Add table function exclude_columns

[kasiafi]

Description

Adds a new table function exclude_columns as a built-in table function.

SELECT *
FROM TABLE(exclude_columns(
                        input => TABLE(orders),
                        columns => DESCRIPTOR(clerk, comment)))

This function is a replacement for syntax SELECT * EXCEPT(...). Such syntax, in a few variants, is supported by different databases (as discussed here https://stackoverflow.com/questions/413819/select-except). However, it is not part of the SQL standard, and so Trino does not support it.

Additional context and related issues

There is a related optimization PR submitted for review: #16012. Technically, the optimization is not a prerequisite, but it could affect the performance a lot by enabling column pruning.

Release notes

( ) This is not user-visible or docs only and no release notes are required.
( ) Release notes are required, please propose a release note for me.
(X) Release notes are required, with the following suggested text:

# General
* Add a built-in table function `exclude_columns()`.

# General
* Allow table functions to return anonymous columns.

Docs included.

[findepi]

CI Pull Request Labeler / Test Report (pull_request_target) https://github.com/trinodb/trino/pull/16584/checks?check_run_id=12085703933 red:

TestKafkaPushdownSmokeTest > testCreateTimePushdown [groups: profile_specific_tests, kafka]
java.sql.SQLException: Query failed (#20230317_115011_00000_p5jy5): No worker nodes available

but pt (default, suite-kafka, ) is green and reports success for TestKafkaPushdownSmokeTest.testCreateTimePushdown method (3 times)

is Pull Request Labeler right or not?

cc @nineinchnick

[kasiafi]

@ebyhr I added the ability to return anonymous columns from a table function (as a prep commit).
Now the function exclude_columns returns columns with the same name, or no name, as the original column.
For the example mentioned above, the result is:

trino:tiny> SELECT 0, 1, *
         -> FROM TABLE(exclude_columns(
         ->                    TABLE(SELECT * FROM (SELECT 3 a), (SELECT 4)),
         ->                    DESCRIPTOR(a)));
 _col0 | _col1 | _col2
-------+-------+-------
     0 |     1 |     4
(1 row)

[kasiafi]

@martint I enabled returning anonymous columns from a table function. I believe we discussed it offline some time ago. Please take a look.

[martint]

Maybe better to be explicit and say "COLUMNS descriptor is null".

BTW, under what conditions would this happen? The argument is marked as required, so the query should fail during analysis if it's missing.

[kasiafi]

Yes, the argument is required, but the user can pass null descriptor: columns => cast(null AS descriptor). There is a test case that covers it.

[martint]

This is something we'll have to fix when we address identifier semantics. Let's add a TODO.

[martint]

Also, since descriptor columns names are canonical, maybe we keep them as is here and canonicalize the TableArgument columns instead when matching (for now, canonicalize to lower-case to preserve current semantics)

[kasiafi]

I don't think we should do this.
If we uppercase one side of comparison (which is common in SQL-standard canonicalization), and lowercase the other side, we have little chance for a match. Effectively, the user would have to write all names in COLUMNS in lowercase and quoted.

[kasiafi]

@martint Updated. Please re-review.