fix(gromacs): preserve first-seen element order

dpdata/formats/gromacs/gro.py

@@ -77,7 +77,12 @@ def file_to_system_data(fname: FileType, format_atom_name=True, **kwargs):
                 posis = np.array(posis)
                 if frame == 1:
                     system["orig"] = np.zeros(3)
-                    system["atom_names"] = list(set(names))
+                    # A .gro file does not carry a separate atom-type table. The
+                    # original reader sorted inferred names alphabetically, but
+                    # the multi-frame refactor accidentally retained an unsorted
+                    # set. Use first occurrence as the explicit contract so type
+                    # IDs remain stable and follow the file's atom order.
+                    system["atom_names"] = list(dict.fromkeys(names))
                     system["atom_numbs"] = [
                         names.count(ii) for ii in system["atom_names"]
                     ]

tests/gromacs/type_order.gro

@@ -0,0 +1,15 @@
+Generated with MDTraj, t= 0.0
+ 12
+    1MOL     LI    1   0.254   0.254   0.024
+    2MOL     CL    2   0.507   0.000   0.000
+    3MOL      P    3   0.507   0.507   0.507
+    4MOL     S1    4   0.896   0.389   0.118
+    5MOL      O    5   0.100   0.200   0.300
+    6MOL      N    6   0.200   0.300   0.400
+    7MOL      C    7   0.300   0.400   0.500
+    8MOL      H    8   0.400   0.500   0.600
+    9MOL      F    9   0.500   0.600   0.700
+   10MOL     BR   10   0.600   0.700   0.800
+   11MOL      I   11   0.700   0.800   0.900
+   12MOL     NA   12   0.800   0.900   1.000
+   1.01400   1.01400   1.01400   0.00000   0.00000   0.00000   0.00000   0.00000   0.00000

tests/test_gromacs_gro.py

@@ -7,6 +7,32 @@
 
 
 class TestGromacsGro(unittest.TestCase):
+    def test_infer_atom_types_in_first_seen_order(self):
+        # Without an explicit type_map, atom names define type IDs in first-seen
+        # order. The fixture uses many distinct names so the former set-based
+        # implementation cannot routinely pass by coincidental hash ordering.
+        system = dpdata.System("gromacs/type_order.gro")
+
+        expected_names = [
+            "Li",
+            "Cl",
+            "P",
+            "S",
+            "O",
+            "N",
+            "C",
+            "H",
+            "F",
+            "Br",
+            "I",
+            "Na",
+        ]
+        self.assertEqual(system["atom_names"], expected_names)
+        self.assertEqual(system["atom_numbs"], [1] * len(expected_names))
+        self.assertEqual(
+            system["atom_types"].tolist(), list(range(len(expected_names)))
+        )
+
     def test_read_file(self):
         system = dpdata.System("gromacs/1h.gro", type_map=["H", "O"])
         self.assertTrue("H" in system["atom_names"])