Fix attribute array comparison

[ukmo-ccbunney]

🚀 Pull Request

Description

Fixes: #6027

Logic for comparing arrays in attributes was relying on old numpy behaviour where an array compared to another any object would return a scalar True or False. Current numpy behaviour is to do a element-wise comparison which throws an error if the arrays are of different size.

Fix is to use the numpy.array_equal.

---

Consult Iris pull request check list

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 89.81%. Comparing base (3b8b33c) to head (06c997e).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #6181      +/-   ##
==========================================
- Coverage   89.82%   89.81%   -0.01%     
==========================================
  Files          88       88              
  Lines       23185    23182       -3     
  Branches     4314     4314              
==========================================
- Hits        20825    20822       -3     
  Misses       1628     1628              
  Partials      732      732              

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

[ukmo-ccbunney]

Note, this change fixes what I believe is a bug here:

iris/lib/iris/_merge.py

Lines 388 to 391 in d3071ff

	diff_attrs = [
	repr(key[1])
	for key in attrs_1
	if np.all(attrs_1[key] != attrs_2[key])

Line 391 should be checking for any differences in the array elements, not all.

It also cleans up a now unnecessary try...except block that I believe was only needed for older versions of numpy.

iris/lib/iris/common/mixin.py

Lines 108 to 111 in d3071ff

	try:
	match = bool(match)
	except ValueError:
	match = match.all()

I have extended the existing unit test to check for correct handling inequality.
However, I question the validity of this test

iris/lib/iris/tests/unit/common/mixin/test_LimitedAttributeDict.py

Lines 45 to 49 in d3071ff

	values = dict(one=np.arange(1), two=np.arange(1), three=np.arange(1))
	left = LimitedAttributeDict(dict(one=0, two=0, three=0))
	right = LimitedAttributeDict(**values)
	self.assertEqual(left, right)
	self.assertEqual(left, values)

which now fails as it is testing a scalar value against a single element array. With the old logic, this would work, but it is flawed as np.all( np.array([0]) == 0 ) and np.all( np.zeros(100) == 0 ) would both return True, even though the latter is not what is intended.

I have commented out this test at the moment and would appreciate the opinion of @pp-mo.