fix: Dialect requires derived table alias #12994

peasee · 2024-10-18T06:53:01Z

Which issue does this PR close?

Rationale for this change

Fixes a bug preventing derived logical plans from running in MySQL due to a requirement that every derived table must have an alias:

The [AS] tbl_name clause is mandatory because every table in a FROM clause must have a name.

What changes are included in this PR?

Includes a dialect update to specify whether derived tables require aliases. For dialects that do, set a static alias. I've simply selected the name of the operation prefixed with derived_, like derived_sort, etc.

Are these changes tested?

Yes

Are there any user-facing changes?

No

* fix: Add Dialect option for requiring table aliases * feat: Add CustomDialectBuilder for requires_table_alias * docs: Spelling * refactor: rename requires_derived_table_alias * refactor: rename requires_derived_table_alias

goldmedal

Thanks @peasee for working on this. I left some suggestions here.

datafusion/sql/src/unparser/plan.rs

goldmedal · 2024-10-18T15:21:13Z

datafusion/sql/tests/cases/plan_to_sql.rs

+            expected:
+                // top projection sort gets derived into a subquery
+                // for MySQL, this subquery needs an alias
+                "SELECT `j1_min` FROM (SELECT min(`ta`.`j1_id`) AS `j1_min`, min(`ta`.`j1_id`) FROM `j1` AS `ta` ORDER BY min(`ta`.`j1_id`) ASC) AS `derived_sort` LIMIT 10",


I found the test only covers the derived_sort case but some cases aren't covered, such as derived_projection, derived_limit,...
Could you add more tests for them?

I added another test case for a more complex SQL that results in 2 nested derives for derived_sort and derived_distinct in the same query.

I'm not too sure what SQL to use to trigger derived_limit or derived_projection though... any ideas that I could include?

For derived_projection, I found the SQL can trigger:

select j1_id from (select 1 as j1_id) -> SELECT `j1_id` FROM (SELECT 1 AS `j1_id`) AS `derived_projection`

However, I guess it's not a valid SQL for MySQL, right? I tried it in DataFusion and DuckDB, they accept it but Postgres doesn't allow it. Maybe, it could be a DataFusion dialect to MySQL dialect case.

For derived_limit, a similar case can trigger it:

select * from (select * from j1 limit 10) -> SELECT * FROM (SELECT * FROM `j1` LIMIT 10) AS `derived_limit`

It is valid for MySQL, I just tested it out in my CLI. I've added both as new cases, thanks!

goldmedal · 2024-10-19T09:34:29Z

By the way, I tried to pull your branch and run the tests on my laptop but the tests always fail. After merging with the latest main branch, the tests are passed.

…-derived-table-alias

goldmedal

It's a good improvement for the derived table. 👍 It makes sense to me.
I noticed some databases don't support this kind of unnamed subquery, see #12896 (comment)
This PR would be helpful.

goldmedal · 2024-10-19T13:19:46Z

Maybe @phillipleblanc and @sgrebnov want to take a look at this PR.

phillipleblanc · 2024-10-19T14:55:01Z

Looks good to me, thanks @goldmedal and @peasee

alamb · 2024-10-20T12:48:47Z

@goldmedal perhaps you would like to merge this PR as a way to verify your permissions are configured correctly?

goldmedal · 2024-10-20T12:54:07Z

@goldmedal perhaps you would like to merge this PR as a way to verify your permissions are configured correctly?

Sure, I plan to merge this PR if no more comments tomorrow.

peasee · 2024-10-21T03:13:30Z

@goldmedal perhaps you would like to merge this PR as a way to verify your permissions are configured correctly?

Sure, I plan to merge this PR if no more comments tomorrow.

Let's hold off on merging this for now. I think I've introduced a regression that I'd like to test more first before merging.

…lias

peasee · 2024-10-21T03:21:37Z

@goldmedal perhaps you would like to merge this PR as a way to verify your permissions are configured correctly?

Sure, I plan to merge this PR if no more comments tomorrow.

Let's hold off on merging this for now. I think I've introduced a regression that I'd like to test more first before merging.

Okay, looks like my PR didn't introduce the regression but there's a somewhat related bug currently on main. Take the following SQL:

SELECT j1_id FROM (SELECT ta.j1_id AS j1_id FROM j1 ta);

This gets rewritten in the GenericDialect to:

SELECT ta.j1_id FROM (SELECT ta.j1_id FROM j1 AS ta)

Which is invalid, because the subquery is un-aliased. I tested it in DuckDB to confirm:

D SELECT ta.j1_id FROM (SELECT ta.j1_id FROM j1 ta);
Binder Error: Referenced table "ta" not found!
Candidate tables: "unnamed_subquery"
LINE 1: SELECT ta.j1_id FROM (SELECT ta.j1_id FROM j1 ...

Let's merge my PR, and I can raise a new issue for this other bug @goldmedal ?

goldmedal · 2024-10-21T06:08:02Z

This gets rewritten in the GenericDialect to:
SELECT ta.j1_id FROM (SELECT ta.j1_id FROM j1 AS ta)

It's valid for DataFusion

> create table j1(j1_id int);
0 row(s) fetched. 
Elapsed 0.085 seconds.

> SELECT ta.j1_id FROM (SELECT ta.j1_id FROM j1 AS ta);
+-------+
| j1_id |
+-------+
+-------+
0 row(s) fetched. 
Elapsed 0.042 seconds.

Generally, GenericDialect fits the DataFusion syntax. This behavior makes sense to me.

Which is invalid, because the subquery is un-aliased. I tested it in DuckDB to confirm:
D SELECT ta.j1_id FROM (SELECT ta.j1_id FROM j1 ta);
Binder Error: Referenced table "ta" not found!
Candidate tables: "unnamed_subquery"
LINE 1: SELECT ta.j1_id FROM (SELECT ta.j1_id FROM j1 ...
Let's merge my PR, and I can raise a new issue for this other bug @goldmedal ?

How about MySQLDialect? It enables requires_derived_table_alias. Which would be the result?

SELECT ta.j1_id FROM (SELECT ta.j1_id FROM j1 ta) derived_projection

or

SELECT j1_id FROM (SELECT ta.j1_id FROM j1 ta) `derived_projection`

If it's the first one, I think it's an issue because it isn't valid. We can file an issue for it.
If it's the second one, I think the user can customize the dialect::requires_derived_table_alias for the DuckDB syntax purpose.

peasee · 2024-10-21T06:15:25Z

If it's the first one, I think it's an issue because it isn't valid. We can file an issue for it.

My bad, I should've clarified better. It is the first one with MySqlDialect. I had just used GenericDialect as the example, but it affects both:

SELECT `ta`.`j1_id` FROM (SELECT `ta`.`j1_id` FROM `j1` AS `ta`) AS `derived_projection`

goldmedal · 2024-10-21T06:21:38Z

If it's the first one, I think it's an issue because it isn't valid. We can file an issue for it.

My bad, I should've clarified better. It is the first one with MySqlDialect. I had just used GenericDialect as the example, but it affects both:
SELECT `ta`.`j1_id` FROM (SELECT `ta`.`j1_id` FROM `j1` AS `ta`) AS `derived_projection`

I see. I think it's an issue but we can fix it in the follow-up PR. Let's merge this PR first.
Could you help to file an issue to trace it?

peasee · 2024-10-21T06:31:13Z

I see. I think it's an issue but we can fix it in the follow-up PR. Let's merge this PR first. Could you help to file an issue to trace it?

#13027

goldmedal · 2024-10-21T06:34:48Z

Thanks @peasee and @phillipleblanc @alamb for reviewing. 👍

fix: Dialect requires table alias (#46)

689b5b5

* fix: Add Dialect option for requiring table aliases * feat: Add CustomDialectBuilder for requires_table_alias * docs: Spelling * refactor: rename requires_derived_table_alias * refactor: rename requires_derived_table_alias

goldmedal reviewed Oct 18, 2024

View reviewed changes

review: Rewrite match to if, add another test case

3569311

github-actions bot added the sql SQL Planner label Oct 18, 2024

test: Update RHS expected

ec46696

peasee added 2 commits October 19, 2024 21:41

Merge remote-tracking branch 'origin/main' into feat/dialect-requires…

ed6ed6d

…-derived-table-alias

test: Update tests with more cases

433ac84

goldmedal approved these changes Oct 19, 2024

View reviewed changes

phillipleblanc approved these changes Oct 19, 2024

View reviewed changes

Merge branch 'apache:main' into feat/dialect-requires-derived-table-a…

c040175

…lias

peasee mentioned this pull request Oct 21, 2024

Table references in subqueries cause invalid references in top-level projection #13027

Open

goldmedal merged commit b42d9b8 into apache:main Oct 21, 2024
24 checks passed

alamb mentioned this pull request Oct 21, 2024

Oct 21, 2024: This week in DataFusion #13035

Closed

4 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix: Dialect requires derived table alias #12994

fix: Dialect requires derived table alias #12994

peasee commented Oct 18, 2024 •

edited

Loading

goldmedal left a comment

goldmedal Oct 18, 2024

peasee Oct 18, 2024

goldmedal Oct 19, 2024

peasee Oct 19, 2024

goldmedal commented Oct 19, 2024

goldmedal left a comment •

edited

Loading

goldmedal commented Oct 19, 2024

phillipleblanc commented Oct 19, 2024

alamb commented Oct 20, 2024

goldmedal commented Oct 20, 2024

peasee commented Oct 21, 2024

peasee commented Oct 21, 2024

goldmedal commented Oct 21, 2024

peasee commented Oct 21, 2024

goldmedal commented Oct 21, 2024

peasee commented Oct 21, 2024

goldmedal commented Oct 21, 2024

fix: Dialect requires derived table alias #12994

fix: Dialect requires derived table alias #12994

Conversation

peasee commented Oct 18, 2024 • edited Loading

Which issue does this PR close?

Rationale for this change

What changes are included in this PR?

Are these changes tested?

Are there any user-facing changes?

goldmedal left a comment

Choose a reason for hiding this comment

goldmedal Oct 18, 2024

Choose a reason for hiding this comment

peasee Oct 18, 2024

Choose a reason for hiding this comment

goldmedal Oct 19, 2024

Choose a reason for hiding this comment

peasee Oct 19, 2024

Choose a reason for hiding this comment

goldmedal commented Oct 19, 2024

goldmedal left a comment • edited Loading

Choose a reason for hiding this comment

goldmedal commented Oct 19, 2024

phillipleblanc commented Oct 19, 2024

alamb commented Oct 20, 2024

goldmedal commented Oct 20, 2024

peasee commented Oct 21, 2024

peasee commented Oct 21, 2024

goldmedal commented Oct 21, 2024

peasee commented Oct 21, 2024

goldmedal commented Oct 21, 2024

peasee commented Oct 21, 2024

goldmedal commented Oct 21, 2024

peasee commented Oct 18, 2024 •

edited

Loading

goldmedal left a comment •

edited

Loading