refactor(sqllogictests): port group by test to sqllogic

[aprimadi]

Which issue does this PR close?

Port group by tests from group_by.rs.

Work towards #4495

Rationale for this change

N/A

What changes are included in this PR?

N/A

Are these changes tested?

Yes

Are there any user-facing changes?

No

[alamb]

Thank you @aprimadi -- we really appreciate the help

[aprimadi]

Not sure if we should port group_by_dictionary test case @alamb.

It seems that if we port it into sqllogic, the test would be longer. While doing it programmatically seems to be very concise.

[aprimadi]

This is probably enough for now, all the other tests use a one-time-use test data.

[alamb]

I don't think you need rust code to do this setup anymore

You can create arbitrary Arrow types that have no corresponding SQL types with the arrow_cast function -- https://arrow.apache.org/datafusion/user-guide/sql/data_types.html

For example, here is one way to do it:
https://github.com/apache/arrow-datafusion/blob/7759d96e892e7c159136af35ebc0def49ed3b3b1/datafusion/core/tests/sqllogictests/test_files/timestamps.slt#L113-L129

So you can make a value of Time32(Seconds) with something like

❯ select arrow_cast(5000::int, 'Time32(Second)');
+-------------+
| Int64(5000) |
+-------------+
| 01:23:20    |
+-------------+

[aprimadi]

Alright, thank you for the feedback @alamb. Didn't know that. I'm done for the day, will probably work more on this tomorrow if nothing comes up.

[aprimadi]

@alamb not sure how to port group_by_date_trunc and group_by_dictionary without writing setup code. Do we want to migrate those also?

[jiangzhx]

My idea is mainly based on these two test cases and what they cover.

1. group_by_date_trunc based on the method name, is to test date_trunc. I think we can use an existing test data and move to sqllogictests
2. group_by_dictionary based on the code, is testing DictionaryArray. In my opinion, it should not be migrated to sqllogictest.

[aprimadi]

I think both group_by_date_trunc and group_by_dictionary test should not be moved.

For group_by_date_trunc the data is procedurally generated and it also generates multiple partition file.

For group_by_dictionary, I totally agree with you @jiangzhx

IMO, the data setup should be as close to the code that test it except if it's being used multiple times.

[aprimadi]

Btw,, those are the only two tests from group_by.rs not being moved to sqllogic.

[alamb]

For group_by_date_trunc the data is procedurally generated and it also generates multiple partition file.

For group_by_dictionary, I totally agree with you @jiangzhx

I agree -- btw I am hoping eventually to be able to create partitioned files using COPY TO ... #5654

[alamb]

Looks good to me . Thank you @aprimadi

It looks like we can eventually consolidate group_by and aggregate.slt 🤔

https://github.com/apache/arrow-datafusion/blob/main/datafusion/core/tests/sqllogictests/test_files/aggregate.slt

[alamb]

I really appreciate the help @aprimadi ❤️