Local functions compared to lambda expressions

• 06/27/2016
• 4 minutes to read
• Contributors
  • 
  • 
  • 
  • 
  • 
  • all

At first glance, local functions and lambda expressions are very similar. In many cases, the choice between using lambda expressions and local functions is a matter of style and personal preference. However, there are real differences in where you can use one or the other that you should be aware of.

Let's examine the differences between the local function and lambda expression implementations of the factorial algorithm. First the version using a local function:

public static int LocalFunctionFactorial(int n)
{
    return nthFactorial(n);

    int nthFactorial(int number) => (number < 2) ? 
        1 : number * nthFactorial(number - 1);
}

Contrast that implementation with a version that uses lambda expressions:

public static int LambdaFactorial(int n)
{
    Func<int, int> nthFactorial = default(Func<int, int>);

    nthFactorial = (number) => (number < 2) ? 
        1 : number * nthFactorial(number - 1);

    return nthFactorial(n);
}

The local functions have names. The lambda expressions are anonymous methods that are assigned to variables that are Func or Action types. When you declare a local function, the argument types and return type are part of the function declaration. Instead of being part of the body of the lambda expression, the argument types and return type are part of the lambda expression's variable type declaration. Those two differences may result in clearer code.

Local functions have different rules for definite assignment than lambda expressions. A local function declaration can be referenced from any code location where it is in scope. A lambda expression must be assigned to a delegate variable before it can be accessed (or called through the delegate referencing the lambda expression.) Notice that the version using the lambda expression must declare and initialize the lambda expression, nthFactorial before defining it. Not doing so results in a compile time error for referencing nthFactorial before assigning it. These differences mean that recursive algorithms are easier to create using local functions. You can declare and define a local function that calls itself. Lambda expressions must be declared, and assigned a default value before they can be re-assigned to a body that references the same lambda expression.

Definite assignment rules also affect any variables that are captured by the local function or lambda expression. Both local functions and lambda expression rules demand that any captured variables are definitely assigned at the point when the local function or lambda expression is converted to a delegate. The difference is that lambda expressions are converted to delegates when they are declared. Local functions are converted to delegates only when used as a delegate. If you declare a local function and only reference it by calling it like a method, it will not be converted to a delegate. That rule enables you to declare a local function at any convenient location in its enclosing scope. It's common to declare local functions at the end of the parent method, after any return statements.

Third, the compiler can perform static analysis that enables local functions to definitely assign captured variables in the enclosing scope. Consider this example:

int M()
{
    int y;
    LocalFunction();
    return y;

    void LocalFunction() => y = 0;
}

The compiler can determine that LocalFunction definitely assigns y when called. Because LocalFunction is called before the return statement, y is definitely assigned at the return statement.

The analysis that enables the example analysis enables the fourth difference. Depending on their use, local functions can avoid heap allocations that are always necessary for lambda expressions. If a local function is never converted to a delegate, and none of the variables captured by the local function is captured by other lambdas or local functions that are converted to delegates, the compiler can avoid heap allocations.

Consider this async example:

public Task<string> PerformLongRunningWorkLambda(string address, int index, string name)
{
    if (string.IsNullOrWhiteSpace(address))
        throw new ArgumentException(message: "An address is required", paramName: nameof(address));
    if (index < 0)
        throw new ArgumentOutOfRangeException(paramName: nameof(index), message: "The index must be non-negative");
    if (string.IsNullOrWhiteSpace(name))
        throw new ArgumentException(message: "You must supply a name", paramName: nameof(name));

    Func<Task<string>> longRunningWorkImplementation = async () =>
    {
        var interimResult = await FirstWork(address);
        var secondResult = await SecondStep(index, name);
        return $"The results are {interimResult} and {secondResult}. Enjoy.";
    };

    return longRunningWorkImplementation();
}

The closure for this lambda expression contains the address, index and name variables. In the case of local functions, the object that implements the closure may be a struct type. That struct type would be passed by reference to the local function. This difference in implementation would save on an allocation.

The instantiation necessary for lambda expressions means extra memory allocations, which may be a performance factor in time-critical code paths. Local functions do not incur this overhead. In the example above, the local functions version has 2 fewer allocations than the lambda expression version.

public Task<string> PerformLongRunningWork(string address, int index, string name)
{
    if (string.IsNullOrWhiteSpace(address))
        throw new ArgumentException(message: "An address is required", paramName: nameof(address));
    if (index < 0)
        throw new ArgumentOutOfRangeException(paramName: nameof(index), message: "The index must be non-negative");
    if (string.IsNullOrWhiteSpace(name))
        throw new ArgumentException(message: "You must supply a name", paramName: nameof(name));

    return longRunningWorkImplementation();

    async Task<string> longRunningWorkImplementation()
    {
        var interimResult = await FirstWork(address);
        var secondResult = await SecondStep(index, name);
        return $"The results are {interimResult} and {secondResult}. Enjoy.";
    }
}

One final advantage not demonstrated in this sample is that local functions can be implemented as iterators, using the yield return syntax to produce a sequence of values. The yield return statement is not allowed in lambda expressions.

While local functions may seem redundant to lambda expressions, they actually serve different purposes and have different uses. Local functions are more efficient for the case when you want to write a function that is called only from the context of another method.