fix:*-like creation routines take kwargs #2992
Conversation
github-actions
bot
added
the
needs release notes
github-actions
bot
removed
the
needs release notes
|
requesting review from @jhamman since I think you touched this code most recently |
| @@ -109,10 +109,10 @@ def _get_shape_chunks(a: ArrayLike | Any) -> tuple[ChunkCoords | None, ChunkCoor | |||
| return shape, chunks | |||
|
|
|||
|
|
|||
| def _like_args(a: ArrayLike, kwargs: dict[str, Any]) -> dict[str, Any]: | |||
| def _like_args(a: ArrayLike) -> dict[str, object]: | |||
There was a problem hiding this comment.
Why the change to object? And if you're feeling eager, this could maybe be a typed dict?
class LikeArgs(TypedDict):
shape: tuple[int, ...]
chunks: tuple[int, ...]
dtype: np.typing.DTypeLike
order: ...
but that's probably out of scope for this PR.
| like_kwargs = _like_args(a) | kwargs | ||
| if isinstance(a, (AsyncArray | Array)): | ||
| like_kwargs.setdefault("fill_value", a.metadata.fill_value) |
There was a problem hiding this comment.
To make sure I understand the precedence, it's
- User keyword arguments (via
| kwargs) _like_kwargs(a)fill_valueviaa.metadata.fill_value
? It'd be worth confirming we have each of those cases covered with tests.
| like_kwargs = _like_args(a) | kwargs | ||
| if isinstance(a, (AsyncArray | Array)): | ||
| like_kwargs.setdefault("fill_value", a.metadata.fill_value) | ||
| return await empty(**like_kwargs) # type: ignore[arg-type] |
There was a problem hiding this comment.
Why the type ignore here? I guess the LikeArgs typed dict might solve that.
| LikeFuncName = Literal["zeros_like", "ones_like", "empty_like", "full_like", "open_like"] | ||
|
|
||
|
|
||
| @pytest.mark.parametrize("func_name", get_args(LikeFuncName)) |
There was a problem hiding this comment.
Maybe there's a reason to structure it like this, but my usual preference is to use functions instead of names. So the actual params would be [zarr.api.asynchronous.zeros_like, zarr.api.asynchronous.one_like, ...].
In main, array creation routines that take an existing array, like
full_like,zeros_like, etc, don't allow users to override certain array properties (like shape and dtype) via keyword arguments. E.g.,full_like(arr, shape=(10,))will ignore theshapekeyword argument and produce an array with the same shape asarr.This PR makes these array creation routines sensitive to keyword arguments like
shape,dtype,chunks, etc.closes #2989