Can't collapse over a mesh dimension

[pp-mo]

A cube cannot currently be collapsed over a MeshCoord (or mesh dim),
since the collapse operation expects a bounds shape of (N, 2) and it can't handle (N, 4).

Example:

>>> import iris.tests.stock.mesh as ism
>>> cube = ism.sample_mesh_cube()
>>> cube.collapsed('latitude', iris.analysis.MEAN)
/home/h05/itpp/git/iris/iris_main/lib/iris/cube.py:3696: UserWarning: Collapsing spatial coordinate 'latitude' without weighting
  warnings.warn(msg.format(coord.name()))
/home/h05/itpp/git/iris/iris_main/lib/iris/coords.py:2220: UserWarning: Collapsing a non-contiguous coordinate. Metadata may not be fully descriptive for 'i_mesh_face'.
  warnings.warn(msg.format(self.name()))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/h05/itpp/git/iris/iris_main/lib/iris/cube.py", line 3744, in collapsed
    collapsed_cube.replace_coord(coord.collapsed(local_dims))
  File "/home/h05/itpp/git/iris/iris_main/lib/iris/coords.py", line 2215, in collapsed
    elif not self.is_contiguous():
  File "/home/h05/itpp/git/iris/iris_main/lib/iris/coords.py", line 2018, in is_contiguous
    contiguous, _ = self._discontiguity_in_bounds(rtol=rtol, atol=atol)
  File "/home/h05/itpp/git/iris/iris_main/lib/iris/coords.py", line 1944, in _discontiguity_in_bounds
    self._sanity_check_bounds()
  File "/home/h05/itpp/git/iris/iris_main/lib/iris/coords.py", line 1900, in _sanity_check_bounds
    raise ValueError(
ValueError: Invalid operation for 'latitude', with 4 bound(s). Contiguous bounds are only defined for 1D coordinates with 2 bounds.
>>> 

So, the problem isn't really specific to the peculiarities of MeshCoords, it is about the shape of the bounds :
(N. 2) is expected and nothing else works.
However, it would make perfect sense to discard the bounds here and proceed (with a warning).

In the case of a MeshCoord, we can see that this does produce a perfectly sensible result.

>>> cube = cube[..., 0:]
>>> print(cube)
mesh_phenom / (unknown)             (level: 2; i_mesh_face: 3)
    Dimension coordinates:
        level                             x               -
        i_mesh_face                       -               x
    Auxiliary coordinates:
        latitude                          -               x
        longitude                         -               x
        mesh_face_aux                     -               x
>>> cube.collapsed('latitude', iris.analysis.MEAN)
   . . .
ValueError: Invalid operation for 'latitude', with 4 bound(s). Contiguous bounds are only defined for 1D coordinates with 2 bounds.
>>> 
>>> cube.coord('latitude').bounds = None
>>> cube.coord('longitude').bounds = None
>>> cube_1d = cube.collapsed('latitude', iris.analysis.MEAN)
>>> print(cube_1d)
mesh_phenom / (unknown)             (level: 2)
    Dimension coordinates:
        level                             x
    Scalar coordinates:
        i_mesh_face                 1, bound=(0, 2)
        latitude                    3201, bound=(3200, 3202)
        longitude                   3101 degrees_east, bound=(3100, 3102) degrees_east
        mesh_face_aux               0.0, bound=(0.0, 0.0)
    Cell methods:
        mean                        latitude
>>> 

[rcomer]

That particular call to is_contiguous only exists to generate a warning about metadata

iris/lib/iris/coords.py

Lines 2215 to 2222 in 62047f7

	elif not self.is_contiguous():
	msg = (
	"Collapsing a non-contiguous coordinate. "
	"Metadata may not be fully descriptive for {!r}."
	)
	warnings.warn(msg.format(self.name()))
	if self.has_bounds():

I have seen this warning many times, and have never found it useful. What happens if you just don't bother with the contiguity check?

[schlunma]

In ESMValTool, we recently started improving our support for unstructured grids. This issue here hampers the application of our area_statistics operation (basically collapsing over latitude and longitude) to data on unstructured grids.

I tested commenting the elif statement linked by @rcomer above, which seems to fix the problem.

Would it make sense to change the lines above to something like

@@ -2228,12 +2228,22 @@ class Coord(_DimensionalMetadata):
                     "Metadata may not be fully descriptive for {!r}."
                 )
                 warnings.warn(msg.format(self.name()))
-            elif not self.is_contiguous():
-                msg = (
-                    "Collapsing a non-contiguous coordinate. "
-                    "Metadata may not be fully descriptive for {!r}."
-                )
-                warnings.warn(msg.format(self.name()))
+            else:
+                try:
+                    self._sanity_check_bounds()
+                except ValueError as exc:
+                    msg = (
+                        f"Cannot check if coordinate is contiguous: "
+                        f"{str(exc)}."
+                    )
+                    warnings.warn(msg.format(self.name()))
+                else:
+                    if not self.is_contiguous():
+                        msg = (
+                            "Collapsing a non-contiguous coordinate. "
+                            "Metadata may not be fully descriptive for {!r}."
+                        )
+                        warnings.warn(msg.format(self.name()))
 
             if self.has_bounds():
                 item = self.core_bounds()

?

I could open a PR if that makes sense.

[rcomer]

There has been talk elsewhere (e.g. #4762 (comment)) of reducing the number of warnings Iris throws. Personally my vote would be to remove this if loop (and the two warnings it throws) altogether.

[schlunma]

Sounds much easier! Can I open a pull request with that?

[rcomer]

@schlunma I'd be delighted to merge such a PR, but I don't think I have the authority to decide that by myself. Hopefully someone else from @SciTools/peloton can weigh in. I guess it could be considered a breaking change, but I can't imagine anyone is relying on these warnings to tell them that their "metadata might not be fully descriptive".

[trexfeathers]

@SciTools/peloton like @schunma's suggestion, thanks! Please submit something and we'll get it reviewed

[schlunma]

See #4870.