Skip to main content

General Pattern Matching with Visitors

My previous definition of IVisitor hard-coded the return value as a type of Val, but one can generalize a visitor to any kind pattern-matching function by adding a generic type parameter to the interface:
//the pattern matching visitor
public interface IVisitor<T>
{
//return type is now parameterized
T App(Exp e0, Exp e1);
}

This of course requires a modification to the class variants as well:
//the type representing an expression application
public sealed class App : Exp
{
Exp e0;
Exp e1;

public App(Exp e0, Exp e1)
{
this.e0 = e0;
this.e1 = e1;
}

//add a generic constraint to the Visit method, so the client can
//specify the return type

public override T Visit<T>(IVisitor<T> v)
{
return v.App(e0, e1);
}
}

So instead of IVisitor being hard-coded as an evaluator, I can now write an IVisitor that performs transformations on the expression tree (such as rewriting optimizations), or a visitor that implements some arbitrary predicate over the expression tree (such as type-checking, type inference, etc.).

Like pattern matching functions over ordinary parameterized algebraic data types, the return value of the visitor must be T or a subtype of T.

In fact, this visitor pattern, minus the decomposition as in pattern matching, is featured in Generalized Algebraic Data Types and Object-Oriented Programming; see also the LTU discussion.

According to the paper, generics as in C# are almost capable of encoding GADTs. Unfortunately, they lack some flexibility in declaring type dependencies between class type parameters, and method type parameters. The paper provides some very interesting C# examples of type-safe type checking, evaluation, phantom types, and representation types. This generics black magic warrants further study for the advanced C# student.

Comments

Popular posts from this blog

Blue-eyed Islander Puzzle - an analysis

Many people find themselves stumped by the so-called Blue-Eyed Islanders puzzle . There is also much controversy over its supposed solution. I'm going to analyze the problem and the solution, and in the process, explain why the solution works. To begin, let's modify the problem slightly and say that there's only 1 blue-eyed islander. When the foreigner makes his pronouncement, the blue-eyed islander looks around and sees no other blue eyes, and being logical, correctly deduces that his own eyes must be blue in order for the foreigner's statement to make sense. The lone blue-eyed islander thus commits suicide the following day at noon. Now comes the tricky part, and the source of much confusion. Let's say there are 2 blue-eyed islanders, Mort and Bob. When the foreigner makes his pronouncement, Mort and Bob look around and see only each other. Mort and Bob thus both temporarily assume that the other will commit suicide the following day at noon. Imagine their chagrin...

Easy Reverse Mode Automatic Differentiation in C#

Continuing from my last post on implementing forward-mode automatic differentiation (AD) using C# operator overloading , this is just a quick follow-up showing how easy reverse mode is to achieve, and why it's important. Why Reverse Mode Automatic Differentiation? As explained in the last post, the vector representation of forward-mode AD can compute the derivatives of all parameter simultaneously, but it does so with considerable space cost: each operation creates a vector computing the derivative of each parameter. So N parameters with M operations would allocation O(N*M) space. It turns out, this is unnecessary! Reverse mode AD allocates only O(N+M) space to compute the derivatives of N parameters across M operations. In general, forward mode AD is best suited to differentiating functions of type: R → R N That is, functions of 1 parameter that compute multiple outputs. Reverse mode AD is suited to the dual scenario: R N → R That is, functions of many parameters t...

Extensible, Statically Typed Pratt Parser in C#

I just completed a statically typed Pratt-style single-state extensible lexer+parser, otherwise known as a top-down operator precedence parser, for the Sasa library . The implementation is available in the Sasa.Parsing dll, under Sasa.Parsing.Pratt . Two simple arithmetic calculators are available in the unit tests . This implementation is novel in two ways: Aside from an alleged implementation in Ada , this is the only statically typed Pratt parser I'm aware of. Pratt parsers typically require a pre-tokenized input before parsing semantic tokens, but I've eliminated this step by using the symbol definitions to drive a longest-match priority-based lexer. Symbols by default match themselves, but you can optionally provide a scanning function used to match arbitrary string patterns. The symbol selected for the current position of the input is the symbol that matches the longest substring. If two symbols match equally, then the symbol with higher precedence is selected. The design...