TP

Pattern matching in action using C# 6

On year ago, on this very day, I wrote about the open-sourcing of C# 6.0. Thanks to a very special information leak, I learned about this about a week before Microsoft officially announced it. However, my information were slightly incorrect - rather then releasing the much improved version of the language, Microsoft continued working on language version internally called "Small C#", which is now available as "C# 6" in the Visual Studio 2015 preview.

It is my understanding that, with this release, Microsoft is secretly testing the reaction of the developer audience to some of the amazing features that F# developers loved and used for the last 7 years and that are coming to C# very soon. To avoid shock, these are however carefuly hidden!

In this blog post, I'm going to show you pattern matching which is probably the most useful hidden C# feature and its improvements in C# 6. For reasons that elude me, pattern matching in C# 6 is called exception filters and has some unfortunate restrictions. But we can still use it to write nice functional code!

UPDATE: In case you are reading this article later than on the day when it was published, let me just point out that this was released on 1 April 2015. Keep that in mind before putting the code in production. Have fun ;-).

Formatting simple math expressions with F#

For easier understanding, I'll first introduce the idea of pattern matching using F#. I'll use a fairly standard functional programming example - formatting and evaluation of simple algebraic expression. We want to work with expressions like \(x * (1 + 2)\), so we'll need variables, constants, addition and multiplication. In F#, we can define a discriminated union to model the four cases:

1: 
2: 
3: 
4: 
5: 
type Expression = 
  | Variable of string
  | Constant of int
  | Add of Expression * Expression
  | Mul of Expression * Expression

This says that an Expression can be one of four different cases. If it is a variable, it contains the name (as a string), if it is addition, it contains two sub-expressions and so on.

Now, writing formatting is quite easy, because we can write a recursive function format that takes an expression and uses the match construct to handle the different cases:

1: 
2: 
3: 
4: 
5: 
6: 
let rec format e = 
  match e with
  | Variable s -> s
  | Constant n -> string n
  | Add(l, r) -> sprintf "(%s + %s)" (format l) (format r)
  | Mul(l, r) -> sprintf "%s*%s" (format l) (format r)

The code is quite straightforward. When we get a variable, we just return its name. When we get a constant, we turn the number to a string and return it. Addition and multiplication is a little more interesting - we call format recursively on the two sub-expressions and then build a composed string.

Defining discriminated unions in C#

Unfortunately, C# does not give us a simple way to define custom discriminated union types (you can send a pull request from this repo to this repo, although you might get labelled as troll). However, there is one discriminated union hiding in C#!

To find it, we need a bit of programming language theory (which has nothing to do with monads, by the way). The Types and Programming Languages book (page 177) says the following about ML exceptions:

The same idea can be refined to leave room for user-defined exceptions by taking \(T_{exn}\) to be an extensible variant type. ML adopts this idea, providing a single extensible variant type called exn.

So, interestingly, you can see ML (and F#) exceptions as a single discriminated union and custom exceptions as cases of this union! If we stretch the idea a bit, we can use it and extend the only discriminated union that is available in C# by defining our expressions as custom exceptions:

 1: 
 2: 
 3: 
 4: 
 5: 
 6: 
 7: 
 8: 
 9: 
10: 
11: 
12: 
13: 
14: 
public class Variable : Exception {
  public string Name { get; set; } 
}
public class Constant : Exception {
  public int Value { get; set; } 
}
public class Add : Exception {
  public Exception Left { get; set; }
  public Exception Right { get; set; } 
}
public class Multiply : Exception {
  public Exception Left { get; set; }
  public Exception Right { get; set; } 
}

This is a bit tedious, but I'm pretty sure that Resharper can (with some nice plugin) generate this for you from the original F# code. If you are a Clean Coder, be sure to split this definition across 4 different files.

Formatting math expressions in C#

Now we have our discriminated union, so let's have a look how we can rewrite the expression formatting code in C#. The amazing thing about this first example is that it does not even need C# 6 - which means you can adopt this technique today! But because I want to be cooler and show off my cutting edge coding skills, I'll also use string interpolation from C# 6.

The idea is quite simple, we get the expression as an argument of type Exception, we throw it and we use the limited form of pattern matching provided by C# catch construct:

public static string Format(Exception e) {
  try { throw e; }
  catch (Constant n) { return n.Value.ToString(); }
  catch (Variable v) { return v.Name; }
  catch (Add a) { return $"({Format(a.Left)} + {Format(a.Right)})"; }
  catch (Multiply a) { return $"{Format(a.Left)} * {Format(a.Right)}"; }
}

This amazing technique is actually quite close to what you can do with F#. Each catch clause corresponds to one clause of the match construct in F#. Another nice thing is that this not just checks that the value has the right type, say Multiply, but it also type casts the input to the right type for free - so in the body, we can access a.Left and a.Right directly.

Now, there are some limitations - because the Exception discriminated union is open, the C# compiler cannot do exhaustiveness checks. That is, F# will warn you if you forget a case but C# will not. For completeness let's see how this works using a sample input:

1: 
2: 
3: 
4: 
5: 
6: 
7: 
var expr = new Multiply {
  Left = new Variable { Name = "x" },
  Right = new Add {
    Left = new Constant { Value = 1 },
    Right = new Constant { Value = 2 } }
};
Console.WriteLine(Format(expr));

To run this, you do not even need C# 6, so you can try it right now - and you should see that the code not only does not throw an exception but gives the correct result which is "x * (1 + 2)".

Evaluating expressions with exception filters

As I mentioned, pattern matching is significantly improved in C# 6 thanks to exception filters. To demonstrate this feature, I'll extend our example with a simple expression evaluator. In addition to an expression, the method also takes a dictionary that defines values for the variables in the expression.

When evaluating Variable, we need to distinguish two different cases - if the variable is in the dictionary, we just return its value. Otherwise, we report an error. This can be done elegantly with exception filters:

1: 
2: 
3: 
4: 
5: 
6: 
7: 
8: 
9: 
public static int Evaluate(Exception e, IDictionary<string, int> vars) {
  int res;
  try { throw e; }
  catch (Constant n) { return n.Value; }
  catch (Variable v) when (vars.TryGetValue(v.Name, out res)) { return res; }
  catch (Variable _) { throw new ArgumentException("Variable not found!"); }
  catch (Add a) { return Evaluate(a.Left, vars) + Evaluate(a.Right, vars); }
  catch (Multiply a) { return Evaluate(a.Left, vars) + Evaluate(a.Right, vars); }
}

As you can see, exception filters give us some more of the pattern matching power from F#. Here, we can use two patterns (Variable v) when (...) handles the case when a variable value is defined in the dictionary vars and the pattern (Variable _) is used to handle all remaining cases.

To test this, we can call the Evaluate function as follows:

1: 
2: 
Console.WriteLine(Evaluate(expr, 
  new Dictionary<string, int> { { "x", 4 } }));

To run this, you'll actually need a version of C# that supports exception filters - either Visual Studio 2015 preview, or latest build of Roslyn. If you run it, you'll get 12 as the result, as expected.

Summary

Pattern matching is one of the most useful concepts in F# and functional programming, because it lets you express complex logic in a very clear way with just a few lines of code. Unfortunately, the full power of pattern matching is not yet available in C#.

As a C# developer, you have basically two options.

Why support pattern matching only on exceptions? Don't ask me! It seems a bit silly to me too - if I was in charge of C# language design, I would obviously add pattern matching on the JSON.NET JContainer classes, because that would be useful at least for some things.

union case Expression.Variable: string -> Expression
Multiple items
val string : value:'T -> string

Full name: Microsoft.FSharp.Core.Operators.string

--------------------
type string = System.String

Full name: Microsoft.FSharp.Core.string
union case Expression.Constant: int -> Expression
Multiple items
val int : value:'T -> int (requires member op_Explicit)

Full name: Microsoft.FSharp.Core.Operators.int

--------------------
type int = int32

Full name: Microsoft.FSharp.Core.int

--------------------
type int<'Measure> = int

Full name: Microsoft.FSharp.Core.int<_>
union case Expression.Add: Expression * Expression -> Expression
type Expression =
  | Variable of string
  | Constant of int
  | Add of Expression * Expression
  | Mul of Expression * Expression

Full name: Csharp-pattern-matching.Expression
union case Expression.Mul: Expression * Expression -> Expression
val format : e:Expression -> string

Full name: Csharp-pattern-matching.format
val e : Expression
val s : string
val n : int
val l : Expression
val r : Expression
val sprintf : format:Printf.StringFormat<'T> -> 'T

Full name: Microsoft.FSharp.Core.ExtraTopLevelOperators.sprintf

Published: Wednesday, 1 April 2015, 12:41 PM
Author: Tomas Petricek
Typos: Send me a pull request!
Tags: c#, fun, functional programming