BUG: Fix display with nested NumPy arrays

[basnijholt]

Currently, the following will fail to display:

import xarray as xr
import numpy as np
x = np.empty((2, 2), dtype=object)
for i in range(2):
    for j in range(2):
        x[i, j] = np.zeros(2)  # Set to 1D array of size 2
ds = xr.DataArray(x)
ds

Click to see the error message

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
File ~/Work/pipefunc/.venv/lib/python3.13/site-packages/IPython/core/formatters.py:347, in BaseFormatter.__call__(self, obj)
    345     method = get_real_method(obj, self.print_method)
    346     if method is not None:
--> 347         return method()
    348     return None
    349 else:

File ~/Work/pipefunc/.venv/lib/python3.13/site-packages/xarray/core/common.py:188, in AbstractArray._repr_html_(self)
    186 if OPTIONS["display_style"] == "text":
    187     return f"<pre>{escape(repr(self))}</pre>"
--> 188 return formatting_html.array_repr(self)

File ~/Work/pipefunc/.venv/lib/python3.13/site-packages/xarray/core/formatting_html.py:323, in array_repr(arr)
    315 arr_name = f"'{arr.name}'" if getattr(arr, "name", None) else ""
    317 header_components = [
    318     f"<div class='xr-obj-type'>{obj_type}</div>",
    319     f"<div class='xr-array-name'>{arr_name}</div>",
    320     format_dims(dims, indexed_dims),
    321 ]
--> 323 sections = [array_section(arr)]
    325 if hasattr(arr, "coords"):
    326     sections.append(coord_section(arr.coords))

File ~/Work/pipefunc/.venv/lib/python3.13/site-packages/xarray/core/formatting_html.py:232, in array_section(obj)
    226 collapsed = (
    227     "checked"
    228     if _get_boolean_with_default("display_expand_data", default=True)
    229     else ""
    230 )
    231 variable = getattr(obj, "variable", obj)
--> 232 preview = escape(inline_variable_array_repr(variable, max_width=70))
    233 data_repr = short_data_repr_html(obj)
    234 data_icon = _icon("icon-database")

File ~/Work/pipefunc/.venv/lib/python3.13/site-packages/xarray/core/formatting.py:304, in inline_variable_array_repr(var, max_width)
    302     return var._data._repr_inline_(max_width)
    303 if getattr(var, "_in_memory", False):
--> 304     return format_array_flat(var, max_width)
    305 dask_array_type = array_type("dask")
    306 if isinstance(var._data, dask_array_type):

File ~/Work/pipefunc/.venv/lib/python3.13/site-packages/xarray/core/formatting.py:223, in format_array_flat(array, max_width)
    220 # every item will take up at least two characters, but we always want to
    221 # print at least first and last items
    222 max_possibly_relevant = min(max(array.size, 1), max(math.ceil(max_width / 2.0), 2))
--> 223 relevant_front_items = format_items(
    224     first_n_items(array, (max_possibly_relevant + 1) // 2)
    225 )
    226 relevant_back_items = format_items(last_n_items(array, max_possibly_relevant // 2))
    227 # interleave relevant front and back items:
    228 #     [a, b, c] and [y, z] -> [a, z, b, y, c]

File ~/Work/pipefunc/.venv/lib/python3.13/site-packages/xarray/core/formatting.py:212, in format_items(x)
    209     elif np.logical_not(time_needed).all():
    210         timedelta_format = "date"
--> 212 formatted = [format_item(xi, timedelta_format) for xi in x]
    213 return formatted

File ~/Work/pipefunc/.venv/lib/python3.13/site-packages/xarray/core/formatting.py:193, in format_item(x, timedelta_format, quote_strings)
    191     return repr(x) if quote_strings else x
    192 elif hasattr(x, "dtype") and np.issubdtype(x.dtype, np.floating):
--> 193     return f"{x.item():.4}"
    194 else:
    195     return str(x)

ValueError: can only convert an array of size 1 to a Python scalar

Whenever there are size==1 arrays, it currently does work:

import xarray as xr
import numpy as np
x = np.empty((2, 2), dtype=object)
for i in range(2):
    for j in range(2):
        x[i, j] = np.zeros((1, 1, 1))  # Set to 3D array of size 1
ds = xr.DataArray(x)
ds

Should I add these examples as regression tests anywhere, if so, where?

For context, I am running into this issue in https://github.com/pipefunc/pipefunc where one can put the resulting objects into xarray.Datasets.
For example, see the docs here https://pipefunc.readthedocs.io/en/latest/examples/physics-simulation/ (search for xarray.Dataset).

• Closes #xxxx
• Tests added
• User visible changes (including notable bug fixes) are documented in whats-new.rst
• New functions/methods are listed in api.rst

[welcome]

Thank you for opening this pull request! It may take us a few days to respond here, so thank you for being patient.
If you have questions, some answers may be found in our contributing guidelines.

[basnijholt]

Now test_slice_in_title_single_item_array is failing with:

  - d = [10.009]
  ?     -    ^^^
  + d = 10.01
  ?         ^

This test was introduced in #5948 and touches the exact same code that I changed. I think the test that is failing now is wrong and should be updated to d = [10.009] instead.

[dcherian]

OK please update the test in that case. Merging in main should fix the other "array api" failures

[basnijholt]

@dcherian, I updated the tests, all is passing except some (I believe) unrelated failure.

[welcome]

Congratulations on completing your first pull request! Welcome to Xarray! We are proud of you, and hope to see you again!