Add supported for incorrectly ordered but the correct set of columns in from_pyarrow

Author: moeyensj

quivr/tables.py

@@ -196,6 +196,18 @@ def from_pyarrow(
                 # Only metadata differs; replace without touching data buffers
                 table = table.replace_schema_metadata(desired_schema.metadata)
             # else: already perfectly matches; keep as-is
+        elif set(table.schema.names) == set(schema.names):
+            # Columns are the same but ordered differently; reorder by name first
+            reordered = table.select(list(schema.names))
+            if reordered.schema.equals(schema, check_metadata=False):
+                # Only metadata may differ after reorder
+                if not reordered.schema.equals(desired_schema, check_metadata=True):
+                    table = reordered.replace_schema_metadata(desired_schema.metadata)
+                else:
+                    table = reordered
+            else:
+                # Types/nullability differ; cast after reorder to aligned names
+                table = reordered.cast(desired_schema)
         else:
             # Fallback: perform Arrow cast to coerce to the desired schema (may raise if names missing)
             table = table.cast(desired_schema)

test/test_tables.py

@@ -2051,3 +2051,14 @@ def test_bench_from_pyarrow_needs_cast(benchmark, N):
     table = pa.table({"x": x, "y": y})
 
     benchmark(Pair.from_pyarrow, table)
+
+
+@pytest.mark.parametrize("N", [100_000, 1_000_000, 10_000_000])
+@pytest.mark.benchmark(group="from-pyarrow")
+def test_bench_from_pyarrow_reorder_only(benchmark, N):
+    x = pa.array(np.random.randint(low=-10000, high=1000, size=N), type=pa.int64())
+    y = pa.array(np.random.randint(low=-10000, high=1000, size=N), type=pa.int64())
+    # Build with matching types but columns in opposite order
+    table = pa.table({"y": y, "x": x})
+
+    benchmark(Pair.from_pyarrow, table)