A Whirlwind Tour of C++

Author: Elizabeth Dinella Editor: Josh Hanson

C++ was originally created in the 1980s as an extension of the C language to add support for classes. Since its inception, C++ has grown enormously to include a host of features such as exception handling, templates, an expansion of library functions, stronger type checking, inheritance, and other object-oriented concepts. This article will serve as a primer for C++ in this course as well as a source of resources for a more complete documentation.

This article is organized into three parts: Classes & Objects, General Use C++, and C++ for LLVM. Classes & Objects will cover classes in C++, access modifiers, and object-oriented concepts such as single and multiple inheritance, and polymorphism. C++ for LLVM will serve as an introduction to the LLVM API which will be heavily used in the assignments for this course. Conversely, General Use C++ will detail features that are not specific to the LLVM API, but will likely be frequently used in the assignments for this course.

You are encouraged to try out different C++ features in this Online C++ IDE as you read through the rest of this article.

Part 1: Classes & Objects

Classes are an expanded concept of data structures: like data structures, they can contain data members, but they can also contain functions as members. An object is an instantiation of a class. In terms of variables, a class would be the type, and an object would be the variable.

Readers that are familiar with other languages such as Java are presumably familiar with many of the object-oriented concepts discussed in this section. Nevertheless, there are some notable features that may be new to readers such as multiple inheritance, compilation, destructors, and manual object management.

C++ Class Syntax

Class implementation is usually split between two files: a header and an implementation file. The header file typically contains the class declaration including list of member variables and member function headers. A function header includes the name of the function and tells the compiler what type of data it expects to receive and what type of data it will return. The function body with the actual instructions to execute when called are written in the implementation file. The header and implementation files typically share the same name and end in “.h” and “.cpp” respectively.

Here is an example of a Time class split between a header and implementation file. In this example, notice that the header file is referenced in the implementation file using:

#include "Time.h"

This line tells the C++ compiler to use in Time.cpp the function and class declarations written inside of Time.h. Suppose that we additionally include a file that itself also includes Time.h. Now, we have multiple definitions of the members and functions defined in Time.h. In order to prevent this, C++ has a mechanism to prevent headers from being compiled more than once. These mechanisms, called “include guards”, are illustrated in the following lines in Time.h:

#ifndef TIME_H

#define TIME_H
.
.
.
#endif

These lines tell the compiler to compile the Time.h functions only if they have not been compiled yet, to prevent double declaration.

C++ Scope Resolution Operator

To access a member function, member variable, or member class, you can use the double colon (::) operator. Examples include ClassName::functionName, ClassName::variableName, namespace::ClassName, or EnumName::Member. Within class scope, the member functions and variables are accessible without explicit scope resolution; but note that in implementation files, the scope resolution operator must be used when defining functions that are declared in the header file.

Class Compilation

The compilation of a C++ program consists of two stages: compilation and linking. Compilation includes preprocessing (macro and include expansion), transforming the program into an intermediate language such as LLVM IR, optimizing the intermediate program, and generating object code such as x86 assembly. Linking puts together one or many object code files into an executable program. In this stage, the compiler matches function calls with their definitions and ensures that each function that is called has exactly one definition. Linking is a common source of errors for users that are new to C++ classes.

Access Specifiers

Access specifiers identify access rights for the members they are applied to. Access can be either private, public, or protected. By default, a class has private access to all of its members. This means that members of the class are only accessible from within members of the same class. Conversely, a public member variable is accessible from everywhere that the object is visible. A protected member variable is accessible from members of the same class and members of its child classes.

These rules change slightly when the friend keyword is introduced. This feature will be detailed in the Friendship section below.

Constructors and Destructors

Each class has a specific constructor function that is called each time a new object of the class is created. The constructor has the same name as the class and is typically used to initialize member variables and/or perform some setup. The default constructor is a special constructor that is called when an object of the class is declared but is not initialized with any arguments. For example, consider the following declarations of the Rectangle class from here:

Rectangle r1; // ok, default constructor called

Rectangle r2(); // oops, default constructor NOT called

The default constructor is called for r1. Note that r1 is not even constructed with an empty set of parentheses. This is because the empty set of parentheses make r2 a function declaration instead of an object declaration: it is a function that takes no arguments and returns a value of type Rectangle.

Another special constructor type is the copy constructor. This is executed when an object of the same type is passed to the constructor. By default, it performs a shallow copy of all member variables. If a class has a pointer type member variable, a shallow copy may not be enough. Multiple classes may share the same object the pointer refers to which can cause issues if this is not what the developer intended. If a different semantics is preferred, the developer can write their own copy constructor with the signature:

Rectangle::Rectangle(const Rectangle& x)

Likewise, the destructor function is called each time an object of the class is destroyed. It has the same name as the class (and the constructor), but is preceded with the tilde sign (~). It is typically used to perform some cleanup of heap allocated memory.

When a class has dynamic member variables, these member variables are default-constructed when an instance of the class is created (when the constructor is called), and then re-constructed when they are initialized. To avoid this wasteful double-construction of member variables, constructors can use initialization lists before their bodies, ensuring that the member variables (width and height below) are only constructed once:

Rectangle::Rectangle(int x, int y) : width(x), height(y) { }

this keyword

The keyword this represents a pointer to the object whose member function is being executed. It is used within a class’s member function to refer to the object itself.

It is often used to resolve ambiguity between a member variable of the object executing the call and an object of the same class passed as a parameter. An example illustrating this is shown in the member function isitme.

Static members

A static member of a class is a variable that is shared between all objects of the class. Likewise, a static member function is not associated with any specific object. Therefore, they do not have access to a this pointer and are instead accessed by the class’s name, like so:

ClassName::staticMemberVariable

Friendship

The friend keyword defines access relationships between functions and classes. If a class A is a “friend” of class B, member functions in class A can access private and protected members of class B. Friendship is defined in the class giving access (Class B in this example). An example of friend classes is shown here.

Similarly, if a function is a “friend” of class B, it can access private and protected members of class B within its function body. An example of a friend function is shown here.

Inheritance

Inheritance creates an “is-a” relationship between the derived and base class. The derived class inherits member variables and functions from the base class. It can also include its own member variables and functions. An example can be found here. The Rectangle class is derived from the base class, Shape.

Recall that derived classes can access protected members in the base class.

As shown in the above-linked example, inheritance is specified using access modifiers: public, protected, or private:

class Rectangle: public Shape { ... }

A public inheritance (as above) behaves as expected: the variables inherited from the base class keep their access status. Conversely, in private inheritance, all inherited members become private. In protected inheritance, public members become protected and all other members are unchanged.

Multiple Inheritance and Virtualism

As described in the introduction, C++ was originally created to add support for classes to C. Thus, C++ includes many object-oriented concepts that did not exist in C. In this section, we assume that the reader has a basic understanding of classes and inheritance.

Unlike Java, C++ supports multiple inheritance. This means that an object can have one or more parent classes. At first glance this seems like a useful feature that should exist in all programming languages that support inheritance. However, multiple inheritance is often a source of confusion and implementation troubles.

Consider the infamous “Diamond Problem”:

Diamond Inheritance

Both B and C inherit from A and thus contain A’s member variables. This could lead to ambiguities and duplicate copies of A’s member variables in class D. To solve this, we can use Virtual Inheritance as follows:

Class B : public virtual A { }

Class C : public virtual A { }

This structure ensures that a class D that inherits from B and C would only contain one copy of A’s member variables. Read Understanding Virtual Tables in C++ to gain a deeper understanding of how this works under the hood.

Polymorphism

Class inheritance provides many useful features. One of these, is that a pointer to a derived class is type-compatible with a pointer of the base class. More concretely, if class B is derived from base class A, a pointer to class A can also refer to an object of type class B. This is intuitive as class B has an “is-a” relationship to class A.

Part 2: General C++

Templates

C++, like other statically typed languages, requires the developer to specify types in nearly all declarations. However, this can lead to unnecessary repetition of code that is somewhat type independent. Consider the code for the STL vector. Regardless of the element type, the code to add an element, remove an element, calculate size, and even sort are likely identical. Templates provide a way to write programs that are generic or independent of any particular type. Below is an example of a templated swap function. So long as the types of n1 and n2 are equivalent, this function will be applied successfully.

template <typename T> void Swap(T& n1, T& n2) {
    T temp;
    temp = n1;
    n1 = n2;
    n2 = temp;
}

C++ Type Checking

In the introduction, we briefly noted that C++ has stronger type checking than C. In this section, we will describe how the stronger type checking manifests in differences in enums and void pointers. In C, an enum is simply an alias to integer types. However, the C++ enum defines an actual type. So, the following will result in a type error in C++, but not in C.

enum direction { North, South, West, East };
direction myDir = 3;

Another source of type unsafety in C is the implicit promotion of void pointers. Although malloc returns a void*, the following program compiled with gcc produces no errors or warnings.

#include <stdlib.h>

int main() {
    int* arr = malloc(5 * sizeof(int));
}

However, compiling with g++ produces the following error:

void_ptr.cpp:5:5: error: assigning to 'int *' from incompatible type 'void *'
        arr = malloc(5 * sizeof(int));
              ^~~~~~~~~~~~~~~~~~~~~~~
1 error generated.

In C++, casting the void* to int* is required.

Command Line Arguments

In this course, we will often pass additional information from the user when the program is run. The syntax to implement this in C++ is exactly the same as in C. However, it is worth reviewing as it will be used quite frequently in the assignments.

int main(int argc, char* argv[]) {
    if (std::string(argv[1]) == std::string("blue")) {
        std::cout << "my favorite color is blue too!" << std::endl;
    }
}

References and Pointers

C and C++ support pointers. A pointer is a variable that stores a memory address. In order to access the memory location that it stores or points to, it must be dereferenced with the * operator. The shorthand to dereference a pointer and access a member function or variable is

ptr->member

instead of the more verbose version:

(*ptr).member

In contrast, a reference is an alias to another variable. References are often used to pass objects to a function to persist modifications. By default, objects are passed by value, meaning they are copied at call time; any modifications of the object that occur in the function body will be applied to the copy rather than the original value. References allow you to bypass this behavior to work directly on the original objects rather than on their copies, or simply to avoid making expensive copies.

Const

A variable declared as const is constant and cannot be modified. If a function is declared as const, its body may not modify the calling object.

In addition to mutability safety, the const and reference constructs can be combined to employ efficient parameter passing. As noted above, passing an object by reference is faster as the runtime environment does not need to make a copy. However, if passed as reference, modifications will persist to the original object. Passing an object by constant reference allows to make efficient function calls while ensuring that the object will not be modified.

Local type inference

The type of variables can often be deduced automatically within a local scope using the auto keyword. This mainly just allows C++ code to be more concise, and can also save time when refactoring.

For-each loops

In addition to the traditional for loop structure, C++ also allows for a for each loop that works cleanly with iterators. The syntax is as follows:

for (T element : collection) {
    ...
}

The loop iterates through each element of the collection. This allows for the natural instructions: “for each … in …. do ….”.

It is common to use for-each loops in combination with type inference when the type of each element is too verbose to write out explicitly. In some cases you can also specify reference semantics so that you’re accessing the elements themselves rather than copies of each element.

For example, after executing the below lines:

std::string s = "hello";
for (auto& c : s) {
    c = 'a';
}

the string s will be set to "aaaaa".

new and delete keywords

C++ includes dynamic memory features that did not exist in C. The traditional malloc and free keywords are still available in C++; however, it is widely accepted that new and delete should be used in their place.

new and delete have slightly different syntax for scalar types and arrays.

int* foo = new int;        // allocates memory for a single int
delete foo;                // deletes memory pointed to by foo
int* baz = new int[100];   // allocates memory for 100 int array
delete [] baz;             // deletes memory pointed to by baz

delete with [] indicates that the entire array should be deleted rather than a single element.

Note: normally when an object’s scope is left, its destructor will be called. However, if the object was dynamically declared using the new keyword, then it will persist in memory beyond the end of its scope. The cleanup of dynamic memory is the developer’s responsibility and is often a source of memory leaks.

C++ Standard Library

The standard libraries in C++ are a superset of the standard C libraries. In addition, the C++ libraries include a diverse and powerful set of features. In this document, we will detail, arguably, the most important features: iterators, strings, I/O streams, and data structures.

Iterators

Iterators provide an interface to traverse containers (vectors, sets, maps, etc.) and view / edit specific elements stored in such containers. In many ways, iterators are a generalization of pointers. Iterators in C++ abstracts the traversing of a container from the container itself.

// Accessing the elements of a vector without using iterators
for (j = 0; j < 4; ++j) {
    cout << v[j] << " ";
}
// Accessing the elements of a vector using iterators
for (i = v.begin(); i != v.end(); ++i) {
    cout << *i << " ";
}
Strings

This powerful library simplifies the tedious character array manipulation tasks left to the developer in C. C++ strings are a wrapper around the array of characters that would be explicitly manipulated in C.

Developers can:

  • Create and delete strings using constructor and deconstructors
  • Assign values to strings using the = operator
  • Compare contents of strings using the == operator
  • Iterate over strings using C++ iterators
  • Access particular characters using the [] operator
  • Erase a substring from a string
I/O Streams

The IOStream library is intended to replace the stdio library for an easier and more flexible input and output developer experience. A stream has some source or sink. The IOStream library supports standard input, standard output, standard error, a file, or an array of characters. The >> and << operators are conveniently overloaded to simplify the use of IOStreams. The following example shows the basics of writing to the console and a file:

#include <iostream>
#include <fstream>

int main() {
    /**
     * Writing to the console:
     * cout is defined to access to screen by default
     * endl (rather than std::end) appends a newline (\n)
     */
    std::cout << "Hello World!" << std::endl;

    //writing to a file
    std::ofstream myfile;
    myfile.open("out.txt");
    myfile << "Hello File!" << std::endl;
    myfile.close();
}

However, according to the LLVM style guide, IOStreams are forbidden. Instead, developers are encouraged to use LLVM’s raw_ostream. For more details, we refer the reader to the C++ for LLVM section below.

Data Structures

The C++ Standard Library (STL) defines many useful containers that will feel familiar to readers with experience in Java or Python. The STL includes vectors, lists, queues, stacks, sets, and maps. Each of these containers can be traversed using iterators. Additionally, many containers share common functions. This manifests in a seamless developer experience regardless of the container or element type. For example, size(), empty(), max_size(), ==, !=, swap, a default constructor, copy constructor, and assignment constructor are shared by all STL containers. Here is a detailed list of STL containers and properties.

Smart Pointers

The C++ Standard Library (STL) defines a few different types of smart pointers that take care of their own dynamic memory allocation and deallocation, making them safer than raw pointers, which can cause memory leaks. Smart pointers automatically clean up their underlying owned memory when they leave scope. This is done implicitly in their destructors. Here are the three main types of smart pointers:

  • std::unique_ptr - Use this when you don’t intend to hold multiple references to the same object. For example, use it for a pointer to memory which gets allocated on entering some scope and de-allocated on exiting the scope. Since only one unique pointer can hold a reference to an object, use std::move() to transfer object ownership from one unique pointer to another.

  • std::shared_ptr - Use this when you do want to refer to your object from multiple places, and do not want your object to be de-allocated until all these references are themselves gone. Note that shared pointers do some internal bookkeeping called reference counting to count the number of active references to the object that they hold. This can be a performance drawback.

  • std::weak_ptr - Use this when you do want to refer to your object from multiple places - for those references for which it’s ok to ignore and deallocate (so they’ll just note the object is gone when you try to dereference).

Part 3: C++ for LLVM

LLVM data structures

C++ STL data structures are for general-purpose use and have performance that is platform dependent.

In order to create very fast specialized but platform independent structures, LLVM created its own data structures. To the user, these structures feel similar to STL data structures, but have some notable differences. For each of the following, the developer can choose to use the STL data structure or the LLVM specialized structure. This choice should be made with efficiency and their use cases in mind.

In general, STL structures are variable size. To conserve memory, on creation, the underlying array does not have any space allocated for elements. Once the user calls push_back or insert, memory is allocated for the element. Each time the user inserts an element to a container that is at capacity, the memory allocated for the structure is doubled. As the number of elements increases, the frequency of allocation calls decreases. So, for a large n, the STL version is often a good choice of data structure. For a small number of elements, LLVM provides alternatives.

  • LLVM SmallVector is optimized for a small n. It is created with some number of elements in place. In this way, it avoids allocation when the actual number of elements is below that threshold. Inserts below that threshold will be much faster to a SmallVector than an STL vector.

  • LLVM DenseMap is an unsorted alternative to the STL map. The STL map guarantees that iteration order of the container is the same as the insertion order. The LLVM DenseMap keeps keys and values next to each other in memory in order to speed up lookup.

  • LLVM StringMap is a specialized structure with only strings as keys. This is useful as string keys are difficult to support efficiently. Long strings are inefficient to compare and copy. The LLVM StringMap supports an arbitrarily long key value. Like the LLVM DenseMap, the iteration order is not guaranteed.

  • LLVM SmallSet is an unsorted alternative to the STL set. Because STL sets have a defined order, searches are O(log n). Conversely, in the LLVM SmallSet searches are O(n). Additionally, as the LLVM SmallSet does not have a defined order, it cannot be iterated over. For a small n that does not need to be regularly queried, LLVM SmallSet could be an efficient choice.

raw_ostream

The raw_ostream is an LLVM alternative to iostream. Instead of using std::cout the developer should pipe output to outs(). Analogously, the alternative to std::cerr is errs().

int main() {
    outs() << "Hello World!\n"
}

Dynamic Casting

A dynamic cast converts pointers to a more specific type in its class hierarchy at runtime.

Consider the “is-a” relationship between an LLVM AllocaInst and an LLVM Instruction. In the assignments, we will often write control flow similar to the following structure:

if (auto *AI = dyn_cast<AllocationInst>(Val)) {
    // ...
}

This allows us to handle specific instructions (BinaryOperator, CallInst, AllocaInst, etc.) differently. The dyn_cast operator checks if the parameter is an instance of the templated type. If so, it returns a pointer to the downcasted class. If not, dyn_cast returns a null pointer.

Note: that you must use dyn_cast, not the C++ built-in dynamic_cast, when casting objects from the LLVM library.

Acknowledgements

We thank Kihong Heo, Greg Kofman, and Pardis Pashakhanloo for suggestions to improve this article.