[ARROW-8142] [C++] Casting a chunked array with 0 chunks critical failure

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.17.0
• Component/s: C++
• Labels: pull-request-available
• External issue URL: https://github.com/apache/arrow/issues/24347
• Assignee: Ben Kietzman
• Reporter: Florian Jetter
• Created: 18/Mar/20 08:59
• Updated: 11/Jan/23 07:58
• Resolved: 25/Mar/20 18:54

Description

When casting a schema of an empty table from dict encoded to non-dict encoded type a critical error is raised and not handled causing the interpreter to shut down.

This only happens after a parquet roundtrip

 

import pyarrow as pa
import pandas as pd
import pyarrow.parquet as pq

df = pd.DataFrame({"col": ["a"]}).astype({"col": "category"}).iloc[:0]
table = pa.Table.from_pandas(df)
field = table.schema[0]
new_field = pa.field(field.name, field.type.value_type, field.nullable, field.metadata)

buf = pa.BufferOutputStream()
pq.write_table(table, buf)
reader = pa.BufferReader(buf.getvalue().to_pybytes())
table = pq.read_table(reader)

schema = table.schema.remove(0).insert(0, new_field)
new_table = table.cast(schema)
assert new_table.schema == schema
 

 

Output

WARNING: Logging before InitGoogleLogging() is written to STDERR
F0318 09:55:14.266649 299722176 table.cc:47] Check failed: (chunks.size()) > (0) cannot construct ChunkedArray from empty vector and omitted type 

 

Issue Links

• relates to: ARROW-7907 [Python] Conversion to pandas of empty table with timestamp type aborts
• links to: GitHub Pull Request #6668