convert promote argument to fill_value

Author: toddrjen

xarray/core/computation.py

@@ -23,7 +23,7 @@
 
 import numpy as np
 
-from . import dtypes, duck_array_ops, utils
+from . import duck_array_ops, utils
 from .alignment import deep_align
 from .merge import merge_coordinates_without_align
 from .nanops import dask_array
@@ -1347,7 +1347,7 @@ def _calc_idxminmax(
     func: Callable,
     dim: Optional[Hashable],
     skipna: Optional[bool],
-    promote: Optional[bool],
+    fill_value: Any = None,
     keep_attrs: Optional[bool],
     **kwargs: Any,
 ):
@@ -1377,19 +1377,6 @@ def _calc_idxminmax(
         # Need to skip NaN values since argmin and argmax can't handle them
         allna = array.isnull().all(dim)
         array = array.where(~allna, 0)
-        hasna = allna.any()
-    else:
-        allna = None
-        hasna = False
-
-    # If promote is None we only promote if there are NaN values.
-    if promote is None:
-        promote = hasna
-
-    if not promote and hasna and array.coords[dim].dtype.kind not in na_dtypes:
-        raise TypeError(
-            "NaN values present for NaN-incompatible dtype and Promote=False"
-        )
 
     # This will run argmin or argmax.
     indx = func(array, dim=dim, axis=None, keep_attrs=False, skipna=skipna, **kwargs)
@@ -1405,13 +1392,8 @@ def _calc_idxminmax(
             indx,
         ]
 
-    # Promote to a dtype that can handle NaN values if needed.
-    newdtype, fill_value = dtypes.maybe_promote(res.dtype)
-    if promote and newdtype != res.dtype:
-        res = res.astype(newdtype)
-
-    # Put the NaN values back in after removing them, if necessary.
-    if hasna and allna is not None:
+    if skipna or (skipna is None and array.dtype.kind in na_dtypes):
+        # Put the NaN values back in after removing them
         res = res.where(~allna, fill_value)
 
     # The dim is gone but we need to remove the corresponding coordinate.

xarray/core/dataarray.py

@@ -3509,7 +3509,7 @@ def idxmin(
         self,
         dim: Hashable = None,
         skipna: bool = None,
-        promote: bool = None,
+        fill_value: Any = np.NaN,
         keep_attrs: Optional[bool] = False,
         **kwargs: Any,
     ) -> "DataArray":
@@ -3532,11 +3532,11 @@ def idxmin(
             skips missing values for float dtypes; other dtypes either do not
             have a sentinel missing value (int) or skipna=True has not been
             implemented (object, datetime64 or timedelta64).
-        promote : bool or None, default True
-            If True (default) dtypes that do not support NaN values will be
-            automatically promoted to those that do.  If False a NaN in the
-            results will raise a TypeError.  If None the result will only be
-            promoted if a NaN is actually present.
+        fill_value : Any, default NaN
+            Value to be filled in case all of the values along a dimension are
+            null.  By default this is NaN.  The fill value and result are
+            automatically converted to a compatible dtype if possible.
+            Ignored if `skipna` is False.
         keep_attrs : bool, default False
             If True, the attributes (`attrs`) will be copied from the original
             object to the new one.  If False (default), the new object will be
@@ -3598,7 +3598,7 @@ def idxmin(
             func=lambda x, *args, **kwargs: x.argmin(*args, **kwargs),
             dim=dim,
             skipna=skipna,
-            promote=promote,
+            fill_value=fill_value,
             keep_attrs=keep_attrs,
             **kwargs,
         )
@@ -3607,7 +3607,7 @@ def idxmax(
         self,
         dim: Hashable = None,
         skipna: bool = None,
-        promote: bool = None,
+        fill_value: Any = np.NaN,
         keep_attrs: Optional[bool] = False,
         **kwargs: Any,
     ) -> "DataArray":
@@ -3630,11 +3630,11 @@ def idxmax(
             skips missing values for float dtypes; other dtypes either do not
             have a sentinel missing value (int) or skipna=True has not been
             implemented (object, datetime64 or timedelta64).
-        promote : bool or None, default True
-            If True (default) dtypes that do not support NaN values will be
-            automatically promoted to those that do.  If False a NaN in the
-            results will raise a TypeError.  If None the result will only be
-            promoted if a NaN is actually present.
+        fill_value : Any, default NaN
+            Value to be filled in case all of the values along a dimension are
+            null.  By default this is NaN.  The fill value and result are
+            automatically converted to a compatible dtype if possible.
+            Ignored if `skipna` is False.
         keep_attrs : bool, default False
             If True, the attributes (`attrs`) will be copied from the original
             object to the new one.  If False (default), the new object will be
@@ -3696,7 +3696,7 @@ def idxmax(
             func=lambda x, *args, **kwargs: x.argmax(*args, **kwargs),
             dim=dim,
             skipna=skipna,
-            promote=promote,
+            fill_value=fill_value,
             keep_attrs=keep_attrs,
             **kwargs,
         )

xarray/core/dataset.py

@@ -6098,7 +6098,7 @@ def idxmin(
         self,
         dim: Hashable = None,
         skipna: bool = None,
-        promote: bool = None,
+        fill_value: Any = np.NaN,
         keep_attrs: bool = False,
         **kwargs: Any,
     ) -> "Dataset":
@@ -6121,11 +6121,11 @@ def idxmin(
             skips missing values for float dtypes; other dtypes either do not
             have a sentinel missing value (int) or skipna=True has not been
             implemented (object, datetime64 or timedelta64).
-        promote : bool or None, default True
-            If True (default) dtypes that do not support NaN values will be
-            automatically promoted to those that do.  If False a NaN in the
-            results will raise a TypeError.  If None the result will only be
-            promoted if a NaN is actually present.
+        fill_value : Any, default NaN
+            Value to be filled in case all of the values along a dimension are
+            null.  By default this is NaN.  The fill value and result are
+            automatically converted to a compatible dtype if possible.
+            Ignored if `skipna` is False.
         keep_attrs : bool, default False
             If True, the attributes (`attrs`) will be copied from the original
             object to the new one.  If False (default), the new object will be
@@ -6187,7 +6187,7 @@ def idxmin(
                 "idxmin",
                 dim=dim,
                 skipna=skipna,
-                promote=promote,
+                fill_value=fill_value,
                 keep_attrs=keep_attrs,
                 **kwargs,
             ),
@@ -6197,7 +6197,7 @@ def idxmax(
         self,
         dim: Hashable = None,
         skipna: bool = None,
-        promote: bool = None,
+        fill_value: Any = np.NaN,
         keep_attrs: Optional[bool] = False,
         **kwargs: Any,
     ) -> "Dataset":
@@ -6220,11 +6220,11 @@ def idxmax(
             skips missing values for float dtypes; other dtypes either do not
             have a sentinel missing value (int) or skipna=True has not been
             implemented (object, datetime64 or timedelta64).
-        promote : bool or None, default True
-            If True (default) dtypes that do not support NaN values will be
-            automatically promoted to those that do.  If False a NaN in the
-            results will raise a TypeError.  If None the result will only be
-            promoted if a NaN is actually present.
+        fill_value : Any, default NaN
+            Value to be filled in case all of the values along a dimension are
+            null.  By default this is NaN.  The fill value and result are
+            automatically converted to a compatible dtype if possible.
+            Ignored if `skipna` is False.
         keep_attrs : bool, default False
             If True, the attributes (`attrs`) will be copied from the original
             object to the new one.  If False (default), the new object will be
@@ -6248,7 +6248,7 @@ def idxmax(
                 "idxmax",
                 dim=dim,
                 skipna=skipna,
-                promote=promote,
+                fill_value=fill_value,
                 keep_attrs=keep_attrs,
                 **kwargs,
             ),