Fix plots bug due to new pandas casting

changes/983.fix.rst

@@ -0,0 +1 @@
+bug fix for pandas >= 3 migration

src/sisl/viz/plotters/xarray.py

@@ -8,13 +8,48 @@
 from typing import Literal, Optional, Union
 
 import numpy as np
+import pandas as pd
 from xarray import DataArray, Dataset
 
 from . import plot_actions
 
 # from sisl.viz.nodes.processors.grid import get_isos
 
 
+def _correct_pandas_string_none(dic: dict[str, DataArray]):
+    """Correct string arguments in xarray arrays to avoid casting to nan
+
+    This is a migration notice from pandas >= 3.
+    They change their str implementation to a custom (new default) dtype.
+    And the result is that `None` will be parsed to `np.nan` as opposed
+    to the old `None -> None` parsing.
+    """
+
+    # xarray uses pandas for casting types, and pandas decided to change
+    # the default behavior and parse [None, ""], to [nan, ""]. Passing
+    # NaN to plotting functions when they are expecting None breaks things.
+    # Here we make sure that we parse all NaN to None
+    # Pandas docs on this:
+    # https://pandas.pydata.org/docs/user_guide/migration-3-strings.html
+    old_infer_string = pd.options.future.infer_string
+    pd.options.future.infer_string = False
+
+    for key, value in dic.items():
+        value = dic[key]
+        try:
+            dic[key] = value.where(value.notnull(), other=None)
+        except:
+            pass
+
+    pd.options.future.infer_string = old_infer_string
+
+
+if int(pd.__version__.split(".")[0]) >= 3:
+
+    def _correct_pandas_string_none(dic: dict[str, str]):
+        pass
+
+
 def _process_xarray_data(
     data: Union[DataArray, Dataset],
     x: Union[str, None] = None,
@@ -461,6 +496,8 @@
         "border_width",
         "border_color",
     )
+    _correct_pandas_string_none(style)
+
     for key in style_keys:
         lines_style[key] = style.get(key)