Enable automatic type deduction for columns built from lists of dataframes.

torcharrow/dtypes.py

@@ -624,6 +624,12 @@ def infer_dtype_from_value(value):
         for t in value:
             dtypes.append(infer_dtype_from_value(t))
         return prt(value, Tuple(dtypes))
+
+    from torcharrow.velox_rt.dataframe_cpu import DataFrameCpu
+
+    if isinstance(value, DataFrameCpu):
+        return prt(value, List(value.dtype))
+
     raise AssertionError(f"unexpected case {value} of type {type(value)}")
 
 
@@ -729,8 +735,14 @@ def common_dtype(l: DType, r: DType) -> ty.Optional[DType]:
     if is_list(l) and is_list(r):
         k = common_dtype(l.item_dtype, r.item_dtype)
         return List(k).with_null(l.nullable or r.nullable) if k is not None else None
+    if is_struct(l) and is_struct(r):
+        if l.fields == r.fields:
+            return Struct(l.fields, l.nullable or r.nullable)
+        else:
+            return None
     if l.with_null() == r.with_null():
         return l if l.nullable else r
+
     return None
 
 

torcharrow/test/test_list_column.py

@@ -208,6 +208,19 @@ def base_test_fixed_size_list(self):
             f"Unexpected failure reason: {str(ex.exception)}",
         )
 
+    def base_test_column_from_dataframe_list(self):
+        a = ta.dataframe({"a": [1, 2, 3, 4], "b": [5, 6, 7, 8]})
+        b = ta.column([a, a])
+        self.assertEqual(
+            list(b),
+            [[(1, 5), (2, 6), (3, 7), (4, 8)], [(1, 5), (2, 6), (3, 7), (4, 8)]],
+        )
+        self.assertEqual(
+            b.dtype,
+            dt.List(dt.Struct([dt.Field("a", dt.int64), dt.Field("b", dt.int64)])),
+        )
+        self.assertTrue(isinstance(b, ta.velox_rt.list_column_cpu.ListColumnCpu))
+
 
 if __name__ == "__main__":
     unittest.main()

torcharrow/test/test_list_column_cpu.py

@@ -46,6 +46,9 @@ def test_map_reduce_etc(self):
     def test_fixed_size_list(self):
         self.base_test_fixed_size_list()
 
+    def test_column_from_dataframe_list(self):
+        self.base_test_column_from_dataframe_list()
+
 
 if __name__ == "__main__":
     unittest.main()