feat: Add array_min function support

[erenavsarogullari]

Which issue does this PR close?

Closes #14416.

What changes are included in this PR?

Currently, Spark, Snowflake and Presto support array_min function. This can also be useful for DataFusion.

array_min(make_array(3,1,4,2)) => 1
array_min(make_array('h','e','l','l',NULL,'o')) => e
array_min(make_array(NULL,NULL)) => NULL
select input, array_min(input) from 
   (select make_array(d - 1, d, d + 1) input from (values (10), (NULL)) t(d)) =>
----
[9, 10, 11] 9
[NULL, NULL, NULL] NULL

Spark: https://docs.databricks.com/en/sql/language-manual/functions/array_min.html
Snowflake: https://docs.snowflake.com/en/sql-reference/functions/array_min
Presto: https://prestodb.io/docs/current/functions/array.html#array_min-x-x

All potential use-cases have been covered like different data_types, empty array, NULL etc.

Also, planning to add array_max function as follow-up.

Are these changes tested?

Added new UT cases to verify array_min function in terms of different source arrays.

Are there any user-facing changes?

Yes, new SQL function is supported and documentation has also be updated.

[jayzhan211]

First of all, I'm not sure whether this function should be in datafusion core or datafusion-functions-extra. It seems this is not the "core" function that is supported in both Postgres or DuckDB.

Since we are going to support Spark function, maybe we should move this function inside it #5600

[findepi]

DuckDB has list_max, and our array semantics are supposed to model Duck's list semantics, thus it makes sense to add to DataFusion core.

Review-wise, let's do array_max well in #14470
and then return to this PR. It doesn't make sense to review the two in parallel, since most of the comments will be the same. For example, this PR still uses sort to get minimal element.

I would actually recommend closing this PR and creating a new afresh once array_max gets in, to avoid using old copy of the code. For

[jayzhan211]

DuckDB has list_max, and our array semantics are supposed to model Duck's list semantics, thus it makes sense to add to DataFusion core.

Review-wise, let's do array_max well in #14470 and then return to this PR. It doesn't make sense to review the two in parallel, since most of the comments will be the same. For example, this PR still uses sort to get minimal element.

I would actually recommend closing this PR and creating a new afresh once array_max gets in, to avoid using old copy of the code. For

DuckDB offers many array functions, but that doesn’t mean we need to port all of them to DataFusion Core. Our focus should be on functions that are already supported in PostgreSQL (which are a must-have) or those with significant user interest that justify ongoing maintenance in DataFusion Core.

[erenavsarogullari]

Thanks @jayzhan211 and @findepi for the reviews.
Updated this PR in terms of previous feedback from array_max PR: #14470
Please also find my comment for module selection for both functions: #14470 (comment)

[erenavsarogullari]

I have applied incoming review feedbacks from array_max PR: #14470 to here as well. Just FYI.

[cht42]

hey @erenavsarogullari, are you still tracking merging this ?

[erenavsarogullari]

hey @erenavsarogullari, are you still tracking merging this ?

Yes, this PR is ready to be merged if there is not any other concern. array_max function (#14470) has been merged. array_min function is conjugate of it.

[rluvaton]

Please implement return_type_from_args instead of this so you can return that the input is not nullable in case of non nullable input

[alamb]

Marking as draft as I think this PR is no longer waiting on feedback and I am trying to make it easier to find PRs in need of review. Please mark it as ready for review when it is ready for another look