[ARROW-11763] [C++] Dict index type ALWAYS gets coerced to int32 when saving to parquet

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 3.0.0
• Component/s: C++
• External issue URL: https://github.com/apache/arrow/issues/27616
• Assignee: Unassigned
• Reporter: ARF
• Created: 24/Feb/21 16:45
• Updated: 11/Jan/23 08:21

Description

On saving a pyarrow Dictionary-type column to parquet, any non-int32 index gets coerced to an int32 index without warning:

import pyarrow as pa
from pyarrow import parquet as pq

schema = pa.schema({
    'foo': pa.dictionary(pa.int8(), pa.string(), ordered=False),
})

def make_trivial_dict_array(dict_type, value, size):
    return 

table = pa.Table.from_pydict({
    'foo': pa.DictionaryArray.from_arrays(
        pa.nulls(1, schema.field('foo').type.index_type).fill_null(0),
        ['bar'])
})

pq.write_table(table, 'test_dict_int8.parquet', version='2.0', data_page_version='2.0')

print(f"dict index type before saving to parquet: {table.schema.field('foo').type.index_type}")

del table

table = pq.read_table('test_dict_int8.parquet')
print(f"dict index type after saving to parquet: {table.schema.field('foo').type.index_type}")

Output:

dict index type before saving to parquet: int8
dict index type after saving to parquet: int32

While this is surprising for smaller index types, coercing an int64 index to an int32 index without warning the user seems like asking for trouble.