docs: add changelog for v2023.12 specification revision

CHANGELOG.md

@@ -4,6 +4,138 @@
 
 This changelog is organized by specification version and notes all changes with respect to the previous version. Within the section for a specific version (e.g., v2022.12), separate sections are used for (a) changes to existing APIs and requirements, (b) new APIs and new requirements, and (c) errata.
 
+## v2023.12
+
+### Updates
+
+> Updates to existing APIs and requirements.
+
+#### Normative
+
+- Clarify expectations concerning exception handling ([gh-613](https://github.com/data-apis/array-api/pull/613))
+- Clarify that the constant `newaxis` is an alias of `None` ([gh-687](https://github.com/data-apis/array-api/pull/687))
+- Add design discussion on lazy versus eager implementations ([gh-708](https://github.com/data-apis/array-api/pull/708))
+- Revise guidance to require a minimum upper bound for supported ranks ([gh-702](https://github.com/data-apis/array-api/pull/702))
+- Relax design requirements for positional and keyword-only arguments ([gh-730](https://github.com/data-apis/array-api/pull/730))
+- Add recommendation to `__dlpack__` for handling read-only arrays ([gh-749](https://github.com/data-apis/array-api/pull/749))
+
+#### APIs
+
+- `__bool__`: allow lazy implementations to raise from intrinsically eager functions ([gh-652](https://github.com/data-apis/array-api/pull/652))
+- `__complex__`: allow lazy implementations to raise from intrinsically eager functions ([gh-652](https://github.com/data-apis/array-api/pull/652))
+- `__dlpack__`: add `max_version` keyword argument to support versioning ([gh-602](https://github.com/data-apis/array-api/pull/602))
+- `__dlpack__`: add `dl_device` and `copy` keyword arguments ([gh-741](https://github.com/data-apis/array-api/pull/741))
+- `__float__`: allow lazy implementations to raise from intrinsically eager functions ([gh-652](https://github.com/data-apis/array-api/pull/652))
+- `__index__`: allow lazy implementations to raise from intrinsically eager functions ([gh-652](https://github.com/data-apis/array-api/pull/652))
+- `__int__`: allow lazy implementations to raise from intrinsically eager functions ([gh-652](https://github.com/data-apis/array-api/pull/652))
+- `astype`: add support for an optional `device` keyword argument ([gh-665](https://github.com/data-apis/array-api/pull/665))
+- `from_dlpack`: require exceptions to address unsupported use cases ([gh-709](https://github.com/data-apis/array-api/pull/709))
+- `from_dlpack`: add support for `copy` and `device` keywords ([gh-741](https://github.com/data-apis/array-api/pull/741))
+- `max`: clarify that the order of signed zeros is unspecified ([gh-751](https://github.com/data-apis/array-api/pull/751))
+- `min`: clarify that the order of signed zeros is unspecified ([gh-751](https://github.com/data-apis/array-api/pull/751))
+- `take`: explicitly leave out-of-bounds behavior unspecified ([gh-701](https://github.com/data-apis/array-api/pull/701))
+- `tensordot`: allow negative axes ([gh-625](https://github.com/data-apis/array-api/pull/625))
+- `to_device`: clarify behavior when a provided `device` object corresponds to the same device on which an array instance resides ([gh-742](https://github.com/data-apis/array-api/pull/742))
+- `unique_all`: clarify the shape of the array containing unique values and the order of returned counts ([gh-752](https://github.com/data-apis/array-api/pull/752))
+- `unique_counts`: clarify the shape of the array containing unique values and the order of returned counts ([gh-752](https://github.com/data-apis/array-api/pull/752))
+- `unique_inverse`: clarify the shape of the array containing unique values ([gh-752](https://github.com/data-apis/array-api/pull/752))
+- `unique_values`: clarify the shape of the returned array ([gh-752](https://github.com/data-apis/array-api/pull/752))
+
+#### Extensions
+
+> Updates to APIs and requirements included as part of specification extensions.
+
+- `fft.*`: clarify behavior of the `n` and `s` keyword arguments and the expected output array shape ([gh-720](https://github.com/data-apis/array-api/pull/720) and [gh-746](https://github.com/data-apis/array-api/pull/746); backported to v2022.12 revision of Array API specification)
+
+* * *
+
+### Additions
+
+> New APIs and requirements added to the specification.
+
+#### APIs
+
+The following APIs were added to the specification:
+
+- `__array_namespace_info__`: namespace with Array API namespace inspection utilities ([gh-689](https://github.com/data-apis/array-api/pull/689))
+- `clip`: clamp each element of an input array to a specified range ([gh-715](https://github.com/data-apis/array-api/pull/715))
+- `copysign`: compose a floating-point value with the magnitude of a `x1_i` and the sign of `x2_i` ([gh-693](https://github.com/data-apis/array-api/pull/693))
+- `cumulative_sum`: calculate the cumulative sum ([gh-653](https://github.com/data-apis/array-api/pull/653))
+- `hypot`: compute the square root of the sum of squares for each element in an array ([gh-703](https://github.com/data-apis/array-api/pull/703))
+- `maximum`: compute the maximum value for each element of an array relative to the respective element in another array ([gh-713](https://github.com/data-apis/array-api/pull/713))
+- `minimum`: compute the minimum value for each element of an array relative to the respective element in another array ([gh-713](https://github.com/data-apis/array-api/pull/713))
+- `moveaxis`: move array axes to new positions, while leaving other axes in their original positions ([gh-656](https://github.com/data-apis/array-api/pull/656))
+- `repeat`: repeat each element of an array a specified number of times ([gh-690](https://github.com/data-apis/array-api/pull/690))
+- `searchsorted`: find the indices into `x1` such that, if the corresponding elements in `x2` were inserted before the indices, the order of `x1`, when sorted in ascending order, would be preserved ([gh-699](https://github.com/data-apis/array-api/pull/699))
+- `signbit`: determine whether the sign bit is set for each element of an array ([gh-705](https://github.com/data-apis/array-api/pull/705))
+- `tile`: construct an array by tiling an input array ([gh-692](https://github.com/data-apis/array-api/pull/692))
+- `unstack`: split an array into a sequence of arrays along a given axis ([gh-604](https://github.com/data-apis/array-api/pull/604))
+
+#### Inspection APIs
+
+The following inspection APIs were added to the specification:
+
+- `capabilities`: return a dictionary of array library capabilities ([gh-689](https://github.com/data-apis/array-api/pull/689))
+- `default_device`: return the default device ([gh-689](https://github.com/data-apis/array-api/pull/689))
+- `default_dtypes`: return a dictionary containing default data types ([gh-689](https://github.com/data-apis/array-api/pull/689))
+- `dtypes`: return a dictionary support Array API data types ([gh-689](https://github.com/data-apis/array-api/pull/689))
+- `devices`: return a list of supported devices ([gh-689](https://github.com/data-apis/array-api/pull/689))
+
+* * *
+
+### Breaking Changes
+
+The following is a list of breaking changes relative to the previous version of the specification:
+
+- `prod`: when provided a floating-point array, the function must return a floating-point array having the same data type ([gh-744](https://github.com/data-apis/array-api/pull/744))
+- `sum`: when provided a floating-point array, the function must return a floating-point array having the same data type ([gh-744](https://github.com/data-apis/array-api/pull/744))
+- `vecdot`: only allow a negative integer for the `axis` keyword argument ([gh-740](https://github.com/data-apis/array-api/pull/740))
+
+#### Extensions
+
+The following is a list of breaking changes in specification extensions relative to the previous version of the specification:
+
+- `fft.fft`: require the input array to have a complex-valued floating-point data type and require that the output array have the same data type as the input array ([gh-720](https://github.com/data-apis/array-api/pull/720); backported to v2022.12 revision of Array API specification)
+- `fft.fftn`: require the input array to have a complex-valued floating-point data type and require that the output array have the same data type as the input array ([gh-720](https://github.com/data-apis/array-api/pull/720); backported to v2022.12 revision of Array API specification)
+- `fft.hfft`: require the input array to have a complex-valued floating-point data type and require that the output array have a real-valued data type having the same precision as the input array ([gh-720](https://github.com/data-apis/array-api/pull/720); backported to v2022.12 revision of Array API specification)
+- `fft.ifft`: require the input array to have a complex-valued floating-point data type and require that the output array have the same data type as the input array ([gh-720](https://github.com/data-apis/array-api/pull/720); backported to v2022.12 revision of Array API specification)
+- `fft.ifftn`: require the input array to have a complex-valued floating-point data type and require that the output array have the same data type as the input array ([gh-720](https://github.com/data-apis/array-api/pull/720); backported to v2022.12 revision of Array API specification)
+- `fft.irfft`: require the output array have a real-valued floating-point data type having the same precision as the input array ([gh-720](https://github.com/data-apis/array-api/pull/720); backported to v2022.12 revision of Array API specification)
+- `fft.irfftn`: require the output array have a real-valued floating-point data type having the same precision as the input array ([gh-720](https://github.com/data-apis/array-api/pull/720); backported to v2022.12 revision of Array API specification)
+- `fft.fftfreq`: require the output array have the default real-valued floating-point data type ([gh-720](https://github.com/data-apis/array-api/pull/720); backported to v2022.12 revision of Array API specification)
+- `fft.rfftfreq`: require the output array have the default real-valued floating-point data type ([gh-720](https://github.com/data-apis/array-api/pull/720); backported to v2022.12 revision of Array API specification)
+- `linalg.cross`: broadcast only along non-compute axes and only allow a negative integer for the `axis` keyword argument ([gh-740](https://github.com/data-apis/array-api/pull/740))
+- `linalg.trace`: when provided a floating-point array, the function must return a floating-point array having the same data type ([gh-744](https://github.com/data-apis/array-api/pull/744))
+
+* * *
+
+### Errata
+
+The following is a list of fixes and points of clarification with regard to the previous version of the specification:
+
+- `__getitem__`: clarify typing to allow `None` in indexing ([gh-674](https://github.com/data-apis/array-api/pull/674) and [gh-687](https://github.com/data-apis/array-api/pull/687))
+- `__ge__`: clarify that the operation is only defined for arrays having real-valued data types ([gh-736](https://github.com/data-apis/array-api/pull/736))
+- `__gt__`: clarify that the operation is only defined for arrays having real-valued data types ([gh-736](https://github.com/data-apis/array-api/pull/736))
+- `__le__`: clarify that the operation is only defined for arrays having real-valued data types ([gh-736](https://github.com/data-apis/array-api/pull/736))
+- `__lt__`: clarify that the operation is only defined for arrays having real-valued data types ([gh-736](https://github.com/data-apis/array-api/pull/736))
+- `abs`: fix typo in return value description ([gh-633](https://github.com/data-apis/array-api/pull/633))
+- `asarray`: fix typo in `device` keyword argument description ([gh-681](https://github.com/data-apis/array-api/pull/681))
+- `conj`: fix typo in parameter description ([gh-706](https://github.com/data-apis/array-api/pull/706))
+- `finfo_object`: fix missing `dtype` attribute ([gh-639](https://github.com/data-apis/array-api/pull/639))
+- `fft.*`: fix various typing issues ([gh-720](https://github.com/data-apis/array-api/pull/720))
+- `iinfo_object`: fix missing `dtype` attribute ([gh-639](https://github.com/data-apis/array-api/pull/639))
+- `linalg.qr`: fix typo in function description ([gh-661](https://github.com/data-apis/array-api/pull/661))
+- `linalg.cholesky`: fix typo in function description ([gh-677](https://github.com/data-apis/array-api/pull/677))
+- `linalg.svd`: fix return type ([gh-619](https://github.com/data-apis/array-api/pull/619))
+- `prod`: clarify type promotion behavior when `dtype=None` ([gh-666](https://github.com/data-apis/array-api/pull/666))
+- `sum`: clarify type promotion behavior when `dtype=None` ([gh-666](https://github.com/data-apis/array-api/pull/666))
+- `take`: fix typing for optional `axis` keyword argument ([gh-644](https://github.com/data-apis/array-api/pull/644))
+- `tensordot`: fix typo in parameter description ([gh-622](https://github.com/data-apis/array-api/pull/622))
+- `trace`: clarify type promotion behavior when `dtype=None` ([gh-666](https://github.com/data-apis/array-api/pull/666))
+- `vecdot`: fix definition of complex inner product ([gh-723](https://github.com/data-apis/array-api/pull/723))
+
+* * *
+
 ## v2022.12
 
 ### Updates

src/array_api_stubs/_2022_12/indexing_functions.py

@@ -14,7 +14,7 @@ def take(x: array, indices: array, /, *, axis: Optional[int] = None) -> array:
         input array.
     indices: array
         array indices. The array must be one-dimensional and have an integer data type.
-    axis: int
+    axis: Optional[int]
         axis over which to select values. If ``axis`` is negative, the function must determine the axis along which to select values by counting from the last dimension.
 
         If ``x`` is a one-dimensional array, providing an ``axis`` is optional; however, if ``x`` has more than one dimension, providing an ``axis`` is required.

src/array_api_stubs/_draft/array_object.py

@@ -247,6 +247,9 @@ def __bool__(self: array, /) -> bool:
 
         .. versionchanged:: 2022.12
             Added boolean and complex data type support.
+
+        .. versionchanged:: 2023.12
+            Allowed lazy implementations to error.
         """
 
     def __complex__(self: array, /) -> complex:
@@ -285,6 +288,9 @@ def __complex__(self: array, /) -> complex:
         The Python language requires the return value to be of type ``complex``. Lazy implementations are therefore not able to return any kind of lazy/delayed object here and should raise a ``ValueError`` instead.
 
         .. versionadded:: 2022.12
+
+        .. versionchanged:: 2023.12
+            Allowed lazy implementations to error.
         """
 
     def __dlpack__(
@@ -451,10 +457,13 @@ def __dlpack__(
         memory isn't modified).
 
         .. versionchanged:: 2022.12
-            Added BufferError.
+           Added BufferError.
 
         .. versionchanged:: 2023.12
-            Added the ``max_version``, ``dl_device``, and ``copy`` keywords.
+           Added the ``max_version``, ``dl_device``, and ``copy`` keywords.
+
+        .. versionchanged:: 2023.12
+           Added recommendation for handling read-only arrays.
         """
 
     def __dlpack_device__(self: array, /) -> Tuple[Enum, int]:
@@ -539,6 +548,9 @@ def __float__(self: array, /) -> float:
 
         .. versionchanged:: 2022.12
             Added boolean and complex data type support.
+
+        .. versionchanged:: 2023.12
+            Allowed lazy implementations to error.
         """
 
     def __floordiv__(self: array, other: Union[int, float, array], /) -> array:
@@ -664,6 +676,9 @@ def __index__(self: array, /) -> int:
         **Lazy implementations**
 
         The Python language requires the return value to be of type ``int``. Lazy implementations are therefore not able to return any kind of lazy/delayed object here and should raise a ``ValueError`` instead.
+
+        .. versionchanged:: 2023.12
+            Allowed lazy implementations to error.
         """
 
     def __int__(self: array, /) -> int:
@@ -711,6 +726,9 @@ def __int__(self: array, /) -> int:
 
         .. versionchanged:: 2022.12
             Added boolean and complex data type support.
+
+        .. versionchanged:: 2023.12
+            Allowed lazy implementations to error.
         """
 
     def __invert__(self: array, /) -> array:
@@ -1192,6 +1210,9 @@ def to_device(
 
         -   When a provided ``device`` object corresponds to the same device on which an array instance resides, implementations may choose to perform an explicit copy or return ``self``.
         -   If ``stream`` is provided, the copy operation should be enqueued on the provided ``stream``; otherwise, the copy operation should be enqueued on the default stream/queue. Whether the copy is performed synchronously or asynchronously is implementation-dependent. Accordingly, if synchronization is required to guarantee data safety, this must be clearly explained in a conforming array library's documentation.
+
+        .. versionchanged:: 2023.12
+           Clarified behavior when a provided ``device`` object corresponds to the device on which an array instance resides.
         """
 
 

src/array_api_stubs/_draft/creation_functions.py

@@ -289,6 +289,9 @@ def func(x, y):
             ...
 
 
+    .. versionchanged:: 2023.12
+       Required exceptions to address unsupported use cases.
+
     .. versionchanged:: 2023.12
        Added device and copy support.
     """

src/array_api_stubs/_draft/data_type_functions.py

@@ -57,6 +57,9 @@ def astype(
 
     .. versionchanged:: 2022.12
        Added complex data type support.
+
+    .. versionchanged:: 2023.12
+       Added device keyword argument support.
     """
 
 

src/array_api_stubs/_draft/elementwise_functions.py

@@ -806,6 +806,8 @@ def clip(
     - If both ``min`` and ``max`` are ``None``, the elements of the returned array must equal the respective elements in ``x``.
     - If a broadcasted element in ``min`` is greater than a corresponding broadcasted element in ``max``, behavior is unspecified and thus implementation-dependent.
     - If ``x`` and either ``min`` or ``max`` have different data type kinds (e.g., integer versus floating-point), behavior is unspecified and thus implementation-dependent.
+
+    .. versionadded:: 2023.12
     """
 
 
@@ -880,6 +882,8 @@ def copysign(x1: array, x2: array, /) -> array:
     - If ``x2_i`` is greater than ``0``, the result is ``NaN`` with a sign bit of ``0``.
     - If ``x2_i`` is ``NaN`` and the sign bit of ``x2_i`` is ``1``, the result is ``NaN`` with a sign bit of ``1``.
     - If ``x2_i`` is ``NaN`` and the sign bit of ``x2_i`` is ``0``, the result is ``NaN`` with a sign bit of ``0``.
+
+    .. versionadded:: 2023.12
     """
 
 
@@ -1410,6 +1414,8 @@ def hypot(x1: array, x2: array, /) -> array:
        IEEE 754-2019 requires support for subnormal (a.k.a., denormal) numbers, which are useful for supporting gradual underflow. However, hardware support for subnormal numbers is not universal, and many platforms (e.g., accelerators) and compilers support toggling denormals-are-zero (DAZ) and/or flush-to-zero (FTZ) behavior to increase performance and to guard against timing attacks.
 
        Accordingly, conforming implementations may vary in their support for subnormal numbers.
+
+    .. versionadded:: 2023.12
     """
 
 
@@ -1929,6 +1935,8 @@ def maximum(x1: array, x2: array, /) -> array:
     For floating-point operands,
 
     -   If either ``x1_i`` or ``x2_i`` is ``NaN``, the result is ``NaN``.
+
+    .. versionadded:: 2023.12
     """
 
 
@@ -1960,6 +1968,8 @@ def minimum(x1: array, x2: array, /) -> array:
     For floating-point operands,
 
     -   If either ``x1_i`` or ``x2_i`` is ``NaN``, the result is ``NaN``.
+
+    .. versionadded:: 2023.12
     """
 
 
@@ -2390,6 +2400,8 @@ def signbit(x: array, /) -> array:
     - If ``x_i`` is a negative (i.e., less than ``0``) finite number, the result is ``True``.
     - If ``x_i`` is ``NaN`` and the sign bit of ``x_i`` is ``0``, the result is ``False``.
     - If ``x_i`` is ``NaN`` and the sign bit of ``x_i`` is ``1``, the result is ``True``.
+
+    .. versionadded:: 2023.12
     """