[SPARK-58252][SQL] Throw GROUP_BY_POS_OUT_OF_RANGE for out-of-range ordinal in pipe AGGREGATE GROUP BY

[LuciferYang]

What changes were proposed in this pull request?

An out-of-range GROUP BY ordinal in the SQL pipe AGGREGATE operator threw a raw java.lang.IndexOutOfBoundsException instead of a proper AnalysisException. For example:

select 3 as x, 4 as y
|> aggregate sum(y) group by 5;

resolvePipeAggregateExpressionOrdinal resolved the ordinal with an unguarded inputs(index - 1), so an out-of-range position (or 0) surfaced the internal JVM exception.

This PR adds the same bounds guard the regular GROUP BY ordinal resolver (resolveGroupByExpressionOrdinal) already uses, throwing GROUP_BY_POS_OUT_OF_RANGE (sqlState 42805). The UnresolvedPipeAggregateOrdinal node is now created with the ordinal literal's origin (via CurrentOrigin.withOrigin), so the error's queryContext points at the offending ordinal rather than the whole pipe statement, matching regular GROUP BY.

Why are the changes needed?

A raw IndexOutOfBoundsException escaping the analyzer for valid-syntax SQL is an internal error leak. User-facing errors should carry an error class, sqlState, and query context. The pipe AGGREGATE path was the only pipe operator with its own GROUP BY ordinal handling and the only one missing this guard; regular GROUP BY <ordinal> has thrown GROUP_BY_POS_OUT_OF_RANGE with precise context for years.

Does this PR introduce any user-facing change?

Yes, on the error path only. For an out-of-range or non-positive ordinal in pipe AGGREGATE GROUP BY:

Before:

java.lang.IndexOutOfBoundsException: Index 4 out of bounds for length 2

After:

[GROUP_BY_POS_OUT_OF_RANGE] GROUP BY position 5 is not in select list (valid range is [1, 2]). SQLSTATE: 42805

with a query context pointing at the ordinal. Valid (in-range) ordinals are unaffected.

How was this patch tested?

Added positive-range-adjacent negative cases to pipe-operators.sql (group by 5 above range, group by 0 non-positive) in the aggregation negative-tests section; regenerated the golden files. The regenerated goldens assert the error class, sqlState, index/size parameters, and the precise queryContext (fragment pointing at the ordinal). Existing positive pipe-ordinal cases (including the exact upper bound group by x, 2, 3) regenerated with no diff, confirming zero happy-path impact.

Was this patch authored or co-authored using generative AI tooling?

Generated-by: Claude Code (Opus 4.8)

[LuciferYang]

cc @cloud-fan @dtenedor

[uros-b]

LGTM, thank you @LuciferYang!

[dongjoon-hyun]

+1, LGTM.

This is much better although the new error message could be a little confusing in a way. Since it's inevitable when we reuse the existing error class, I agree with this PR.

[dongjoon-hyun]

Merge Summary:

• merged into master 4462cb1
• merged into branch-4.x 9d4cb16

Posted by merge_spark_pr.py

[dongjoon-hyun]

Merged for Apache Spark 4.3.0 whose Feature Freeze is scheduled on 1st August.

[LuciferYang]

Thank you @dongjoon-hyun and @uros-b