Merge pull request #886 from davidhassell/collapse-cell-measures

`cf.Field.collapse` fails when cell measure has external data

Author: davidhassell

Changelog.rst

@@ -9,6 +9,9 @@ Version NEXTVERSION
 * Allow multiple conditions for the same axis in `cf.Field.subspace`
   and `cf.Field.indices`
   (https://github.com/NCAS-CMS/cf-python/issues/881)
+* Fix bug in `cf.Field.collapse` that causes a ``ValueError`` to be raised
+  for missing external cell measures data
+  (https://github.com/NCAS-CMS/cf-python/issues/885)
 * New dependency: ``distributed>=2025.5.1``
 
 ----

cf/field.py

@@ -3708,7 +3708,12 @@ def weights(
                     weights_axes.discard(xaxis)
                     weights_axes.discard(yaxis)
                     if not Weights.cell_measure(
-                        self, "area", comp, weights_axes, methods=methods
+                        self,
+                        "area",
+                        comp,
+                        weights_axes,
+                        methods=methods,
+                        auto=True,
                     ):
                         Weights.area_XY(
                             self,
@@ -5472,52 +5477,60 @@ def collapse(
 
             weights: optional
                 Specify the weights for the collapse axes. The weights
-                are, in general, those that would be returned by this
-                call of the field construct's `weights` method:
-                ``f.weights(weights, axes=axes, measure=measure,
-                scale=scale, radius=radius, great_circle=great_circle,
-                components=True)``. See the *axes*, *measure*,
-                *scale*, *radius* and *great_circle* parameters and
-                `cf.Field.weights` for details, and note that the
-                value of *scale* may be modified depending on the
-                value of *measure*.
+                are created internally as the output of this call of
+                the field construct's `weights` method:
+                ``f.weights(weights, components=True, axes=axes,
+                measure=measure, scale=scale, radius=radius,
+                great_circle=great_circle)``.
 
-                .. note:: By default *weights* is `None`, resulting in
-                          **unweighted calculations**.
+                See the `cf.Field.weights` and the *axes*, *measure*,
+                *scale*, *radius*, and *great_circle* parameters for
+                details; and note that the value of *scale* may be
+                modified depending on the value of *measure*.
+
+                .. warning:: By default *weights* is `None`, resulting
+                             in **unweighted calculations**.
+
+                .. note:: Setting *weights* to `True` is generally a
+                          good way to ensure that all collapses are
+                          appropriately weighted according to the
+                          field construct's metadata. In this case, if
+                          it is not possible to create weights for any
+                          axis then an exception will be raised.
 
                 .. note:: Unless the *method* is ``'integral'``, the
                           units of the weights are not combined with
                           the field's units in the collapsed field.
 
-                If the alternative form of providing the collapse method
-                and axes combined as a CF cell methods-like string via the
-                *method* parameter has been used, then the *axes*
-                parameter is ignored and the axes are derived from the
-                *method* parameter. For example, if *method* is ``'T:
-                area: minimum'`` then this defines axes of ``['T',
+                .. note:: A pre-calculated weights array or dictionary
+                          may also be provided as the *weights*
+                          parameter. See `cf.Field.weights` for
+                          details
+
+                If the collapse method and axes have been provided as
+                a CF cell methods-like string via the *method*
+                parameter, then the *axes* parameter is ignored and
+                the axes for weights instead inferred from that
+                string. For instance, if *method* is ``'T: area:
+                minimum'`` then this defines axes of ``['T',
                 'area']``. If *method* specifies multiple collapses,
-                e.g. ``'T: minimum area: mean'`` then this implies axes of
-                ``'T'`` for the first collapse, and axes of ``'area'`` for
-                the second collapse.
-
-                .. note:: Setting *weights* to `True` is generally a good
-                          way to ensure that all collapses are
-                          appropriately weighted according to the field
-                          construct's metadata. In this case, if it is not
-                          possible to create weights for any axis then an
-                          exception will be raised.
-
-                          However, care needs to be taken if *weights* is
-                          `True` when cell volume weights are desired. The
-                          volume weights will be taken from a "volume"
-                          cell measure construct if one exists, otherwise
-                          the cell volumes will be calculated as being
-                          proportional to the sizes of one-dimensional
-                          vertical coordinate cells. In the latter case
-                          **if the vertical dimension coordinates do not
-                          define the actual height or depth thickness of
-                          every cell in the domain then the weights will
-                          be incorrect**.
+                e.g. ``'T: minimum area: mean'`` then this implies
+                axes of ``'T'`` for the first collapse, and axes of
+                ``'area'`` for the second collapse.
+
+                .. warning:: Care needs to be taken if *weights* is
+                             set to `True` when cell volume weights
+                             are desired. The volume weights will be
+                             taken from a "volume" cell measure
+                             construct if one exists, otherwise the
+                             cell volumes will be calculated as being
+                             proportional to the sizes of
+                             one-dimensional vertical coordinate
+                             cells. In the latter case **if the
+                             vertical dimension coordinates do not
+                             define the actual height or depth
+                             thickness of every cell in the domain
+                             then the weights will be incorrect**.
 
                 *Parameter example:*
                   To specify weights based on the field construct's

cf/weights.py

@@ -1508,15 +1508,16 @@ def cell_measure(
                 return False
 
             raise ValueError(
-                f"Can't find weights: No {measure!r} cell measure"
+                f"Can't find weights for {f!r}: No {measure!r} cell measure"
             )
 
         elif len_m > 1:
             if auto:
                 return False
 
             raise ValueError(
-                f"Can't find weights: Multiple {measure!r} cell measures"
+                f"Can't find weights for {f!r}: Multiple {measure!r} cell "
+                "measures"
             )
 
         key, clm = m.popitem()
@@ -1526,9 +1527,11 @@ def cell_measure(
                 return False
 
             raise ValueError(
-                f"Can't find weights: Cell measure {m!r} has no data, "
-                "possibly because it is external. "
-                "Consider setting cell_measures=False"
+                f"Can't find weights for {f!r}: Cell measure {clm!r} has no "
+                "data, possibly because it is in a missing external file. "
+                "Consider setting cell_measures=False or, if applicable, "
+                "re-reading the dataset whilst providing the external "
+                "file (see cf.read for details)."
             )
 
         clm_axes0 = f.get_data_axes(key)
@@ -1543,7 +1546,8 @@ def cell_measure(
                     return False
 
                 raise ValueError(
-                    "Multiple weights specifications for "
+                    f"Can't find weights for {f!r}: Multiple weights "
+                    "specifications for "
                     f"{f.constructs.domain_axis_identity(axis)!r} axis"
                 )