sql (fix): Resolve UNNEST earlier to prevent cyclic references

[takezoe]

Case 1

SELECT id, t.name FROM A CROSS JOIN UNNEST (name) AS t (name)

This SQL is valid but fails to resolve with the following error:

wvlet.airframe.sql.SQLError: [SyntaxError] line 1:45 name is ambiguous:
- *name:string <- A.name
- *name:?

Because the output attribute of AliasedRelation (t (name)) is included in the source attributes when resolve UNNEST (name) here:

airframe/airframe-sql/src/main/scala/wvlet/airframe/sql/analyzer/TypeResolver.scala

Lines 319 to 321 in 49d31ba

	case r: Relation =>
	r.transformUpExpressions { case x: Expression => resolveExpression(context, x, r.inputAttributes) }
	}

• r is CrossJoin
• r.inputAttributes is left.outputAttributes + right.outputAttributes
• right is AliasedRelation(Unnest)

I think output attributes of AliasedRelation shouldn't be in the scope when Unnest is resolved but I didn't come up with an easy fix for this issue.

Case 2

SELECT id, t.name FROM A CROSS JOIN UNNEST (A.name) AS t (name)

This SQL is also valid but fails to resolve with the following error:

Found unresolved expressions in:
[sql]
SELECT id, t.name FROM A CROSS JOIN UNNEST (A.name) AS t (name)
[plan]
[Project]: (id:long, name:string, name:string) => (id:long, name:?)
  - *id:long <- A.id
  - UnresolvedAttribute(t.name)
  [CrossJoin]: (id:long, name:string, name:string) => (id:long, name:string, name:string)
    - NaturalJoin
    [TableScan] default.A:  => (id:long, name:string)
    [AliasedRelation]:  => (name:string)
      - ResolvedIdentifier(Id(t))
      [Unnest]:  => (name:string)
        - *A.name:string <- A.name

Maybe this is because the resolved UNNEST column has a qualifier. This case might not be UNNEST specific.

[xerial]

I also don't have a clear idea of how we should represent Unnest. Need to check how UNNEST is handled in Trino or SparkSQL.

[takezoe]

• Spark doesn't seem to support UNNEST (It has EXPLODE function instead)
• Trino seems to create an analysis for UNNEST first. This prevents cyclic references in UNNEST expression

[codecov]

Codecov Report

Attention: 1 lines in your changes are missing coverage. Please review.

Comparison is base (09287a3) 82.71% compared to head (8a82fd8) 82.60%.
Report is 7 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3312      +/-   ##
==========================================
- Coverage   82.71%   82.60%   -0.11%     
==========================================
  Files         355      355              
  Lines       14836    14861      +25     
  Branches     2510     2537      +27     
==========================================
+ Hits        12271    12276       +5     
- Misses       2565     2585      +20     

Files	Coverage Δ
...ala/wvlet/airframe/sql/analyzer/TypeResolver.scala	93.33% <87.50%> (-0.31%)	⬇️

... and 4 files with indirect coverage changes

---

Continue to review full report in Codecov by Sentry.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 09287a3...8a82fd8. Read the comment docs.

[takezoe]

The following case cannot be resolved without this fix:

select t.id from (select id from A) as t(id)

[takezoe]

Excluding unresolved attributes is a workaround to prevent cyclic references easily.

[xerial]

Ok. Thanks for the explanation. I understand the fix