fix: select SIESTA pseudopotentials per task

Coding-Agent: Codex
Codex-Version: codex-cli 0.149.0
Model: gpt-5.6-sol
Reasoning-Effort: xhigh

Author: njzjz-bot

dpgen/generator/run.py

@@ -3710,6 +3710,10 @@ def make_fp_siesta(iter_index, jdata):
     iter_name = make_iter_name(iter_index)
     work_path = os.path.join(iter_name, fp_name)
     fp_pp_files = jdata["fp_pp_files"]
+    type_map = jdata["type_map"]
+    if len(type_map) != len(fp_pp_files):
+        raise RuntimeError("fp_pp_files must correspond one-to-one with type_map")
+    pp_by_element = dict(zip(type_map, fp_pp_files))
     if "user_fp_params" in jdata.keys():
         fp_params = jdata["user_fp_params"]
         user_input = True
@@ -3720,7 +3724,10 @@ def make_fp_siesta(iter_index, jdata):
     for ii in fp_tasks:
         os.chdir(ii)
         sys_data = dpdata.System("POSCAR").data
-        ret = make_siesta_input(sys_data, fp_pp_files, fp_params)
+        # A workflow may share a global type map while an individual system
+        # contains only a subset of those elements.
+        task_pp_files = [pp_by_element[name] for name in sys_data["atom_names"]]
+        ret = make_siesta_input(sys_data, task_pp_files, fp_params)
         with open("input", "w") as fp:
             fp.write(ret)
         os.chdir(cwd)

tests/generator/test_make_fp.py

@@ -991,6 +991,42 @@ def test_make_fp_siesta(self):
         _check_potcar(self, 0, jdata["fp_pp_path"], jdata["fp_pp_files"])
         shutil.rmtree("iter.000000")
 
+    def test_make_fp_siesta_with_element_subset(self):
+        """Each SIESTA input should select PPs for its present elements."""
+        setUpModule()
+        if os.path.isdir("iter.000000"):
+            shutil.rmtree("iter.000000")
+        with open(param_siesta_file) as fp:
+            jdata = json.load(fp)
+        md_descript = []
+        for _ in range(2):
+            system_deviations = []
+            for _ in range(3):
+                system_deviations.append(np.arange(0, 0.29, 0.29 / 10))
+            md_descript.append(system_deviations)
+
+        try:
+            _make_fake_md(0, md_descript, [0] * 6, jdata["type_map"])
+            make_fp(0, jdata, {})
+
+            tasks = glob.glob(os.path.join("iter.000000", "02.fp", "task.*"))
+            self.assertGreater(len(tasks), 0)
+            for task in tasks:
+                with open(os.path.join(task, "input")) as fp:
+                    input_text = fp.read()
+                self.assertIn("NumberOfSpecies   1", input_text)
+                species_block = input_text.split(
+                    "%block Chemical_Species_label\n", maxsplit=1
+                )[1].split("%endblock Chemical_Species_label", maxsplit=1)[0]
+                species_lines = [
+                    line for line in species_block.splitlines() if line.strip()
+                ]
+                self.assertEqual(1, len(species_lines))
+                self.assertEqual("C", species_lines[0].split()[-1])
+        finally:
+            if os.path.isdir("iter.000000"):
+                shutil.rmtree("iter.000000")
+
 
 class TestMakeFPVasp(unittest.TestCase):
     def test_make_fp_vasp(self):