feat: Request all arrays from uproot at once inside dask task

pyproject.toml

@@ -38,7 +38,7 @@ classifiers = [
 ]
 dependencies = [
   "awkward>=2.6.3",
-  "uproot>=5.3.0",
+  "uproot!=5.3.3,>=5.3.0",
   "dask[array]>=2024.3.0;python_version>'3.8'",
   "dask[array]>=2023.4.0;python_version<'3.9'",
   "dask-awkward>=2024.3.0",

src/coffea/nanoevents/factory.py

@@ -86,6 +86,19 @@ def __getitem__(self, index):
         return self._mapping[self._func(index)]
 
 
+class _OnlySliceableAs:
+    """A workaround for how PreloadedSourceMapping works"""
+
+    def __init__(self, array, expected_slice):
+        self._array = array
+        self._expected_slice = expected_slice
+
+    def __getitem__(self, s):
+        if s != self._expected_slice:
+            raise RuntimeError(f"Mismatched slice: {s} vs. {self._expected_slice}")
+        return self._array
+
+
 class _map_schema_uproot(_map_schema_base):
     def __init__(
         self, schemaclass=BaseSchema, metadata=None, behavior=None, version=None
@@ -141,17 +154,23 @@ def load_buffers(
             f"{start}-{stop}",
         )
         uuidpfn = {partition_key[0]: tree.file.file_path}
-        mapping = UprootSourceMapping(
-            TrivialUprootOpener(uuidpfn, interp_options),
-            start,
-            stop,
-            cache={},
-            access_log=None,
-            use_ak_forth=True,
+        arrays = tree.arrays(
+            keys,
+            entry_start=start,
+            entry_stop=stop,
+            ak_add_doc=interp_options["ak_add_doc"],
             decompression_executor=decompression_executor,
             interpretation_executor=interpretation_executor,
+            how=dict,
         )
-        mapping.preload_column_source(partition_key[0], partition_key[1], tree)
+        source_arrays = {
+            k: _OnlySliceableAs(v, slice(start, stop)) for k, v in arrays.items()
+        }
+        mapping = PreloadedSourceMapping(
+            PreloadedOpener(uuidpfn), start, stop, access_log=None
+        )
+        mapping.preload_column_source(partition_key[0], partition_key[1], source_arrays)
+
         buffer_key = partial(self._key_formatter, tuple_to_key(partition_key))
 
         # The buffer-keys that dask-awkward knows about will not include the

src/coffea/nanoevents/mapping/preloaded.py

@@ -1,6 +1,9 @@
 import json
 import warnings
 
+import awkward
+import numpy
+
 from coffea.nanoevents.mapping.base import BaseSourceMapping, UUIDOpener
 from coffea.nanoevents.util import quote, tuple_to_key
 
@@ -76,12 +79,31 @@ def preload_column_source(self, uuid, path_in_source, source):
         key = self.key_root() + tuple_to_key((uuid, path_in_source))
         self._cache[key] = source
 
-    def get_column_handle(self, columnsource, name):
+    def get_column_handle(self, columnsource, name, allow_missing):
+        if allow_missing:
+            return columnsource[name] if name in columnsource else None
         return columnsource[name]
 
-    def extract_column(self, columnhandle, start, stop, **kwargs):
-        # make sure uproot is single-core since our calling context might not be
-        return columnhandle[start:stop]
+    def extract_column(
+        self, columnhandle, start, stop, allow_missing, use_ak_forth=True
+    ):
+        if allow_missing and columnhandle is None:
+            return awkward.contents.IndexedOptionArray(
+                awkward.index.Index64(numpy.full(stop - start, -1, dtype=numpy.int64)),
+                awkward.contents.NumpyArray(numpy.array([], dtype=bool)),
+            )
+        elif not allow_missing and columnhandle is None:
+            raise RuntimeError(
+                "Received columnhandle of None when missing column in file is not allowed!"
+            )
+        the_array = columnhandle[start:stop]
+        if allow_missing:
+            the_array = awkward.contents.IndexedOptionArray(
+                awkward.index.Index64(numpy.arange(stop - start, dtype=numpy.int64)),
+                awkward.contents.NumpyArray(the_array),
+            )
+
+        return the_array
 
     def __len__(self):
         return self._stop - self._start

src/coffea/nanoevents/transforms.py

@@ -477,7 +477,8 @@ def nestedindex(stack):
 
 def item(stack):
     field = stack.pop()
-    stack.append(stack.pop()[field])
+    array = stack.pop()
+    stack.append(array[field])
 
 
 def eventindex(stack):