Set Algorithms

A Set is Not Restricted to Numbers

The examples in most of this lesson use integers, but a set can contain any type of object. For example, $\{red, green, blue\}$ is also a set.

Later in this lesson, we'll show how we can use the algorithms with sets containing any type of data, including our custom types.

Sets do Not Need to be Stored in Set Containers

The sets we're discussing here do not need to be stored in a container specifically designed for sets, such as a std::unordered_set or a std::set.

A set is a more generic concept than a type of container. Whilst a set can be contained in a std::set, it doesn't need to be. A set can be stored in an array, linked list, or a wide range of other possible containers.

These Algorithms Require the Inputs to be Forward Iterable

The range-based algorithms in this lesson require the containers holding the set to be forward ranges. The iterator-based variations require the container to provide forward iterators.

Most of the containers we've seen are compatible, including std::array, std::vector, std::list, std::set and more.

These Algorithms Require the Inputs to be Sorted

The algorithms additionally require the input sets to be sorted. By default, they assume the sets are sorted in ascending order. That means if A < B, then A is before B in the container.

Later in the lesson, we'll see how we can override that assumption. This lets the algorithms work regardless of how our inputs are sorted, and even with types that aren't even sortable.

Let's introduce the algorithms, which are all available within the <algorithm> header:

1#include <algorithm>

The most basic set algorithm checks if a set contains another set.

1std::ranges::includes(A, B)

This will return true if A "includes" B. In other words, it checks if all of the objects in B are also in A.

1#include <algorithm>
2#include <set>
3#include <iostream>
4
5int main(){
6  std::set A{1, 2, 3};
7  std::set B{1, 2};
8
9  if (std::ranges::includes(A, B)) {
10    std::cout << "A includes B";
11  }
12}

1A includes B

Note that the argument order matters here. If A includes B, that doesn't necessarily mean that B includes A:

1#include <algorithm>
2#include <set>
3#include <iostream>
4
5int main(){
6  std::set A{1, 2, 3};
7  std::set B{1, 2};
8
9  if (std::ranges::includes(A, B)) {
10    std::cout << "B is a subset of A";
11  }
12
13  if (!std::ranges::includes(B, A)) {
14    std::cout << "\\nA is NOT a subset of B";
15  }
16}

1B is a subset of A
2A is NOT a subset of B

The union of two sets includes any objects that are in either of the sets. For example, the union of $\{ 1, 2, 3 \}$ and $\{ 3, 4, 5 \}$ is $\{ 1, 2, 3, 4, 5 \}$

Within textbooks and other resources, the symbol $\cup$ is sometimes used to denote the union operation. For example, $\{ 1, 2, 3 \} \cup \{3,4,5\} = \{1,2,3,4,5\}$

The std::ranges::set_union algorithm provides an implementation of this. Its basic usage involves three arguments - the two sets we want to use, and an iterator to where we want the union to be created.

We need to ensure there is enough memory in the result location to accommodate the union. The maximum size a union can be will be the combined size of both of the input sets.

Below, we resize a std::vector to match that size, and then use the std::ranges::set_union function to populate it with the union results:

1#include <algorithm>
2#include <vector>
3
4int main(){
5  std::vector A{1, 2, 3};
6  std::vector B{3, 4, 5};
7  std::vector<int> Results;
8  Results.resize(A.size() + B.size());
9
10  std::ranges::set_union(A, B, Results.begin());
11}

After running this code, our Results vector will have a size() of 6 with the union $\{ 1, 2, 3, 4, 5 \}$ in the first 5 slots. We can shrink the result collection to contain only the union members by using the value returned from std::ranges::set_union().

This will be a std::ranges::set_union_result struct containing three iterators, which may be useful for follow-up operations. The three iterators are:

1. in1 - An iterator for the first set, pointing at where the range ended - equivalent to A.end() in this example
2. in2 - An iterator for the second set, pointing at where the range ended - equivalent to B.end() in this example
3. out - An iterator pointing to the last element of the union, within the result collection

These three iterators are typically accessed using structured binding. The third tends to be the most useful in real-world applications, as it gives us an easy way to tell how big the union is, and to shrink the result collection to just contain those elements:

1#include <algorithm>
2#include <iostream>
3#include <vector>
4
5int main(){
6  std::vector A{1, 2, 3};
7  std::vector B{3, 4, 5};
8  std::vector<int> Results;
9  Results.resize(A.size() + B.size());
10
11  auto [AEnd, BEnd, UnionEnd]{
12    std::ranges::set_union(
13      A, B, Results.begin())
14  };
15
16  std::cout << "Union Size: "
17    << UnionEnd - Results.begin()
18    << '\n';
19
20  Results.erase(UnionEnd, Results.end());
21  for (auto x : Results) {
22    std::cout << x << ", ";
23  }
24}

1Union Size: 5
21, 2, 3, 4, 5,

The intersection of two sets includes all the objects that are in both sets. For example, the intersection of $\{ 1, 2, 3 \}$ and $\{ 2, 3, 4 \}$ is $\{ 2, 3 \}$

Within textbooks and other resources, the symbol $\cap$ is sometimes used to denote the intersection operation. For example, $\{ 1, 2, 3 \} \cap \{ 2, 3, 4 \} = \{2,3\}$

The std::ranges::set_intersection and std::set_intersection algorithms provide a way to create this intersection.

As with the union operation, we need to ensure there is enough memory in the result location to accommodate the intersection. The maximum size of a set intersection will be the size of the smaller of the two input sets.

The std::min() function can help us here, returning the smallest value of its arguments:

1#include <algorithm>
2#include <vector>
3
4int main(){
5  std::vector A{1, 2, 3};
6  std::vector B{2, 3, 4};
7  std::vector<int> Results;
8  Results.resize(std::min(A.size(), B.size()));
9
10  std::ranges::set_intersection(
11    A, B, Results.begin()
12  );
13}

Similar to the union algorithms, the std::ranges::set_intersection algorithm returns a std::ranges::set_intersection_result struct, with three members:

1. in1 - An iterator pointing to where the range we provided as the first argument ended. In this example, it is equivalent to A.end()
2. in2 - An iterator pointing to where the range we provided as the second argument ended. In this example, it is equivalent to B.end()
3. out - An iterator pointing to the last element inserted into the result collection

As with the union algorithms, these are commonly accessed using structured binding, and the third variable is the most useful:

1#include <algorithm>
2#include <iostream>
3#include <vector>
4
5int main(){
6  std::vector A{1, 2, 3};
7  std::vector B{2, 3, 4};
8  std::vector<int> Results;
9  Results.resize(std::min(A.size(), B.size()));
10
11  auto [AEnd, BEnd, IntersectionEnd]{
12    std::ranges::set_intersection(
13      A, B, Results.begin())
14  };
15
16  std::cout << "Intersection Size: "
17    << IntersectionEnd - Results.begin()
18    << '\n';
19
20  Results.erase(IntersectionEnd, Results.end());
21  for (auto x : Results) {
22    std::cout << x << ", ";
23  }
24}

1Intersection Size: 2
22, 3,

The difference operation returns all the objects in the first set after any objects in the second set are removed.

Within textbooks and other learning resources, the subtraction symbol $-$ or division symbol $\\$ is sometimes used to denote a difference operation. For example:

• $\{ 1, 2, 3 \} - \{ 3, 4, 5 \} = \{1, 2\}$
• $\{ 3, 4, 5 \} \\ \{ 1, 2, 3 \} = \{4, 5\}$

We need to ensure there is enough memory in the result location to accommodate the difference. The maximum size of a set difference will be the size of the first input argument:

1#include <algorithm>
2#include <vector>
3
4int main(){
5  std::vector A{1, 2, 3};
6  std::vector B{3, 4, 5};
7  std::vector<int> Results;
8  Results.resize(A.size());
9
10  std::ranges::set_difference(
11    A, B, Results.begin()
12  );
13}

The return value of set_difference() is slightly simpler than the previous algorithms, as it returns a struct containing only two iterators:

1. in - An iterator pointing to where the first range ended - equivalent to what is returned by A.end() in this case
2. out - An iterator pointing to the last element inserted into the result collection

As usual, these are commonly accessed using structured binding, and the out variable is the most useful:

1#include <algorithm>
2#include <iostream>
3#include <vector>
4
5int main(){
6  std::vector A{1, 2, 3};
7  std::vector B{3, 4, 5};
8  std::vector<int> Results;
9  Results.resize(A.size());
10
11  auto [AEnd, DifferenceEnd]{
12    std::ranges::set_difference(
13      A, B, Results.begin())
14  };
15
16  std::cout << "Difference Size: "
17    << DifferenceEnd - Results.begin()
18    << '\n';
19
20  Results.erase(DifferenceEnd, Results.end());
21  for (auto x : Results) {
22    std::cout << x << ", ";
23  }
24}

1Difference Size: 2
21, 2,

The intersection of two sets includes all the objects that are in either set, but not both sets. We can think of this as having similar characteristics to an exclusive or (xor) operation. It's also equivalent to calculating both the union and intersection of two sets, and then subtracting the intersection from the union.

For example, the symmetric difference of $\{ 1, 2, 3 \}$ and $\{ 2, 3, 4 \}$ is $\{ 1, 4 \}$

Within textbooks and other resources, the symbol $\triangle$ is sometimes used to denote the symmetric difference. For example, $\{ 1, 2, 3 \} \triangle \{2,3,4\} = \{1,4\}$

We need to ensure there is enough memory in the result location to accommodate the symmetric difference. The maximum size of a symmetric difference will be the combined size of both of the inputs:

1#include <algorithm>
2#include <vector>
3
4int main() {
5  std::vector A { 1, 2, 3 };
6  std::vector B { 2, 3, 4 };
7  std::vector<int> Results;
8  Results.resize(A.size() + B.size());
9
10  std::ranges::set_symmetric_difference(
11    A, B, Results.begin()
12  );
13}

The set_symmetric_difference has the same return type as the union and intersection algorithms. It is a struct containing:

1. in1 - An iterator pointing to where the range we provided as the first argument ended. In this example, it is equivalent to A.end()
2. in2 - An iterator pointing to where the range we provided as the second argument ended. In this example, it is equivalent to B.end()
3. out - An iterator pointing to the last element inserted into the result collection

Below, we show an example of accessing these using structured binding, and using out property for some common follow-up operations:

1#include <algorithm>
2#include <iostream>
3#include <vector>
4
5int main(){
6  std::vector A{1, 2, 3};
7  std::vector B{2, 3, 4};
8  std::vector<int> Results;
9  Results.resize(A.size() + B.size());
10
11  auto [AEnd, BEnd, SymmetricDifferenceEnd]{
12    std::ranges::set_symmetric_difference(
13      A, B, Results.begin())
14  };
15
16  std::cout << "Symmetric Difference Size: "
17    << SymmetricDifferenceEnd - Results.begin()
18    << '\n';
19
20  Results.erase(SymmetricDifferenceEnd,
21                Results.end());
22
23  for (auto x : Results) {
24    std::cout << x << ", ";
25  }
26}

1Symmetric Difference Size: 2
21, 4,

In the introduction, we covered how the set algorithms assume the inputs are sorted in ascending order. That is, if A < B returns true, then A should precede B in the container.

This is customizable by providing an additional argument to the algorithm. This argument should be a function that will receive two objects from our set, either as values or, more commonly, by const reference.

The function should return true if the first parameter comes before the second parameter in our container.

Below, we use this argument to tell our algorithm that our inputs are sorted in descending order, rather than the default assumption of ascending:

1#include <algorithm>
2#include <set>
3#include <iostream>
4
5int main(){
6  std::set A{3, 2, 1};
7  std::set B{2, 1};
8
9  auto Comparer{
10    [](const int& A, const int& B){
11      return A > B;
12    }};
13
14  if (std::ranges::includes(A, B, Comparer)) {
15    std::cout << "A includes B";
16  }
17}

1A includes B

The ability to pass a custom comparator allows our algorithms to work with sets containing any type of object, including those that are not inherently sortable. Below, we use a set algorithm with collections of custom types:

1#include <algorithm>
2#include <iostream>
3#include <vector>
4
5class Character {
6public:
7  Character(){};
8  Character(std::string Name) : Name{Name}{}
9  std::string Name;
10};
11
12int main(){
13  using namespace std::string_literals;
14  std::vector<Character> A{
15    "Aragorn"s, "Frodo"s, "Gimli"s, "Legolas"s};
16  std::vector<Character> B{
17    "Frodo"s, "Legolas"s};
18
19  auto Comparer{
20    [](const Character& A,
21       const Character& B){
22      return A.Name < B.Name;
23    }};
24
25  if (std::ranges::includes(A, B, Comparer)) {
26    std::cout << "A includes B";
27  }
28}

1A includes B

Here is a more complex example, where we create the union instead:

1#include <algorithm>
2#include <iostream>
3#include <vector>
4
5class Character {
6public:
7  Character(){};
8  Character(std::string Name) : Name{Name}{}
9  std::string Name;
10};
11
12int main(){
13  using namespace std::string_literals;
14  std::vector<Character> A{
15    "Aragorn"s, "Frodo"s, "Gimli"s};
16  std::vector<Character> B{
17    "Frodo"s, "Gandalf"s, "Legolas"s};
18  std::vector<Character> Results;
19  Results.resize(A.size() + B.size());
20
21  auto Comparer{
22    [](const Character& A,
23       const Character& B){
24      return A.Name < B.Name;
25    }};
26
27  auto [AEnd, BEnd, UnionEnd]{
28    std::ranges::set_union(
29      A, B, Results.begin(), Comparer)
30  };
31
32  std::cout << "Union Size: "
33    << UnionEnd - Results.begin()
34    << '\n';
35
36  Results.erase(UnionEnd, Results.end());
37  for (auto x : Results) {
38    std::cout << x.Name << ", ";
39  }
40}

1Union Size: 5
2Aragorn, Frodo, Gandalf, Gimli, Legolas,

If we have a custom type where it makes sense for them to be sortable, we can implement support for that within the class itself.

This makes our classes more versatile and removes the need for standalone comparison operators like this scattered through our code. We will demonstrate how to do that later in the course.

Similar to other range-based algorithms, the set algorithms support projection.

This allows our objects to be projected into a different form before the comparison. We do this by passing a function that will receive an object in our container, and return the projection of that object that we want to use.

A common use case for this is when our objects are complex types, and we want the comparison to be done on a variable within that object.

Below, we implement our previous example using projection instead of a custom comparison. Two things to note here:

• We want to use the default comparer, so we pass a blank argument {} in that position
• The subsequent arguments are two projection functions - one to use for the objects in the set we provided as the first argument (A in this example) and one to use for the objects in the set we provided as the first argument (B in this example). In this case, we use the same projection function for both sets, but they can be different

1#include <algorithm>
2#include <iostream>
3#include <vector>
4
5class Character {
6public:
7  Character(){};
8  Character(std::string Name) : Name{Name}{}
9  std::string Name;
10};
11
12int main(){
13  using namespace std::string_literals;
14  std::vector<Character> A{
15    "Aragorn"s, "Frodo"s, "Gimli"s, "Legolas"s};
16  std::vector<Character> B{
17    "Frodo"s, "Legolas"s};
18
19  auto Projector{
20    [](const Character& C){ return C.Name; }};
21
22  if (std::ranges::includes(
23    A, B, {}, Projector, Projector
24  )) { std::cout << "A includes B"; }
25}

1A includes B

The examples in this lesson all use the C++20 ranges library, but variations of all of these algorithms that work with iterator pairs are available instead.

To use the iterator versions, we need to:

• Remove ranges from the namespace qualification - for example, std::ranges::includes becomes std::includes
• Replace every range argument with two arguments, forming an iterator pair. For example, A becomes A.begin(), A.end()

The following example shows this:

1#include <algorithm>
2#include <set>
3#include <iostream>
4
5int main(){
6  std::set A{1, 2, 3};
7  std::set B{1, 2};
8
9  if (std::includes(A.begin(), A.end(),
10                    B.begin(), B.end())) {
11    std::cout << "A includes B";
12  }
13}

1A includes B

Note that projection is part of the ranges library. As such, the iterator variations of these algorithms do not support projection.

In this lesson, we explored set algorithms by delving into their implementation, usage, and customization.

Main Points

• Sets can contain any type of object, not just numbers, and do not need to be stored in specific set containers.
• Set algorithms require their input ranges to be sorted and forward iterable.
• The C++ Standard Library provides five main set algorithms: includes(), set_union(), set_intersection(), set_difference(), and set_symmetric_difference(), each serving a unique purpose in set manipulation.
• Custom comparers can be used to customize the sorting and comparison behavior of set algorithms, allowing them to work with custom types or non-default sorting criteria.
• Projection enables complex types to be compared based on specific attributes or transformed representations.
• Both range-based and iterator-based versions of set algorithms are available, catering to different requirements.