Fix failure when matching empty dictionary

[martint]

Fixes #9424

Release note:

# Iceberg, Hive, Delta
* Fix failure when reading Parquet data that contains only null values. ({issue}`9424`)

[findepi]

The CI didn't run here.

[guerremdq]

Any plan to merge this soon?

[xtto]

Hi, with that fix applied Trino returns us 0 rows, when perform IS NULL prediction on the column which is always NULL within the whole partition.

SELECT *
FROM test_tbl
WHERE a_partition = '01'
AND always_null_column IS NULL

 always_null_column | a_partition
--------------------+-------------
(0 rows)

SELECT *
FROM test_tbl
WHERE a_partition = '01'
AND always_null_column IS NOT NULL

 always_null_column | a_partition
--------------------+-------------
(0 rows)

SELECT *
FROM test_tbl
WHERE a_partition = '01'
-- AND always_null_column IS NULL

 always_null_column | a_partition
--------------------+-------------
 NULL               | 01
 NULL               | 01
 NULL               | 01
 NULL               | 01
 NULL               | 01

[martint]

@kantonczak, can you try with the latest updates to this PR?

[Gingernaut]

+1 on this issue and fix, this issue is preventing me from migrating from Presto to Trino.

[rushton]

@martint the issue can be reproduced by writing a parquet file with pyarrow:

import pyarrow
from pyarrow import parquet

with open("foo.parquet", "wb") as fp:
    parquet.write_table(pyarrow.Table.from_arrays([pyarrow.array([None,None,None,None])], names=["x"]), fp)

I will send the parquet file in trino slack.

Repro query:

CREATE TABLE test_null_parquet (x VARCHAR) WITH (format='PARQUET', external_location='s3a://.../foo.parquet');

SELECT * FROM test_null_parquet WHERE x IS NULL;

Query 20220817_185451_00007_yq9np, FAILED, 1 node
Splits: 5 total, 0 done (0.00%)
1.76 [0 rows, 0B] [0 rows/s, 0B/s]

Query 20220817_185451_00007_yq9np failed: Error opening Hive split s3a://.../nick/foo/myparquet.parquet (offset=0, length=408): cannot use empty rangeList

[raunaqmorarka]

lgtm, minor comments