C++ allows you to write clean, safe and fast code, and writing it is even easier than ever thanks to new features in C++11 and C++14. This article highlights a couple of tips.
Comments & ‘Cleverness’
Code gets read a lot more often than it gets written, so make sure that your code can be read and understood in the future, by yourself and by other developers. This means that you have to add comments, and that you should avoid trying to make code as terse and ‘clever’ as possible; because that hurts readability.
Memory Management
In modern C++ you should avoid any explicit calls to delete operators or functions to deallocate memory such as free(). Code that relies on such explicit calls is susceptible to memory leaks. For example:
void foo(size_t size) {
char* buffer = new char[size];
// ...
delete [] buffer;
}
The code is straightforward; it allocates memory, does something with the memory, and then deletes the memory. It uses a ‘naked’ pointer, which you have to manage yourself. This piece of code causes a memory leak if an exception is thrown before the delete operator is called. You should avoid using naked pointers, instead use encapsulation objects that automatically handle the memory. For example, the above code should be written using a container such as a vector:
void foo(size_t size) {
std::vector buffer(size);
// ...
}
Using a container, there is no need to free the memory, the destructor of the container will handle all that for you. Similarly, if you need dynamically allocated objects, do not store them in naked pointers, instead use smart pointers such as std::unique_ptr, which is not reference counted, or if you need shared ownership, std::shared_ptr, which is reference counted. unique_ptr will automatically delete the object when the unique_ptr goes out of scope. An object wrapped in a shared_ptr is automatically deleted when the reference count goes to zero.
When you use unique_ptr, use std::make_unique<>(). Similarly, use std::make_shared<>() when you use shared_ptr.
One key point to remember: never use std::auto_ptr, it has officially been deprecated because of its problems. One of its problems is that it is forbidden to store them in STL containers because auto_ptr does not meet C++98/03’s requirements for container elements. unique_ptr and shared_ptr don’t have any of these problems and are perfectly safe to use in STL containers.
The auto Keyword
Use auto to let the compiler deduce the type of variables automatically. This helps a lot with complicated types that are hard or even impossible to write. For example:
std::map<std::string, std::vector> myMap;
for (std::map<std::string, std::vector>::iterator iter = begin(myMap);
iter != end(myMap); ++iter)
{
//...
}
This is complicated, difficult to write and error prone.
Note that since C++11 it’s recommended to use the non-member functions std::begin(), end(), cbegin(), cend(), rbegin(), rend(), crbegin(), and crend() instead of the member functions.
Using the auto keyword the above can be rewritten in a very clean way:
for (auto iter = begin(myMap); iter != end(myMap); ++iter) { ... }
Using the range-based for loop, the code can be written even more elegantly:
for (auto&& item : myMap) { ... }
Or:
for (auto& item : myMap) { ... }
Or:
for (const auto& item : myMap) { ... }
Never write the following, unless your container only contains primitive types:
for (auto item : myMap) { ... }
If you use this for containers with non-primitive elements performance will suffer because item will be a copy of each of the elements in the container.
nullptr
Do not use NULL for pointers, use the nullptr keyword. NULL is a macro for 0 which can sometimes result in problems that are hard to track down. nullptr is really a null pointer type and is not just a macro for 0.
Move Semantics
Move semantics since C++11 helps with performance, allowing objects to be moved instead of copied. It is based on rvalue references, represented as &&. C++ automatically uses move semantics in cases where the source object will cease to exist. You can explicitly move an object by using std::move(). The STL also supports move semantics. To add move semantics to your classes you only need to implement a move constructor and move assignment operator which have the following prototype:
MyObject(MyObject&& src); // Move constructor
MyObject& operator=(MyObject&& rhs); // Move assignment operator
Lambda Expressions
Lambda expressions are a powerful addition to the C++ language. One huge benefit is that they make it much easier to use STL algorithms. For example, you can use the count_if() algorithm in combination with a simple lambda expression to count all elements in a vector that are less than 0.
std::vector vec;
auto count = std::count_if(cbegin(vec), cend(vec),
[](int i){ return i < 0; });
Investing time in learning about lambda expressions will pay itself off quickly.
Threading
You should read up on the threading functionalities introduced in C++11. One useful and often overlooked function is std::call_once(), used to make sure that a specific callable entity is executed exactly one time, no matter how many threads call it simultaneously. This can for example be used to make a compact and thread safe implementation of the singleton pattern.
Strings
Text matching, parsing, and searching can be tricky and complicated, especially when it should work with Unicode strings. You should avoid writing your own functions to perform such operations, instead use classes from <regex> such as std::wregex.
C++11 supports raw string literals. They are extremely useful in the context of regular expressions. For example, the following regular expression searches for spaces, newlines, and backslashes:
string s = "( |\\n|\\\\)";
The last 4 backslashes are correct. If you want to match a backslash, it needs to be escaped in the regular expression, thus \\. If you want to write \\ in a normal string literal, you need to escape both backslashes, resulting in 4 of them. Using raw string literals you simply write:
string s = R"(( |\n|\\))";
R”( and )” denote the boundary of the raw string literal.
Random Numbers
If your application needs to generate random numbers, don’t use the old C style srand()/rand() functions. Their randomness is not good. Instead, use <random>, which contains classes to generate random numbers, such as a Mersenne Twister, and supports different mathematical distributions, for example a uniform distribution, a Poisson distribution, and so on.
Further Reading
Take a look at “Professional C++ 2nd Edition” if you are determined to master C++, including all C++11 features.
It has been a lot of work, but I’m proud to announce my fourth edition of “Professional C++” 🙂
