feat(isthmus): configurable fallback to dynamic function mapping

isthmus/build.gradle.kts

@@ -105,6 +105,7 @@ dependencies {
   testImplementation(platform(libs.junit.bom))
   testImplementation(libs.junit.jupiter)
   testRuntimeOnly(libs.junit.platform.launcher)
+  testRuntimeOnly(libs.slf4j.jdk14)
   implementation(libs.guava)
   implementation(libs.protobuf.java.util) {
     exclude("com.google.guava", "guava")

isthmus/src/main/java/io/substrait/isthmus/FeatureBoard.java

@@ -32,4 +32,22 @@ public Casing unquotedCasing() {
   public boolean allowDynamicUdfs() {
     return false;
   }
+
+  /**
+   * Controls whether to automatically create mappings for all unmapped functions using
+   * SimpleExtensionToSqlOperator.
+   *
+   * <p>When enabled, functions from extension YAML files that are not explicitly mapped in
+   * FunctionMappings will be automatically mapped to Calcite SqlOperators. This allows custom and
+   * dynamic functions to be used in SQL queries without manual mapping configuration.
+   *
+   * <p>This feature is disabled by default for backward compatibility.
+   *
+   * @return true if automatic fallback to dynamic function mapping should be enabled; false
+   *     otherwise (default)
+   */
+  @Value.Default
+  public boolean autoFallbackToDynamicFunctionMapping() {
+    return false;
+  }
 }

isthmus/src/main/java/io/substrait/isthmus/SimpleExtensionToSqlOperator.java

@@ -45,9 +45,45 @@ public static List<SqlOperator> from(
       SimpleExtension.ExtensionCollection collection,
       RelDataTypeFactory typeFactory,
       TypeConverter typeConverter) {
-    // TODO: add support for windows functions
     return Stream.concat(
-            collection.scalarFunctions().stream(), collection.aggregateFunctions().stream())
+            Stream.concat(
+                collection.scalarFunctions().stream(), collection.aggregateFunctions().stream()),
+            collection.windowFunctions().stream())
+        .map(function -> toSqlFunction(function, typeFactory, typeConverter))
+        .collect(Collectors.toList());
+  }
+
+  /**
+   * Converts a list of functions to SqlOperators. Handles scalar, aggregate, and window functions.
+   *
+   * @param functions list of functions to convert
+   * @param typeFactory the Calcite type factory
+   * @return list of SqlOperators
+   */
+  public static List<SqlOperator> from(
+      List<? extends SimpleExtension.Function> functions, RelDataTypeFactory typeFactory) {
+    return from(functions, typeFactory, TypeConverter.DEFAULT);
+  }
+
+  /**
+   * Converts a list of functions to SqlOperators. Handles scalar, aggregate, and window functions.
+   *
+   * <p>Each function variant is converted to a separate SqlOperator. Functions with the same base
+   * name but different type signatures (e.g., strftime:ts_str, strftime:ts_string) are ALL added to
+   * the operator table. Calcite will try to match the function call arguments against all available
+   * operators and select the one that matches. This allows functions with multiple signatures to be
+   * used correctly without explicit deduplication.
+   *
+   * @param functions list of functions to convert
+   * @param typeFactory the Calcite type factory
+   * @param typeConverter the type converter
+   * @return list of SqlOperators
+   */
+  public static List<SqlOperator> from(
+      List<? extends SimpleExtension.Function> functions,
+      RelDataTypeFactory typeFactory,
+      TypeConverter typeConverter) {
+    return functions.stream()
         .map(function -> toSqlFunction(function, typeFactory, typeConverter))
         .collect(Collectors.toList());
   }

isthmus/src/main/java/io/substrait/isthmus/SqlToSubstrait.java

@@ -32,20 +32,52 @@ public SqlToSubstrait(FeatureBoard features) {
   public SqlToSubstrait(SimpleExtension.ExtensionCollection extensions, FeatureBoard features) {
     super(features, extensions);
 
+    List<SqlOperator> dynamicOperators = new java.util.ArrayList<>();
+
     if (featureBoard.allowDynamicUdfs()) {
       SimpleExtension.ExtensionCollection dynamicExtensionCollection =
           ExtensionUtils.getDynamicExtensions(extensions);
       if (!dynamicExtensionCollection.scalarFunctions().isEmpty()
           || !dynamicExtensionCollection.aggregateFunctions().isEmpty()) {
         List<SqlOperator> generatedDynamicOperators =
             SimpleExtensionToSqlOperator.from(dynamicExtensionCollection, this.factory);
-        this.operatorTable =
-            SqlOperatorTables.chain(
-                SubstraitOperatorTable.INSTANCE, SqlOperatorTables.of(generatedDynamicOperators));
-        return;
+        dynamicOperators.addAll(generatedDynamicOperators);
+      }
+    }
+
+    if (featureBoard.autoFallbackToDynamicFunctionMapping()) {
+      List<SimpleExtension.ScalarFunctionVariant> unmappedScalars =
+          io.substrait.isthmus.expression.FunctionConverter.getUnmappedFunctions(
+              extensions.scalarFunctions(),
+              io.substrait.isthmus.expression.FunctionMappings.SCALAR_SIGS);
+      List<SimpleExtension.AggregateFunctionVariant> unmappedAggregates =
+          io.substrait.isthmus.expression.FunctionConverter.getUnmappedFunctions(
+              extensions.aggregateFunctions(),
+              io.substrait.isthmus.expression.FunctionMappings.AGGREGATE_SIGS);
+      List<SimpleExtension.WindowFunctionVariant> unmappedWindows =
+          io.substrait.isthmus.expression.FunctionConverter.getUnmappedFunctions(
+              extensions.windowFunctions(),
+              io.substrait.isthmus.expression.FunctionMappings.WINDOW_SIGS);
+
+      if (!unmappedScalars.isEmpty()) {
+        dynamicOperators.addAll(SimpleExtensionToSqlOperator.from(unmappedScalars, this.factory));
       }
+      if (!unmappedAggregates.isEmpty()) {
+        dynamicOperators.addAll(
+            SimpleExtensionToSqlOperator.from(unmappedAggregates, this.factory));
+      }
+      if (!unmappedWindows.isEmpty()) {
+        dynamicOperators.addAll(SimpleExtensionToSqlOperator.from(unmappedWindows, this.factory));
+      }
+    }
+
+    if (!dynamicOperators.isEmpty()) {
+      this.operatorTable =
+          SqlOperatorTables.chain(
+              SubstraitOperatorTable.INSTANCE, SqlOperatorTables.of(dynamicOperators));
+    } else {
+      this.operatorTable = SubstraitOperatorTable.INSTANCE;
     }
-    this.operatorTable = SubstraitOperatorTable.INSTANCE;
   }
 
   /**

isthmus/src/main/java/io/substrait/isthmus/SubstraitRelNodeConverter.java

@@ -87,6 +87,8 @@
 import org.apache.calcite.sql.parser.SqlParser;
 import org.apache.calcite.tools.Frameworks;
 import org.apache.calcite.tools.RelBuilder;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
 
 /**
  * RelVisitor to convert Substrait Rel plan to Calcite RelNode plan. Unsupported Rel node will call
@@ -95,6 +97,8 @@
 public class SubstraitRelNodeConverter
     extends AbstractRelVisitor<RelNode, SubstraitRelNodeConverter.Context, RuntimeException> {
 
+  private static final Logger LOGGER = LoggerFactory.getLogger(SubstraitRelNodeConverter.class);
+
   protected final RelDataTypeFactory typeFactory;
 
   protected final ScalarFunctionConverter scalarFunctionConverter;
@@ -120,9 +124,9 @@ public SubstraitRelNodeConverter(
     this(
         typeFactory,
         relBuilder,
-        createScalarFunctionConverter(extensions, typeFactory, featureBoard.allowDynamicUdfs()),
-        new AggregateFunctionConverter(extensions.aggregateFunctions(), typeFactory),
-        new WindowFunctionConverter(extensions.windowFunctions(), typeFactory),
+        createScalarFunctionConverter(extensions, typeFactory, featureBoard),
+        createAggregateFunctionConverter(extensions, typeFactory, featureBoard),
+        createWindowFunctionConverter(extensions, typeFactory, featureBoard),
         TypeConverter.DEFAULT);
   }
 
@@ -165,11 +169,11 @@ public SubstraitRelNodeConverter(
   private static ScalarFunctionConverter createScalarFunctionConverter(
       SimpleExtension.ExtensionCollection extensions,
       RelDataTypeFactory typeFactory,
-      boolean allowDynamicUdfs) {
+      FeatureBoard featureBoard) {
 
-    List<FunctionMappings.Sig> additionalSignatures;
+    List<FunctionMappings.Sig> additionalSignatures = new ArrayList<>();
 
-    if (allowDynamicUdfs) {
+    if (featureBoard.allowDynamicUdfs()) {
       java.util.Set<String> knownFunctionNames =
           FunctionMappings.SCALAR_SIGS.stream()
               .map(FunctionMappings.Sig::name)
@@ -180,28 +184,124 @@ private static ScalarFunctionConverter createScalarFunctionConverter(
               .filter(f -> !knownFunctionNames.contains(f.name().toLowerCase()))
               .collect(Collectors.toList());
 
-      if (dynamicFunctions.isEmpty()) {
-        additionalSignatures = Collections.emptyList();
-      } else {
+      if (!dynamicFunctions.isEmpty()) {
         SimpleExtension.ExtensionCollection dynamicExtensionCollection =
             SimpleExtension.ExtensionCollection.builder().scalarFunctions(dynamicFunctions).build();
 
         List<SqlOperator> dynamicOperators =
             SimpleExtensionToSqlOperator.from(dynamicExtensionCollection, typeFactory);
 
-        additionalSignatures =
+        additionalSignatures.addAll(
             dynamicOperators.stream()
                 .map(op -> FunctionMappings.s(op, op.getName()))
-                .collect(Collectors.toList());
+                .collect(Collectors.toList()));
+      }
+    }
+
+    if (featureBoard.autoFallbackToDynamicFunctionMapping()) {
+      List<SimpleExtension.ScalarFunctionVariant> unmappedFunctions =
+          io.substrait.isthmus.expression.FunctionConverter.getUnmappedFunctions(
+              extensions.scalarFunctions(), FunctionMappings.SCALAR_SIGS);
+
+      if (!unmappedFunctions.isEmpty()) {
+        LOGGER.info(
+            "Dynamically mapping {} unmapped scalar functions: {}",
+            unmappedFunctions.size(),
+            unmappedFunctions.stream().map(f -> f.name()).collect(Collectors.toList()));
+
+        List<SqlOperator> dynamicOperators =
+            SimpleExtensionToSqlOperator.from(unmappedFunctions, typeFactory);
+
+        // Note: We use last-wins deduplication here because:
+        // 1. Multiple variants of the same function create separate SqlOperator instances
+        // 2. Calcite's SqlOperator equality is based on name and kind, not identity
+        // 3. RexCalls may use any one of these equivalent operators
+        // 4. We only need ONE SqlOperator registered per function name as a key in signatures map
+        // 5. The FunctionFinder will match all variants based on type signatures
+        java.util.Map<String, SqlOperator> operatorsByName = new java.util.LinkedHashMap<>();
+        for (SqlOperator op : dynamicOperators) {
+          operatorsByName.put(op.getName().toLowerCase(), op);
+        }
+
+        additionalSignatures.addAll(
+            operatorsByName.values().stream()
+                .map(op -> FunctionMappings.s(op, op.getName().toLowerCase()))
+                .collect(Collectors.toList()));
       }
-    } else {
-      additionalSignatures = Collections.emptyList();
     }
 
     return new ScalarFunctionConverter(
         extensions.scalarFunctions(), additionalSignatures, typeFactory, TypeConverter.DEFAULT);
   }
 
+  private static AggregateFunctionConverter createAggregateFunctionConverter(
+      SimpleExtension.ExtensionCollection extensions,
+      RelDataTypeFactory typeFactory,
+      FeatureBoard featureBoard) {
+
+    List<FunctionMappings.Sig> additionalSignatures = new ArrayList<>();
+
+    if (featureBoard.autoFallbackToDynamicFunctionMapping()) {
+      List<SimpleExtension.AggregateFunctionVariant> unmappedFunctions =
+          io.substrait.isthmus.expression.FunctionConverter.getUnmappedFunctions(
+              extensions.aggregateFunctions(), FunctionMappings.AGGREGATE_SIGS);
+
+      if (!unmappedFunctions.isEmpty()) {
+        List<SqlOperator> dynamicOperators =
+            SimpleExtensionToSqlOperator.from(unmappedFunctions, typeFactory);
+
+        // Deduplicate operators by name (last-wins precedence) since multiple variants
+        // of the same function create multiple SqlOperator objects
+        java.util.Map<String, SqlOperator> operatorsByName = new java.util.LinkedHashMap<>();
+        for (SqlOperator op : dynamicOperators) {
+          operatorsByName.put(op.getName().toLowerCase(), op);
+        }
+
+        additionalSignatures.addAll(
+            operatorsByName.values().stream()
+                .map(op -> FunctionMappings.s(op, op.getName().toLowerCase()))
+                .collect(Collectors.toList()));
+      }
+    }
+
+    return new AggregateFunctionConverter(
+        extensions.aggregateFunctions(), additionalSignatures, typeFactory, TypeConverter.DEFAULT);
+  }
+
+  private static WindowFunctionConverter createWindowFunctionConverter(
+      SimpleExtension.ExtensionCollection extensions,
+      RelDataTypeFactory typeFactory,
+      FeatureBoard featureBoard) {
+
+    List<FunctionMappings.Sig> additionalSignatures = new ArrayList<>();
+
+    if (featureBoard.autoFallbackToDynamicFunctionMapping()) {
+      List<SimpleExtension.WindowFunctionVariant> unmappedFunctions =
+          io.substrait.isthmus.expression.FunctionConverter.getUnmappedFunctions(
+              extensions.windowFunctions(), FunctionMappings.WINDOW_SIGS);
+
+      if (!unmappedFunctions.isEmpty()) {
+        List<SqlOperator> dynamicOperators =
+            SimpleExtensionToSqlOperator.from(unmappedFunctions, typeFactory);
+
+        // Deduplicate operators by name (last-wins precedence) since multiple variants
+        // of the same function create multiple SqlOperator objects
+        java.util.Map<String, SqlOperator> operatorsByName = new java.util.LinkedHashMap<>();
+        for (SqlOperator op : dynamicOperators) {
+          operatorsByName.put(op.getName().toLowerCase(), op);
+        }
+
+        additionalSignatures.addAll(
+            operatorsByName.values().stream()
+                .map(op -> FunctionMappings.s(op, op.getName().toLowerCase()))
+                .collect(Collectors.toList()));
+      }
+    }
+
+    return new WindowFunctionConverter(
+        extensions.windowFunctions(), additionalSignatures, typeFactory, TypeConverter.DEFAULT);
+  }
+
   public static RelNode convert(
       Rel relRoot,
       RelOptCluster relOptCluster,