Antlr4 Listeners and Visitors - which to implement?

Question

I'm reading "The Definitive Antlr 4 Reference" and get the idea in relation to how Listeners and Visitors work. The book explains particularly well how Listeners relate to SAX parsers and makes it obvious when methods are going to be called during the implementation of each. I can see also that listeners are quite good for transforming input to output but I would appreciate a short explanation/example as to when to use a Listener and when to use a Visitor (or should they both be used in certain cases?).

My particular intention is to create an interpreter (Cucumber-style / TinyBasic Interpreter with some custom calls) that will check for syntax errors and stop executing on an error from a custom function without recovering - would love to see a complete implementation of such a thing in antlr - if anyone happens to know of one.

Thanks in advance for any advice.

Answer 1

If you plan to directly use the parser output for interpretation, the visitor is a good choice. You have full control of the traversal, so in conditionals only one branch is visited, loops can be visited n times and so on.

If you translate the input to a lower level, e.g. virtual machine instructions, both patterns may be useful.

You might take a look at "Language Implementation Patterns", which covers the basic interpreter implementations.

I mostly use the visitor pattern, as it's more flexible.

Answer 2

Here is quote from the book that I think is relevant:

The biggest difference between the listener and visitor mechanisms is that listener methods are called by the ANTLR-provided walker object, whereas visitor methods must walk their children with explicit visit calls. Forgetting to invoke visit() on a node’s children means those subtrees don’t get visited.

In visitor pattern you have the ability to direct tree walking while in listener you are only reacting to the tree walker.

Answer 3

There's another important difference between these two patterns: a visitor uses the call stack to manage tree traversals, whereas the listener uses an explicit stack allocated on the heap, managed by a walker. This means that large inputs to a visitor could blow out the stack, while a listener would have no trouble.

If your inputs could be potentially unbounded, or you might call your visitor very deep in a call tree, you should use a Listener, rather than a Visitor, or at least validate that the parse tree is not too deep. Some companies' coding practices discourage or even outright forbid non-tail recursion for this reason.