Skip the residual eval max/min divergences on the analytics route

With the ANY-rescue fix, five of CalcitePPLEvalMaxMinFunctionIT's eight tests
now pass on the analytics route. Skip the three that still diverge via the
assumeNotAnalytics(...) registry plus matching excludeTestsMatching entries:
  - EVAL_MAX_MIN_MIXED_TYPES: mixed numeric+string operands throw 'Cannot infer
    return type for GREATEST' (the DataFusion GREATEST/LEAST can't unify
    heterogeneous operand types) — testEvalMax/MinNumericAndString.
  - EVAL_MAX_MIN_INT_WIDENING: max() over int operands reports bigint on the
    route (DataFusion widens integers to Int64) where v2/Calcite reports int —
    testEvalMaxNumeric.

Results (-Dtests.analytics.parquet_indices=true against the analytics route):
  CalcitePPLEvalMaxMinFunctionIT: 0/8 -> 5/8 pass, 3 excluded, 0 fail
v2/Calcite route unchanged: 8/8 pass.

Signed-off-by: Kai Huang <ahkcs@amazon.com>

Author: ahkcs

integ-test/build.gradle

@@ -1149,19 +1149,16 @@ task integTestRemote(type: RestIntegTestTask) {
             excludeTestsMatching '*CalcitePPLInSubqueryIT.testSubsearchMaxOut'
 
             // === Excludes: CalcitePPLEvalMaxMinFunctionIT route divergences ===
-            // Every test exercises eval max()/min() (lowered to GREATEST/LEAST), which has no
-            // return-type inference on the AE route: homogeneous operands type the result column
-            // 'undefined' (values are correct) and mixed numeric+string operands throw
-            // 'Cannot infer return type for GREATEST'. Each test also carries an in-test
-            // assumeNotAnalytics(EVAL_MAX_MIN_RETURN_TYPE) (see AnalyticsRouteLimitation).
-            excludeTestsMatching '*CalcitePPLEvalMaxMinFunctionIT.testEvalMaxNumeric'
-            excludeTestsMatching '*CalcitePPLEvalMaxMinFunctionIT.testEvalMaxString'
+            // The AnalyticsExecutionEngine ANY-rescue (this PR) fixes the homogeneous max()/min()
+            // schema types, so most of the class now passes on the AE route. Three tests still
+            // diverge; each carries an in-test assumeNotAnalytics(...) (see AnalyticsRouteLimitation).
+            //  - Mixed numeric+string operands throw 'Cannot infer return type for GREATEST' (the
+            //    DataFusion GREATEST/LEAST can't unify heterogeneous operand types).
             excludeTestsMatching '*CalcitePPLEvalMaxMinFunctionIT.testEvalMaxNumericAndString'
-            excludeTestsMatching '*CalcitePPLEvalMaxMinFunctionIT.testEvalMinNumeric'
-            excludeTestsMatching '*CalcitePPLEvalMaxMinFunctionIT.testEvalMinString'
             excludeTestsMatching '*CalcitePPLEvalMaxMinFunctionIT.testEvalMinNumericAndString'
-            excludeTestsMatching '*CalcitePPLEvalMaxMinFunctionIT.testEvalMaxIgnoresNulls'
-            excludeTestsMatching '*CalcitePPLEvalMaxMinFunctionIT.testEvalMinIgnoresNulls'
+            //  - max() over int operands reports bigint on the AE route (DataFusion widens integers
+            //    to Int64) where the v2/Calcite path reports int.
+            excludeTestsMatching '*CalcitePPLEvalMaxMinFunctionIT.testEvalMaxNumeric'
         }
     }
 

integ-test/src/test/java/org/opensearch/sql/calcite/remote/CalcitePPLEvalMaxMinFunctionIT.java

@@ -7,7 +7,8 @@
 
 import static org.opensearch.sql.legacy.TestsConstants.TEST_INDEX_DOG;
 import static org.opensearch.sql.legacy.TestsConstants.TEST_INDEX_NULL_MISSING;
-import static org.opensearch.sql.util.AnalyticsRouteLimitation.EVAL_MAX_MIN_RETURN_TYPE;
+import static org.opensearch.sql.util.AnalyticsRouteLimitation.EVAL_MAX_MIN_INT_WIDENING;
+import static org.opensearch.sql.util.AnalyticsRouteLimitation.EVAL_MAX_MIN_MIXED_TYPES;
 import static org.opensearch.sql.util.MatcherUtils.rows;
 import static org.opensearch.sql.util.MatcherUtils.schema;
 import static org.opensearch.sql.util.MatcherUtils.verifyDataRows;
@@ -29,7 +30,8 @@ public void init() throws Exception {
 
   @Test
   public void testEvalMaxNumeric() throws Exception {
-    assumeNotAnalytics(EVAL_MAX_MIN_RETURN_TYPE);
+    // max(1, 3, age) selects an int-valued result; the AE route reports it as bigint.
+    assumeNotAnalytics(EVAL_MAX_MIN_INT_WIDENING);
     JSONObject result =
         executeQuery(
             String.format(
@@ -40,7 +42,6 @@ public void testEvalMaxNumeric() throws Exception {
 
   @Test
   public void testEvalMaxString() throws Exception {
-    assumeNotAnalytics(EVAL_MAX_MIN_RETURN_TYPE);
     JSONObject result =
         executeQuery(
             String.format(
@@ -52,7 +53,8 @@ public void testEvalMaxString() throws Exception {
 
   @Test
   public void testEvalMaxNumericAndString() throws Exception {
-    assumeNotAnalytics(EVAL_MAX_MIN_RETURN_TYPE);
+    // Mixed numeric+string operands -> 'Cannot infer return type for GREATEST' on the AE route.
+    assumeNotAnalytics(EVAL_MAX_MIN_MIXED_TYPES);
     JSONObject result =
         executeQuery(
             String.format(
@@ -66,7 +68,6 @@ public void testEvalMaxNumericAndString() throws Exception {
 
   @Test
   public void testEvalMinNumeric() throws Exception {
-    assumeNotAnalytics(EVAL_MAX_MIN_RETURN_TYPE);
     JSONObject result =
         executeQuery(
             String.format(
@@ -77,7 +78,6 @@ public void testEvalMinNumeric() throws Exception {
 
   @Test
   public void testEvalMinString() throws Exception {
-    assumeNotAnalytics(EVAL_MAX_MIN_RETURN_TYPE);
     JSONObject result =
         executeQuery(
             String.format(
@@ -89,7 +89,8 @@ public void testEvalMinString() throws Exception {
 
   @Test
   public void testEvalMinNumericAndString() throws Exception {
-    assumeNotAnalytics(EVAL_MAX_MIN_RETURN_TYPE);
+    // Mixed numeric+string operands -> 'Cannot infer return type for GREATEST' on the AE route.
+    assumeNotAnalytics(EVAL_MAX_MIN_MIXED_TYPES);
     JSONObject result =
         executeQuery(
             String.format(
@@ -103,7 +104,6 @@ public void testEvalMinNumericAndString() throws Exception {
 
   @Test
   public void testEvalMaxIgnoresNulls() throws Exception {
-    assumeNotAnalytics(EVAL_MAX_MIN_RETURN_TYPE);
     JSONObject result =
         executeQuery(
             String.format(
@@ -125,7 +125,6 @@ public void testEvalMaxIgnoresNulls() throws Exception {
 
   @Test
   public void testEvalMinIgnoresNulls() throws Exception {
-    assumeNotAnalytics(EVAL_MAX_MIN_RETURN_TYPE);
     JSONObject result =
         executeQuery(
             String.format(

integ-test/src/test/java/org/opensearch/sql/util/AnalyticsRouteLimitation.java

@@ -136,17 +136,28 @@ public enum AnalyticsRouteLimitation {
           + " regardless of the cap."),
 
   /**
-   * The PPL {@code max(...)} / {@code min(...)} eval functions (lowered to {@code GREATEST} /
-   * {@code LEAST}) diverge on the analytics-engine route. With homogeneous operands the result
-   * column comes back typed {@code undefined} instead of the concrete type (the values are
-   * correct), so schema assertions fail. With mixed numeric+string operands the route throws {@code
-   * SqlValidatorException: Cannot infer return type for GREATEST}. Both stem from the route not
-   * inferring a return type for GREATEST/LEAST; the v2/Calcite path types them correctly.
+   * eval {@code max(...)} / {@code min(...)} over operands that mix numeric and string types (e.g.
+   * {@code max(14, age, 'Fred', holdersName)}) throws {@code SqlValidatorException: Cannot infer
+   * return type for GREATEST} on the analytics-engine route. The SCALAR_MAX/SCALAR_MIN UDF is
+   * converted to a DataFusion {@code GREATEST}/{@code LEAST} during Substrait conversion, and that
+   * operator can't unify heterogeneous operand types. The v2/Calcite path applies OpenSearch's
+   * mixed-type comparison (strings sort above numbers) and returns a result.
    */
-  EVAL_MAX_MIN_RETURN_TYPE(
-      "eval max()/min() (lowered to GREATEST/LEAST) has no return-type inference on the"
-          + " analytics-engine route: the result column is typed undefined for homogeneous operands"
-          + " and mixed numeric+string operands throw 'Cannot infer return type for GREATEST'.");
+  EVAL_MAX_MIN_MIXED_TYPES(
+      "eval max()/min() over mixed numeric+string operands throws 'Cannot infer return type for"
+          + " GREATEST' on the analytics-engine route: the DataFusion GREATEST/LEAST can't unify"
+          + " heterogeneous operand types."),
+
+  /**
+   * eval {@code max(...)} / {@code min(...)} over all-integer operands reports the result column as
+   * {@code bigint} on the analytics-engine route — DataFusion widens integer results to Int64 —
+   * whereas the v2/Calcite path reports {@code int} when the selected value fits in an int. The
+   * values match; only the schema's integer width differs.
+   */
+  EVAL_MAX_MIN_INT_WIDENING(
+      "eval max()/min() over integer operands reports the result column as bigint on the"
+          + " analytics-engine route (DataFusion widens integers to Int64), whereas the v2/Calcite"
+          + " path reports int.");
 
   private final String reason;