New rule var-missing-data

CHANGES.md

@@ -4,6 +4,9 @@
 
 ### Adjustments and Enhancements
 
+- Added a new core rule `var-missing-data` that checks for the recommended 
+  use of a variable's missing data.
+
 - Added a new core rule `access-latency` that can be used to check the
   time it takes to open a dataset.
 

docs/todo.md

@@ -13,7 +13,6 @@
 ## Desired
 
 - project logo
-- add `core` rule checks recommended use of fill value
 - add `xcube` rule that helps to identify chunking issues 
 - apply rule op args/kwargs validation schema 
 - allow outputting suggestions, if any, that are emitted by some rules

tests/plugins/core/rules/test_var_missing_data.py

@@ -0,0 +1,65 @@
+#  Copyright © 2025 Brockmann Consult GmbH.
+#  This software is distributed under the terms and conditions of the
+#  MIT license (https://mit-license.org/).
+import numpy as np
+import xarray as xr
+
+from xrlint.plugins.core.rules.var_missing_data import VarMissingData
+from xrlint.testing import RuleTest, RuleTester
+
+valid_dataset_0 = xr.Dataset()
+valid_dataset_1 = xr.Dataset(
+    attrs=dict(title="v-data"),
+    coords={"t": xr.DataArray([0, 1, 2], dims="t", attrs={"units": "seconds"})},
+    data_vars={"v": xr.DataArray([10, 20, 30], dims="t", attrs={"units": "m/s"})},
+)
+
+invalid_dataset_0 = valid_dataset_1.copy(deep=True)
+invalid_dataset_0.t.attrs["_FillValue"] = -999
+
+invalid_dataset_1 = valid_dataset_1.copy(deep=True)
+invalid_dataset_1.t.encoding["_FillValue"] = -999
+
+invalid_dataset_2 = valid_dataset_1.copy(deep=True)
+invalid_dataset_2.v.attrs["scaling_factor"] = 0.01
+
+invalid_dataset_3 = valid_dataset_1.copy(deep=True)
+invalid_dataset_3.v.encoding["dtype"] = np.dtype(np.float64)
+
+invalid_dataset_4 = valid_dataset_1.copy(deep=True)
+invalid_dataset_4.v.attrs["valid_range"] = [0, 1]
+
+VarMissingDataTest = RuleTester.define_test(
+    "var-missing-data",
+    VarMissingData,
+    valid=[
+        RuleTest(dataset=valid_dataset_0),
+        RuleTest(dataset=valid_dataset_1),
+    ],
+    invalid=[
+        RuleTest(
+            dataset=invalid_dataset_0,
+            expected=[
+                "Unexpected attribute '_FillValue', coordinates must not have missing data."
+            ],
+        ),
+        RuleTest(
+            dataset=invalid_dataset_1,
+            expected=[
+                "Unexpected encoding '_FillValue', coordinates must not have missing data."
+            ],
+        ),
+        RuleTest(
+            dataset=invalid_dataset_2,
+            expected=["Missing attribute '_FillValue' since data packing is used."],
+        ),
+        RuleTest(
+            dataset=invalid_dataset_3,
+            expected=["Missing attribute '_FillValue', which should be NaN."],
+        ),
+        RuleTest(
+            dataset=invalid_dataset_4,
+            expected=["Valid ranges are not recognized by xarray (as of Feb 2025)."],
+        ),
+    ],
+)

tests/plugins/core/test_plugin.py

@@ -24,6 +24,7 @@ def test_rules_complete(self):
                 "time-coordinate",
                 "var-desc",
                 "var-flags",
+                "var-missing-data",
                 "var-units",
             },
             set(plugin.rules.keys()),

xrlint/plugins/core/__init__.py

@@ -28,6 +28,7 @@ def export_plugin() -> Plugin:
                 "time-coordinate": "error",
                 "var-desc": "warn",
                 "var-flags": "error",
+                "var-missing-data": "warn",
                 "var-units": "warn",
             },
         },

xrlint/plugins/core/rules/var_missing_data.py

@@ -0,0 +1,53 @@
+#  Copyright © 2025 Brockmann Consult GmbH.
+#  This software is distributed under the terms and conditions of the
+#  MIT license (https://mit-license.org/).
+
+import numpy as np
+
+from xrlint.node import VariableNode
+from xrlint.plugins.core.plugin import plugin
+from xrlint.rule import RuleContext, RuleOp
+
+
+@plugin.define_rule(
+    "var-missing-data",
+    version="1.0.0",
+    type="suggestion",
+    description=(
+        "Checks the recommended use of missing data, i.e., coordinate variables"
+        " should not define missing data, but packed data should."
+        " Notifies about the use of valid ranges to indicate missing data, which"
+        " is currently not supported by xarray."
+    ),
+    docs_url="https://cfconventions.org/cf-conventions/cf-conventions.html#units",
+)
+class VarMissingData(RuleOp):
+    def validate_variable(self, ctx: RuleContext, node: VariableNode):
+        array = node.array
+        encoding = array.encoding
+        attrs = array.attrs
+
+        fill_value_source = None
+        if "_FillValue" in encoding:
+            fill_value_source = "encoding"
+        elif "_FillValue" in attrs:
+            fill_value_source = "attribute"
+
+        if fill_value_source is not None and node.name in ctx.dataset.coords:
+            ctx.report(
+                f"Unexpected {fill_value_source} '_FillValue',"
+                f" coordinates must not have missing data."
+            )
+        elif fill_value_source is None and node.name in ctx.dataset.data_vars:
+            scaling_factor = encoding.get("scaling_factor", attrs.get("scaling_factor"))
+            add_offset = encoding.get("add_offset", attrs.get("add_offset"))
+            raw_dtype = encoding.get("dtype")
+            if add_offset is not None or scaling_factor is not None:
+                ctx.report("Missing attribute '_FillValue' since data packing is used.")
+            elif isinstance(raw_dtype, np.dtype) and np.issubdtype(
+                raw_dtype, np.floating
+            ):
+                ctx.report("Missing attribute '_FillValue', which should be NaN.")
+
+        if any((name in attrs) for name in ("valid_min", "valid_max", "valid_range")):
+            ctx.report("Valid ranges are not recognized by xarray (as of Feb 2025).")