fix: serialize LAMMPS HDF5 trajectories

dpgen2/op/run_lmp.py

@@ -202,7 +202,7 @@ def execute(
 
         ret_dict = {
             "log": work_dir / lmp_log_name,
-            "traj": work_dir / lmp_traj_name,
+            "traj": self.get_traj(work_dir / lmp_traj_name),
             "model_devi": self.get_model_devi(work_dir / lmp_model_devi_name),
         }
         plm_output = (
@@ -223,6 +223,10 @@ def execute(
     def get_model_devi(self, model_devi_file):
         return model_devi_file
 
+    def get_traj(self, traj_file):
+        """Return the filesystem trajectory used by the standard backend."""
+        return traj_file
+
     @staticmethod
     def lmp_args():
         doc_lmp_cmd = "The command of LAMMPS"
@@ -412,3 +416,7 @@ def get_output_sign(cls):
 
     def get_model_devi(self, model_devi_file):
         return np.loadtxt(model_devi_file)
+
+    def get_traj(self, traj_file):
+        """Return trajectory text for serialization into an HDF5 dataset."""
+        return traj_file.read_text()

tests/op/test_run_lmp.py

@@ -35,6 +35,7 @@
 )
 from dpgen2.op.run_lmp import (
     RunLmp,
+    RunLmpHDF5,
     get_ele_temp,
     merge_pimd_files,
     set_models,
@@ -149,6 +150,31 @@ def test_extra_outputs(self):
             "Hello -i in.lammps -log log.lammps",
         )
 
+    @patch("dpgen2.op.run_lmp.run_command")
+    def test_hdf5_outputs_dataset_values(self, mocked_run):
+        """Return serializable data instead of filesystem paths for HDF5."""
+
+        def write_outputs(*args, **kwargs):
+            Path(lmp_traj_name).write_text("trajectory data")
+            np.savetxt(lmp_model_devi_name, np.arange(7).reshape(1, 7))
+            return 0, "foo\n", ""
+
+        mocked_run.side_effect = write_outputs
+
+        out = RunLmpHDF5().execute(
+            OPIO(
+                {
+                    "config": {"command": "mylmp"},
+                    "task_name": self.task_name,
+                    "task_path": self.task_path,
+                    "models": self.models,
+                }
+            )
+        )
+
+        self.assertEqual(out["traj"], "trajectory data")
+        np.testing.assert_array_equal(out["model_devi"], np.arange(7))
+
 
 class TestRunLmpDist(unittest.TestCase):
     lmp_config = """variable        NSTEPS          equal 1000