cleanup round 1

Author: mbutrovich

.github/workflows/pr_build_linux.yml

@@ -307,7 +307,7 @@ jobs:
               org.apache.comet.CometFuzzAggregateSuite
               org.apache.comet.CometFuzzIcebergSuite
               org.apache.comet.CometFuzzMathSuite
-              org.apache.comet.CometCodegenDispatchFuzzSuite
+              org.apache.comet.CometCodegenFuzzSuite
               org.apache.comet.DataGeneratorSuite
           - name: "shuffle"
             value: |
@@ -386,7 +386,7 @@ jobs:
               org.apache.comet.expressions.conditional.CometIfSuite
               org.apache.comet.expressions.conditional.CometCoalesceSuite
               org.apache.comet.expressions.conditional.CometCaseWhenSuite
-              org.apache.comet.CometCodegenDispatchSmokeSuite
+              org.apache.comet.CometCodegenSuite
               org.apache.comet.CometCodegenSourceSuite
               org.apache.comet.CometCodegenHOFSuite
           - name: "sql"

.github/workflows/pr_build_macos.yml

@@ -155,7 +155,7 @@ jobs:
               org.apache.comet.CometFuzzAggregateSuite
               org.apache.comet.CometFuzzIcebergSuite
               org.apache.comet.CometFuzzMathSuite
-              org.apache.comet.CometCodegenDispatchFuzzSuite
+              org.apache.comet.CometCodegenFuzzSuite
               org.apache.comet.DataGeneratorSuite
           - name: "shuffle"
             value: |
@@ -233,7 +233,7 @@ jobs:
               org.apache.comet.expressions.conditional.CometIfSuite
               org.apache.comet.expressions.conditional.CometCoalesceSuite
               org.apache.comet.expressions.conditional.CometCaseWhenSuite
-              org.apache.comet.CometCodegenDispatchSmokeSuite
+              org.apache.comet.CometCodegenSuite
               org.apache.comet.CometCodegenSourceSuite
               org.apache.comet.CometCodegenHOFSuite
           - name: "sql"

common/src/main/java/org/apache/comet/codegen/CometBatchKernel.java

@@ -28,13 +28,6 @@
  * {@code BoundReference.genCode} can call {@code this.getUTF8String(ord)} directly) and carries
  * typed input fields baked at codegen time, one per input column. Expression evaluation plus Arrow
  * read/write fuse into one method per expression tree.
- *
- * <p>Input scope: any {@code ValueVector[]}; the generated subclass casts each slot to the concrete
- * Arrow type the compile-time schema specified. Output is a generic {@code FieldVector}; the
- * generated subclass casts to the concrete type matching the bound expression's {@code dataType}.
- * Widen input support by adding vector classes to the getter switch in {@code
- * CometBatchKernelCodegen.emitTypedGetters}; widen output support by adding cases in {@code
- * CometBatchKernelCodegen.allocateOutput} and {@code emitOutputWriter}.
  */
 public abstract class CometBatchKernel extends CometInternalRow {
 
@@ -47,22 +40,22 @@ protected CometBatchKernel(Object[] references) {
   /**
    * Process one batch.
    *
-   * @param inputs Arrow input vectors; length and concrete classes must match the schema the kernel
-   *     was compiled against
+   * @param inputs Arrow input vectors; length and concrete classes match the schema the kernel was
+   *     compiled against
    * @param output Arrow output vector; caller allocates to the expression's {@code dataType}
    * @param numRows number of rows in this batch
    */
   public abstract void process(ValueVector[] inputs, FieldVector output, int numRows);
 
   /**
    * Run partition-dependent initialization. The generated subclass overrides this to execute
-   * statements collected via {@code CodegenContext.addPartitionInitializationStatement}, for
-   * example reseeding {@code Rand}'s {@code XORShiftRandom} from {@code seed + partitionIndex}.
+   * statements collected via {@code CodegenContext.addPartitionInitializationStatement}, e.g.
+   * reseeding {@code Rand}'s {@code XORShiftRandom} from {@code seed + partitionIndex}.
    * Deterministic expressions leave this as a no-op.
    *
-   * <p>The caller must invoke this before the first {@code process} call of each partition. The
-   * generated subclass is not thread-safe across concurrent {@code process} calls, so kernels are
-   * allocated per dispatcher invocation and init is run once on the fresh instance.
+   * <p>The caller invokes this before the first {@code process} call of each partition. The
+   * generated subclass is not thread-safe across concurrent {@code process} calls; the dispatcher
+   * allocates one per partition and serializes calls.
    */
   public void init(int partitionIndex) {}
 }

common/src/main/scala/org/apache/comet/codegen/CometArrayData.scala

@@ -27,23 +27,16 @@ import org.apache.spark.unsafe.types.{CalendarInterval, UTF8String}
 import org.apache.comet.shims.CometInternalRowShim
 
 /**
- * Throwing-default base for [[ArrayData]] in the Arrow-direct codegen kernel. Subclasses override
- * only the getters their element type needs (e.g. `numElements`, `isNullAt`, `getUTF8String` for
- * an `ArrayType(StringType)` input).
+ * Throwing-default `ArrayData` base for the codegen kernel. Subclasses override only the getters
+ * their element type needs.
  *
- * Consumer: `InputArray_${path}` nested classes the input emitter generates per `ArrayType` input
- * column. They back `getArray(ord)` plus the recursion for `Array<Array<...>>` and array-typed
- * map keys / struct fields.
+ * Consumer: per-column `InputArray_${path}` nested classes that back `getArray(ord)` plus the
+ * recursion for `Array<Array<...>>` and array-typed map keys / struct fields.
  *
- * `ArrayData` and [[CometInternalRow]]'s [[InternalRow]] are sibling abstract classes in Spark
- * (both extend `SpecializedGetters`, neither inherits the other), so a base aimed at one cannot
- * serve the other. The dispatch body that '''is''' shared between them lives in
- * [[CometSpecializedGettersDispatch]]. The third sibling, [[CometMapData]], backs `InputMap_*`
- * and routes `keyArray()` / `valueArray()` through `CometArrayData` instances.
- *
- * Mixes in [[CometInternalRowShim]] for the same reason `CometInternalRow` does: Spark 4.x adds
- * abstract `SpecializedGetters` methods (`getVariant`, `getGeography`, `getGeometry`) that both
- * `InternalRow` and `ArrayData` inherit; the per-profile shim provides throwing defaults.
+ * `ArrayData` and `InternalRow` are sibling abstract classes, so a base aimed at one cannot serve
+ * the other. The shared `get(ordinal, dataType)` dispatch lives in
+ * [[CometSpecializedGettersDispatch]]. Mixes in [[CometInternalRowShim]] so Spark 4.x's
+ * `getVariant` / `getGeography` / `getGeometry` get throwing defaults.
  */
 abstract class CometArrayData extends ArrayData with CometInternalRowShim {