Partial collapse of multi-dim string coords: take 2

[rcomer]

🚀 Pull Request

Description

This is the same bugfix as #4294. In there, I also tried to add type hints following my interpretation of the dev guide. This turned out to be complicated and I suspect is the main reason the PR has not made progress.

I think it would be good to fix the bug regardless of typing questions, so here is a version without the type hints which I'm hoping will be an easier review.

Fixes #3653.

If you have a 2D string coordinate on a cube, and collapse one of the dimensions spanned by that coordinate, the coordinate's points (and bounds, if any) are now serialised along the collapsed dimension. This seems to have been missed when partial collapse was implemented for the numeric coordinates.

---

Consult Iris pull request check list

---

Add any of the below labels to trigger actions on this PR:

• benchmark_this Request that this pull request be benchmarked to check if it introduces performance shifts

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 89.78%. Comparing base (5cc6737) to head (bd30394).
Report is 68 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #5955   +/-   ##
=======================================
  Coverage   89.78%   89.78%           
=======================================
  Files          88       88           
  Lines       23056    23060    +4     
  Branches     5027     5031    +4     
=======================================
+ Hits        20700    20704    +4     
  Misses       1624     1624           
  Partials      732      732           

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

[rcomer]

For reference, here is what we currently use for string coordinates in aggregated_by, which handles 2D coords but not bounded string coords.

iris/lib/iris/analysis/__init__.py

Lines 2437 to 2460 in 95b7ffe

	if coord.points.dtype.kind in "SU":
	if coord.bounds is None:
	new_points_list = []
	new_bounds = None
	# np.apply_along_axis does not work with str.join, so we
	# need to loop through the array directly. First move axis
	# of interest to trailing dim and flatten the others.
	work_arr = np.moveaxis(coord.points, dim, -1)
	shape = work_arr.shape
	work_shape = (-1, shape[-1])
	new_shape: tuple[int, ...] = (len(self),)
	if coord.ndim > 1:
	new_shape += shape[:-1]
	work_arr = work_arr.reshape(work_shape)
	for indices in self._groupby_indices:
	for arr in work_arr:
	new_points_list.append("|".join(arr.take(indices)))
	# Reinstate flattened dimensions. Aggregated dim now leads.
	new_points = np.array(new_points_list).reshape(new_shape)
	# Move aggregated dimension back to position it started in.
	new_points = np.moveaxis(new_points, 0, dim)

[stephenworsley]

Looks good.

[rcomer]

Thanks @stephenworsley! 🎉