fix(starrocks): exp.Unnest transpilation

[hellozepp]

Fixes #3962

This PR adds support for the following:

• Default initialize exp.UNNEST to include "unnest" as the default table alias if not specified
• Added expression type of exp.Inline
• use arrays_zip to merge multiple Lateral views
• Added some ut to validate unnest transpile

[hellozepp]

I tried to add the logic of setting default value when there is no table alias, PTAL @georgesittas @VaggelisD

I need to transpile the starrocks sql:

SELECT student, score, unnest FROM tests CROSS JOIN LATERAL UNNEST(scores)

Expected spark sql:

SELECT student, score, unnest FROM tests LATERAL VIEW EXPLODE(scores) unnest AS unnest

[hellozepp]

I found that when unnest has multiple outputs, it will be transpiled into multiple Lateral view statements (actually incorrect), as shown in the following SQL of starrocks:

SELECT id, t.type, t.scores FROM example_table CROSS JOIN LATERAL unnest(split(type, ";"), scores) AS t(type,scores)

Transpiled into spark, the output is:

SELECT id, t.type, t.scores FROM example_table LATERAL VIEW EXPLODE(SPLIT(type, CONCAT('\\Q', ';'))) t AS type LATERAL VIEW EXPLODE(scores) t AS scores

Incorrect reason:
The output of unnest is the combined value of multiple columns, and LATERAL VIEW will be expanded multiple times and cross joined.
We can use arrays_zip to merge multiple Lateral views.

In fact, our users often use the syntax combination LATERAL VIEW INLINE(arrays_zip(col1, col2)).

table function inline doc as following:

databricks: https://docs.databricks.com/en/sql/language-manual/functions/inline.html
spark: https://spark.apache.org/docs/latest/api/sql/#inline
hive:https://stackoverflow.com/questions/25088488/use-inlinearraystruct-struct-in-hive

Expected output:

SELECT id, t.type, t.score FROM example_table LATERAL VIEW INLINE(ARRAYS_ZIP(SPLIT(type, CONCAT('\\Q', ';')), scores)) t AS type, score

But function arrays_zip may not be supported by some engines, such as hive. I added a parameter unnest_using_arrays_zip to avoid hive being affected.

I added a commit. Will be open to any suggestions =)

[VaggelisD]

Hey, thanks for the follow-up PR! Left a few comments in my first pass.

[hellozepp]

@VaggelisD Added new commit, PTAL

[hellozepp]

@VaggelisD Good evening, I made some change, please take a look at it when you have time

[georgesittas]

Thanks for the PR @hellozepp!

[hellozepp]

@georgesittas @VaggelisD Added new commit, PTAL

[georgesittas]

Thanks for the PR @hellozepp - LGTM. Appreciate the patience & quick iterations.

We plan to get this in soon and take a stab at refactoring the hive transformation so that it actually does the right thing. One way to go about this is by simulating the INLINE & ARRAYS_ZIP with a subquery that joins the POSEXPLODEd arrays on their positions and then projects the corresponding columns so they can be accessed in the scope of the original UNNEST.

[hellozepp]

@georgesittas Agree.
Thanks for the previous advice, it helped me a lot =)

[hellozepp]

Sorry, I cleaned up my sqlglot repository because it was a mess. Wait a minute and I'll make another commit. There won't be any new code since our last commit.