Attempt to avoid unbounded refcat lookups in QG generation.

python/lsst/pipe/base/graphBuilder.py

@@ -50,10 +50,12 @@
     NamedValueSet,
     Quantum,
     Registry,
+    SkyPixDimension,
 )
 from lsst.daf.butler.registry import MissingCollectionError, MissingDatasetTypeError
 from lsst.daf.butler.registry.queries import DataCoordinateQueryResults
 from lsst.daf.butler.registry.wildcards import CollectionWildcard
+from lsst.sphgeom import PixelizationABC, RangeSet
 
 # -----------------------------
 #  Imports for other modules --
@@ -415,6 +417,27 @@ def __repr__(self) -> str:
     inputs to this quantum.
     """
 
+    def computeSpatialExtent(self, pixelization: PixelizationABC) -> RangeSet:
+        """Return the spatial extent of this quantum's inputs and outputs in
+        a skypix system.
+
+        Parameters
+        ----------
+        pixelization : `lsst.sphgeom.PixelizationABC`
+            Pixelization system.
+
+        Returns
+        -------
+        extent : `lsst.sphgeom.RangeSet`
+            Ranges of sky pixels that touch this quantum's inputs and outputs.
+        """
+        result = RangeSet()
+        for dataset_type, datasets in itertools.chain(self.inputs.items(), self.outputs.items()):
+            if dataset_type.dimensions.spatial:
+                for data_id in datasets.keys():
+                    result |= pixelization.envelope(data_id.region)
+        return result
+
     def makeQuantum(self, datastore_records: Mapping[str, DatastoreRecordData] | None = None) -> Quantum:
         """Transform the scaffolding object into a true `Quantum` instance.
 
@@ -1318,6 +1341,30 @@ def resolveDatasetRefs(
                             # which just means there are no datasets here.
                             prereq_refs = []
                     else:
+                        where = ""
+                        bind: dict[str, Any] = {}
+                        if not quantum.dataId.graph.spatial:
+                            # This has skypix dimensions (probably a reference
+                            # catalog), but the quantum's data is not spatial
+                            # (it's probably a full-survey sequence point).
+                            # Try to limit the spatial extent to the union of
+                            # the spatial extent of the inputs and outputs.
+                            for dimension in datasetType.dimensions:
+                                if isinstance(dimension, SkyPixDimension):
+                                    extent = quantum.computeSpatialExtent(dimension.pixelization)
+                                    pixels: list[int] = []
+                                    for begin, end in extent:
+                                        pixels.extend(range(begin, end))
+                                    if not pixels:
+                                        _LOG.warning(
+                                            "Prerequisite input %r to task %r may be unbounded.",
+                                            datasetType.name,
+                                            quantum.task.taskDef.label,
+                                        )
+                                    else:
+                                        bind["quantum_extent"] = pixels
+                                        where = f"{dimension.name} IN (quantum_extent)"
+                                    break
                         # Most general case.
                         prereq_refs = [
                             prereq_ref if component is None else prereq_ref.makeComponentRef(component)
@@ -1326,6 +1373,8 @@ def resolveDatasetRefs(
                                 collections=collections,
                                 dataId=quantum.dataId,
                                 findFirst=True,
+                                where=where,
+                                bind=bind,
                             ).expanded()
                         ]