Fix CI test failures

[m-albert]

With this PR I started out just wanting to ensure compatibility with numpy 2 as reported in #392, however different issues came up in the tests (and if unaddressed they make the CI fail), so I decided to address them together in the same PR.

Summary:

1. Use np.ptp instead of the method to ensure compatibility with numpy 2 as reported in NumPy 2 removes ptp method (use function instead) #392.

2. Replace the deprecated tifffile.TiffWriter.save by tifffile.TiffWriter.save, see here.

3. dask-image currently doesn't pass tests with dask>=2025.1.0 because of New Dask Arrow-based strings cause test failures #335 which came back because of changes in dask.dataframe (more details here). This PR applies the proposed workaround. An alternative would be to pin dask<2025.1.0, but probably a very limiting option.

4. (edit) Stop using Mambaforge in the CI. According to https://github.com/conda-forge/miniforge mambaforge is deprecated since 2024 and installers have been retired in 2025.

@jakirkham @GenevieveBuckley

[m-albert]

The CI fails in trying to get mamba from https://github.com/conda-forge/miniforge/releases/latest/download/Mambaforge-Linux-x86_64.sh.

The error might be related to conda-incubator/setup-miniconda#392.

[m-albert]

Okay switching from Mambaforge to Miniforge solved the issue.

[jakirkham]

Thanks Marvin! 🙏

Had a couple comments below

[jakirkham]

Is this enough given compute happens later?

Do we have a way to test this?

[m-albert]

Is this enough given compute happens later?

Yes, consider this code snippet:

import dask.dataframe as dd
import dask.array as da
from dask import delayed
import pandas as pd
import dask.config as dask_config

x = dd.from_delayed(delayed(pd.Series([slice(2)])))

with dask_config.set({'dataframe.convert-string': False}):
    x_cm = dd.from_delayed(delayed(pd.Series([slice(2)])))

print('x', x)
print('x_cm', x_cm)

print('x computed', x.compute())
print('x_cm computed', x_cm.compute())

outputs

x Dask Series Structure:
npartitions=1
    string
       ...
Dask Name: to_string_dtype, 3 expressions
Expr=ArrowStringConversion(frame=FromDelayed(5621810))
x_cm Dask Series Structure:
npartitions=1
    object
       ...
Dask Name: fromdelayed, 2 expressions
Expr=FromDelayed(79a507f)
x computed 0    slice(None, 2, None)
dtype: string
x_cm computed 0    slice(None, 2, None)
dtype: object

Also he tests do perform a compute and fail without the context manager in place, so it should be covered:

dask-image/tests/test_dask_image/test_ndmeasure/test_find_objects.py

Lines 61 to 67 in cb1360a

	computed_result = result.compute()
	assert isinstance(computed_result, pd.DataFrame)
	expected = pd.DataFrame.from_dict(
	{0: {111: slice(1, 3), 222: slice(3, 4), 333: slice(0, 2)},
	1: {111: slice(0, 2), 222: slice(3, 8), 333: slice(7, 10)}}
	)
	assert computed_result.equals(expected)

[m-albert]

Thanks for the review @jakirkham! I addressed your points.

[jakirkham]

(edit) Stop using Mambaforge in the CI. According to https://github.com/conda-forge/miniforge mambaforge is deprecated since 2024 and installers have been retired in 2025

Just a clarification point, they became the same installer. IOW Mambaforge and Miniforge include all the same things. Hence deprecating to clean up the namespace

[jakirkham]

Thanks Marvin! 🙏

LGTM. Let's see what Genevieve thinks 🙂

[m-albert]

Sorry for the slightly complicated PR here! It's a bit annoying that several issues are addressed at the same time, but individually they make the CI fail so I thought I'd address them together.

Happy to split out individual points if you think that makes things easier @jakirkham @GenevieveBuckley.

It'll be good to get back to green CI one way or another, also with regard to #384.

[jakirkham]

Let's go ahead and merge. The changes here look good and having working CI would be a great help for other work

We can follow up on any feedback in a new PR