Also allow storage class conversion on butler.put

The infrastructure already existed to support it and there
is no particular reason why get should convert but put
should not. Doing this provides a way for newer classes
to be used in code but whilst still supporting old
dataset type definitions.

python/lsst/daf/butler/configs/storageClasses.yaml

@@ -110,6 +110,8 @@ storageClasses:
     pytype: lsst.skymap.BaseSkyMap
   PropertySet:
     pytype: lsst.daf.base.PropertySet
+    converters:
+      lsst.pipe.base.TaskMetadata: lsst.daf.base.PropertySet.from_mapping
   PropertyList:
     pytype: lsst.daf.base.PropertyList
   IsrCalib:

python/lsst/daf/butler/datastores/fileDatastore.py

@@ -1058,6 +1058,10 @@ def _write_in_memory_to_artifact(self, inMemoryDataset: Any, ref: DatasetRef) ->
         info : `StoredFileInfo`
             Information describing the artifact written to the datastore.
         """
+        # May need to coerce the in memory dataset to the correct
+        # python type.
+        inMemoryDataset = ref.datasetType.storageClass.coerce_type(inMemoryDataset)
+
         location, formatter = self._prepare_for_put(inMemoryDataset, ref)
         uri = location.uri
 

tests/config/basic/storageClasses.yaml

@@ -21,6 +21,8 @@ storageClasses:
     delegate: lsst.daf.butler.tests.MetricsDelegate
     parameters:
       - slice
+    converters:
+      dict: lsst.daf.butler.tests.MetricsExample.makeFromDict
   StructuredData:
     # Data from a simple Python class
     pytype: lsst.daf.butler.tests.MetricsExample

tests/test_butler.py

@@ -1352,7 +1352,7 @@ def testPruneDatasets(self):
         butler.pruneDatasets([ref1, ref2, ref3], purge=True, unstore=True)
 
     def testPytypeCoercion(self):
-        """Test python type coercion on Butler.get"""
+        """Test python type coercion on Butler.get and put."""
 
         # Store some data with the normal example storage class.
         storageClass = self.storageClassFactory.getStorageClass("StructuredDataNoComponents")
@@ -1366,6 +1366,15 @@ def testPytypeCoercion(self):
         datasetType_ori = butler.registry.getDatasetType(datasetTypeName)
         self.assertEqual(datasetType_ori.storageClass.name, "StructuredDataNoComponents")
 
+        # Put a dict and this should coerce to a MetricsExample
+        test_dict = {"summary": {"a": 1}, "output": {"b": 2}}
+        metric_ref = butler.put(test_dict, datasetTypeName, dataId=dataId, visit=424)
+        test_metric = butler.getDirect(metric_ref)
+        self.assertEqual(test_metric.summary, test_dict["summary"])
+        self.assertEqual(test_metric.output, test_dict["output"])
+        # Remove the dataset because another will be stored later in the test.
+        butler.pruneDatasets([metric_ref], unstore=True, purge=True)
+
         # Now need to hack the registry dataset type definition.
         # There is no API for this.
         manager = butler.registry._managers.datasets