change TranslationalSymmetry.act() to reject non-integer input

Previously, if a tuple of floats was passed as the 'domain' argument to
'act' then the tags of the resulting sites would be floats, rather than
integers. This is because the tags of sites created inside 'act' were
not normalized to integers, as they would usually be, and this causes
bugs in other locations where an integer dtype is assumed.

kwant/lattice.py

@@ -693,6 +693,9 @@ class TranslationalSymmetry(builder.Symmetry):
         return -result if self.is_reversed else result
 
     def act(self, element, a, b=None):
+        element = ta.array(element)
+        if element.dtype is not int:
+            raise ValueError("group element must be a tuple of integers")
         m_part = self._get_site_family_data(a.family)[0]
         try:
             delta = ta.dot(m_part, element)

kwant/tests/test_lattice.py

@@ -232,6 +232,25 @@ def test_norbs():
     assert lat1 != lat2
 
 
+def test_symmetry_act():
+    lat = lattice.square()
+    sym = lattice.TranslationalSymmetry((1, 0), (0, 1))
+    site = lat(0, 0)
+    hopping = (lat(0, 0), lat(1, 0))
+    el = (1, 0)
+
+    # Verify that the dtype of tags of sites returned by 'act' is 'int'
+    for el in [el, ta.array(el, int)]:
+        assert sym.act(el, site).tag.dtype is int
+        assert all(s.tag.dtype is int for s in sym.act(el, *hopping))
+
+    for el in [(1.0, 0), (1.5, 0)]:
+        with raises(ValueError):
+            sym.act(el, site)
+        with raises(ValueError):
+            sym.act(ta.array(el), site)
+
+
 def test_symmetry_subgroup():
     rng = np.random.RandomState(0)
     ## test whether actual subgroups are detected as such