Merge remote-tracking branch 'upstream/2.3.x' into backport-60295

Author: jorisvandenbossche

.circleci/config.yml

@@ -15,7 +15,6 @@ jobs:
       - checkout
       - run: .circleci/setup_env.sh
       - run: |
-          sudo apt-get update && sudo apt-get install -y libegl1 libopengl0
           PATH=$HOME/miniconda3/envs/pandas-dev/bin:$HOME/miniconda3/condabin:$PATH \
           LD_PRELOAD=$HOME/miniconda3/envs/pandas-dev/lib/libgomp.so.1:$LD_PRELOAD \
           ci/run_tests.sh

doc/source/whatsnew/v2.3.0.rst

@@ -119,7 +119,7 @@ Interval
 
 Indexing
 ^^^^^^^^
--
+- Fixed bug in :meth:`Index.get_indexer` round-tripping through string dtype when ``infer_string`` is enabled (:issue:`55834`)
 -
 
 Missing

pandas/_libs/index.pyi

@@ -68,6 +68,9 @@ class MaskedUInt16Engine(MaskedIndexEngine): ...
 class MaskedUInt8Engine(MaskedIndexEngine): ...
 class MaskedBoolEngine(MaskedUInt8Engine): ...
 
+class StringObjectEngine(ObjectEngine):
+    def __init__(self, values: object, na_value) -> None: ...
+
 class BaseMultiIndexCodesEngine:
     levels: list[np.ndarray]
     offsets: np.ndarray  # ndarray[uint64_t, ndim=1]

pandas/_libs/index.pyx

@@ -532,6 +532,32 @@ cdef class ObjectEngine(IndexEngine):
         return loc
 
 
+cdef class StringObjectEngine(ObjectEngine):
+
+    cdef:
+        object na_value
+        bint uses_na
+
+    def __init__(self, ndarray values, na_value):
+        super().__init__(values)
+        self.na_value = na_value
+        self.uses_na = na_value is C_NA
+
+    cdef bint _checknull(self, object val):
+        if self.uses_na:
+            return val is C_NA
+        else:
+            return util.is_nan(val)
+
+    cdef _check_type(self, object val):
+        if isinstance(val, str):
+            return val
+        elif self._checknull(val):
+            return self.na_value
+        else:
+            raise KeyError(val)
+
+
 cdef class DatetimeEngine(Int64Engine):
 
     cdef:

pandas/_libs/lib.pyi

@@ -86,6 +86,7 @@ def maybe_convert_objects(
     safe: bool = ...,
     convert_numeric: bool = ...,
     convert_non_numeric: Literal[False] = ...,
+    convert_string: Literal[False] = ...,
     convert_to_nullable_dtype: Literal[False] = ...,
     dtype_if_all_nat: DtypeObj | None = ...,
 ) -> npt.NDArray[np.object_ | np.number]: ...
@@ -97,6 +98,7 @@ def maybe_convert_objects(
     safe: bool = ...,
     convert_numeric: bool = ...,
     convert_non_numeric: bool = ...,
+    convert_string: bool = ...,
     convert_to_nullable_dtype: Literal[True] = ...,
     dtype_if_all_nat: DtypeObj | None = ...,
 ) -> ArrayLike: ...
@@ -108,6 +110,7 @@ def maybe_convert_objects(
     safe: bool = ...,
     convert_numeric: bool = ...,
     convert_non_numeric: bool = ...,
+    convert_string: bool = ...,
     convert_to_nullable_dtype: bool = ...,
     dtype_if_all_nat: DtypeObj | None = ...,
 ) -> ArrayLike: ...

pandas/_libs/lib.pyx

@@ -2498,6 +2498,7 @@ def maybe_convert_objects(ndarray[object] objects,
                           bint convert_numeric=True,  # NB: different default!
                           bint convert_to_nullable_dtype=False,
                           bint convert_non_numeric=False,
+                          bint convert_string=True,
                           object dtype_if_all_nat=None) -> "ArrayLike":
     """
     Type inference function-- convert object array to proper dtype
@@ -2747,7 +2748,11 @@ def maybe_convert_objects(ndarray[object] objects,
             dtype = StringDtype()
             return dtype.construct_array_type()._from_sequence(objects, dtype=dtype)
 
-        elif using_string_dtype() and is_string_array(objects, skipna=True):
+        elif (
+            convert_string
+            and using_string_dtype()
+            and is_string_array(objects, skipna=True)
+        ):
             from pandas.core.arrays.string_ import StringDtype
 
             dtype = StringDtype(na_value=np.nan)

pandas/core/arrays/arrow/array.py

@@ -1633,7 +1633,11 @@ def _accumulate(
             else:
                 data_to_accum = data_to_accum.cast(pa.int64())
 
-        result = pyarrow_meth(data_to_accum, skip_nulls=skipna, **kwargs)
+        try:
+            result = pyarrow_meth(data_to_accum, skip_nulls=skipna, **kwargs)
+        except pa.ArrowNotImplementedError as err:
+            msg = f"operation '{name}' not supported for dtype '{self.dtype}'"
+            raise TypeError(msg) from err
 
         if convert_to_int:
             result = result.cast(pa_dtype)

pandas/core/dtypes/cast.py

@@ -87,8 +87,8 @@
 
 if TYPE_CHECKING:
     from collections.abc import (
+        Collection,
         Sequence,
-        Sized,
     )
 
     from pandas._typing import (
@@ -1586,7 +1586,7 @@ def _maybe_box_and_unbox_datetimelike(value: Scalar, dtype: DtypeObj):
     return _maybe_unbox_datetimelike(value, dtype)
 
 
-def construct_1d_object_array_from_listlike(values: Sized) -> np.ndarray:
+def construct_1d_object_array_from_listlike(values: Collection) -> np.ndarray:
     """
     Transform any list-like object in a 1-dimensional numpy array of object
     dtype.
@@ -1604,10 +1604,11 @@ def construct_1d_object_array_from_listlike(values: Sized) -> np.ndarray:
     -------
     1-dimensional numpy array of dtype object
     """
-    # numpy will try to interpret nested lists as further dimensions, hence
-    # making a 1D array that contains list-likes is a bit tricky:
+    # numpy will try to interpret nested lists as further dimensions in np.array(),
+    # hence explicitly making a 1D array using np.fromiter
     result = np.empty(len(values), dtype="object")
-    result[:] = values
+    for i, obj in enumerate(values):
+        result[i] = obj
     return result
 
 

pandas/core/indexes/base.py

@@ -884,6 +884,8 @@ def _engine(
             # error: Item "ExtensionArray" of "Union[ExtensionArray,
             # ndarray[Any, Any]]" has no attribute "_ndarray"  [union-attr]
             target_values = self._data._ndarray  # type: ignore[union-attr]
+        elif is_string_dtype(self.dtype) and not is_object_dtype(self.dtype):
+            return libindex.StringObjectEngine(target_values, self.dtype.na_value)  # type: ignore[union-attr]
 
         # error: Argument 1 to "ExtensionEngine" has incompatible type
         # "ndarray[Any, Any]"; expected "ExtensionArray"
@@ -6133,7 +6135,6 @@ def _should_fallback_to_positional(self) -> bool:
     def get_indexer_non_unique(
         self, target
     ) -> tuple[npt.NDArray[np.intp], npt.NDArray[np.intp]]:
-        target = ensure_index(target)
         target = self._maybe_cast_listlike_indexer(target)
 
         if not self._should_compare(target) and not self._should_partial_index(target):
@@ -6695,7 +6696,16 @@ def _maybe_cast_listlike_indexer(self, target) -> Index:
         """
         Analogue to maybe_cast_indexer for get_indexer instead of get_loc.
         """
-        return ensure_index(target)
+        target_index = ensure_index(target)
+        if (
+            not hasattr(target, "dtype")
+            and self.dtype == object
+            and target_index.dtype == "string"
+        ):
+            # If we started with a list-like, avoid inference to string dtype if self
+            # is object dtype (coercing to string dtype will alter the missing values)
+            target_index = Index(target, dtype=self.dtype)
+        return target_index
 
     @final
     def _validate_indexer(

pandas/core/interchange/from_dataframe.py

@@ -6,6 +6,8 @@
 
 import numpy as np
 
+from pandas._config import using_string_dtype
+
 from pandas.compat._optional import import_optional_dependency
 from pandas.errors import SettingWithCopyError
 
@@ -124,8 +126,6 @@ def protocol_df_chunk_to_pandas(df: DataFrameXchg) -> pd.DataFrame:
     -------
     pd.DataFrame
     """
-    # We need a dict of columns here, with each column being a NumPy array (at
-    # least for now, deal with non-NumPy dtypes later).
     columns: dict[str, Any] = {}
     buffers = []  # hold on to buffers, keeps memory alive
     for name in df.column_names():
@@ -324,8 +324,12 @@ def string_column_to_ndarray(col: Column) -> tuple[np.ndarray, Any]:
         # Add to our list of strings
         str_list[i] = string
 
-    # Convert the string list to a NumPy array
-    return np.asarray(str_list, dtype="object"), buffers
+    if using_string_dtype():
+        res = pd.Series(str_list, dtype="str")
+    else:
+        res = np.asarray(str_list, dtype="object")  # type: ignore[assignment]
+
+    return res, buffers  # type: ignore[return-value]
 
 
 def parse_datetime_format_str(format_str, data) -> pd.Series | np.ndarray: