Fix 10 failing tests (structure_mapper confidence, hardcoded path, e2e slot assertions, cpp StateStore call)

metainfer/tasks/gen_infer_framework_cpp/tests/test_cpp_framework_task.py

@@ -693,7 +693,7 @@ def test_correctness_only_mode_stops_after_d_review(self):
             root = Path(td)
             state_dir = root / "state"
             store = StateStore(state_dir)
-            store.init_or_resume(task_id="bounded", task_type=TASK_TYPE)
+            store.init_or_resume(task_id="bounded")
             cfg = OrchestratorConfig(
                 workdir=state_dir,
                 repo_root=root,
@@ -738,7 +738,7 @@ def test_optimize_mode_stops_after_measured_target_and_review(self):
             root = Path(td)
             state_dir = root / "state"
             store = StateStore(state_dir)
-            store.init_or_resume(task_id="target", task_type=TASK_TYPE)
+            store.init_or_resume(task_id="target")
             cfg = OrchestratorConfig(
                 workdir=state_dir,
                 repo_root=root,

metainfer/tasks/harness_evolve/tests/test_publish.py

@@ -14,8 +14,8 @@
     publish_kernels,
 )
 
-_DCU_SEED = Path("/root/zth_agent/MetaInfer/metainfer/tasks/"
-                 "dcu_kernel_auto_opt/harness_default")
+_DCU_SEED = (Path(__file__).resolve().parents[2]
+             / "dcu_kernel_auto_opt" / "harness_default")
 
 
 def _fake_experiment(tmp_path, *, verdict="PROMOTE", held="PASS"):

metainfer/tasks/sglang_trace_analyze/orchestrator/structure_mapper.py

@@ -74,6 +74,9 @@ def _map_one(
             confidence = "low"
     elif layer is None:
         confidence = "medium"
+    else:
+        # Call stack present and the layer was resolved from it.
+        confidence = "high"
 
     return {
         "kernel_name": kernel_name,

metainfer/tests/e2e/test_admin_force_kill.py

@@ -50,7 +50,8 @@ def test_force_release_endpoint_frees_slot(tmp_path: Path) -> None:
     assert body["was_held"] is True
 
     # Slot must be gone.
-    claims = scoreboard.list_claims()
+    # list_claims() reports every GPU (free or held); only held rows count.
+    claims = [c for c in scoreboard.list_claims() if c["status"] == "held"]
     assert all(not (c["node_id"] == "w0" and c["gpu_idx"] == 0) for c in claims)
 
     # cancel.marker must be present in the job_dir so the worker sees it.

metainfer/tests/e2e/test_pp2_contention.py

@@ -56,5 +56,6 @@ def run_round(round_idx: int) -> None:
     assert len(winners_per_round) == rounds, \
         "all rounds must complete (no deadlock)"
     # After all rounds done, no slots remain held.
-    remaining = scoreboard.list_claims()
+    # list_claims() reports every GPU (free or held); only held rows count.
+    remaining = [c for c in scoreboard.list_claims() if c["status"] == "held"]
     assert remaining == [], f"no slot should remain held, got {remaining}"

metainfer/tests/e2e/test_remote_job_flow.py

@@ -40,7 +40,8 @@ def test_full_remote_job_flow_acquire_release_result(tmp_path: Path) -> None:
     def handler(handle: JobHandle, own_node_id: str) -> JobResult:
         # While the job is running, the slot must be held.
         claims = scoreboard.list_claims()
-        acquired = [c for c in claims if c["node_id"] == "w0" and c["gpu_idx"] == 0]
+        acquired = [c for c in claims if c["node_id"] == "w0" and c["gpu_idx"] == 0
+                    and c["status"] == "held"]
         captured["acquired_during_run"] = len(acquired) == 1
         # Write some stdout so tail-readers can verify
         (Path(handle.job_dir) / "stdout.log").write_text("hello e2e\n")
@@ -69,7 +70,7 @@ def handler(handle: JobHandle, own_node_id: str) -> JobResult:
         "GPU slot must be held while job is running"
 
     # Slot must be released after job.
-    claims_after = scoreboard.list_claims()
+    claims_after = [c for c in scoreboard.list_claims() if c["status"] == "held"]
     assert all(not (c["node_id"] == "w0" and c["gpu_idx"] == 0)
                for c in claims_after), "slot must be released"
 

metainfer/tests/e2e/test_worker_crash.py

@@ -47,6 +47,6 @@ def test_worker_dead_surfaces_and_slot_reaped(tmp_path: Path) -> None:
     assert result.status == "worker_dead", \
         f"expected worker_dead, got {result.status}"
     # Slot must be released (RemoteJob's finally calls release_gpus).
-    claims = scoreboard.list_claims()
+    claims = [c for c in scoreboard.list_claims() if c["status"] == "held"]
     assert all(not (c["node_id"] == "w0" and c["gpu_idx"] == 0) for c in claims), \
         f"slot leaked: {claims}"