Remove inaccessible and unused AccessibleArrowColumnVector

[WilliamK112]

Summary

Fixes #11863

AccessibleArrowColumnVector was originally added (in #1622) for a customer who planned to write their own DataSource V2. That customer is not using it, and as noted in the issue, the class is not actually accessible at runtime due to the shimming / parallel-world packaging. The branches that reference it are therefore dead code.

Changes

• Delete sql-plugin/src/main/java/org/apache/spark/sql/vectorized/rapids/AccessibleArrowColumnVector.java.
• Remove its references in HostColumnarToGpu.scala:
  • the import
  • the case av: AccessibleArrowColumnVector branch in arrowColumnarCopy (the case _ already throws for any other vector type)
  • the isInstanceOf[AccessibleArrowColumnVector] check when selecting the Arrow batch builder

Testing

• No remaining references to AccessibleArrowColumnVector in the repo (verified via grep).
• CI to validate compile/build (local environment does not have the full Scala/Maven build set up).

Made with Cursor

[greptile-apps]

Greptile Summary

This PR removes the AccessibleArrowColumnVector Java class and all its call sites in HostColumnarToGpu.scala. The class was originally added for a customer-specific DataSource V2 use case but was never reachable at runtime due to spark-rapids' parallel-world shim packaging.

• Deletes AccessibleArrowColumnVector.java (507 lines), a dead subclass of Spark's ColumnVector that provided public access to the underlying Arrow ValueVector.
• Removes the case av: AccessibleArrowColumnVector branch from arrowColumnarCopy and the isInstanceOf[AccessibleArrowColumnVector] guard in initNewBatch, leaving the standard ArrowColumnVector path and the existing case _ error throw intact.

Confidence Score: 5/5

Safe to merge — removes dead code that was provably unreachable at runtime; no functional paths are altered.

Both changed files make purely subtractive changes to dead-code paths. The case _ error branch in arrowColumnarCopy already covered any non-ArrowColumnVector input, and a grep of the full repo confirms zero remaining references to AccessibleArrowColumnVector. The remaining Arrow and non-Arrow copy paths in HostColumnarToGpu are unchanged.

Files Needing Attention: No files require special attention.

Important Files Changed

Filename	Overview
sql-plugin/src/main/java/org/apache/spark/sql/vectorized/rapids/AccessibleArrowColumnVector.java	Entire file deleted — 507-line dead-code class that was inaccessible at runtime due to shim packaging; no callers remain in the codebase.
sql-plugin/src/main/scala/com/nvidia/spark/rapids/HostColumnarToGpu.scala	Removed import, dead match branch, and isInstanceOf check for AccessibleArrowColumnVector; remaining ArrowColumnVector path and error fallback are unaffected.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[HostToGpuCoalesceIterator.initNewBatch] -->|useArrowCopyOpt && numCols > 0 && arrowTypesSupported| B{column 0 type?}
    B -->|ArrowColumnVector| C[GpuArrowColumnarBatchBuilder]
    B -->|anything else| D[GpuColumnarBatchBuilder]

    E[arrowColumnarCopy] -->|cv match| F{ColumnVector type?}
    F -->|ArrowColumnVector| G[getArrowValueVector via reflection]
    F -->|anything else| H[throw IllegalStateException]

    style B fill:#d4edda,stroke:#28a745
    style F fill:#d4edda,stroke:#28a745

Loading

_{Reviews (3): Last reviewed commit: "Merge branch 'main' into remove-unused-a..." | Re-trigger Greptile}

[nvauto]

NOTE: release/26.08 has been created from main. Please retarget your PR to release/26.08 if it should be included in the release.

[sdrp713]

LGTM

[sdrp713]

build