Fix data ID queries constrained by datasets with empty dimensions.

lsst · Mar 31, 2022 · 198df4b · 198df4b
1 parent cefad88
commit 198df4b
Show file tree

Hide file tree

Showing 2 changed files with 21 additions and 1 deletion.
diff --git a/python/lsst/daf/butler/registry/queries/_builder.py b/python/lsst/daf/butler/registry/queries/_builder.py
@@ -277,7 +277,11 @@ def _build_dataset_constraint_subquery(
             storage.select(
                 *collections,
                 dataId=SimpleQuery.Select,
-                id=None,
+                # If this dataset type has no dimensions, we're in danger of
+                # generating an invalid subquery that has no columns in the
+                # SELECT clause.  Any easy fix is to just select some arbitrary
+                # column that goes unused, like the dataset ID.
+                id=None if storage.datasetType.dimensions else SimpleQuery.Select,
                 run=None,
                 ingestDate=None,
                 timespan=None,

diff --git a/python/lsst/daf/butler/registry/tests/_registry.py b/python/lsst/daf/butler/registry/tests/_registry.py
@@ -1656,6 +1656,22 @@ def testEmptyDimensionsQueries(self):
                     dataIds.findDatasets(schema, collections=[run2, run1], findFirst=True),
                     [dataset2],
                 )
+        # Query for non-empty data IDs with a constraint on an empty-data-ID
+        # dataset that exists.
+        dataIds = registry.queryDataIds(["instrument"], datasets="schema", collections=...)
+        self.checkQueryResults(
+            dataIds.subset(unique=True),
+            [DataCoordinate.standardize(instrument="Cam1", universe=registry.dimensions)],
+        )
+        # Again query for non-empty data IDs with a constraint on empty-data-ID
+        # datasets, but when the datasets don't exist.  We delete the existing
+        # dataset and query just that collection rather than creating a new
+        # empty collection because this is a bit less likely for our build-time
+        # logic to shortcut-out (via the collection summaries), and such a
+        # shortcut would make this test a bit more trivial than we'd like.
+        registry.removeDatasets([dataset2])
+        dataIds = registry.queryDataIds(["instrument"], datasets="schema", collections=run2)
+        self.checkQueryResults(dataIds, [])
 
     def testDimensionDataModifications(self):
         """Test that modifying dimension records via: