Fix failing test with Dask 2025.2: ours issue not theirs

[valeriupredoi]

Description

@sloosvel and I found a failing test with Dask==2025.2 in conda-forge/esmvalcore-feedstock#83 while we were trying to release RC1. @schlunma looked into it here, and concluded the test (ours) needs be fixed, so he's applied the fix - many thanks, Manu!

---

Before you get started

• ☝ Create an issue to discuss what you are going to do

Checklist

It is the responsibility of the author to make sure the pull request is ready to review. The icons indicate whether the item will be subject to the 🛠 Technical or 🧪 Scientific review.

• 🛠 This pull request has a descriptive title and labels
• 🛠 Unit tests have been added
• 🛠 Changes are backward compatible
• 🛠 Any changed dependencies have been added or removed correctly
• 🛠 All checks below this pull request were successful

---

To help with the number pull requests:

• 🙏 We kindly ask you to review two other open pull requests in this repository

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 95.07%. Comparing base (dd1e29c) to head (c3561d7).
Report is 48 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2663      +/-   ##
==========================================
- Coverage   95.08%   95.07%   -0.01%     
==========================================
  Files         255      255              
  Lines       14884    14884              
==========================================
- Hits        14152    14151       -1     
- Misses        732      733       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[schlunma]

I don't think this is bug in Dask, but rather a problem with our test.

We have the following setup:

mask = np.ma.masked_array(np.arange(2), mask=[1, 0])  # [-- 1]
array = da.arange(2)  # [0, 1]

Applying this mask to the array gives [-- --] in Dask 2025.1.0 AND Dask 2025.2.0. Using numpy also gives the same result. However, the expected array is

expected = da.ma.masked_array(np.ones(2), np.arange(2))  # [1.0, --]

which is clearly different to the actual result.

So, I think we should adapt our test, not pin Dask. I have no idea why this wasn't a test fail before. Most likely Dask fixed a bug in v2025.2.0...

Test code (gives always the same result, regardless of Dask version; also gives same result for pure-numpy code):

import dask.array as da
import numpy as np

mask = np.ma.masked_array(np.arange(2), mask=[1, 0])
array = da.arange(2)

x = da.ma.masked_where(mask, array)

print("mask", mask)
print("array", array.compute())
print("result", x.compute())
print("result data", x.compute().data)
print("result mask", x.compute().mask)

[schlunma]

Test ran successful with Dask 2025.2.0 here and with 2025.1.0 locally 🎉

[valeriupredoi]

very many thanks for looking into this, Manu! The test is failing with dask==2025.2 but not with dask==2025.1 so we have to be a wee bit careful about it. Let me test your MRE myself, and thanks for including it!

[schlunma]

Yes, but with this PR it runs in both Dask versions. As I mentioned above, it rather seems that there was a bug in <=2025.1.0 which let the test pass even though the expected result was clearly wrong (looks like a typo to me). This PR here fixes the expected result of our test.

[valeriupredoi]

many thanks, Manu! Indeed - sanity checks all good:

dask==2025.1

mask [-- 1]
array [0 1]
result [-- --]
result data [0 1]
result mask [ True  True]

dask==2025.2

mask [-- 1]
array [0 1]
result [-- --]
result data [0 1]
result mask [ True  True]

dask 2024.8.0

mask [-- 1]
array [0 1]
result [-- --]
result data [0 1]
result mask [ True  True]

[valeriupredoi]

fixed PR title, and description, @schlunma cheers muchly for your detective work here, pls merge at will 🍺

[schlunma]

Cheers! 🍻