Add support for wider range of scalar subqueries

[andygrove]

Is your feature request related to a problem or challenge? Please describe what you are trying to do.

Setup

$ export DATAFUSION_OPTIMIZER_SKIP_FAILED_RULES=false
$ echo "1,2" > test.csv
$ cd datafusion-cli
$ cargo run

❯ create external table test (a int, b int) stored as csv location 'test.csv';

Test

This query works:

❯ select * from test where a = (select max(a) from test t2);
+---+---+
| a | b |
+---+---+
| 1 | 2 |
+---+---+

These non-aggregate subqueries that return a single row do not work:

❯ select * from test where a = (select distinct a from test);
Internal("Optimizer rule 'scalar_subquery_to_join' failed due to unexpected error: scalar subqueries must have a projection at /home/andy/git/apache/arrow-datafusion/datafusion/optimizer/src/scalar_subquery_to_join.rs:185\ncaused by\nError during planning: Could not coerce into Projection! at /home/andy/git/apache/arrow-datafusion/datafusion/expr/src/logical_plan/plan.rs:1121")

❯ select * from test where a = (select a from test limit 1);
Internal("Optimizer rule 'scalar_subquery_to_join' failed due to unexpected error: scalar subqueries must have a projection at /home/andy/git/apache/arrow-datafusion/datafusion/optimizer/src/scalar_subquery_to_join.rs:185\ncaused by\nError during planning: Could not coerce into Projection! at /home/andy/git/apache/arrow-datafusion/datafusion/expr/src/logical_plan/plan.rs:1121")

In both cases, the subqueries do have a projection, but they are wrapped with a Distinct or Limit node.

This query fails with a different error. In this case there is a Filter wrapping the Projection.

❯ select * from test where a = (select a from test where a is not null);
Internal("Optimizer rule 'scalar_subquery_to_join' failed due to unexpected error: scalar subqueries must aggregate a value at /home/andy/git/apache/arrow-datafusion/datafusion/optimizer/src/scalar_subquery_to_join.rs:193\ncaused by\nError during planning: Could not coerce into Aggregate! at /home/andy/git/apache/arrow-datafusion/datafusion/expr/src/logical_plan/plan.rs:1368")

Describe the solution you'd like
Support these subqueries

Describe alternatives you've considered
None

Additional context
None

[andygrove]

This is going to be harder than I realized. We need to check that scalar subqueries return a single row at some point during execution. I am not sure that we can translate queries to joins if there is the possibility of more than one row being returned.

Postgres example:

postgres=# select * from test t1 where a = (select distinct b from test where b > 7);
 a | b 
---+---
(0 rows)

postgres=# select * from test t1 where a = (select distinct b from test where b > 3);
ERROR:  more than one row returned by a subquery used as an expression