[SPARK-59600][SQL] Bind the resolved function owner across star preprocessing and function resolution

[ganeshashree]

What changes were proposed in this pull request?

Direct-star preprocessing (ResolveReferences.expandStarExpression in the fixed-point analyzer and FunctionResolverUtils.handleStarInArguments in the single-pass resolver) and the later ResolveFunctions each resolved the owner of a routed SQL/JSON call (json_array, json_value, json_query, json_exists) independently against the SQL PATH.

This resolves the owner once and binds it:

• Replace the Boolean FunctionResolution.resolvesToStarDisallowedSqlJsonFunction with selectRoutedSqlJsonDirectStarOwner, returning RoutedSqlJsonStarOwner (RejectStockBuiltin / BindShadowOwner(candidate) / NoBinding).
• When a shadow owns the call, expand its star and record the winning SQL PATH candidate on the new UnresolvedFunction.boundOwner (a relative persistent-catalog candidate stays relative until later resolution qualifies it); resolveFunction then resolves only that candidate, failing with UNRESOLVED_ROUTINE if it disappeared, instead of falling back to the stock built-in.
• Stock-builtin rejection, count(*) / count(tbl.*) handling, and injectFunction-replacement semantics are unchanged. The check is shared by both analyzer strategies.

Why are the changes needed?

If a visible temporary or persistent shadow is dropped between the two phases, preprocessing expands the direct star for the shadow while resolution then falls through to the stock built-in, which no longer sees a Star and so skips INVALID_USAGE_OF_STAR_OR_REGEX.

Does this PR introduce any user-facing change?

No, beyond fixing the above analysis inconsistency, which is only reachable under concurrent DDL between a single query's analysis phases.

How was this patch tested?

New deterministic interleaving test in JsonArraySuite, run for both the fixed-point and single-pass analyzers, using a catalog that reports the shadow as existing during preprocessing but fails to load it at resolution. Existing routed-function shadowing, star-rejection, count, persistentCatalogFirst, temp-table-terminal, injected-function, and view-frozen-catalog tests continue to pass.

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

Generated-by: Claude Code (Opus 4.8)

[cloud-fan]

Review summary

The selected-owner design and deterministic two-mode regression coverage look sound overall. I found two non-blocking cleanup issues: the new carrier documentation overstates catalog qualification, and the routed owner check now materializes the full PATH before it can short-circuit.

Findings

2 total: 0 P0, 0 P1, 0 P2, 2 P3.

Nit (P3)

• Describe boundOwner as the selected PATH candidate — sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/unresolved.scala:398 — see inline.
• Build resolution candidates lazily for early-return owner checks — sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/FunctionResolution.scala:507 — see inline.

PR description suggestions

• Replace the PR description's claim that boundOwner records a fully-qualified candidate with wording that it records the selected SQL PATH candidate, since the persistent-first two-part form can remain relative until later resolution.

[cloud-fan]

Review summary

The selected-owner implementation and its two-analyzer regression coverage look sound. The two P3 concerns from the prior review are resolved: boundOwner is now documented as the selected SQL PATH candidate with later qualification made explicit, and resolutionCandidates now uses LazyList so early owner walks do not materialize unused PATH suffixes. I found no new issues in this revision.

Findings

0 total: 0 P0, 0 P1, 0 P2, 0 P3.

No findings.

Re-review status

Prior AI findings: 2 addressed, 0 still present; additional unresolved findings in this review: 0.

New attribution: 0 newly introduced, 0 late catch, 0 previously raised, 0 unattributed.

Remaining prior AI findings

No prior AI findings remain.

[cloud-fan]

Merge Summary:

• merged into master 2247cb2
• merged into branch-4.x c7267a2

Posted by merge_spark_pr.py