7.15 — Introduction to lambdas (anonymous functions)

Consider this snippet of code that we introduced in a previous lesson:

This code searches through an array of strings looking for the first element that contains the substring “nut”. Thus, it produces the result:

Found walnut

And while it works, it could be improved.

The root of the issue here is that std::find_if requires that we pass it a function pointer. Because of that, we are forced to define a function that’s only going to be used once, that must be given a name, and that must be put in the global scope (because functions can’t be nested!). The function is also so short, it’s almost easier to discern what it does from the one line of code than from the name and comments.

Lambdas to the rescue

A lambda expression (also called a lambda or closure) allows us to define an anonymous function inside another function. The nesting is important, as it allows us both to avoid namespace naming pollution, and to define the function as close to where it is used as possible (providing additional context).

The syntax for lambdas is one of the weirder things in C++, and takes a bit of getting used to. Lambdas take the form:

[ captureClause ] ( parameters ) -> returnType
{
    statements;
}

The capture clause and parameters can both be empty if they are not needed.

The return type is optional, and if omitted, auto will be assumed (thus using type inference used to determine the return type). While we previously noted that type inference for function return types should be avoided, in this context, it’s fine to use (because these functions are typically so trivial).

Also note that lambdas have no name, so we don’t need to provide one.

Let’s rewrite the above example using a lambda:

This works just like the function pointer case, and produces an identical result:

Found walnut

Note how similar our lambda is to our containsNut function. They both have identical parameters and function bodies. The lambda has no capture clause (we’ll explain what a capture clause is in the next lesson) because it doesn’t need one. And we’ve omitted the trailing return type in the lambda (for conciseness), but since operator!= returns a bool, our lambda will return a bool too.

Type of a lambda

In the above example, we defined a lambda right where it was needed. This use of a lambda is sometimes called a function literal.

However, writing a lambda in the same line as it’s used can sometimes make code harder to read. Much like we can initialize a variable with a literal value (or a function pointer) for use later, we can also initialize a lambda variable with a lambda definition and then use it later. A named lambda along with a good function name can make code easier to read.

For example, in the following snippet, we’re using std::all_of to check if all elements of an array are even:

We can improve the readability of this as follows:

Note how well the last line reads: “return whether all of the elements in the array are even”

But what is the type of lambda isEven?

As it turns out, lambdas don’t have a type that we can explicitly use. When we write a lambda, the compiler generates a unique type just for the lambda that is not exposed to us.

Although we don’t know the type of a lambda, there are several ways of storing a lambda for use post-definition. If the lambda doesn’t capture anything, we can use a regular function pointer. As soon as the lambda captures anything, a function pointer won’t work anymore. However, std::function can be used for lambdas even if they are capturing something.

The only way of using the lambda’s actual type is by means of auto. auto also has the benefit of having no overhead compared to std::function.

Unfortunately, we can’t always use auto. In cases where the actual lambda is unknown (e.g. because we’re passing a lambda to a function as a parameter and the caller determines what lambda will be passed in), we can’t use auto. In such cases, std::function should be used.

Output

0
1
2

Generic lambdas

For the most part, lambda parameters work by the same rules as regular function parameters.

One notable exception is that since C++14 we’re allowed to use auto for parameters (note: in C++20, regular functions will be able to use auto for parameters too). When a lambda has one or more auto parameter, the compiler will infer what parameter types are needed from the calls to the lambda.

Because lambdas with one or more auto parameter can potentially work with a wide variety of types, they are called generic lambdas.

Let’s take a look at a generic lambda:

Output:

June and July start with the same letter

In the above example, we use auto parameters to capture our strings by const reference. Because all string types allow access to their individual characters via operator[], we don’t need to care whether the user is passing in a std::string, C-style string, or something else. This allows us to write a lambda that could accept any of these, meaning if we change the type of months later, we won’t have to rewrite the lambda.

However, auto isn’t always the best choice. Consider:

Output:

There are 2 months with 5 letters

In this example, using auto would infer a type of const char*. C-style strings aren’t easy to work with (apart from using operator[]). In this case, we prefer to explicitly define the parameter as a std::string_view, which allows us to work with the underlying data much more easily (e.g. we can ask the string view for its length, even if the user passed in a C-style array).

Generic lambdas and static variables

One thing to be aware of is that a unique lambda will be generated for each different type that auto resolves to. The following example shows how one generic lambda turns into two distinct lambdas:

Output

0: hello
1: world
0: 1
1: 2
2: ding dong

In the above example, we define a lambda and then call it with two different parameters (a string literal parameter, and an integer parameter). This generates two different versions of the lambda (one with a string literal parameter, and one with an integer parameter).

Most of the time, this is inconsequential. However, note that if the generic lambda uses static duration variables, those variables are not shared between the generated lambdas.

We can see this in the example above, where each type (string literals and integers) has its own unique count! Although we only wrote the lambda once, two lambdas were generated -- and each has its own version of callCount.

If we wanted callCount to be shared between the lambdas, we’d have to declare it outside of the lambda and capture it by reference so it can be changed by the lambda (we’ll discuss captures in the next lesson).

Return type deduction and trailing return types

If return type deduction is used, a lambda’s return type is deduced from the return-statements inside the lambda. If return type inference is used, all return statements in the lambda must return the same type (otherwise the compiler won’t know which one to prefer).

For example:

This produces a compile error because the return type of the first return statement (int) doesn’t match the return type of the second return statement (double).

In the case where we’re returning different types, we have two options:
1) Do explicit casts to make all the return types match, or
2) explicitly specify a return type for the lambda, and let the compiler do implicit conversions.

The second case is usually the better choice:

That way, if you ever decide to change the return type, you (usually) only need to change the lambda’s return type, and not touch the lambda body.

Standard library function objects

For common operations (e.g. addition, negation, or comparison) you don’t need to write your own lambdas, because the standard library comes with many basic callable objects that can be used instead. These are defined in the <functional> header.

In the following example:

Output

99 90 80 40 13 5

Instead of converting our greater function to a lambda (which would obscure it’s meaning a bit), we can instead use std::greater:

Output

99 90 80 40 13 5

Conclusion

Lambdas and the algorithm library may seem unnecessarily complicated when compared to a solution that uses a loop. However, this combination can allow some very powerful operations in just a few lines of code, and can be more readable than writing your own loops. On top of that, the algorithm library features powerful and easy-to-use parallelism, which you won’t get with loops. Upgrading source code that uses library functions is easier than upgrading code that uses loops.

Lambdas are great, but they don’t replace regular functions for all cases. Prefer a regular functions for non-trivial and reusable cases.

Quiz time

Dan is the best student

Winter
Spring
Fall
Summer

7.16 -- Lambda captures

Index

7.14 -- Ellipsis (and why to avoid them)