test: rename CometCastSuite to CometNativeCastSuite

.github/workflows/pr_build_linux.yml

@@ -381,7 +381,7 @@ jobs:
               org.apache.comet.CometHashExpressionSuite
               org.apache.comet.CometTemporalExpressionSuite
               org.apache.comet.CometArrayExpressionSuite
-              org.apache.comet.CometCastSuite
+              org.apache.comet.CometNativeCastSuite
               org.apache.comet.CometDateTimeUtilsSuite
               org.apache.comet.CometMathExpressionSuite
               org.apache.comet.CometStringExpressionSuite

.github/workflows/pr_build_macos.yml

@@ -197,7 +197,7 @@ jobs:
               org.apache.comet.CometHashExpressionSuite
               org.apache.comet.CometTemporalExpressionSuite
               org.apache.comet.CometArrayExpressionSuite
-              org.apache.comet.CometCastSuite
+              org.apache.comet.CometNativeCastSuite
               org.apache.comet.CometDateTimeUtilsSuite
               org.apache.comet.CometMathExpressionSuite
               org.apache.comet.CometStringExpressionSuite

docs/source/contributor-guide/development.md

@@ -269,10 +269,10 @@ and `-Dtest=none`:
 ./mvnw test -Dtest=none -Dsuites="org.apache.comet.CometArrayExpressionSuite"
 
 # Run multiple suites (comma-separated, fully qualified)
-./mvnw test -Dtest=none -Dsuites="org.apache.comet.CometCastSuite,org.apache.comet.CometArrayExpressionSuite"
+./mvnw test -Dtest=none -Dsuites="org.apache.comet.CometNativeCastSuite,org.apache.comet.CometArrayExpressionSuite"
 
 # Run only tests whose name contains "valid" inside one suite
-./mvnw test -Dtest=none -Dsuites="org.apache.comet.CometCastSuite valid"
+./mvnw test -Dtest=none -Dsuites="org.apache.comet.CometNativeCastSuite valid"
 ```
 
 `-Dtest=none` tells the Surefire (JUnit) plugin to skip its tests; without it, Surefire still
@@ -474,10 +474,10 @@ However if the tests is related to the native side. Please make sure to run `mak
 
 Specify which ScalaTest suites to run with the `suites` argument and disable Surefire's JUnit
 discovery with `-Dtest=none`. For example, to run only the test cases containing _valid_ in
-their name from `org.apache.comet.CometCastSuite`:
+their name from `org.apache.comet.CometNativeCastSuite`:
 
 ```sh
-./mvnw test -Dtest=none -Dsuites="org.apache.comet.CometCastSuite valid"
+./mvnw test -Dtest=none -Dsuites="org.apache.comet.CometNativeCastSuite valid"
 ```
 
 See [Running a Specific ScalaTest Suite](#running-a-specific-scalatest-suite) above for the full

docs/source/contributor-guide/spark_configs_support.md

@@ -104,7 +104,7 @@ across Spark 3.4, 3.5, 4.0, and 4.1. Three expression classes mix in
   `to_date`, and Spark 4's `try_to_timestamp`)
 
 `Cast` between strings and date / timestamp also reads the policy via the default
-formatters but is tested separately by `CometCastSuite` and is out of scope here.
+formatters but is tested separately by `CometNativeCastSuite` and is out of scope here.
 
 **Comet status.** None of the native implementations consult
 `legacyTimeParserPolicy` directly. Comet instead uses native implementations only for

native/spark-expr/src/conversion_funcs/string.rs

@@ -2201,7 +2201,7 @@ mod tests {
     /// The codepoint matrix from
     /// <https://github.com/apache/datafusion-comet/issues/5149>: the ASCII control bytes and
     /// DELETE, plus the non-ASCII codepoints that are whitespace to Unicode but that no Spark
-    /// cast trims. `CometCastSuite` runs the same matrix with Spark itself as the oracle.
+    /// cast trims. `CometNativeCastSuite` runs the same matrix with Spark itself as the oracle.
     fn trim_pads() -> Vec<String> {
         let mut pads: Vec<String> = (0x00u8..=0x20).map(|b| String::from(b as char)).collect();
         pads.push("\u{7f}".to_string());
@@ -2305,7 +2305,7 @@ mod tests {
     /// than change behaviour silently. `timestamp_parser` and `timestamp_ntz_parser` still use
     /// `str::trim`, so they accept the non-ASCII whitespace that Spark's
     /// `SparkDateTimeUtils.getTrimmedStart` / `getTrimmedEnd` leave in place, where Spark returns
-    /// NULL. `CometCastSuite` cannot cover this, because Spark is the oracle there and Comet does
+    /// NULL. `CometNativeCastSuite` cannot cover this, because Spark is the oracle there and Comet does
     /// not fall back -- it silently returns a value.
     #[test]
     fn test_cast_string_to_timestamp_unicode_whitespace_divergence() {

spark/src/main/scala/org/apache/comet/expressions/CometCast.scala

@@ -37,7 +37,7 @@ object CometCast
     with CometTypeShim
     with CodegenDispatchFallback {
 
-  // Shared with CometCastSuite so the asserted reason cannot drift from production.
+  // Shared with CometNativeCastSuite so the asserted reason cannot drift from production.
   private[comet] val negativeScaleDecimalToStringReason: String =
     "Negative-scale decimal requires spark.sql.legacy.allowNegativeScaleOfDecimal=true"
 

spark/src/test/resources/sql-tests/expressions/cast/cast_string_to_date_ansi.sql

@@ -19,7 +19,7 @@
 -- shape of a date but names no real calendar date, rather than returning NULL.
 -- See https://github.com/apache/datafusion-comet/issues/5012.
 --
--- Per-string parity with Spark is covered by CometCastSuite "cast StringType to DateType";
+-- Per-string parity with Spark is covered by CometNativeCastSuite "cast StringType to DateType";
 -- this fixture covers the plan shapes that suite does not exercise.
 
 -- Config: spark.sql.ansi.enabled=true

spark/src/test/resources/sql-tests/expressions/cast/cast_string_trim.sql

@@ -22,7 +22,7 @@
 -- and NBSP / U+3000 are trimmed for nothing. See conversion_funcs::trim in the native crate and
 -- https://github.com/apache/datafusion-comet/issues/5149.
 --
--- Per-codepoint parity across all three eval modes is covered by the CometCastSuite
+-- Per-codepoint parity across all three eval modes is covered by the CometNativeCastSuite
 -- "whitespace trim parity" tests; this fixture covers the same regimes over a Parquet column.
 
 statement

spark/src/test/scala/org/apache/comet/CometCastSuite.scala → spark/src/test/scala/org/apache/comet/CometNativeCastSuite.scala

@@ -40,7 +40,28 @@ import org.apache.comet.expressions.{CometCast, CometEvalMode}
 import org.apache.comet.rules.CometScanTypeChecker
 import org.apache.comet.serde.{Compatible, Incompatible, Unsupported}
 
-class CometCastSuite extends CometTestBase with AdaptiveSparkPlanHelper {
+/**
+ * Spark-parity coverage for Comet's **native** `Cast` implementation, and for the
+ * `CometCast.isSupported` matrix that decides which casts reach it.
+ *
+ * Scope: for every `(from, to)` type pair that `CometCast` reports as `Compatible` or
+ * `Incompatible`, run the cast against Spark as the oracle and assert both that the results match
+ * and — via `checkSparkAnswerAndOperator` — that the native kernel actually executed. Support
+ * decisions themselves (`Compatible` / `Incompatible` / `Unsupported`, and the reason strings)
+ * are asserted directly against `CometCast.isSupported`.
+ *
+ * Out of scope: the JVM codegen dispatch path. `CometCast` mixes in `CodegenDispatchFallback`, so
+ * a cast that `isSupported` rejects still runs inside the Comet operator by way of the
+ * Arrow-direct codegen dispatcher, evaluating Spark's own `Cast` against Arrow vectors. That path
+ * is covered by `CometCodegenSuite` and friends. Where this suite touches it at all, it does so
+ * only to pin that an `Unsupported` cast keeps the enclosing operator native rather than falling
+ * back to Spark; it does not attempt to cover dispatch semantics.
+ *
+ * Because of that split, a cast marked `Unsupported` here is not untested overall — it is tested
+ * through the dispatcher instead. Adding a native cast implementation therefore means moving a
+ * pair out of the `Unsupported` assertions and into the parity matrix below.
+ */
+class CometNativeCastSuite extends CometTestBase with AdaptiveSparkPlanHelper {
 
   import testImplicits._
 
@@ -1853,6 +1874,9 @@ class CometCastSuite extends CometTestBase with AdaptiveSparkPlanHelper {
   }
 
   test("cast ArrayType(DateType) to unsupported ArrayType routes through codegen dispatch") {
+    // Boundary case rather than dispatch coverage: these pairs have no native cast, so all this
+    // asserts is that `Unsupported` keeps the operator native via `CodegenDispatchFallback`
+    // instead of falling back to Spark. Dispatch semantics belong in the codegen suites.
     val fromType = ArrayType(DateType)
     val unsupportedElementTypes =
       Seq(BooleanType, ByteType, ShortType, LongType, FloatType, DoubleType, DecimalType(10, 2))