[Backport 2.19-dev] Support mvmap eval function

core/src/main/java/org/opensearch/sql/calcite/CalcitePlanContext.java

@@ -8,6 +8,7 @@
 import static org.opensearch.sql.calcite.utils.OpenSearchTypeFactory.TYPE_FACTORY;
 
 import java.sql.Connection;
+import java.util.ArrayList;
 import java.util.HashMap;
 import java.util.List;
 import java.util.Map;
@@ -61,6 +62,16 @@ public class CalcitePlanContext {
 
   @Getter public Map<String, RexLambdaRef> rexLambdaRefMap;
 
+  /**
+   * List of captured variables from outer scope for lambda functions. When a lambda body references
+   * a field that is not a lambda parameter, it gets captured and stored here. The captured
+   * variables are passed as additional arguments to the transform function.
+   */
+  @Getter private List<RexNode> capturedVariables;
+
+  /** Whether we're currently inside a lambda context. */
+  @Getter @Setter private boolean inLambdaContext = false;
+
   private CalcitePlanContext(FrameworkConfig config, SysLimit sysLimit, QueryType queryType) {
     this.config = config;
     this.sysLimit = sysLimit;
@@ -70,6 +81,24 @@ private CalcitePlanContext(FrameworkConfig config, SysLimit sysLimit, QueryType
     this.rexBuilder = new ExtendedRexBuilder(relBuilder.getRexBuilder());
     this.functionProperties = new FunctionProperties(QueryType.PPL);
     this.rexLambdaRefMap = new HashMap<>();
+    this.capturedVariables = new ArrayList<>();
+  }
+
+  /**
+   * Private constructor for creating a context that shares relBuilder with parent. Used by clone()
+   * to create lambda contexts that can resolve fields from the parent context.
+   */
+  private CalcitePlanContext(CalcitePlanContext parent) {
+    this.config = parent.config;
+    this.sysLimit = parent.sysLimit;
+    this.queryType = parent.queryType;
+    this.connection = parent.connection;
+    this.relBuilder = parent.relBuilder; // Share the same relBuilder
+    this.rexBuilder = parent.rexBuilder; // Share the same rexBuilder
+    this.functionProperties = parent.functionProperties;
+    this.rexLambdaRefMap = new HashMap<>(); // New map for lambda variables
+    this.capturedVariables = new ArrayList<>(); // New list for captured variables
+    this.inLambdaContext = true; // Mark that we're inside a lambda
   }
 
   public RexNode resolveJoinCondition(
@@ -101,8 +130,13 @@ public Optional<RexCorrelVariable> peekCorrelVar() {
     }
   }
 
+  /**
+   * Creates a clone of this context that shares the relBuilder with the parent. This allows lambda
+   * expressions to reference fields from the current row while having their own lambda variable
+   * mappings.
+   */
   public CalcitePlanContext clone() {
-    return new CalcitePlanContext(config, sysLimit, queryType);
+    return new CalcitePlanContext(this);
   }
 
   public static CalcitePlanContext create(
@@ -134,4 +168,42 @@ public static boolean isLegacyPreferred() {
   public void putRexLambdaRefMap(Map<String, RexLambdaRef> candidateMap) {
     this.rexLambdaRefMap.putAll(candidateMap);
   }
+
+  /**
+   * Captures an external variable for use inside a lambda. Returns a RexLambdaRef that references
+   * the captured variable by its index in the captured variables list. The actual RexNode value is
+   * stored in capturedVariables and will be passed as additional arguments to the transform
+   * function.
+   *
+   * @param fieldRef The RexInputRef representing the external field
+   * @param fieldName The name of the field being captured
+   * @return A RexLambdaRef that can be used inside the lambda to reference the captured value
+   */
+  public RexLambdaRef captureVariable(RexNode fieldRef, String fieldName) {
+    // Check if this variable is already captured
+    for (int i = 0; i < capturedVariables.size(); i++) {
+      if (capturedVariables.get(i).equals(fieldRef)) {
+        // Return existing reference - offset by number of lambda params (1 for array element)
+        return rexLambdaRefMap.get("__captured_" + i);
+      }
+    }
+
+    // Add to captured variables list
+    int captureIndex = capturedVariables.size();
+    capturedVariables.add(fieldRef);
+
+    // Create a lambda ref for this captured variable
+    // The index is offset by the number of lambda parameters (1 for single-param lambda)
+    // Count only actual lambda parameters, not captured variables
+    int lambdaParamCount =
+        (int)
+            rexLambdaRefMap.keySet().stream().filter(key -> !key.startsWith("__captured_")).count();
+    RexLambdaRef lambdaRef =
+        new RexLambdaRef(lambdaParamCount + captureIndex, fieldName, fieldRef.getType());
+
+    // Store it so we can find it again if the same field is referenced multiple times
+    rexLambdaRefMap.put("__captured_" + captureIndex, lambdaRef);
+
+    return lambdaRef;
+  }
 }

core/src/main/java/org/opensearch/sql/calcite/CalciteRexNodeVisitor.java

@@ -298,6 +298,20 @@ public RexNode visitLambdaFunction(LambdaFunction node, CalcitePlanContext conte
                               TYPE_FACTORY.createSqlType(SqlTypeName.ANY))))
               .collect(Collectors.toList());
       RexNode body = node.getFunction().accept(this, context);
+
+      // Add captured variables as additional lambda parameters
+      // They are stored with keys like "__captured_0", "__captured_1", etc.
+      List<RexNode> capturedVars = context.getCapturedVariables();
+      if (capturedVars != null && !capturedVars.isEmpty()) {
+        args = new ArrayList<>(args);
+        for (int i = 0; i < capturedVars.size(); i++) {
+          RexLambdaRef capturedRef = context.getRexLambdaRefMap().get("__captured_" + i);
+          if (capturedRef != null) {
+            args.add(capturedRef);
+          }
+        }
+      }
+
       RexNode lambdaNode = context.rexBuilder.makeLambdaCall(body, args);
       return lambdaNode;
     } catch (Exception e) {
@@ -391,6 +405,7 @@ public RexNode visitFunction(Function node, CalcitePlanContext context) {
       context.setInCoalesceFunction(true);
     }
 
+    List<RexNode> capturedVars = null;
     try {
       for (UnresolvedExpression arg : args) {
         if (arg instanceof LambdaFunction) {
@@ -409,6 +424,8 @@ public RexNode visitFunction(Function node, CalcitePlanContext context) {
             lambdaNode = analyze(arg, lambdaContext);
           }
           arguments.add(lambdaNode);
+          // Capture any external variables that were referenced in the lambda
+          capturedVars = lambdaContext.getCapturedVariables();
         } else {
           arguments.add(analyze(arg, context));
         }
@@ -419,6 +436,15 @@ public RexNode visitFunction(Function node, CalcitePlanContext context) {
       }
     }
 
+    // For transform/mvmap functions with captured variables, add them as additional arguments
+    if (capturedVars != null && !capturedVars.isEmpty()) {
+      if (node.getFuncName().equalsIgnoreCase("mvmap")
+          || node.getFuncName().equalsIgnoreCase("transform")) {
+        arguments = new ArrayList<>(arguments);
+        arguments.addAll(capturedVars);
+      }
+    }
+
     if ("LIKE".equalsIgnoreCase(node.getFuncName()) && arguments.size() == 2) {
       RexNode defaultCaseSensitive =
           CalcitePlanContext.isLegacyPreferred()

core/src/main/java/org/opensearch/sql/calcite/QualifiedNameResolver.java

@@ -64,12 +64,30 @@ private static RexNode resolveInNonJoinCondition(
       QualifiedName nameNode, CalcitePlanContext context) {
     log.debug("resolveInNonJoinCondition() called with nameNode={}", nameNode);
 
-    return resolveLambdaVariable(nameNode, context)
-        .or(() -> resolveFieldDirectly(nameNode, context, 1))
-        .or(() -> resolveFieldWithAlias(nameNode, context, 1))
-        .or(() -> resolveFieldWithoutAlias(nameNode, context, 1))
-        .or(() -> resolveRenamedField(nameNode, context))
-        .or(() -> resolveCorrelationField(nameNode, context))
+    // First try to resolve as lambda variable
+    Optional<RexNode> lambdaVar = resolveLambdaVariable(nameNode, context);
+    if (lambdaVar.isPresent()) {
+      return lambdaVar.get();
+    }
+
+    // Try to resolve as regular field
+    Optional<RexNode> fieldRef =
+        resolveFieldDirectly(nameNode, context, 1)
+            .or(() -> resolveFieldWithAlias(nameNode, context, 1))
+            .or(() -> resolveFieldWithoutAlias(nameNode, context, 1))
+            .or(() -> resolveRenamedField(nameNode, context));
+
+    if (fieldRef.isPresent()) {
+      // If we're in a lambda context and this is not a lambda variable,
+      // we need to capture it as an external variable
+      if (context.isInLambdaContext()) {
+        log.debug("Capturing external field {} in lambda context", nameNode);
+        return context.captureVariable(fieldRef.get(), nameNode.toString());
+      }
+      return fieldRef.get();
+    }
+
+    return resolveCorrelationField(nameNode, context)
         .or(() -> replaceWithNullLiteralInCoalesce(context))
         .orElseThrow(() -> getNotFoundException(nameNode));
   }

core/src/main/java/org/opensearch/sql/expression/function/BuiltinFunctionName.java

@@ -78,6 +78,7 @@ public enum BuiltinFunctionName {
   MVZIP(FunctionName.of("mvzip")),
   SPLIT(FunctionName.of("split")),
   MVDEDUP(FunctionName.of("mvdedup")),
+  MVMAP(FunctionName.of("mvmap")),
   FORALL(FunctionName.of("forall")),
   EXISTS(FunctionName.of("exists")),
   FILTER(FunctionName.of("filter")),

core/src/main/java/org/opensearch/sql/expression/function/CollectionUDF/TransformFunctionImpl.java

@@ -73,6 +73,12 @@ public static Object eval(Object... args) {
     List<Object> target = (List<Object>) args[0];
     List<Object> results = new ArrayList<>();
     SqlTypeName returnType = (SqlTypeName) args[args.length - 1];
+
+    // Check for captured variables: args structure is [array, lambda, captured1, captured2, ...,
+    // returnType]
+    // If there are more than 3 args (array, lambda, returnType), we have captured variables
+    boolean hasCapturedVars = args.length > 3;
+
     if (args[1] instanceof org.apache.calcite.linq4j.function.Function1) {
       org.apache.calcite.linq4j.function.Function1 lambdaFunction =
           (org.apache.calcite.linq4j.function.Function1) args[1];
@@ -90,9 +96,27 @@ public static Object eval(Object... args) {
       org.apache.calcite.linq4j.function.Function2 lambdaFunction =
           (org.apache.calcite.linq4j.function.Function2) args[1];
       try {
-        for (int i = 0; i < target.size(); i++) {
-          results.add(
-              transferLambdaOutputToTargetType(lambdaFunction.apply(target.get(i), i), returnType));
+        if (hasCapturedVars) {
+          // Lambda has captured variables - pass the first captured variable as second arg
+          // LIMITATION: Currently only the first captured variable (args[2]) is supported.
+          // Supporting multiple captured variables would require either:
+          // 1. Packing args[2..args.length-1] into an Object[] and modifying lambda generation
+          //    to accept a container as the second parameter, or
+          // 2. Using higher-arity function interfaces (Function3, Function4, etc.)
+          // For now, lambdas that capture multiple external variables may not work correctly.
+          Object capturedVar = args[2];
+          for (Object candidate : target) {
+            results.add(
+                transferLambdaOutputToTargetType(
+                    lambdaFunction.apply(candidate, capturedVar), returnType));
+          }
+        } else {
+          // Original behavior: lambda with index (x, i) -> expr
+          for (int i = 0; i < target.size(); i++) {
+            results.add(
+                transferLambdaOutputToTargetType(
+                    lambdaFunction.apply(target.get(i), i), returnType));
+          }
         }
       } catch (Exception e) {
         throw new RuntimeException(e);

core/src/main/java/org/opensearch/sql/expression/function/PPLFuncImpTable.java

@@ -154,6 +154,7 @@
 import static org.opensearch.sql.expression.function.BuiltinFunctionName.MVDEDUP;
 import static org.opensearch.sql.expression.function.BuiltinFunctionName.MVINDEX;
 import static org.opensearch.sql.expression.function.BuiltinFunctionName.MVJOIN;
+import static org.opensearch.sql.expression.function.BuiltinFunctionName.MVMAP;
 import static org.opensearch.sql.expression.function.BuiltinFunctionName.MVZIP;
 import static org.opensearch.sql.expression.function.BuiltinFunctionName.NOT;
 import static org.opensearch.sql.expression.function.BuiltinFunctionName.NOTEQUAL;
@@ -1044,6 +1045,7 @@ void populate() {
       registerOperator(MVAPPEND, PPLBuiltinOperators.MVAPPEND);
       registerOperator(MVDEDUP, SqlLibraryOperators.ARRAY_DISTINCT);
       registerOperator(MVZIP, PPLBuiltinOperators.MVZIP);
+      registerOperator(MVMAP, PPLBuiltinOperators.TRANSFORM);
       registerOperator(MAP_APPEND, PPLBuiltinOperators.MAP_APPEND);
       registerOperator(MAP_CONCAT, SqlLibraryOperators.MAP_CONCAT);
       registerOperator(MAP_REMOVE, PPLBuiltinOperators.MAP_REMOVE);

docs/user/ppl/functions/collection.md

@@ -725,6 +725,74 @@ fetched rows / total rows = 1/1
 +--------------------------+
 ```
 
+## MVMAP
+
+### Description
+
+Usage: mvmap(array, expression) iterates over each element of a multivalue array, applies the expression to each element, and returns a multivalue array with the transformed results. The field name in the expression is implicitly bound to each element value.
+Argument type: array: ARRAY, expression: EXPRESSION
+Return type: ARRAY
+Example
+
+```ppl
+source=people
+| eval array = array(1, 2, 3), result = mvmap(array, array * 10)
+| fields result
+| head 1
+```
+
+Expected output:
+
+```text
+fetched rows / total rows = 1/1
++------------+
+| result     |
+|------------|
+| [10,20,30] |
++------------+
+```
+
+```ppl
+source=people
+| eval array = array(1, 2, 3), result = mvmap(array, array + 5)
+| fields result
+| head 1
+```
+
+Expected output:
+
+```text
+fetched rows / total rows = 1/1
++---------+
+| result  |
+|---------|
+| [6,7,8] |
++---------+
+```
+
+Note: For nested expressions like ``mvmap(mvindex(arr, 1, 3), arr * 2)``, the field name (``arr``) is extracted from the first argument and must match the field referenced in the expression.
+
+The expression can also reference other single-value fields:
+
+```ppl
+source=people
+| eval array = array(1, 2, 3), multiplier = 10, result = mvmap(array, array * multiplier)
+| fields result
+| head 1
+```
+
+Expected output:
+
+```text
+fetched rows / total rows = 1/1
++------------+
+| result     |
+|------------|
+| [10,20,30] |
++------------+
+```
+
+
 ## MVZIP
 
 ### Description