Remove old Spark 3.3.0 shim sources

[gerashegalov]

Related to #14834.

Description

This PR is one reviewable layer in the unshim stack introduced by #15025. It removes old Spark 3.3.0 shim sources that are now provided by shared helpers or the new helper modules. This is scoped to the OSS Spark 3.3.0 cleanup.

Stack context

• Stack introduction: Add default common unshim packaging flow #15025
• Previous layer: Move common shim helpers to shared sources #15052
• Next layer: Remove old Spark 3.3 DBR shim sources #15053

Testing and validation notes

• No standalone behavior change is intended in this layer. It is covered by the full-stack packaging/build validation described in Add default common unshim packaging flow #15025 and the existing tests for the affected subsystem.
• The full split stack was verified to be tree-equivalent to the pre-split stack top.

Checklists

Documentation

• Updated for new or modified user-facing features or behaviors
• No user-facing change

Testing

• Added or modified tests to cover new code paths
• Covered by existing tests
  (Covered by the validation notes in the PR description.)
• Not required

Performance

• Tests ran and results are added in the PR description
• Issue filed with a link in the PR description
• Not required

[greptile-apps]

Greptile Summary

This PR removes Spark 3.3.0-specific shim sources that are now covered by shared helper modules introduced earlier in the unshim stack. It also makes several backward-compatible adjustments to the remaining spark330 files to enable binary-class deduplication with newer shims.

• Deletions (11 files): DateTimeUtilsShims, SequenceSizeTooLongErrorBuilder, SparkShimServiceProvider, ConvUtils, ShuffleManagerShims, FileCommitProtocolShims, OriginContextShim, SparkSessionUtils, SparkUpgradeExceptionShims, TrampolineConnectShims, and ShuffleClientShims are removed because they are now provided by shared helper modules in the stack.
• Binary-dedupe alignment: Remaining spark330 files adopt intentional line-number padding, string-literal enum comparisons in LegacyBehaviorPolicyShim, and inline SLF4J loggers (replacing with Logging mixins) so compiled bytecode can be recognized as identical to the common-module class.
• Reflection-based ORC protobuf dispatch (OrcProtoWriterShim): Replaced direct CodedOutputStream use with a lazy, reflection-driven protoApis registry that discovers both org.apache.orc.protobuf and com.google.protobuf at runtime, allowing the same compiled class to work across both protobuf variants.

Confidence Score: 4/5

Structured cleanup of spark330-specific shim sources; no new end-user logic is introduced, and the reflection-based ORC dispatch and Logging-to-SLF4J conversions are technically correct.

The reflection-based OrcProtoWriterShim and the GpuGroupedPythonRunnerFactory API surface changes are the two areas that deserve a second look. The reflection dispatch is sound — protoApis is lazily populated once and the per-instance cache is always flushed before reassignment — but the single-slot cache design is worth documenting. The removal of the case class companion apply and the argNames default value are intentional within the validated stack but are silent source-compatibility breaks for any downstream consumer.

OrcProtoWriterShim.scala (reflection-based protobuf dispatch) and GpuGroupedPythonRunnerFactory.scala (case class to class, default arg removal) warrant the closest review; deleted files and Logging-to-SLF4J conversions are straightforward.

Important Files Changed

Filename	Overview
sql-plugin/src/main/spark330/scala/com/nvidia/spark/rapids/shims/OrcProtoWriterShim.scala	Replaced direct protobuf CodedOutputStream usage with a reflection-based approach to support both org.apache.orc.protobuf and com.google.protobuf variants; adds lazy protoApis discovery and per-instance API caching in proxiedFor.
sql-plugin/src/main/spark330/scala/com/nvidia/spark/rapids/shims/NullIntolerantShim.scala	Extended NullIntolerantShim with nullIntolerant: Boolean and added a new GpuLiteralShim abstract class with a jsonFields implementation covering Date/Timestamp serialization.
sql-plugin/src/main/spark330/scala/com/nvidia/spark/rapids/shuffle/RapidsShuffleIterator.scala	Dropped Spark Logging mixin; replaced with explicit SLF4J logger and private log helpers for binary-dedupe; also fixes GpuSemaphore.acquireIfNecessary explicit eta-expansion.
sql-plugin/src/main/spark330/scala/org/apache/spark/sql/rapids/RapidsCachingReader.scala	Dropped Spark Logging mixin; replaced with inline SLF4J logger with logInfo/logDebug helpers only, consistent with actual call sites in this class.
sql-plugin/src/main/spark330/scala/org/apache/spark/sql/rapids/execution/python/shims/GpuGroupedPythonRunnerFactory.scala	Changed from case class to class extends Serializable and removed the default argNames = None; removes auto-generated companion apply so callers must use new or be updated in the stack.
sql-plugin/src/main/spark330/scala/com/nvidia/spark/rapids/shims/SparkShims.scala	Updated getDataWriteCmds to use new GpuOverrides.dataWriteCmdFromShim + CreateDataSourceTableAsSelectRules.dataWriteCmd helper pattern instead of the inline dataWriteCmd call.
sql-plugin/src/main/scala/com/nvidia/spark/rapids/shims/Spark320PlusShims.scala	Removed unused with Logging mixin and its import; no method bodies in this trait directly used the Logging helpers.
sql-plugin/src/main/scala/com/nvidia/spark/rapids/shims/gpuWindows.scala	Changed ParsedBoundary apply to new ParsedBoundary constructor call since the case class companion no longer exists in the newer shim.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[writeAndFlush called] --> B{apiFor obj}
    B -- None --> C[throw IllegalArgumentException]
    B -- Some api --> D{proxiedApi == api?}
    D -- Yes --> E[reuse cached proxied]
    D -- No --> F[invoke api.newInstance with orcOutStream]
    F --> G[update proxiedApi + proxied]
    G --> E
    E --> H[api.writeTo.invoke obj proxied]
    H --> I[api.flush.invoke proxied]
    I --> J[orcOutStream.flush]

    subgraph protoApis [lazy protoApis init]
        K[try org.apache.orc.protobuf] --> L{found?}
        L -- Yes --> M[add ProtoApi]
        L -- No --> N[skip]
        O[try com.google.protobuf] --> P{found?}
        P -- Yes --> Q[add ProtoApi]
        P -- No --> R[skip]
    end

    B -.->|first call triggers| protoApis

Loading

Comments Outside Diff (1)

1. sql-plugin/src/main/spark330/scala/com/nvidia/spark/rapids/shims/OrcProtoWriterShim.scala, line 230-239 (link)

   Per-instance API cache resets on type switch

   proxiedFor caches only the most-recently-used ProtoApi per OrcProtoWriterShim instance. If writeAndFlush is called alternately with messages from two different protobuf variants (one org.apache.orc.protobuf, one com.google.protobuf), a new CodedOutputStream wrapping orcOutStream is allocated on every call. The old stream is already flushed before reassignment, so no data is lost, but in a mixed-variant scenario the cache never hits. In practice ORC writers use a single protobuf variant per JVM, so this is unlikely to matter, but the single-slot, last-used nature of the cache is worth documenting.

   Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

_{Reviews (1): Last reviewed commit: "Remove old Spark 3.3.0 shim sources" | Re-trigger Greptile}

[greptile-apps]

Removal of case class companion apply and default argNames

Converting from case class to class removes the auto-generated companion apply factory method. Any call site using GpuGroupedPythonRunnerFactory(args...) without new will no longer compile. Removing = None from argNames also means every call site must now supply the argument explicitly. Both are intentional within the validated stack, but represent silent source-compatibility breaks for any downstream or out-of-tree consumer.