As is tradition in every computer science text, the book starts with a “hello world” program designed to highlight core behaviour of the language. Here’s the source, modified – per my own tradition – to print a slightly modified message:
// helloworld.cpp
#include <iostream>
int main() {
std::cout << "Hello cruel world" << std::endl;
return 0;
} That’s all well and good, but I can’t do anything useful with this until I compile it, and for that to happen I need a C++ compiler. As it happens I already have g++ on my system, but – for no particular reason – I’ve decided to use clang. Installing clang on ubuntu is pretty straightforward:
sudo apt install clangNow that I have a compiler, I need to actually compile it. Here’s the command:
clang++ --std=c++20 helloworld.cpp -o helloworldThe --std=c++20 flag tells clang what version of C++ I’m using, and the -o flag is used to specify the output file.
In practice, this isn’t the commmand I actually use. I don’t want my binaries to build into in the same folder as my source code, so I keep all the source code in src and the binaries in bin. Additionally, I’ve done a bit of tinkering and installed clang 15, and I want to use that as my compiler rather than clang 14 (which is what ubuntu ends up installing with the command above). I could tinker with links to set my default clang to version 15, but at some level I think I prefer the command to explicitly specify the compiler version, so my command would actually look like this:
clang++-15 --std=c++20 ./src/helloworld.cpp -o ./bin/helloworldAs a side benefit, structuring the project this way makes it much easier to avoid accidentally commiting binary files to the git repository. All I have to do is add the bin folder to my .gitignore file. Fantastic.
That being said, I don’t actually want to type this command for every source file. Instead, because I’m a fundamentally lazy person, I’ve written a Makefile that takes care of that for me, and also renders the markdown files to html. So really the only command I ever use is simply:
makeAnyway, once the source has been compiled, I can invoke the executable at the terminal, and out pops the message:
Hello cruel worldExcellent. The basics are working.
Thankfully, I’ve written enough C++ code in the past that nothing about this surprises me. A few very basic syntactic notes in case I ever happen to share this with someone else coming from R.
//;main() function is special: that’s the entry point to your program. This is the function that gets called whenever you invoke the executable at the command line.int main() { // blah } to specify that the output is an integerdplyr::filter), and :: functions similarly here. When I write std::cout, I’m roughly saying “cout, which lives in the std namespace”.#include line is a “preprocessor directive”, used to specify meta-information about the program. In this case, it’s telling the preprocessor to take everything in the <iostream> header file and make it available to the program. Without it, I can’t do any input/outputimport <iostream>; instead of using the #include preprocessor directive. This is because C++20 introduced support for modules. I’m being a bit old-fashioned here, but for a good reason. When I installed clang with the command shown earlier, what I actually got on my system is clang version 14, and the support for modules in clang 14 is limited. I kind of don’t want to get bogged down with those details here, so I’m going to use #include directives instead of trying to get modules working.Going a little deeper:
std::cout refers to “standard out” (stdout), basically the place where we write output to the console. The metaphor used in the book is to think of it as a “chute” where you toss text.<< operator is used to “toss data into the chute”, so std::cout << "hello" passes the text string "hello" to the standard output. As the helloworld program illustrates, you can concatenate multiple << operationsstd::endl represents the end-of-line.If you don’t particularly want to namespace every command, as per std::cout, you can tell the compiler to make the names in a particular namespace available with a using directive (not dissimilar to calling library() in R). So I could have written my helloworld program like this:
// helloworld-using.cpp
#include <iostream>
using namespace std;
int main() {
cout << "Hello cruel world" << endl;
return 0;
} In general though it’s not a good idea because that leads to namespace conflicts pretty quickly.
The next part of the chapter walks you through variables, operators, types, and so on. Most of this feels very familiar and standard. Yes, C++ is strongly typed and requires variable declaration. This I know. I’ve gotten quite used to writing C++ code like this:
// declares variables but does not initalise
int a;
double b;
bool c;
// declares and initialises variables
// (this uses "uniform initialisation" syntax)
int x = 2;
double y = 2.54;
bool z = true;Operators generally feel familiar from other languages. One thing I’ve missed while working in R is the increment and decrement operators, ++ and --. I have to admit I love these:
// increment and decrement with + and -
x = x + 1;
y = y - 1;
// this is the same
x++
y--The nomenclature used in C++ when talking about changing variable types is a little more precise than it usually is in R. A cast is when you explicitly convert from one type to another. In contrast, the term coercion is used to describe an implicit cast.
The book gives this as example code. The idea being that you should be able to reason through the steps that the program is following, what cast operations are taking place, and thereby predict what it will print out at the end.
// typecasting.cpp
#include <iostream>
int main() {
// variable declarations
int someInteger;
short someShort;
long someLong;
float someFloat;
double someDouble;
// some operations that involve casts
someInteger = 256;
someInteger++;
someShort = static_cast<short>(someInteger);
someLong = someShort * 10000;
someFloat = someLong + 0.785f;
someDouble = static_cast<double>(someFloat) / 100000;
// print output and return
std::cout << someDouble << std::endl;
return 0;
} Okay, I’ll give it a go. Stepping through it line by line…
someInteger which has value 256.++ operator increments this to 257.static_cast() to explicitly cast it to a signed 16 bit integer and store this as someShort.someLong is typed as a long integer (also 32 bit), coercion takes place. The output is cast implicitly to long there’s an implicit cast happening here (called coercion), and the result is stored as a long integer with value 2570000.0.785f) to a long integer and storing the result as a float. So the output someFloat has value 2570000.785.someFloat to a double precision floating point number, divide it by 100000, and assign the result to someDouble. That gives us a value of 25.70000785Okay, let’s check:
25.7We’ve lost a little precision in the printed output, however, because the program prints 25.7 to stdout.
Next up: “strongly typed enumerated types”. Enumerated types are essentially the C++ version of R factors, or Python dictionaries I guess. The key idea is to have a discrete set of labelled values. Internally the objects are encoded as integer values, but those numbers are masked so you can’t do silly things with them. So, for example I could encode gender (somewhat crudely) with an enumerated type as follows:
enum class Gender { male, female, nonbinary, other, unspecified };To declare and initialise some gender variables I would do this:
enum class Gender { male, female, nonbinary, other, unspecified };
Gender danielle_gender { Gender::female };
Gender benjamin_gender { Gender::male };The internal coding is revealed by this program:
// enumerated-types.cpp
#include <iostream>
int main() {
enum class Gender { male, female, nonbinary, other, unspecified };
Gender danielle_gender { Gender::female };
Gender benjamin_gender { Gender::male };
std::cout << "Danielle gender: " << static_cast<int>(danielle_gender) << std::endl;
std::cout << "Benjamin gender: " << static_cast<int>(benjamin_gender) << std::endl;
return 0;
}
When we run this one, the output looks like this:
Danielle gender: 1
Benjamin gender: 0Moving along. The next kind of objects the book considers are structs, which allow you to group one or more existing objects (which may themselves be of different types) into a new type. They are, more or less, the C++ analog of lists in R.
It’s not uncommon for C++ code to define a class within a header file which can be made available to the program by an #include directive (or, in C++20, imported as a module). Sticking reasonably closely to what is in the book here, I’ll write a header file that defines an Employee class:
// employee.h
struct Employee {
char firstInitial;
char lastInitial;
int employeeNumber;
int salary;
}; In the book, the code is a little more elaborate because it explicitly defines a module, but since my compiler has incomplete support for C++20 features, I’m keeping it simple.
A thing to note here is that C++ doesn’t supply a string type out of the box, which is – I would imagine – the reason why this class sidesteps that awkwardness and encodes only the firstInitial and lastInitial as fields in an Employee struct. By doing that we can get away with using a char here. There are of course ways to specify strings, but it’s a bit tangential to the discussion.
Anyway, the key thing is that we access the fields of a struct using . as the code below illustrates:
Employee danielle;
danielle.firstInitial = 'D';
danielle.lastInitial = 'N';
danielle.employeeNumber = 69;
danielle.salary = 123456;To see it in action we can write a short program. Again we have a bit of an issue because the book uses C++20 features that my compiler doesn’t support. Specifically, it uses the new C++20 <format> module to handle printing, which I don’t have access to. Rather than mess about with compilers to get the new hotness working or introduce an external dependency, I decided to use stringstream objects provided by the <sstream> header file.
So my program looks like this:
// employee.cpp
#include <iostream>
#include <sstream>
#include "employee.h"
int main() {
// define the employee record
Employee danielle;
danielle.firstInitial = 'D';
danielle.lastInitial = 'N';
danielle.employeeNumber = 69;
danielle.salary = 123456;
// write to stdout
std::stringstream ss;
ss.str("");
ss << "Employee: " << danielle.firstInitial << danielle.lastInitial;
std::cout << ss.str() << std::endl;
ss.str("");
ss << "Employee number: #" << danielle.employeeNumber;
std::cout << ss.str() << std::endl;
ss.str("");
ss << "Employee salary: $" << danielle.salary;
std::cout << ss.str() << std::endl;
return 0;
} When I run this program I get this as output:
Employee: DN
Employee number: #69
Employee salary: $123456That all seems to make sense. What’s next?
Looking ahead, it seems like the book is about to start discussing conditional branching, functions, loops, and all that good stuff. Fair enough. But conditional branching based on testing some boolean expression is never much fun unless you have a something meaningful to test. Ideally you’d have something where the conditional might evaluate differently. So, since we’re all statisticians here (I mean, it’s just me here and I’m a statistician), this feels like a good moment to divert from the text and write some code that generates Poisson random numbers. To do this I’ll use <random>:
// poisson-sample.cpp
#include <iostream>
#include <random>
int main() {
// set seed using time, define PRNG with Mersenne Twister
long unsigned int seed = static_cast<long unsigned int>(time(0));
std::mt19937_64 mersenne {seed};
// sample_poisson() draws from Poisson(4.1) and returns an integer.
std::poisson_distribution<int> sample_poisson(4.1);
// draw poisson sample (passing the PRNG as argument) and write to stdout
std::cout << "poisson sample: " << sample_poisson(mersenne) << std::endl;
return 0;
}
To check that it works, let’s run it:
poisson sample: 5The output is probabilistic. Repeated invocations produce outout like this:
poisson sample: 2
poisson sample: 6
poisson sample: 2
poisson sample: 5
poisson sample: 2
poisson sample: 4
poisson sample: 4You get the idea.
Okay, back to the development of ideas in the book. Next up is conditional branching, which comes in a few different forms.
if/else statementsthe basic syntax is deeply familiar to anyone who has programmed in… pretty much any language I guess. The first example in the book shows a cascading if statement like this:
if (value > 4) {
// do something
} else if (value > 2) {
// do something else
} else {
// do something else
}However, the nice thing about having some code to sample random numbers is that I can write something a little less boring here. Here’s a small program that samples a Poisson random variate, and prints a message to the terminal that adds a little comment about where the sample falls relative to the distribution mode:
// poisson-conditional.cpp
#include <iostream>
#include <random>
int main() {
// define a poisson distribution
long unsigned int seed = static_cast<long unsigned int>(time(0));
std::mt19937_64 mersenne {seed};
std::poisson_distribution<int> sample_poisson(4.1);
// sample a value and write first part of message
int value = sample_poisson(mersenne);
std::cout << "The sampled value of " << value;
// remainder of message depends on the value
if (value == 4) {
std::cout << " is the modal value." << std::endl;
} else if (value < 4) {
std::cout << " is below the mode." << std::endl;
} else {
std::cout << " is above the mode." << std::endl;
}
return 0;
}
The sampled value of 5 is above the mode.Again, here’s some output I get when I repeatedly invoke the poisson-conditional function:
The sampled value of 11 is above the mode.
The sampled value of 4 is the modal value.
The sampled value of 6 is above the mode.
The sampled value of 5 is above the mode.
The sampled value of 4 is the modal value.
The sampled value of 3 is below the mode.Exciting times in the life of Danielle.
if-else statements with initialisersAt this point the book introduces a concept I wasn’t familiar with. You can include an initialiser within an if/else block, where you can define variables that exist only within the scope of that block. The basic syntax looks like this:
if (<initialiser>; <conditional_expression>) {
<if_body>
} else if (<else_if_expression>) {
<else_if_body>
} else {
<else_body>
}The book doesn’t give an example of this at this early stage because, rather sensibly, the author hasn’t gone down the weird little digression into Poisson variates that I did. But, having done so, it’s easy to write a variant of the previous program that creates the random value within the initialiser:
// poisson-initialised-conditional.cpp
#include <iostream>
#include <random>
int main() {
// define a poisson distribution
long unsigned int seed = static_cast<long unsigned int>(time(0));
std::mt19937_64 mersenne {seed};
std::poisson_distribution<int> sample_poisson(4.1);
// conditional statement with an initialiser
if (int x = sample_poisson(mersenne); x == 4) {
std::cout << x << " is the modal value." << std::endl;
} else if (x < 4) {
std::cout << x << " is below the mode." << std::endl;
} else {
std::cout << x << " is above the mode." << std::endl;
}
return 0;
}
5 is above the mode.In this code, the variable x exists only within the scope of the if/else statement. Here’s a few results from running this program several times:
7 is above the mode.
4 is the modal value.
3 is below the mode.
8 is above the mode.switch/case statements// gender-switch.cpp
#include <iostream>
int main() {
enum class Gender { male, female, nonbinary, other, unspecified };
Gender danielle_gender { Gender::female };
std::cout << "Danielle's gender is ";
switch (danielle_gender) {
case Gender::female:
case Gender::male:
std::cout << "within the gender binary" << std::endl;
break;
case Gender::nonbinary:
case Gender::other:
std::cout << "outside the gender binary" << std::endl;
break;
case Gender::unspecified:
std::cout << "unspecified" << std::endl;
}
}
When this code is executed, we get this:
Danielle's gender is within the gender binaryThe reason this happens is due to fallthrough behaviour. When a case expression that matches the switch expression is reached, all subsequent statements are executed until a break statement is reached. So even though there’s nothing to execute immediately following the Gender::female case expression, there’s no break there either, so the flow “falls through” to the next case. In other words, inthis code male and female genders will both produce “within the gender binary” as the printed message, whereas nonbinary and other genders will produce “outside the gender binary” as the message. If the gender is unspecified, none of the case expressions will match against the switch expression, so the default message at the end is printed.
Just like with if/else blocks, switch blocks can use initialisers:
// gender-switch-2.cpp
#include <iostream>
int main() {
enum class Gender { male, female, nonbinary, other, unspecified };
switch (Gender x { Gender::unspecified }; x) {
case Gender::female:
case Gender::male:
std::cout << "Within the gender binary" << std::endl;
break;
case Gender::nonbinary:
case Gender::other:
std::cout << "Outside the gender binary" << std::endl;
break;
case Gender::unspecified:
std::cout << "Gender unspecified" << std::endl;
}
}
The output here is:
Gender unspecifiedYou can also use ? and : to create a conditional expression. The general syntax is
<condition> ? <value_if_condition_true> : <value_if_condition_false>;Here’s a simple example:
std::cout << ((p_value < .05) ? "reject null" : "retain null");Cool Continuing on.
There’s not much to say here. C++ operators are mostly the same as anywhere else:
>, >=, <, <===!=!&&||Absolutely riveting content, honestly. I’m on the edge of my seat. About the only thing worth mentioning is that C++ uses short circuit rules to evaluate logical expressions. A logical expression is evaluated only until the result is guaranteed. Later steps are not evaluated. For instance:
bool result { boolean_eval_to_false || true || boolean_is_irrelevant }In this case boolean_eval_to_false is evaluated (returning false). That doesn’t guarantee the final result though, because the expressions are joined by ||. However, the next expression is true, which means that result must equal true. Consequently the expression as a whole short-circuits, and boolean_is_irrelevant is never evaluated. This trick is used a lot for efficiency: if you put the computationally cheap tests first and the computationally expensive tests last, you may be able to avoid ever having to perform the expensive tests.
So far all the programs I’ve looked at have included a main() function and nothing else. Every time I’ve declared my main() function the code has looked like this:
int main() {
// meaningful content here
// return to user
return 0;
}By writing the declaration in this way we are asserting the main() function always returns an int value, and indeed when we look at the return statement at the end, it always does. Admittedly, I’ve never used this return value for anything, but it’s a requirement in C++ that the main() function return an integer.
Other functions aren’t constrained in the same way, and in fact C++ functions don’t have to return anything if the only reason to call the function is for its side effects. To do that we set the output type to void. Here’s an example:
// char-code.cpp
#include <iostream>
void print_ascii_code(char c) {
std::cout << c << " has integer code " << static_cast<int>(c) << std::endl;
}
int main() {
print_ascii_code('d');
print_ascii_code('a');
print_ascii_code('n');
print_ascii_code('i');
}
When I run this program I get this:
d has integer code 100
a has integer code 97
n has integer code 110
i has integer code 105(The integers correspond to the ASCII codes for each character).
auto keyword to ask the compiler to figure out the output type for you. So, for instance, this works and returns an integer:auto add_numbers(int x, int y) {
return x + y;
}__func__ that contains the function name. As noted in the book, this can be helpful for logging purposes. Here’s a slightly expanded version of the example used in the book:// add-with-logging.cpp
#include <iostream>
int add_numbers(int x, int y) {
std::cout << __func__ << "(" << x << ", " << y << ")" << std::endl;
return x + y;
}
int main() {
int a = 1;
int b = 2;
int c = 3;
int sum1 = add_numbers(a, b);
int sum2 = add_numbers(sum1, c);
std::cout << "result: " << sum2 << std::endl;
return 0;
}
When this executes, we see the following written to stdout:
add_numbers(1, 2)
add_numbers(3, 3)
result: 6// add-with-overloading.cpp
#include <iostream>
int add_numbers(int x, int y) {
std::cout << __func__ << "(" << x << ", " << y << ")" << std::endl;
return x + y;
}
double add_numbers(double x, double y) {
std::cout << __func__ << "(" << x << ", " << y << ")" << std::endl;
return x + y;
}
double add_numbers(int x, double y) {
std::cout << __func__ << "(" << x << ", " << y << ")" << std::endl;
return static_cast<double>(x) + y;
}
double add_numbers(double x, int y) {
std::cout << __func__ << "(" << x << ", " << y << ")" << std::endl;
return x + static_cast<double>(y);
}
int main() {
int int_a = 1;
int int_b = 2;
double dbl_c = 3.45;
double dbl_d = 6.78;
int int_ab = add_numbers(int_a, int_b);
double dbl_cd = add_numbers(dbl_c, dbl_d);
double dbl_abcd = add_numbers(int_ab, dbl_cd);
std::cout << "result: " << dbl_abcd << std::endl;
return 0;
}
Running this code produces the following output:
add_numbers(1, 2)
add_numbers(3.45, 6.78)
add_numbers(3, 10.23)
result: 13.23It works, but it can be a bit risky to abandon type stability. In this case it’s quite easy to reason about the output type of a call to add_numbers() simply by inspecting the input types, but in my experience things go bad quite quickly when you don’t take type stability seriously.
[[maybe_unused]] to indicate that, for instance, a function argument might not be used. There are other attributes like [[deprecated]] for functions, and – though I skipped that bit earlier, attributes like [[fallthrough]] for switch statements. To be honest, at this point I’m only lightly reading the sections on attributes.At this point, the book moves to a discussion of arrays. Specifically, it first talks about old-school C-style arrays before moving onto the C++ std::array type. Given that I try my very best never to use C-style arrays in my C++ code, I think I’ll skip straight to std::array in these notes. C++ style array objects are provided by the <array> library. The size of the array must be specified in advance, and all elements of the array must have the same type. Indexing starts at 0.
Here’s a simple example:
// array-danielle.cpp
#include <iostream>
#include <array>
int main() {
std::array<char, 8> danielle = { 'D', 'a', 'n', 'i', 'e', 'l', 'l', 'e' };
std::cout << "Danielle has " << danielle.size() << " letters." << std::endl;
} The nice thing here is that C++ arrays have a .size() method which makes it easy to do all sort of operations with them. The current example is a bit minimal: I’m just using it to count the letters in my name. Speaking of which, here’s the output:
Danielle has 8 letters.In many situations you don’t know in advance how many elements you need to store. To help with that, C++ provides the std::vector type via the <vector> library. Vectors are flexible containers that can grow and shrink at run time, so you don’t need to specify how long they will be and you don’t have to faff about with memory managemtn. Because vectors are more useful than arrays (in my experience) I’ll give an example that roughly mirrors a situation I’ve had to deal with in real life: when you’re collecting observations and you don’t know in advance how many observations might arise. In real applications the source of this variation usually comes from the outside source, but for this example I’ll rely on my new best friend, the <random> library.
I’ll switch to statistical notation here. In the code below I assume the number of observations \(n\) follows a Poisson distribution with rate parameter \(\lambda\):
\[ n \sim \mbox{Poisson}(\lambda) \]
and I’ll assume each value \(v\) is a Beta variate between 0 and 1:
\[ v \sim \mbox{Beta}(a, b) \]
Annoyingly, the <random> library does not supply functions for beta distributions out of the box but happily I have not forgotten my basic variate relations. To sample \(v\) from a Beta distribution I can draw two Gamma variates \(x\) and \(y\) as follows:
\[ \begin{array}{rcl} x & \sim & \mbox{Gamma}(a, 1) \\ y & \sim & \mbox{Gamma}(b, 1) \end{array} \]
and set \(v = x/(x + y)\). Problem solved.
Anyway, here’s the code:
// beta-sample.cpp
#include <iostream>
#include <vector>
#include <random>
void print_message(double value, double a, double b) {
std::cout << "beta(" << a << "," << b << ") sample: " << value << std::endl;
}
std::vector<double> draw_betas(double rate, double a, double b) {
// distributions
std::gamma_distribution<double> gamma_a(a, 1.0);
std::gamma_distribution<double> gamma_b(b, 1.0);
std::poisson_distribution<int> poisson(rate);
// mersenne twister numbers
std::random_device rd;
std::mt19937 mt(rd());
// draw poisson sample to determine number of betas
int n = poisson(mt);
// draw beta samples and return
std::vector<double> beta_variates {};
double x, y;
for (int i = 0; i < n; i++) {
x = gamma_a(mt);
y = gamma_b(mt);
beta_variates.push_back(x / (x + y));
}
return beta_variates;
}
int main() {
const double a = 2.0; // shape parameter 1
const double b = 1.0; // shape parameter 2
const double rate = 2.4; // rate for poisson dist
// draw samples
std::vector<double> betas = draw_betas(rate, a, b);
// print messages and return
std::cout << "collected " << betas.size() << " samples" << std::endl;
for (int i = 0; i < betas.size(); i++) {
print_message(betas[i], a, b);
}
return 0;
}
Technically I’m getting slightly ahead of the book here because it hasn’t talked about loops yet, but whatever. A for loop is more or less the same thing everywhere.
Anyway, let’s take a quick look at some output. Sometimes this code creates a betas vector containing four values:
collected 4 samples
beta(2,1) sample: 0.293596
beta(2,1) sample: 0.920161
beta(2,1) sample: 0.971502
beta(2,1) sample: 0.502786Sometimes the betas vector has no values:
collected 4 samplesSometimes we sample two values, so the output looks like this:
collected 2 samples
beta(2,1) sample: 0.729145
beta(2,1) sample: 0.414129And so on.
At this point the book discusses a couple of topics that I’m not going to bother writing my own code for:
C++ provides a std::pair class via the <utility> library. It’s used to group two values that can be of different types. It has methods .first() and .second() to extract individual elements. I can imagine this is useful as a tool for representing name-value pairs, for instance.
There is an std::optional class provided by <optional> and can either hold a value of a specific type, or nothing. There’s a method .has_value() that returns a boolean specifying whether a value has een stored, and a .value() method that returns the stored value. Apparently you can also use the dereferencing operator to do the same thing, so if we have an optional object called maybe then maybe.value() and *maybe do the same thing. There’s also a .value_or() method that returns the stored value if one exists, or else returns whatever is passed to .value_or(). The obvious application that comes to mind for me is that optional objects could be used to handle missing values in statistical contexts.
Structured bindings were introduced in C++17 and allow you to declare and assign multiple variables at once, using values taken from an appropriate object: arrays, structs, pairs, and tuples all work for this. The main use case I can see for this that it provides a clean way to allow function to return multiple values. On the function side, you wrap the outputs into an appropriate structure (the example below uses a tuple). Then when the function is called elsewhere, use the structured bindings syntax to assign the elements of the wrapping structure (e.g., the tuple) directly into their own variables. for example, let’s suppose I’ve written some_function() that returns a 3-tuple. I can assign the elements of that tuple to variables x, y and z like so:
auto [x, y, z] = some_function();Back in my MATLAB days I used to do this all the time. In R it’s slightly trickier to do that unless you’re using specialised packages that provide multiple assignment functionality. Anyway, here’s a contrived example:
// structured-binding-asl.cpp
#include <iostream>
#include <tuple>
#include <string>
// somewhat absurd function used to illustrate the point
std::tuple<int, char, std::string> asl() {
return {45, 'F', "Sydney"};
}
int main() {
// use structured bindings to declare and assign multiple
// variables from the output returned by the function call
auto [age, sex, location] = asl();
// messages
std::cout << "age: " << age << std::endl;
std::cout << "sex: " << sex << std::endl;
std::cout << "location: " << location << std::endl;
return 0;
}
For the sake of keeping these notes family-friendly I have restricted myself to “a/s/l”, even though the term “structured bindings” itself suggests the possibility of a rather more… expansive… query that might be adopted here.
Aaaaaaanyway… the results:
age: 45
sex: F
location: SydneyThe book now turns to loops. I’m trying very hard not to start skimming in case there’s something secific to C++ that I need to know, but honestly this all feels very, very familiar. To keep my boredom to a minimum I’ll try to find examples that I find entertaining.
while loopFirst up, a while loop calculating the sequence of integers \(x_1, x_2, \ldots\)
\[ x_{n+1} = \left\{ \begin{array}{rl} x_n/2 & \mbox{if $x_n$ is even} \\ 3x_n +1 & \mbox{if $x_n$ is odd} \\ \end{array} \right. \]
terminating at the first \(n\) such that \(x_n = 1\). This is of course an implementation of the collatz conjecture which proposes that this collatz sequence, for every integer-valued choice of initial value \(x_0\), eventually terminates in 1. Code:
// collatz.cpp
#include <iostream>
int main() {
int value = 39;
std::cout << value;
while (value != 1) {
value = (value % 2 == 0) ? (value / 2) : (3 * value + 1);
std::cout << " -> " << value;
}
std::cout << std::endl;
return 0;
} The results for \(x_0 = 39\), where I’ve manually added linebreaks:
39 -> 118 -> 59 -> 178 -> 89 -> 268 -> 134 -> 67 -> 202 -> 101 -> 304 -> 152 -> 76 -> 38 -> 19 -> 58 -> 29 -> 88 -> 44 -> 22 -> 11 -> 34 -> 17 -> 52 -> 26 -> 13 -> 40 -> 20 -> 10 -> 5 -> 16 -> 8 -> 4 -> 2 -> 1do/while loopThe do/while loop is inherently boring, but is very occasionally handy if it is important to ensure that the code block is executed at least once. This pattern makes sense when, for example, you might need to repeatedly check something until it produces valid output. In that situation a while loop does make sense, but you really have to run at least once test in order to validate… whatever it is you’re trying to validate.
As a truly absurd example:
// validation-check.cpp
#include <iostream>
#include <ctime>
bool valid_time() {
std::time_t elapsed = std::time(nullptr);
bool is_valid = elapsed % 2 == 0;
if (is_valid) {
std::cout << elapsed << " seconds since the epoch" << std::endl;
}
return is_valid;
}
int main() {
bool valid;
int i = 0;
do {
i++;
} while (!valid_time());
std::cout << "attempts = " << i << std::endl;
return 0;
} This code checks the current time, measured in number of seconds since the unix epoch. If that number is even, it deems the time to be “valid” and prints the elapsed seconds to stdout. If that number is odd, it deems the time to be “invalid” and refuses to terminate the do/while loop. That leads to some entertaining behaviour. About half the time you’ll get output like this where it succeeds the first time:
1687250594 seconds since the epoch
attempts = 1But the other half of time the program fails about 200 million times before finally succeeding:
1687250596 seconds since the epoch
attempts = 200744719Suffice it to say, although the idea of using do/while loops to implement validation checks makes sense, this is… um… an unhinged example. But I found it funny so I ran with it.
for loopOne of the earlier examples gave an example of a for loop that was semi-serious, so I feel justified in being absurd again:
// na-na-hey-hey.cpp
#include <iostream>
int main() {
for(int i = 0; i < 8; i++) { std::cout << "na "; }
for(int i = 0; i < 3; i++) { std::cout << "hey "; }
std::cout << "goodbye" << std::endl;
return 0;
}
na na na na na na na na hey hey hey goodbyefor loopThe range-based for loop iterates directly over the elements of a container. It works for a fairly wide range of possible containers: anything that has .begin() and .end() methods that return iterators will work. Example:
// array-iterator.cpp
#include <iostream>
#include <array>
int main() {
std::array<int, 6> fib = { 1, 1, 2, 3, 5, 8 };
for(int f : fib) { std::cout << f << std::endl; }
return 0;
} 1
1
2
3
5
8Note that in this code, at each step of the iteration the variable f stores a copy of the relevant element of fib. However, it’s possible to do range-based for loops without making copies by using a reference variable. The book promises to discuss this later in the chapter.
Initialiser lists are designed to make it easy to write functions that accept a variable number of arguments, and are provided by the <initializer_list> library. All values in an initialiser list must be the same type. Here’s an example that computes the mean of an arbitrary number of double values:
// mean-value.cpp
#include <iostream>
#include <initializer_list>
double mean(std::initializer_list<double> values) {
double tot = 0;
for (double v : values) {
tot += v;
}
return tot / values.size();
}
int main() {
double x_bar = mean({ 2.3, 1.5, 7.8, 11.0 });
double y_bar = mean({ 102.5, 59.1, 98.2 });
std::cout << "mean x: " << x_bar << std::endl;
std::cout << "mean y: " << y_bar << std::endl;
return 0;
}
mean x: 5.65
mean y: 86.6The book talks about strings next, but it’s mostly a promissory note here. It mentions that the std::string type from <string> works more or less the way you’d expect for a string type. Here’s a quick example:
// simple-string.cpp
#include <iostream>
#include <vector>
#include <string>
int main() {
std::vector<std::string> name = { "Daniela", "Jasmine", "Navarro", "Bullock" };
for(std::string n : name) { std::cout << n << std::endl; }
return 0;
} Daniela
Jasmine
Navarro
BullockThe next section of the book introduces object oriented programming (OOP) features in C++. At this point I’m about 40 pages in, and now starting to hit material that isn’t already familiar to me. Yay! Time to learn new things!
Some preliminaries:
Okay, after a bit of reading, the basic ideas seem clear:
Here’s my first attempt at defining a C++ class. One of my loves in life is gardening, so I’ll define a class Species that stores the taxonomic (binomial) name of a plant, and (optionally) stores the common name. Obviously this is too simple for real world use, because plants often have more than one common name, and the binomial name for a species is often insufficient to represent the taxonomic relations that we might care about. And of course it also fails to capture any of the practical information about a plant! But whatever. That’s not the bloody point. Anyway, here’s the code:
// species-first-pass.cpp
#include <iostream>
#include <string>
#include <optional>
class Species {
private:
// internal data structure
std::string name_binomial;
std::optional<std::string> name_common;
public:
// class constructor with one input
Species(std::string binomial) {
setBinomialName(binomial);
}
// class constructor with two inputs
Species(std::string binomial, std::optional<std::string> common) {
setBinomialName(binomial);
setCommonName(common);
}
// methods to set names
void setBinomialName(std::string name) { name_binomial = name; }
void setCommonName(std::optional<std::string> name) { name_common = name; }
// methods to retrieve names
std::string getBinomialName() { return name_binomial; }
std::optional<std::string> getCommonName() { return name_common; }
// print method
void print() {
std::cout << name_binomial;
if (name_common.has_value()) {
std::cout << " (" << name_common.value() << ")";
}
std::cout << std::endl;
}
};
int main() {
Species yellow_plant { "acacia amoena" };
Species purple_plant { "hardenbergia violacea", "happy wanderer" };
yellow_plant.print();
purple_plant.print();
return 0;
}
When I run this program, I get this as the output:
acacia amoena
hardenbergia violacea (happy wanderer)Neat. The output always prints the binomial name of the species, and appends the common name in parentheses if one exists. Additionally, because the .setCommonName() method is public, should I later happen to discover to my delight that the acacia amoena is commonly referred to as the boomerang wattle, I can update the record like so:
yellow_plant.setCommonName("boomerang wattle");Fab. Moving on.
The program I wrote above is simple enough that it could probably stay as a single script, but that quickly becomes unworkable for larger programs, especially if the class needs to be make available in multiple parts of the code base. For that reason it’s conventional to define the class using one or more dedicated files, where the file name is identical to the class name. The specific convention seems to be:
Species.h contains the class definitionSpecies.cpp defines the class methodsI would then save the source code that actually uses these classes for something as species-second-pass.cpp or whatever, and incorporate the class using #include directives.
Next up the book discusses C++ scopes. It all feels very familiar. Every variable and function belongs to a scope and is usually visible only within that scope. Examples:
if, switch, or for are scoped to the code block for that loop/conditional.namespace declarations create variables scoped to that namespace. Variables within a namespace can be explicitly referenced from outside the scope using the :: operator (which is what I’ve been doing with the std namespace throughout this code)Here’s an adaptation of the example used in the book:
// scope-resolution.cpp
#include <iostream>
// value() is scoped to the Five class
class Five {
public:
int value() { return 5; }
};
// value() belongs to the global scope
int value() { return 10; }
// value() belongs to the twenty namespace
namespace twenty {
int value() { return 20; }
}
int main() {
Five five;
std::cout << five.value() << std::endl; // prints 5
std::cout << value() << std::endl; // prints 10
std::cout << twenty::value() << std::endl; // prints 20
return 0;
}
5
10
20In early versions of C++, variable initialisation statements looked different depending on what kind of object is being initialised. Since C++11, however, uniform initialisation has been available so you can initialise variables using the same syntax regardless of type. The book gives the following example:
// circles.cpp
struct CircleStruct {
int x, y;
double radius;
};
class CircleClass {
public:
CircleClass(int x, int y, double radius)
: m_x { x }, m_y { y }, m_radius { radius } {}
private:
int m_x, m_y;
double m_radius;
};
int main() {
// these both use uniform initialisation
CircleStruct circle1 { 10, 10, 2.5 };
CircleClass circle2 { 10, 10, 2.5};
// pre C++11, you had to do this:
CircleStruct circle3 = { 10, 10, 2.5 };
CircleClass circle4(10, 10, 2.5);
// note that this is not uniform initialisation
int a = 3;
int b(3);
// these are both uniform
int c = { 3 };
int d { 3};
return 0;
}
There are some subtle differences between uniform and non-uniform initialisation, particularly in reference to narrowing. In old style initialisation, this works…
int x = 3.14;but because x is an integer the value stored is 3. This is the narrowing phenomenon. If you use uniform initialisation, narrowing is forbidden. This produces a compiler error:
int x { 3.14 };There are tools to explicitly perform narrowing if that’s what’s needed: the book mentions the gsl::narrow_cast() function in the Guidelines Support Library (GSL).
In general, the book recommends uniform initialisation.
Notes on memory management in C++. There are two different kinds of memory allocation in C++: the stack and the free store. The metaphor used to describe the stack is a “deck of cards”. The top “card” (or, stack frame) represents the current scope (e.g., function currently being executed). Variables declared within that scope are stored in this stack frame. Parameters passed to a function call are copied into the stack associated with the function scope. When the scope goes away (e.g., function completes), so too does the corresponding stack frame. Variables allocated to stack memory don’t require any special management: C++ automatically takes care of the allocation and deallocation of memory.
The free store is different. It corresponds to an area of memory that is independent of the stack, and you can place variables there that will persist even when the scope in which they were declared goes away. Memory on the free store needs to be manually managed: it has no notion of frames, and there is no automatic deallocation. (…Unless, apparently, you use smart pointers, which the book foreshadows will be discussed later).
Okay, so… noting that smart pointers are coming later in the book, I think all I’ll do here is quickly summarise the key things. If you want to place an integer on the free store, we first declare a pointer. A pointer is a reference to a location in memory. So when I do this:
int* my_int_pointer { nullptr };what I’m doing is specifying that my_int_pointer is going to reference the location of an integer value (the * in int* indicates that a pointer is being declared). At this stage, the pointer doesn’t actually reference a specific location: the nullptr (“null pointer”) value is a special value that doesn’t correspond to a valid location, and evaluates to false if used in a logical expression.
To allocate the memory and have the pointer reference that location in memory, we use new:
my_int_pointer = new int;At this stage, a block of memory has been allocated for the integer and my_int_pointer now references that location. To access that location, either to retrieve the value or assign a value to it, we dereference the pointer:
*my_int_pointer = 8;My example program:
// pointer-free-store.cpp
#include <iostream>
int main() {
int* int_ptr { new int }; // declare pointer & allocate memory
*int_ptr = 8; // assign value to the allocated memory
std::cout << *int_ptr + 2 << std::endl; // retrieve and print
return 0;
} When executed, it prints 10 to stdout:
10C++ also allows pointers to variables on the stack (and even pointers to pointers, but whatever). The key thing here is to use & (“address of”) to return a pointer to a stack-allocated variable. So this works and also prints 10 to stdout:
// pointer-stack.cpp
#include <iostream>
int main() {
int value { 8 }; // variable on the stack
int* int_ptr { &value }; // declare pointer to it
std::cout << *int_ptr + 2 << std::endl; // retrieve and print
return 0;
} 10There’s a little bit of syntactic sugar for pointers to structs and classes. Copying the code directly from the book. Assume there’s a function getEmployee() that returns a pointer to an Employee instance. Then we could write the salary to stdout like so:
Employee* anEmployee { getEmployee() };
std::cout << (*anEmployee).salary << std::cout;The -> operator allows a shortcut:
Employee* anEmployee { getEmployee() };
std::cout << anEmployee->salary << std::cout;Okay fair enough. Moving on.
The book here has a short section on dynamically allocated arrays, but I’m going to skip over that for now. The main thing is that it advises not to use malloc() and free() from C for this, and instead use new and delete or new[] and delete[]. Noted :-)
const keywordThe const keyword is used to indicate that something is not permitted to change within the program. It can be used in a few ways. The simplest case is declaring that the value of a variable must not be changed:
const int versionNumberMajor { 2 };
const int versionNumberMinor { 1 };
const str::string productName { "My fabulous product" };The same idea can be applied to pointers:
// the pointed-to value cannot be changed
const int* ptr;
int const* ptr; // same meaning
// the pointer itself cannot be changed
int* const ptr { nullptr };
ptr = new int; // this won't compile
// the pointer itself cannot be changed
int* const ptr { new int[10] };You can also do things like this:
void func(const int param) {
// this code cannot change the value of param
}You can also use const to declare that certain methods of a class are not permitted to change the member data. So the relevant bit of the code might look like this:
public:
int methodToComputeSomething() const; // cannot modify internal data
void methodToSetInternalData(); // can modifiy internal dataIt’s generally considered best practice to follow the “const correctness” principle, and always declare member functions that do not change any data members as const. The terminology here is:
const member function, which cannot modify data, is called an inspectorconst member function, which does modify data, is called a mutatorA reference is an alias for another variable: essentially another name given to the same object. Changes to the reference variable are reflected in the original variable, and vice versa. The mental model suggested in the book is that you can think of references as implicit pointers, where you don’t have to take care of addressing and dereferencing yourself. Declaring the reference variable is done using &, like this:
int x { 10 }; // original variable
int& x_ref { x }; // reference variableThe equivalence of the original and the reference is illustrated in this simple program:
// simple-reference.cpp
#include <iostream>
int main() {
int x { 10 }; // original
int& x_ref { x }; // reference
x_ref++; // change the reference, *and* the original
std::cout << x << std::endl; // prints 11
x++; // change the original, *and* the reference
std::cout << x_ref << std::endl; // prints 12
return 0;
} 11
12References must always be initialised when declared. This won’t compile:
int& empty_ref; Along similar lines, you cannot change the mapping once a reference variable is initialised (i.e., it always refers to the same original variable, and you can’t move it to a new one). This is illustrated in this program:
// immovable-reference.cpp
#include <iostream>
int main() {
int x { 3 };
int y { 4 };
int& x_ref { x };
x_ref = y; // changes the value of x to match y
std::cout << x << std::endl;
return 0;
} 4You are allowed to specify a “reference to const”, as illustrated in the second line of the code snippet below:
double val { 1.234 }; // ok
const double& val_ref_const { val }; // okThe creates something akin to a “read only” reference. You can access the value of val by using val_ref_const, and you can change the value of both by modifying val:
val = 2.345; // ok, changes val and val_ref_constWhat you can’t do, however, is change the value of val by modifying val_ref_const:
val_ref_const = 3.456; // fails, does not compileThis turns out to be super helpful when passing by reference.
The main use for references is to avoid making copies of values when passing arguments to a function. You could do this with pointers, of course, but pointers are messier, so it’s generally better to do it with references. There can be performance gains by not making unnecessary copies, but there are also other neat things you can do:
// swap.cpp
#include <iostream>
void swap(int& first, int& second) {
int temp { first };
first = second;
second = temp;
}
int main() {
int x { 10 }, y { 20 };
std::cout << "original x value is " << x << std::endl;
std::cout << "original y value is " << y << std::endl;
swap(x, y);
std::cout << "swapped x is now " << x << std::endl;
std::cout << "swapped y is now " << y << std::endl;
return 0;
} original x value is 10
original y value is 20
swapped x is now 20
swapped y is now 10Notice that when passing a reference, it’s possible to modify the original variables (outside the scope of the function) by performing operations on the references. That was a handy feature in the swap() example, but normally it’s undesirable. A more typical scenario is one in which you don’t want the function to possess the ability to modify the out-of-scope variables that have been passed through the function arguments (because the only reason you’ve chosen to pass-by-reference rather than pass-by-value is to avoid making copies). In this situation, the best bet is to pass-by-reference-to-const. Because the reference-to-const cannot modify the original variable, it is now impossible for the function to accidentally modify the originals.
Here’s the idea:
// pass-by-reference-to-const.cpp
#include <iostream>
#include <string>
// str_print() declares a reference-to-const as the argument
void str_print(const std::string& x) {
std::cout << x << std::endl;
}
int main() {
std::string str { "hello cruel world" };
str_print( str ); // passing a string variable works
str_print( "goodbye cruel world" ); // so does passing a literal
return 0;
}
hello cruel world
goodbye cruel worldThe divide() function in this code throws an invalid-argument exception if the denominator is zero. The code within main() implements a try-catch block to handle such an exception if thrown:
// try-catch.cpp
#include <iostream>
#include <stdexcept>
// a divide() function that throws an error for divide-by-zero
double divide(double numerator, double denominator) {
if (denominator == 0) {
throw std::invalid_argument { "Denominator cannot be 0." };
}
return numerator / denominator;
}
int main() {
try {
std::cout << divide(13, 2) << std::endl;
std::cout << divide(13, 0) << std::endl;
std::cout << divide(13, 3) << std::endl;
} catch (const std::invalid_argument& exception) {
std::cout << "Exception caught: " << exception.what() << std::endl;
}
return 0;
} 6.5
Exception caught: Denominator cannot be 0.Note that when the exception is thrown the try code block immediately terminates and is passed to the catch code block. The third division is never attempted.
The final sections in the chapter discuss the auto keyword and the decltype keyword, before finishing with an example of a bigger C++ program. For now I’m going to skip those in these notes and move onto the next chapter.
const keyword