feat: 3154 from_parquet should be able to read partial columns (#3156)

* feat: 3154 - Adding unit test and fix

* Taking this opportunity to fix an issue with 2772 test

This had been bugging me, and with a better understanding now I can improve
the unit tests. Mostly for future readability and correctness.

* Removing commented-out test return value

---------

Co-authored-by: Ianna Osborne <[email protected]>

src/awkward/_connect/pyarrow/table_conv.py

@@ -1,3 +1,5 @@
+# BSD 3-Clause License; see https://github.com/scikit-hep/awkward/blob/main/LICENSE
+
 from __future__ import annotations
 
 import json
@@ -128,11 +130,34 @@ def native_arrow_field_to_akarraytype(
     fields = _fields_of_strg_type(storage_type)
     if len(fields) > 0:
         # We need to replace storage_type with one that contains AwkwardArrowTypes.
-        awkwardized_fields = [
-            native_arrow_field_to_akarraytype(field, meta)  # Recurse
-            for field, meta in zip(fields, metadata["subfield_metadata"])
-        ]
+        sub_meta = metadata["subfield_metadata"]
+        awkwardized_fields = None  # Temporary
+        if len(sub_meta) == len(fields):
+            awkwardized_fields = [
+                native_arrow_field_to_akarraytype(field, meta)  # Recurse
+                for field, meta in zip(fields, metadata["subfield_metadata"])
+            ]
+        elif len(fields) < len(sub_meta):
+            # If a user has read a partial column, we can have fewer Arrow fields than the original.
+            sub_meta_dict = {sm["field_name"]: sm for sm in sub_meta}
+            awkwardized_fields = []
+            for field in fields:
+                if field.name in sub_meta_dict:
+                    awkwardized_fields.append(
+                        native_arrow_field_to_akarraytype(
+                            field, sub_meta_dict[field.name]
+                        )
+                    )
+                else:
+                    raise ValueError(
+                        f"Cannot find Awkward metadata for sub-field {field.name}"
+                    )
+        else:
+            raise ValueError(
+                f"Not enough fields in Awkward metadata. Have {len(sub_meta)} need at least {len(fields)}."
+            )
         storage_type = _make_pyarrow_type_like(storage_type, awkwardized_fields)
+
     ak_type = AwkwardArrowType._from_metadata_object(storage_type, metadata)
     return pyarrow.field(ntv_field.name, type=ak_type, nullable=ntv_field.nullable)
 

tests/test_2772_parquet_extn_array_metadata.py

@@ -3,6 +3,7 @@
 
 from __future__ import annotations
 
+import io
 import os
 
 import numpy as np
@@ -137,6 +138,22 @@ def test_array_conversions(akarray, as_dict):
     rt_array = ak.from_arrow(as_extn, highlevel=True)
     assert to_list(rt_array) == to_list(akarray)
 
+    # Deeper test of types
+    akarray_high = ak.Array(akarray)
+    if akarray_high.type.content.parameters.get("__categorical__", False) == as_dict:
+        # as_dict is supposed to go hand-in-hand with __categorical__: True, and if it
+        #  does not, we do not round-trip perfectly. So only test when this is set correctly.
+        assert rt_array.type == akarray_high.type
+
+        ak_type_str_orig = io.StringIO()
+        ak_type_str_rtrp = io.StringIO()
+        akarray_high.type.show(stream=ak_type_str_orig)
+        rt_array.type.show(stream=ak_type_str_rtrp)
+        if ak_type_str_orig.getvalue() != ak_type_str_rtrp.getvalue():
+            print("  Original type:", ak_type_str_orig.getvalue())
+            print("  Rnd-trip type:", ak_type_str_rtrp.getvalue())
+        assert ak_type_str_orig.getvalue() == ak_type_str_rtrp.getvalue()
+
 
 def test_table_conversion():
     ak_tbl_like = ak.Array(

tests/test_3154_parquet_subcolumn_select.py

@@ -0,0 +1,40 @@
+# BSD 3-Clause License; see https://github.com/scikit-hep/awkward/blob/main/LICENSE
+# ruff: noqa: E402
+
+from __future__ import annotations
+
+import os
+
+import pytest
+
+import awkward as ak
+
+pa = pytest.importorskip("pyarrow")
+pq = pytest.importorskip("pyarrow.parquet")
+
+
+def test_parquet_subcolumn_select(tmp_path):
+    ak_tbl = ak.Array(
+        {
+            "a": [
+                {"lbl": "item 1", "idx": 11, "ids": [1, 2, 3]},
+                {"lbl": "item 2", "idx": 12, "ids": [51, 52]},
+                {"lbl": "item 3", "idx": 13, "ids": [61, 62, 63, 64]},
+            ],
+            "b": [
+                [[111, 112], [121, 122]],
+                [[211, 212], [221, 222]],
+                [[311, 312], [321, 322]],
+            ],
+        }
+    )
+    parquet_file = os.path.join(tmp_path, "test_3514.parquet")
+    ak.to_parquet(ak_tbl, parquet_file)
+
+    selection = ak.from_parquet(parquet_file, columns=["a.ids", "b"])
+    assert selection["a"].to_list() == [
+        {"ids": [1, 2, 3]},
+        {"ids": [51, 52]},
+        {"ids": [61, 62, 63, 64]},
+    ]
+    assert selection["b"].to_list() == ak_tbl["b"].to_list()