Feature request: Support "unnesting" Array columns in the select clause

Use case: Aggregate queries on Array columns

Feature description: We need to be able to run aggregations on array columns. For example, if we have a table which looks like this…

+----+---------------+
| id | tags          |
+----+---------------+
| 1  | ["abc", "def"]|
| 2  | ["abc", "xyz"]|
+----+---------------+

…we need to be able to find counts for the unique values found in the tags column. For example (using UNNEST as a placeholder for this functionality):

select count(id) as tag_count, UNNEST(tags) as tag
from my_table group by tag order by tag_count desc;

Which would produce:

+----+---------------+
| tag_count | tag    |
+----+---------------+
| 2  | "abc"         |
| 1  | "def"         |
| 1  | "xyz"         |
+----+---------------+

Other products typically provide an UNNEST function which turns an array column into a set of rows so that aggregates over the array column will work. For example:

https://www.postgresql.org/docs/9.2/static/functions-array.html https://www.mapd.com/docs/latest/mapd-core-guide/dml/#array-support https://docs.snowflake.net/manuals/user-guide/querying-semistructured.html#using-the-flatten-function-to-parse-arrays

[mfussenegger]

Would something like the following work? If not, why?

Currently it doesn’t, but you’re right that we could make it work. We’ll have a look at enabling that as well.

[mfussenegger]

We’ve just merged support for table functions in the select list (https://github.com/crate/crate/pull/7716). This will be available in 3.2 and can be used for your use case:

create table t (id int primary key, tags array(string));
insert into t (id, tags) values (1, ['abc', 'def']);
insert into t (id, tags) values (2, ['abc', 'xyz']);

select count(*) as tag_count, tag from (
    select unnest(tags) as tag from t
) tt group by tag order by 1 desc;
+-----------+-----+
| tag_count | tag |
+-----------+-----+
|         2 | abc |
|         1 | xyz |
|         1 | def |
+-----------+-----+