Trouble getting fancy with ARRAY_AGG (DISTINCT ARRAY_AGG)

[bellwether-softworks]

Describe the bug

ARRAY_AGG usage in SQL results in an error when used in conjunction with DISTINCT parameter alongside another aggregate field.

To Reproduce

1. Establish example data:

CREATE TABLE example(id INT, parent_id INT, tag VARCHAR) AS VALUES
    (1, 0, 'bob'),
    (2, 0, 'cat'),
    (3, 1, 'tom'),
    (4, 1, 'cat'),
    (5, 1, 'tom');

2. Execute query using ARRAY_AGG and DISTINCT parameter:

SELECT
        parent_id,
        COUNT(id) AS count_of,
        ARRAY_AGG(DISTINCT tag) AS tags
    FROM example
    GROUP BY parent_id;

Executing the above results in the following message:

ArrowError(ExternalError(Internal("Inconsistent types in ScalarValue::iter_to_array. Expected Utf8, got List([tom,cat])")))

Expected behavior

Desired output should be similar to the following:

parent_id	count_of	tags
1	3	[tom, cat]
0	2	[bob, cat]

Additional context

Omitting either the COUNT field, or the DISTINCT clause in ARRAY_AGG, allows the query to complete successfully.

The above was initially discovered in v17.0.0 and verified to still be presenting in v26.0.0.

[parkma99]

It probably panic in here https://github.com/apache/arrow-datafusion/blob/8be5a8ced87ccf4fb09d33f1f52c525027696e2b/datafusion/common/src/scalar.rs#L2317.

I am confusing where the sv defined.

Same question in https://github.com/apache/arrow-datafusion/blob/8be5a8ced87ccf4fb09d33f1f52c525027696e2b/datafusion/common/src/scalar.rs#L2285

cc @alamb

[alamb]

I am confusing where the sv defined.

In the code you referenced this is basically the catch all meaning "if the type doesn't match one of the other branches"

[alamb]

Thanks for the report @bellwether-softworks -- I added this to #2326

[parkma99]

In the code you referenced this is basically the catch all meaning "if the type doesn't match one of the other branches"

Sorry I make a stupid question.😂

[alamb]

Sorry I make a stupid question.😂

Not at all -- we are all learning here together!