feat(isthmus): support for safe casting (#236)

BREAKING CHANGE: method ExpressionCreator.cast(Type, Expression) has been removed

---------

Signed-off-by: mbwhite <whitemat@uk.ibm.com>
Co-authored-by: Victor Barua <victor.barua@datadoghq.com>

.gitignore

@@ -7,3 +7,4 @@ gen
 *.iml
 out/**
 *.iws
+.vscode

core/src/main/java/io/substrait/expression/ExpressionCreator.java

@@ -407,10 +407,6 @@ public static Expression.WindowFunctionInvocation windowFunction(
         .build();
   }
 
-  public static Expression cast(Type type, Expression expression) {
-    return cast(type, expression, Expression.FailureBehavior.UNSPECIFIED);
-  }
-
   public static Expression cast(
       Type type, Expression expression, Expression.FailureBehavior failureBehavior) {
     return Expression.Cast.builder()

core/src/main/java/io/substrait/expression/proto/ProtoExpressionConverter.java

@@ -174,7 +174,9 @@ public Expression from(io.substrait.proto.Expression expr) {
             .build();
       }
       case CAST -> ExpressionCreator.cast(
-          protoTypeConverter.from(expr.getCast().getType()), from(expr.getCast().getInput()));
+          protoTypeConverter.from(expr.getCast().getType()),
+          from(expr.getCast().getInput()),
+          Expression.FailureBehavior.fromProto(expr.getCast().getFailureBehavior()));
       case SUBQUERY -> {
         switch (expr.getSubquery().getSubqueryTypeCase()) {
           case SET_PREDICATE -> {

isthmus/src/main/java/io/substrait/isthmus/expression/CallConverters.java

@@ -24,13 +24,22 @@ public class CallConverters {
   public static Function<TypeConverter, SimpleCallConverter> CAST =
       typeConverter ->
           (call, visitor) -> {
-            if (call.getKind() != SqlKind.CAST) {
-              return null;
+            Expression.FailureBehavior failureBehavior;
+            switch (call.getKind()) {
+              case CAST:
+                failureBehavior = Expression.FailureBehavior.THROW_EXCEPTION;
+                break;
+              case SAFE_CAST:
+                failureBehavior = Expression.FailureBehavior.RETURN_NULL;
+                break;
+              default:
+                return null;
             }
 
             return ExpressionCreator.cast(
                 typeConverter.toSubstrait(call.getType()),
-                visitor.apply(call.getOperands().get(0)));
+                visitor.apply(call.getOperands().get(0)),
+                failureBehavior);
           };
 
   /**

isthmus/src/main/java/io/substrait/isthmus/expression/ExpressionRexConverter.java

@@ -4,6 +4,7 @@
 import io.substrait.expression.AbstractExpressionVisitor;
 import io.substrait.expression.EnumArg;
 import io.substrait.expression.Expression;
+import io.substrait.expression.Expression.FailureBehavior;
 import io.substrait.expression.Expression.SingleOrList;
 import io.substrait.expression.Expression.Switch;
 import io.substrait.expression.FieldReference;
@@ -478,8 +479,9 @@ private String convert(FunctionArg a) {
 
   @Override
   public RexNode visit(Expression.Cast expr) throws RuntimeException {
+    var safeCast = expr.failureBehavior() == FailureBehavior.RETURN_NULL;
     return rexBuilder.makeAbstractCast(
-        typeConverter.toCalcite(typeFactory, expr.getType()), expr.input().accept(this));
+        typeConverter.toCalcite(typeFactory, expr.getType()), expr.input().accept(this), safeCast);
   }
 
   @Override

isthmus/src/main/java/io/substrait/isthmus/expression/FunctionConverter.java

@@ -486,7 +486,7 @@ private static List<Expression> coerceArguments(List<Expression> arguments, Type
   private static Expression coerceArgument(Expression argument, Type type) {
     var typeMatches = isMatch(type, argument.getType());
     if (!typeMatches) {
-      return ExpressionCreator.cast(type, argument);
+      return ExpressionCreator.cast(type, argument, Expression.FailureBehavior.THROW_EXCEPTION);
     }
     return argument;
   }

isthmus/src/test/java/io/substrait/isthmus/CalciteCallTest.java

@@ -65,7 +65,10 @@ public void coerceNumericOp() {
         func -> {
           // check that there is a cast for the incorrect argument type.
           assertEquals(
-              ExpressionCreator.cast(TypeCreator.REQUIRED.I64, ExpressionCreator.i32(false, 20)),
+              ExpressionCreator.cast(
+                  TypeCreator.REQUIRED.I64,
+                  ExpressionCreator.i32(false, 20),
+                  Expression.FailureBehavior.THROW_EXCEPTION),
               func.arguments().get(0));
         },
         false); // TODO: implicit calcite cast

isthmus/src/test/java/io/substrait/isthmus/ExpressionConvertabilityTest.java

@@ -6,6 +6,7 @@
 
 import io.substrait.dsl.SubstraitBuilder;
 import io.substrait.expression.Expression;
+import io.substrait.expression.ExpressionCreator;
 import io.substrait.expression.proto.ExpressionProtoConverter;
 import io.substrait.extension.ExtensionCollector;
 import io.substrait.isthmus.expression.ExpressionRexConverter;
@@ -105,6 +106,24 @@ public void switchExpression() {
         expression);
   }
 
+  @Test
+  public void castFailureCondition() {
+    Rel rel =
+        b.project(
+            input ->
+                List.of(
+                    ExpressionCreator.cast(
+                        R.I64,
+                        b.fieldReference(input, 0),
+                        Expression.FailureBehavior.THROW_EXCEPTION),
+                    ExpressionCreator.cast(
+                        R.I32, b.fieldReference(input, 0), Expression.FailureBehavior.RETURN_NULL)),
+            b.remap(1, 2),
+            b.namedScan(List.of("test"), List.of("col1"), List.of(R.STRING)));
+
+    assertFullRoundTrip(rel);
+  }
+
   void assertExpressionEquality(Expression expected, Expression actual) {
     // go the extra mile and convert both inputs to protobuf
     // helps verify that the protobuf conversion is not broken