（1）+ inv1 编辑（2），之后 dom1/agg1 与探针全被门挡住。
+        result, client = self._run(2)
+        self.assertEqual(client.created, 2)
+        self.assertIn("dom1", result.mode_result.fell_back)
+        self.assertIn("agg1", result.mode_result.fell_back)
+        self.assertNotIn("inv1", result.mode_result.fell_back)
+
+
+class TestValidateRunTag(unittest.TestCase):
+    def test_valid(self):
+        for tag in ("h2-20261029-real1", "a", "a.b_c-1", "run_1"):
+            validate_run_tag(tag)
+
+    def test_invalid(self):
+        for tag in ("-bad", "has/slash", "has space", "s1-foo", "s12"):
+            with self.assertRaises(ValueError, msg=tag):
+                validate_run_tag(tag)
+        validate_run_tag("")  # 空串表示稍后生成，不校验
+
+
+class TestPreflight(unittest.TestCase):
+    def _dirs(self, agents="Use scripts/summarize.py.\n"):
+        tmp = tempfile.TemporaryDirectory()
+        self.addCleanup(tmp.cleanup)
+        root = Path(tmp.name)
+        (root / "base").mkdir()
+        if agents is not None:
+            (root / "base" / "AGENTS.md").write_text(agents)
+        (root / "source").mkdir()
+        (root / "source" / "ledger.csv").write_bytes(LEDGER)
+        return root
+
+    def test_ok(self):
+        root = self._dirs()
+        errors, _ = preflight_checks(
+            base_harness_dir=root / "base", source_dir=root / "source",
+            prompt="do it", run_tag="h2-ok")
+        self.assertEqual(errors, [])
+
+    def test_errors(self):
+        root = self._dirs(agents="")
+        errors, _ = preflight_checks(
+            base_harness_dir=root / "base", source_dir=root / "missing",
+            prompt="  ", run_tag="s1-bad")
+        self.assertEqual(len(errors), 4)  # AGENTS 为空、source 缺失、提示为空、tag 非法
+        errors, _ = preflight_checks(base_harness_dir=root / "nope", prompt="x")
+        self.assertTrue(any("不存在" in e for e in errors))
+
+
 class TestHttpClient(unittest.TestCase):
     def test_client_against_local_server(self):
         seen = {"keys": [], "uploads": [], "tasks": []}