Exception Types

The C++ standard library comes with a hierarchy of exception classes. The standard library exception types inherit from std::exception, but we'll typically be using more specific derived types. Derived types include classes like std::logic_error and std::runtime_error.

These have yet more specific subclasses - for example, std::logic_error has a subclass called std::invalid_argument

These give us a standardized type of object to throw within our functions. How we classify our exceptions generally doesn't matter, but it's helpful to be consistent.

The standard library exceptions are available after including <stdexcept>:

1#include <stdexcept>

Throwing exception objects works in the same way as throwing any other object. The standard library types (aside from the basic std::exception type) have a constructor that we can pass a string to, explaining the issue:

1#include <iostream>
2#include <stdexcept>
3
4int Divide(int x, int y) {
5  if (y == 0) {
6    throw std::invalid_argument {
7      "Cannot divide by zero"
8    };
9  }
10  return x/y;
11}

Similarly, we catch them like any other. Objects that inherit from std::exception have a what() method that allows us to access a description of the exception:

1#include <iostream>
2#include <stdexcept>
3
4int Divide(int x, int y) {
5  if (y == 0) {
6    throw std::invalid_argument {
7      "Cannot divide by zero"
8    };
9  }
10  return x/y;
11}
12
13int main() {
14  try {
15    Divide(4,0);
16  } catch (std::invalid_argument& e) {
17    std::cout << "Invalid Argument: "
18              << e.what();
19  }
20}

1Invalid Argument: Cannot divide by zero

In the above examples, note that we catch the errors by reference, denoted by the inclusion of the & character:

1catch (std::invalid_argument& e)

By convention, we should generally be catching errors by reference, and by const reference if preferred.

This is important as, just like functions arguments, exceptions will be passed to catch blocks by value otherwise. This has twe undesirable effects.

1. It causes the usual performance degradation involved in copying an object unnecessarily.
2. When dealing with inheritance, as it can result in data loss. If we throw a derived type like std::invalid_argument but catch it by the value of a base type like std::exception, we lose any class variables that are not part of the base std::exception type.

This data loss is sometimes referred to as object slicing - any variables from the more specific types are sliced off when we copy the object to a base type that doesn't contain those variables.

We have an example of this below. Note the two try-catch blocks are almost identical - the only difference is the first block is catching by reference, whilst the second is catching by value:

1#include <iostream>
2#include <stdexcept>
3
4int Divide(int x, int y) {
5  if (y == 0) {
6    throw std::invalid_argument {
7      "Cannot divide by zero"
8    };
9  }
10  return x/y;
11}
12
13int main() {
14  try {
15    Divide(2, 0);
16  } catch (std::exception& e) {
17    std::cout << "Error 1: " << e.what();
18  }
19
20  try {
21    Divide(2, 0);
22  } catch (std::exception e) {
23    std::cout << "\nError 2: " << e.what();
24  }
25}

Because the second code is copying a std::invalid_argument into the simpler std::exception, it has lost all the data that is not part of the std::exception type.

In most compilers, this will include the custom error message we set, so the second catch block cannot determine the nature of the error:

1Error 1: Cannot divide by zero
2Error 2: std::exception

Given the nature of inheritance, multiple catch statements may match the specific type of exception thrown. When this happens, the first valid catch statement is what is activated. Consider this example:

1int main() {
2  try {
3    throw std::logic_error("Oops!");
4  } catch (std::exception& e) {
5    std::cout << "Exception: " << e.what();
6  } catch (std::logic_error& e) {
7    std::cout << "Logic Error: " << e.what();
8  }
9}

Here, we are throwing a std::logic_error exception. However, because std::logic_error is a subclass of std::exception, what we are throwing is also a std::exception.

As a result, the first catch statement can handle the error. Our output is:

1Exception: std::exception

In this case, the catcher for std::logic_error is a useless piece of code. It could never be reached because, under the rules of inheritance, if an object is a std::logic_error it is also a std::exception,

So, std::logic_error exceptions will always be handled by the first catch block. Our tools may warn us as such:

1main.cpp:7:12: warning: exception of type 'std::logic_error'
2will be caught by earlier handler

Because of this "first catcher wins" behavior, we should ensure the more specific catchers are first:

1int main() {
2  try {
3    throw std::logic_error("Oops!");
4  } catch (std::logic_error& e) {
5    std::cout << "Logic Error: " << e.what();
6  } catch (std::exception& e) {
7    std::cout << "Exception:" << e.what();
8  }
9}

1Logic Error: Oops!

We are not restricted to just throwing the std::exception types - we can throw and catch any object type.

This allows us to add fields to our errors that are specific to the needs of our application. Below, we introduce a user-defined AuthenticationError type:

1#include <iostream>
2
3class AuthenticationError {
4 public:
5  AuthenticationError(
6    std::string Email, std::string Password)
7      : Email(Email), Password(Password) {}
8
9  std::string Message{"A user failed to log in"};
10  std::string Email;
11  std::string Password;
12};
13
14int main() {
15  try {
16    throw AuthenticationError {
17      "test@email.com", "something-wrong"
18    };
19  } catch (AuthenticationError& e) {
20    std::cout << e.Message
21              << "\n  E-Mail: " << e.Email
22              << "\n  Password: " << e.Password;
23  }
24}

1A user failed to log in
2  E-Mail: test@email.com
3  Password: something-wrong

Typically, when creating custom exception types, we will want to maintain a hierarchy, similar to std::exception. This allows us to create more generic catch statements, by having them catch the more generic base types.

It's also common for teams just to use the standard library's implementation for this purpose, and to insert their types into the std::exception hierarchy.

When doing that, we should adopt the convention of making our error message available through the what() function, which we can override:

1#include <string>
2
3class AuthenticationError
4  : public std::exception {
5 public:
6  AuthenticationError(
7    std::string Email, std::string Password)
8      : Email(Email), Password(Password) {}
9
10  const char* what() const noexcept override {
11    return m_errorMessage.c_str();
12  };
13
14  std::string Email;
15  std::string Password;
16  std::string m_errorMessage{
17    "A user failed to log in"};
18};

Now, our exceptions can be caught by specific catch statements, tailored to handle our specific type:

1#include <iostream>
2
Unchanged Code|Show
3class AuthenticationError{/*...*/}
20
21int main() {
22  try {
23    throw AuthenticationError{
24      "test@email.com", "wrong"};
25  } catch (AuthenticationError& e) {
26    std::cout
27      << "AuthenticationError Handler:\n"
28      << e.what()
29      << "\n  EMail: " << e.Email
30      << "\n  Password: " << e.Password;
31  }
32}

1AuthenticationError handler:
2A user failed to log in
3  EMail: test@email.com
4  Password: wrong

But they can also be caught by higher-level, more general handlers:

1#include <iostream>
2
Unchanged Code|Show
3class AuthenticationError{/*...*/}
20
21int main() {
22  try {
23    throw AuthenticationError{
24      "test@email.com", "wrong"};
25  } catch (std::exception& e) {
26    std::cout
27      << "std::exception Handler:\n"
28      << e.what();
29  }
30}

1std::exception Handler:
2A user failed to log in

In this lesson, we introduced the notion of exception types, We covered std::exception and its hierarchy, alongside the creation and handling of custom exception types.

Key Takeaways

• Understanding the standard library has an exception hierarchy, based on std::exception and with derived types like std::logic_error and std::runtime_error.
• Techniques for throwing and catching exceptions, emphasizing the importance of catching by reference to avoid object slicing.
• The concept of inheritance in exception handling, ensuring that more specific exceptions are caught before more general ones.
• Best practices for creating and using custom exception types, integrating them into the std::exception hierarchy, and overriding the what() method.
• The application of the noexcept specifier in exception handling, indicating functions that won't throw exceptions.