conjugate array-like in herm_conj

If value passed to `herm_conj()` function can't be conjugated as an
array or a number, attempt to convert it to a tinyarray is performed.
In that way array-like list of lists/tuples can also be conjugated
correctly.

Fixes #406

doc/source/pre/whatsnew/1.4.rst

@@ -346,3 +346,10 @@ Changes in Kwant 1.4.2
   1.2.6.  SInce Qsymm is an optional dependency of Kwant, this
   backwards-incompatible change only affects Kwant users who use it.
 - Minor bugs were fixed in the KPM module.
+
+Changes after Kwant 1.4.2
+-------------------------
+- Builder properly handles Hermitian conjugation of array-like values.
+  Before, if the value was provided not as a true array but, for example, as a
+  list of lists, it was not conjugated, which could lead to generation of a
+  non-Hermitian Hamiltonian matrix without emitting any error.

kwant/builder.py

@@ -13,6 +13,7 @@ import collections
 import copy
 from functools import total_ordering, wraps, update_wrapper
 from itertools import islice, chain
+import numbers
 import inspect
 import tinyarray as ta
 import numpy as np
@@ -499,8 +500,27 @@ def herm_conj(value):
     """
     if hasattr(value, 'conjugate'):
         value = value.conjugate()
-        if hasattr(value, 'transpose'):
+    elif not isinstance(value, numbers.Number):
+        # Fallback for the case of array-like: see issue 406
+        # https://gitlab.kwant-project.org/kwant/kwant/-/issues/406
+        # If we get a list of lists or another array-like, conjugate()
+        # attribute won't be present. To be able to conjugate them, we try to
+        # convert them to tinyarrays
+        try:
+            value = ta.array(value).conjugate()
+        except Exception:
+            pass
+
+    if hasattr(value, 'shape'):
+        if len(value.shape) > 2:
+            is_ta = isinstance(value, (
+                ta.ndarray_int, ta.ndarray_float, ta.ndarray_complex))
+            value = np.swapaxes(value, -1, -2)
+            if is_ta:
+                value = ta.array(value)
+        else:
             value = value.transpose()
+
     return value