This document outlines some core C++ concepts from a high level. Compilable source files demonstrating the concepts accompany each category, and can be built using the instructions below.
This is not meant to exhaustively document what you can do with C++. Instead, we aim to show a subset of the language that is simple, productive, and--yes--even fun to work with.
Table of Contents
- Building the Samples
- The Basics
- Object Creation
- Parameter Passing
- Types and Polymorphism
- Namespaces and Aliases
- Templates
- Type Inference
- Functional Programming
- Collections
- Asynchronous Programming
- Next Steps
The samples are built with a simple Makefile. So far, they have only been tested with clang on OSX. All code should work in VS2013 and with recent versions of gcc, as well.
# Build all samples
$ cd samples
$ make all
# Run a sample
$ ./SampleName.sampleWe tell the compiler to use C++14 so we can use std::make_unique, which (like many C++14 features) completes and adds consistency to features in C++11. We specify the C++ version we want by passing an option to the compiler in the Makefile.
In Xcode, you can set your “C++ Language Dialect” to C++14. In your build settings, it will look like the following:
The samples are generally organized in four sections. First, there is a comment explaining what the sample does. Second, we have the code that does the work. Third, the main function calls the functions to do work (sometimes, the second section just lives within the main function). Finally, we try to define any implementation details that aren't particularly relevant to the sample after main().
C++ code is generally broken into two files: the header (*.h) and the implementation (*.cpp). Header files tell the compiler what functions are available. Implementation files tell the compiler how those functions work. These two pieces are also called declarations (in header files) and definitions (in implementation files). When building C++ projects, it helps to keep these two ideas in mind, as you may run into issues with either missing declarations (e.g. no method by that name) or missing definitions (e.g. linker error: no definition for some method).
You can get access to functions written by others by including the header files that declare those functions in your own files. If you know that a function exists but are receiving compiler errors, make sure you have included the header file that declares the function. We tell the compiler to include files by name:
#include "projectFile.h"Compilers only want to see declarations once, but they will try to look at your header files every time you include them. To keep everyone happy, we explicitly tell the compiler to only read declarations from our header files once. We do this by writing the following at the beginning of our header files:
#pragma onceC++ provides two main methods of abstraction: classes and functions. A class can store data and provide functions to access and manipulate that data. Functions take data as input and return data as output.
Functions are declared like so:
float sum( float a, float b );Classes can control who has access to their data by declaring members either public or private. Classes are defined like so:
class ClassName {
public:
// Declare interface in public block.
private:
// Declare member variables in private block.
};Structs behave just like classes, only their members are public by default. We typically use them to indicate that a class is a simple container of public data, and use classes when we are going to provide an interface that manages the data.
Objects can live in two main places: on the stack, or in dynamic memory. Objects that live on the stack are destroyed automatically when their name falls out of scope. Dynamic objects will live forever if unmanaged, so we use stack-allocated pointer objects to manage their lifetime.
Objects that live on the stack are declared simply by type and name. They live for as long as the scope in which they are declared. That scope could be the body of a function, the lifetime of the parent class instance, or the block of a for-loop.
Stack allocation is the preferred way to store your variables as it is simple and does the right thing. It looks like the following:
Object object;
int number = 5;
auto value = 20.0f;Dynamic allocation is useful when you want to share a single object between multiple users and for handling polymorphism. When memory is allocated dynamically, you receive a pointer to that memory. Since we want to control how long that memory lives, we manage the pointer with a pointer object that lives on the stack.
C++ provides two particularly useful pointer objects: std::shared_ptr and std::unique_ptr. They are created with the std::make_shared and std::make_unique functions, respectively.
When we create managed dynamic objects, we are really creating two objects: (1) the pointer object on the stack, and (2) the pointed-to object in dynamic memory.
The pointer object is used as a handle (or pointer) to the dynamic object, and automatically manages the memory (de)allocation of the dynamic object.
Creating pointer objects looks like the following:
std::shared_ptr<Object> objectRef = std::make_shared<Object>();
std::shared_ptr<Object> derivedRef = std::make_shared<DerivedType>();
std::unique_ptr<Object> objectURef = std::make_unique<Object>();Generally, we will want to get the type information from the right side of our assignment (rather than describe it twice). The following code produces identical results to the above:
auto objectRef = std::make_shared<Object>();
auto derivedRef = std::shared_ptr<Object>( std::make_shared<DerivedType>() );
auto objectURef = std::make_unique<Object>();Since the dynamic memory Object in the above code is managed by a stack-allocated pointer object, we know the dynamic memory will be freed when our pointer object falls out of scope.
The principle behind various smart pointer objects is that they manage the lifetime of the pointed-to object. As an example, here is a simplified implementation of a unique pointer. The destructor of UniquePointer (~UniquePointer) is called when it falls out of scope, guaranteeing the pointed-to object is deleted at an appropriate time.
template <typename Type>
class UniquePointer {
public:
UniquePointer( Type *iPointer )
: pointer( iPointer )
{}
~UniquePointer() {
delete pointer;
}
private:
Type *pointer;
};The std::unique_ptr and shared_ptr objects are more sophisticated versions of the preceding code.
There are three basic rules for passing parameters
- Pass built-in types by value (float, double, int).
- Pass read-only parameters by const T&.
- Pass read-write parameters by T&.
In code, this ends up looking like the following:
float passByValue( float iValue );
void useObjectInfo( const Type &iObject );
void modifyObject( Type &iObject );
float number = 10.0f;
number = passByValue( number );
Type instance;
shared_ptr<Type> dynamicInstance;
useObjectInfo( instance );
useObjectInfo( *dynamicInstance );
modifyObject( instance );
modifyObject( *dynamicInstance );With the introduction of move-only types in C++11, we get a fourth rule that only occasionally comes into effect:
- When you have a move-only type that you want to pass ownership of to a new name, pass by T&&.
This occurs most commonly when you want to pass ownership of a std::unique_ptr (or an object that has a std::unique_ptr as a member) into a function. In that case you move the unique_ptr into the function, explicitly giving up control over the pointed-to object.
void takeOwnershipOfPointer( std::unique_ptr<Type> &&iPtr );
unique_ptr<Type> ptr;
takeOwnershipOfPointer( std::move( ptr ) );There is currently some debate around whether it is preferable to pass by T or T&& when passing a move-only type to a function that will take ownership of the object. T&& makes it clear that the caller should no longer own the object, and it makes sure the object is only moved once.
C++ provides a large number of built-in data types. You can also create your own data types as structs or classes.
Structs are useful for holding plain old data. Their members are public by default.
// Data holds a number of pieces of public data.
struct Data {
int member;
int another;
std::string name;
};Classes are used when objects should have more functionality than simply holding data. Their members are private by default.
Data in a class can be given a default initial value in the declaration. For built-in numeric types, we provide default values in order to avoid bugs from uninitialized data.
// Create a class named Type.
class Type {
public:
void doSomething();
private:
// Data members are declared private.
int someMember = 10;
float another = 5.0f;
};
// Create an object of type Type.
Type thing;Classes and structs can inherit from other classes. Base classes should have a virtual destructor. Any methods that the derived class can override must be marked as virtual in the base class and with override in the derived class.
// Base class has some virtual methods, and a virtual destructor.
class Base {
public:
// Mark destructor as virtual so derived destructors get called.
virtual ~Base() = default;
// draw() can be overridden since it is virtual.
virtual void draw() const;
// doSomething() cannot be overriden, since it is not virtual.
void doSomething();
};
// Derived inherits the public interface from Base
class Derived : public Base {
public:
// virtual method that is overridden is marked with override.
void draw() const override;
};When dealing with large projects or multiple sets of code, it is useful to group sections of code together under a common name. Similarly, it can be useful to have a shorthand for complex types.
Namespaces allow us to group sections of code together and write natural names for the types within each section.
In general, we declare a namespace for the project we are working on. This helps prevent naming conflicts between different sections of our code. It also makes it clear where different pieces of functionality are coming from when refactoring and debugging.
We declare namespaces like so:
namespace project {
void function();
} // namespace projectAnd access members in the namespace as follows:
In a header file (.h):
project::function();In an implementation file (.cpp):
using namespace project;
function();
Aliases allow us to derive clear names from verbose types where needed. Often, you can use auto to capture a verbose type. However, when you need to store the type as a class member or receive it as a function parameter, aliases can help keep things legible.
When declaring aliases, make sure you are inside your own namespace or in an implementation file to avoid naming conflicts elsewhere.
We write aliases as follows:
using ObjectRef = std::shared_ptr<Object>;
using Callback = std::function<void (float, const std::string &)>;
// Templates can be aliased, too:
template<typename T> using Handle = std::shared_ptr<T>;
using ThingHandle = Handle<Thing>;Templates allow us to write code that can operate on any type. They enable compile-time polymorphism: the template is filled out by the compiler for each type that uses it.
While you may not write many templates in day-to-day programming, you will use them whenever you create a dynamically allocated object with make_shared or store objects in an STL container.
Template functions and types do their best to derive their type from the parameters you pass to them or their constructors. Sometimes, however, the compiler can't effectively deduce all the template types from the parameters received. In those cases, you must help the compiler along by filling in the template parameters yourself.
When needed, template parameters are passed in using angle-brackets:
std::vector<Object> collection;
std::shared_ptr<Object> reference;
auto result = std::max<float>( 10, 20.0 );C++ is a statically typed language, which means that every object’s type is set at compile time. Because everything needs to have a known type, it is possible for the compiler to infer the type of one object from the known type of another object. C++ uses type inference when we work with templates and when we use the auto keyword.
auto tells C++ to use the type of the right-hand-side of an expression as the type of the left-hand-hide of that same expression. It allows us to avoid redundant declarations and also to store variables that would otherwise have unwieldy type names. Like any type declaration, auto can be qualified with const, &, or *. That way, const auto& lets us get a reference to the rhs instead of making a copy.
The following two lines of code produce identical results, but I find the second line much easier to read and comprehend.
std::chrono::system_clock::time_point now = std::chrono::system_clock::now();
auto now = std::chrono::system_clock::now();C++ has a lot of things that can behave like a function. Fortunately, it also has a function object that can refer to any of them.
std::function<> is a template that receives a function signature as its parameters. Function objects created by filling out the template can refer to any callable type that matches the signature.
Some possible function object declarations:
std::function<void ()> void_fn;
std::function<float (float)> lerp_fn;When a function object doesn't refer to a function, testing it like a boolean returns false. Make sure you have a function to call from your function object before using it.
if( void_fn ) {
void_fn();
}
if( lerp_fn ) {
auto result = lerp_fn( 5.0f );
}Lambda expressions produce function objects.
Lambdas are closely related to functions. Functions receive parameters when they are called; Lambdas receive parameters when you create them, and return a function.
Lambda syntax varies a bit from other functions. It begins with the capture block [] which is where you pass parameters to the lambda. Following that is the familiar () of a function, where you declare the parameters that the resulting function will receive. Finally, the familiar function block {} contains the code the function will run.
std::function<void ()> fn = [] () {};In general, the compiler can deduce the return type based on what you write inside the function body. In the cases where it can't, you can write the return type using the trailing return style:
auto fn = [] () -> string { return someOtherFunction(); };Parameters passed to the lambda take the same form as parameters passed to a function. If named in the argument block, parameters will be passed by value. If preceded by an ampersand, they will be passed by reference.
int i = 5;
int j = 0;
// Capture i by value and j by reference.
// i will always have the same value in our lambda.
auto fn = [i, &j] () {
j += i;
};
i = 0; // doesn't affect fn's i, since we captured by value.
fn(); // j now equals 5, since variable is referenced in fn.As mentioned earlier, lambdas create function objects. Below, we will create our own function object from scratch and then write a lambda that produces an equivalent function object. Hopefully it will illuminate what goes on behind the scenes of lambda creation and also demonstrate how useful lambda syntax is.
class Sum {
public:
// The constructor and member variables are analogous to the lambda capture block '[]'
Sum( float iBase )
: base( iBase )
{}
// The () operator is equivalent to the lambda function definition.
float operator() (float iValue) { return base + iValue; }
private:
float base;
}
float base = 11.0f;
auto sumFn = Sum( base );
auto lambdaEquivalent = [base] (float iValue) { return base + iValue; }
// We can use either function object in the same way:
sumFn( 5.0f ); // => 16.0f;
lambdaEquivalent( 5.0f ); // => 16.0f
C++ provides a large number of container types in the STL.
The container interface is very similar across types, though certain containers only implement parts of the interface.
Some common methods include:
size()- returns how many elements the collection contains.empty()- returns true if the collection is empty.clear()- removes everything from the collection.erase()- removes a range of elements from the collection.assign()- replace the contents of the collection with an element or range.insert()- add an element or range to the collection at a specified position.at()- returns the element at the specified index or key.
std::vector<Type>
Vectors are the go-to sequential container. They can store a variable number of elements and provide fast iteration and index-based lookup.
Vectors store their contents in a contiguous block of memory and resize themselves as needed. When a vector is resized, it copies the old contents to a new block of memory.
Due to the way it is laid out, we typically only modify the end of a vector.
vector<float> numbers;
numbers.push_back( 10.0f );
numbers.push_back( 0.5f );std::array<Type, Size>
Arrays store a fixed number of elements. In contrast to a C array, C++’s array type knows how long it is.
std::array<float, 5> array;
array.at( 4 ) = 10.0f;std::list<Type>
If you frequently need to modify the front of a container, consider using a std::list. It is implemented as a doubly-linked-list, which makes it cheap to change at arbitrary positions, but it is comparatively slow for iteration and index-based lookup compared to std::vector.
Associative containers let you look up contents by a key other than their index (and may not have an index). The data is associated with a key of your choosing.
Common methods:
count()- returns the number of times the key appears in the collection. (0 or 1 for the containers we use.)find()- returns the iterator for the key/value pair in the collection. (collection::end() if not found)
std::map<Key, Value>
A map stores key-value pairs, allowing you to look up contents by a key other than an index. Maps are stored as a tree structure.
std::map<std::string, float> numbers;
numbers["five"] = 5.0f;
numbers.insert( make_pair( "ten", 10.0f ) );
float val = numbers.at( "five" );std::unordered_map<Key, Value>
An unordered_map is a map structure that is implemented using a hash table. With large collections, it will generally be faster to look up elements in an unordered_map than a map.
You use unordered_maps the same way you use maps.
std::set<Key>
A set is a collection that guarantees each value it contains will only be stored once. They are useful for keeping track of things like selections a user made or possible next moves in a game.
There are many approaches to and reasons for asynchronous programming. Generally, we use asynchronous handling when we want to get some big chunk of work done, but need to keep our animations running smoothly and UI responsive while it happens.
C++ constructs for enabling asynchronous programming include async calls with futures, and threads with atomics and mutexes. Futures and async not only reduce the chances for race conditions, they also enable you to control the re-entry point for asynchronous code. Threads are useful for continuously-running functions, and especially when you don't need results from the function back on the main thread.
std::async runs a function in parallel with your current code. It returns a future that you can use to check on the progress of the function.
In addition to letting you check whether your asynchronous function has completed, futures let you get the return value of the function when it is complete.
// Run a function asynchronously.
std::future<string> data = std::async( std::launch::async, some_function );
// Time passes as we do something else...
// Check in on our data.
if( data.valid() ) {
auto status = data.wait_for( 1ns );
// If we are all done here,
if( status == std::future_status::ready ) {
// Use our data.
string c = data.get();
}
}Note that you must store the future returned by std::async if you want the code to actually run asynchronously. This is true even if you don't care about the return value.
C++11 provides a std::thread object that executes a function in parallel to the main thread. If you want to have a continuously running process with minimal communication back to the main thread, it can make sense to use a thread.
If you want to do some work in parallel to your main thread and get the result back, it usually makes more sense to use a std::future.
std::thread t;
t = std::thread( some_function );If you want to share data between threads, you need to synchronize it using some means. C++ provides two options for synchronizing data: the std::atomic<> template and std::mutex.
Atomic operations cannot be divided into multiple steps, which makes them safe to use across multiple threads. The std::atomic template wraps a piece of data so that operations on that data are atomic.
std::atomic<float> sharedData;Mutexes can be used to protect data as well. Only one thread can lock a mutex at a time, so content correctly protected by a mutex can be safe to share between threads.
std::mutex dataMutex;
DataType data;
// Returns a copy of data.
// Lock to make sure data doesn't change during copy.
DataType getData() {
std::lock_guard lock( dataMutex );
return data;
}
// Change a member of data.
// Lock to make sure nothing else changes data until after we finish.
void changeDataMember( float iA ) {
std::lock_guard lock( dataMutex );
data.a = iA;
}The concepts shown above and in the sample code are informed by Scott Meyer’s Effective Cpp books as well as Herb Sutter’s Guru of the Week column. Additionally, Bjarne Stroustrup’s recent A Tour of C++ was influential in setting the tone of the samples. If you are interested in learning more in breadth and/or depth, have a look at those sources.
For more breadth and depth, start here:
For detailed reference materials, start here: