|
1 | 1 | # Changelog
|
2 | 2 |
|
| 3 | +## 1.11.0 (2025-XX-XX) |
| 4 | + |
| 5 | +### Major Changes |
| 6 | + |
| 7 | +This release targets the 2024.12 Array API revision. This includes |
| 8 | + |
| 9 | + - `__array_api_version__` for the wrapped APIs is now set to `2024.12`; |
| 10 | + - Wrappers for `count_nonzero`; |
| 11 | + - Wrappers for `cumulative_prod`; |
| 12 | + - Wrappers for `take_along_axis` (with the exception of Dask); |
| 13 | + - Wrappers for `diff`; |
| 14 | + - `__capabilities__` dict contains a `max_dimensions` key; |
| 15 | + - Python scalars are accepted as arguments to `result_type`; |
| 16 | + - `fft.fftfreq` and `fft.rfftfreq` functions now accept an optional `dtype` |
| 17 | + argument to control the output data type. |
| 18 | + |
| 19 | +Note that these wrappers, as well as other 2024.12 features, are relatively undertested |
| 20 | +in this release, and may have rough edges. Please report any issues you encounter |
| 21 | +in [the issue tracker](https://github.com/data-apis/array-api-compat/issues). |
| 22 | + |
| 23 | +New functions to test properties of arrays: |
| 24 | + - `is_writeable_array` (benefits NumPy, JAX, Sparse) |
| 25 | + - `is_lazy_array` (benefits JAX, Dask, ndonnx) |
| 26 | + |
| 27 | +Improved support for JAX: |
| 28 | + - Work arounds for `.device` attribute and `to_device` function |
| 29 | + not working correctly within `jax.jit` |
| 30 | + |
| 31 | +### Minor Changes |
| 32 | + |
| 33 | +- Several improvements to `dask.array` wrappers: |
| 34 | + |
| 35 | + - `size` returns None for arrays of unknown shapes. |
| 36 | + - `astype(..., copy=True)` always copies, independently of the Dask version. |
| 37 | + - implementations of `sort` and `argsort` are now available. Note that these |
| 38 | + implementations are relatively crude, and might be memory intensive. |
| 39 | + - `asarray` no longer accidentally materializes the Dask graph |
| 40 | + - `torch` wrappers contain unsigned integer dtypes of widths >8 bits, `uint16`, |
| 41 | + `uint32` and `uint64` if PyTorch version is at least 2.3. Note that the |
| 42 | + unsigned integer support is incomplete in PyTorch itself, see |
| 43 | + [gh-253](https://github.com/data-apis/array-api-compat/pull/253). |
| 44 | + |
| 45 | +### Authors |
| 46 | + |
| 47 | +The following users contributed to this release: |
| 48 | + |
| 49 | +Athan Reines |
| 50 | +Guido Imperiale |
| 51 | +Evgeni Burovski |
| 52 | +Guido Imperiale |
| 53 | +Lucas Colley |
| 54 | +Ralf Gommers |
| 55 | +Thomas Li |
| 56 | + |
| 57 | + |
3 | 58 | ## 1.10.0 (2024-12-25)
|
4 | 59 |
|
5 | 60 | ### Major Changes
|
|
0 commit comments