Use Visitor or Listener with ANTLR4 when returning objects of different types

Question

I translate one language into another with ANTLR4. For example when I read numerical literals I can return an Integer or Double .

@Override
public Integer visitIntegerValue(Parser.IntegerValueContext ctx) {
    return Integer.valueOf(ctx.getText());
}

@Override
public Double visitDoubleValue(Parser.DoubleValueContext ctx) {
    return Double.valueOf(ctx.getText());
}

Ultimately if you extend this approach further and introduce other constructs like strings and conditions, the only reasonable type for the visitor is class Visitor extends BaseVisitor<Object> , but it leads to a code heavily spiced with instanceof . For example

@Override
public CollectionQuery visitCondition(Parser.ConditionContext ctx) {
    Property property = (Property) visit(ctx.property());
    String operator = (String) visit(ctx.operator());
    Object value = visit(ctx.amount());
    Object condition;
    if (value instanceof String && operator.equals("$regex")) {
        condition = Pattern.compile((String) value, Pattern.CASE_INSENSITIVE);
    }
    ...
}

While I don't mind this kind of 'dynamism', I would like to know if this is a maintainable way to proceed or are there other techniques that I should use instead, like creating proper hierarchy of target language constructs.

Answer 1

To get an impression how a custom post-processing could look like.

some ANTLR code

topMostRule : childRule+ EOL;
childRule : variantOne | variantTwo;
variantOne : 'A';
variantTwo : '1';
...

Pseudo code for a custom postprocessing (more C# like than Java / not the real method names ANTLR uses):

public class MyCustomPostprocessor
{
    private IntermediateResults lookupTable; // use private fields for lookup tables etc.

    public ResultType process(Parser.TopMostRuleContext ctx)
    {
        // inspect the children
        var children = new List<object>();
        foreach (var rule in ctx.ChildRules)
        {
            switch (rule.Type)
            {
            case typeof (Parser.ChildRuleContext):
                var result = process(rule);
                children.Add(result);
            else
                throw new NotImplementedException("Don't know what to do with " + rule.Type.ToString());
            }

            // use the information gathered so far to form the result
            return new ResultType (children);
        }
    }


    public object process (Parser.ChildRuleContext)
    {
        foreach (var rule in ctx.ChildRules)
        {
            switch (rule.Type)
            {
            case typeof (Parser.VariantOneContext):
                var result = process(rule);
                return result;
            case typeof (Parser.VariantTwoContext):
                var result = process(rule);
                return result;
            else
                throw new NotImplementedException("Don't know what to do with " + rule.Type.ToString());
            }

        }
    }

    public string process (Parser.VariantOneContext ctx)
    {
        return ctx.GetText();
    }

    public int process (Parser.VariantTwoContext ctx)
    {
        return Int.Parse(ctx.GetText());
    }


}

Answer 2

One suggestion is to have a visitor per return type:

public class IntegerVisitor extends BaseListener<Integer> {
  @Override
  public Integer visitIntegerValue(Parser.IntegerValueContext ctx) {
    return Integer.valueOf(ctx.getText());
  }
}

public class DoubleVisitor extends BaseListener<Double> {
  @Override
  public Double visitDoubleValue(Parser.DoubleValueContext ctx) {
    return Integer.valueOf(ctx.getText());
  }
}

This makes more sense when you're visiting vastly different things (for example if you were parsing using a java grammar you might have a MethodVisitor and a ClassVisitor etc. See an example here: See an example here

Answer 3

I suggest creating a Value class for wrapping the different kinds of objects, and then use this value class as generic type for the visitor.

Visitor<Value> visitor;

public class Value {
    private Object value;

    public Value(Object object) {
        this.value = object
        if (!(isDouble() || isInteger))
            throw new IllegalArgumentException();
    }


    public boolean isDouble() {
        return value instanceof Double;
    }
    public Double asDouble() {
        return (Double) value;
    }

    public boolean isInteger() {
        return value instanceof Integer;
    }

    public Integer asInteger() {
        return (Integer) value;
    }


    @Override
    public int hashCode() {
        // generate hascode
    }

    @Override
    public boolean equals(Object object) {
        // equals
    }
}