fix is_time to avoid memory overload

clisops/utils/dataset_utils.py

@@ -5,6 +5,7 @@
 
 import cf_xarray as cfxr  # noqa
 import cftime
+import dask.array as da
 import fsspec
 import numpy as np
 import xarray as xr
@@ -61,6 +62,13 @@ def get_coord_by_type(
     elif isinstance(ds, xr.Dataset):
         # Not all coordinate variables are always classified as such
         coord_vars = list(ds.coords) + list(ds.data_vars)
+        # make sure we skip the main variable!
+        try:
+            var = get_main_variable(ds)
+        except ValueError:
+            warnings.warn(f"No variable found for dataset '{ds}'.")
+        else:
+            coord_vars.remove(var)
     else:
         raise TypeError("Not an xarray.Dataset or xarray.DataArray.")
     for coord_id in coord_vars:
@@ -95,11 +103,12 @@ def get_coord_by_type(
 
         # Select coordinate with most dims (matching with main variable dims)
         for coord_id in coords:
-            if all([dim in main_var_dims for dim in ds.coords[coord_id].dims]):
-                if return_further_matches:
-                    return coord_id, [x for x in coords if x != coord_id]
-                else:
-                    return coord_id
+            if coord_id in ds.coords:
+                if all([dim in main_var_dims for dim in ds.coords[coord_id].dims]):
+                    if return_further_matches:
+                        return coord_id, [x for x in coords if x != coord_id]
+                    else:
+                        return coord_id
         # If the decision making fails, pass the first match
         if return_further_matches:
             return coords[0], coords[1:]
@@ -207,13 +216,38 @@ def is_level(coord):
     return False
 
 
+def _is_time(coord):
+    """
+    Check if a coordinate uses cftime datetime objects.
+    Handles Dask-backed arrays for lazy evaluation.
+    """
+    if coord.size == 0:
+        return False  # Empty array
+
+    if isinstance(coord.dtype.type(), cftime.datetime):
+        return True
+
+    # Safely get first element without loading entire array
+    first_value = coord.isel({dim: 0 for dim in coord.dims}).values
+
+    # Compute only if it's a Dask array
+    if isinstance(first_value, da.Array):
+        first_value = first_value.compute()
+
+    return isinstance(first_value.item(0), cftime.datetime)
+
+
 def is_time(coord):
     """
     Determines if a coordinate is time.
 
     :param coord: coordinate of xarray dataset e.g. coord = ds.coords[coord_id]
     :return: (bool) True if the coordinate is time.
     """
+    if False and coord.ndim >= 2:
+        # skip variables with more than two dimensions: lat_bnds, lon_bnds, time_bnds, t, ...
+        return False
+
     if "time" in coord.cf.coordinates and coord.name in coord.cf.coordinates["time"]:
         return True
 
@@ -226,14 +260,11 @@ def is_time(coord):
     if np.issubdtype(coord.dtype, np.datetime64):
         return True
 
-    if isinstance(np.atleast_1d(coord.values)[0], cftime.datetime):
-        return True
-
     if hasattr(coord, "axis"):
         if coord.axis == "T":
             return True
 
-    return False
+    return _is_time(coord)
 
 
 def is_realization(coord):