Advanced C++ modern programming
Purpose
This repo contains all the code and notes for the UC3M Advanced C++ Modern Programming course.
Notes
What is C++?
- If your application is slow, quit your dynamic polimorphism.
- Metaprogramming does not solve the 90% of the use cases.
- You can do Functional Programming.
- Memory leaks -> C++11 use smart pointers so you'll rarely see memory leaks. Daniel usually does a
grep "new"
(to delete them) so there are no memory leaks. - Buffer overflows -> use a good static analyzer.
- Macros should be deleted from existence.
- Pragmas, not even once. This is 99% non useful black magic.
- Variadic Macros... vade retro Satan.
How does it looks like?
- Uniform initiation
std::vector<int> v {1,2,3,4,5};
- Range based loop
for(auto& x: v) {
std::cout << x;
}
Uniform Initialization in C++
In C++98 the initilization wasn't uniform, so in C++11 they changed it so it was uniform, easier to read, understand and write.
In one sentence: all is initialized with {}
.
This is the only way to go generic.
When can I use '='
To initialize a variable with copy-initialization (as oposed to direct-initialiation).
Uniform and explicit initialization
- Uniform initialization with direct initialization allows invoking explicit constructor.
- However, using copy initialization does not allow using explicit constructors.
std::ofstream f1("file1.txt") // ok
std::ofstream f1 = "file1.txt" // error
Initializer lists
std::initializer_list<T>
is a type from the standard library which can be useful while initializing complex classes.
- Use
copy
from the std as much as you can. Stop usingfor
loops. - While overloading constructor, the compiler will prefer the empty constructor over the initializer_list constructor when taking zero arguments.
Warning: there are some caveats to this. If there are two constructors for a class like this:
X(int a)
X(std::initializer_list<int> list)
To use X(int a)
you must use the X whatever(9)
to use it. If you use X whatever {9}
you will call the constructor with the initializer_list. This happens in the standard library with the std::vector.
std::vector(9) // creates a vector with 9 elements initialized to 0
std::vector{3} // creates a vector with 1 element initialized to 3
Compile-time assertions
In C++11 they introduced static assers, which are "normal" asserts that while compiling, they trigger itselfs to inform you.
The static asserts are specially usefull in combination with templates in generic programming.
Null pointer
In C++03 the null pointer constant is 0. So this code:
void f (int a);
void f(char* ptr);
sucks:
f(0); // invokes the first function
f(static_cast<char*>(0)) // invokes the second function
Now, we have the NULL "pointer constant", which in reality is a macro:
- NULL == static_cast(0)
So now in C++11 we have the keyword nullptr.
char* ptr = nullptr; // ok
int n = nullptr; // error
int z = 0;
if (pc == 0) { z = 1; } // ok
if (pc == nullptr) { z = 1; } // ok
Generic Code
This wont compile:
template (typename T)
void f (T *t)
f(nullptr); // Error T = ?
But this does:
f(static_cast<double *>(nullptr));
That is because the nullptr has a type of it's own.
Type inference
You know that the STL sucks... but for that there's a solution: auto
. While using the keyword auto
, the compiler should infer the variable's type.
std::vector<int> v;
auto i = v.begin() // typename std::vector<T>::const_iterator i = v.begin()
Range based for loops
There are two ways to iterate over a container:
x.begin()
&x.end()
begin(x)
&end(x)
Between the two, the first one is the preferred.
Aligment
Some low level tricks:
class particle { /∗...∗/ };
typedef std::aligned_storage<sizeof(particle),alignof(particle)>::type buffer;
buffer p1;
new (&p1) particle;
This code:
- Create a buffer type aligned to the size of the class particle.
- Creates a buffer (p1).
- Uses
new
to assign the newly created particle to the buffer p1 address.
Wait, what?
You see, the new
statement in C++ does the following:
- Allocates memory for the soon to be created object. Like a
malloc
on vitamines. - Invokes the constructor function of the object and assigns the newly created object to the memory addres allocated beforehand.
The previous code exploit the second part of the new
statement, by declaring the memory address where the object should be declared between the new
statement and the object class.
address;
new (address) class;
Move semantics
The compiler has two magic weapons: constant folding and inlining. You are not smarter than the compiler and it's tricky optimizations. Period.
Given how C/C++ works with larger objects, there was a trend to void pass-by-value and return-by-value, so it transformed into pass-by-reference and return-by-reference.
So now we have the following situation: we don't want to copy, but passing only references is also a bit ugly. So we have something "better"! C++11 added the idea of move.
sample s{1000};
sample t{std::move(s)}; // s is empty and now t has the content
s = std::move(t); // t is empty and now s has the content
Just like Rust! But wait, there are some caveats... because move
doesn't move!!! it's the combination between move
and =
.
using name_list = std::vector<std::string>;
name_list generate_names();
void f() {
name_list list;
list = generate_names(); // This is a 'copy' operation in C++03, but it's a 'move' operation in C++11
}
Now more interesting:
std::string get_name() {
std::string name{"Julio"};
name += " Iglesias"
return name;
}
Because name
is a temporal variable and will be destroyed outside the function, C++11 will move the value of name
instead of copying it.
Good news! Most of the STL is move
compatible:
using wordlist = std::vector<std::string>;
wordlist read(std::string filename);
void f() { wordlist wordlist wordlist wordlist wordlist wordlist
w1; // Empty word list
w2{w1}; // Copy
w3 = w1; // Copy w4{read("quijote.txt")}; // Move w5 = read("quijote.txt"); // Move w6 = std::move(w4); // Move
w1 = w4; // Copy
w1 = std::move(w4); // Move
}
Bad news: by default, the user defined classes are not move
compatible. To make your class move
compatible, you'll need a move constructor:
A move operation shoud:
- Release resources hold by target.
- Take ownership of resources hold by source.
- Leave source in a valid state.
// declaration
class sample {
public:
// copy
sample & operator=(const sample & s);
// move
sample & operator=(sample && s);
private:
int size_;
double ∗ vec_;
}
// implementation
sample::sample(sample && s) : size_{s.size}, vec_{s.vec_} {
s.size_ = 0;
s.vec_ = nullptr;
}
Use:
sample read(std::string filename);
void f() {
sample s1; // Empty sample
sample s2{read("marks.txt")}; // Move
s1 = s2; // Copy
s1 = std::move(s2); // Move
}
Forwarding
R-Value and L-Value Black Magic Fuckery:
bool is_temporary(int const & x) { return false; }
bool is_temporary(int && x) { return true; }
template <typename T>
void print(T && x) {
if (is_temporary(x)) std::cout << "Temporary\n";
else std::cout << "Not temporary\n";
}
void f() {
print(5); // Prints Not temporary int x = 42;
print(x); // Prints Not temporary
}
Hold up a minute...
bool is_temporary(int const & x) { return false; }
bool is_temporary(int && x) { return true; }
template <typename T>
void print(T && x) {
if (is_temporary(std::forward<T>(x))) std::cout << "Temporary\n";
else std::cout << "Not temporary\n";
}
void f() {
print(5); // Prints Not temporary int x = 42;
print(x); // Prints Not temporary
}
Conclusion: std::forward is black magic and playing with r/l-values is confusing and hurts.
Exception Safety
If you mark your function as noexcept you are stting that the code is not ready to handle exceptions.
template <typename T>
T suma(T x, T y) noexcept {
return x+y;
}
noexcept violations
- Constructor of std::vector might throw if memory allocation fails.
- std::vector::at() might throw if access is out of range.
int cube(int x) {
return x*x*x;
}
std::vector<int> apply_cube (const std::vector<int> &v) noexcept {
std::vector<int> r(v.size()); // might throw exception
for(size_t i = 0; i < v.size(); ++i) {
r.at(i) = cube(v.at(i)); // might throw exception
}
return r;
}
To "fix" it:
bool cond = noexcept(v.at(0));
void f(std::vector<int> & v) noexcept(noexcept(v.at(0)));
- A destructor should not throw exceptions. Because of this, all destructors in STL are noexcept.
- Move operations should not throw exceptions.
Compile time constants: constexpr
- Expressions evaluated at compile time (pdf had an errata) -> constexpr
- Values that can't be modified (immutable) -> const
Why would I want to use this feature?
PERFORMANCE BIATCH! (batman-bitchslap-meme)
Why the hell should you use C++ in the first place? Because you want that sweet, sweet performance and flexibility. But rust... RUST IS NOT ISO, IT DOESN'T HAS A STANDARD AND IS CONTROLLED BY AN ENTITY! ok, ok... jeez....
So why do we want to use compile-time expresions? To calculate things we want ahead of time. Supose that you need to calculate or initialize something every time and that this calculation costs you 2ms because it access the system memory, some hundred of CPU cycles, etc... You can do that in advance even before the code runs the first time.
Non Static Data Member initialization
- Extends static data members to non static data members
- Allows to specify a default value for a data member in the class declaration.
class position {
private:
int x {1};
int y {1};
};
// the same... but different
class position {
private:
int x;
int y;
public:
position() : x{1}, y{1} {}
};
Inline namespaces
The use of inline namespaces allows you to omit part of the namespace name in favor of readiness. For example:
std::V1::cout
std::V2::coud
// now if V1 has inline
std::cout // it's the V1 cout
std::V2::cout // it's the V2 cout
decltype and declval
- decltype
- Gets the type of a declaration
- Useful complement to auto
- Specially useful in generic programming
T x;
auto t = x;
U y;
auto z = y;
auto a = x + y; // a is the same type as (x + y)
decltype(x + y) b; // b is the same type as (x + y)
b = x + y; // now b has (x + y)
The previous example wasn't so much, but here is more:
template <class T, class U>
auto add(const std::vector<T> & v1, const std::vector<U> & v2)
−> std::vector<decltype(T{}+U{})> {
std::vector<decltype(v1[0]+v2[0])> r; auto i=v1.begin();
auto j=v2.begin();
auto end = v1.end();
for (;i!=end;++i, ++j) {
r.push_back((∗i)+(∗j));
}
return r;
}
Here we return a vector of (T + U). But why not declaring the type before where you say 'auto'?. Because in that point, we still don't know T nor U.
Lambda expressions
Useful when you want to declare a function, but you only use it only once. Mostly, when you do some value transformation using the STL.
// C++98
struc square {
double operator()(double x) {return x*x};
}
...
void f(const std::vector<double> & v, std::vector<double> w) {
std::transform(v.begin(), v.end(), std::back_inserter(w), square());
}
Now:
// C++11
void f(const std::vector<double< & v, std::vector<double> & w) {
std::transform(v.begin(), v.end(), std::back_inserter(w), [](double x) { return x*x; });
// [](double x) { return x*x; } would be the lambda
}
Some examples:
double s = 0.0;
void f(int i) {
char c = ’a’; static int n = 0;
vector<int> v { 1, 2, 3};
for_each(v.begin(), v.end(), [](int x) { cout << x << "\n"; }); // OK for_each(v.begin(), v.end(),
for_each(v.begin(), v.end(), [](int x) { cout << s << "\n"; }); // OK. s is global for_each(v.begin(), v.end(),
for_each(v.begin(), v.end(), [](int x) { cout << i << "\n"; }); // Error. i non accessible for_each(v.begin(), v.end(),
for_each(v.begin(), v.end(), [](int x) { cout << c << "\n"; }); // Error. c non accessible for_each(v.begin(), v.end(),
for_each(v.begin(), v.end(), [](int x) { cout << n << "\n"; }); // OK. n is static
}
To access the local variables, you'll need to specify the =
operator in the squares.
for_each(v.begin(), v.end(), [=](int x) { cout << c << "\n"; }); // c is now accessible for_each(v.begin(), v.end(),
To access specific variables, you'll need to pass them in the squares:
int c {0};
for_each(v.begin(), v.end(), [c](int x) { cout << c << "\n"; }); // c is now accessible for_each(v.begin(), v.end(),
Function wrappers
This is some black magic of C++: with lambdas, we cannot do recursion... until now that we can use std::function
:
std::function<int(int)> factorial = [&](int x) { return (x==1) ? 1 : (x ∗ factorial(x−1)); };
Variadic Templates
// Classes
template <typename ... Args>
class tuple;
// Functions
void print (const char *s);
template <typename T, typename ... Args>
void print(const char * s, T v, Args ... args); // Args: template parameter pack; args: function argument pack
print("Hi %: you were born %−%−% in %.", "Daniel", 4, 9, 1969, "Spain");
Primitive literals and user defined literals
- Literal operator: transforms a literal with a given suffix into a value of a specific type:
operato"" suffix
- A literal operator may be calculated at compile time
constexpr complex<double> operator"" _i(long double imag)
{
return {0.imag};
}
constexpr complex<double> x1 {5.0_i};
complex<double> x2 {2 + 3.6_i}
Some examples:
class big_number {
public:
big_number(const char * c);
};
big_number operator""_bn(const char * c)
{
return big_number{c};
}
void f(big_number x);
f(13298374293740276058736958764958173465_bn);
Strict ownership
- Unique pointers:
- Simplify management of data members that traditionally were raw pointers
- If all the pointers in a class are unique pointers, I don't need to specify a destructor. Only the copy operations.
class base {
public:
virtual ~base();
virtual void f();
};
class derived: public base {
public:
virtual void f();
};
void g() {
unique_ptr<derived> pd{new derived};
unique_ptr<base> pb{pd}; // Error. Cannot copy unique_ptr<base> pb2{std::move(pd)}; // OK
pb2−>f(); // Virtual call
// ...
}
If you only use unique pointer and never use new (so you'll never use delete), probably you'll never have to worry about memory leaks.
make_unique
was forgotten to add in c++11, but the fix is simple:
template <typename T, typename ... Args>
unique_ptr<T> make_unique(Args && ... args) {
return unique_ptr<T>{new T{std::forward<Args>(args)...}};
}
auto p = std::make_unique<record>(id, name);
Tuples
As with the pairs, do not overuse them, if you do, maybe you need a struct
instead of the tuple
.
Some pythonic examples:
std::tuple<int, int> divide(int x, int y) {
return std::make_tuple(x/y, x%y);
}
int quotient, remainder;
std::tie(quotient, remainder) = divide(10,3);
The tuples allows us to return more than one value. Tuples can be compared with ==
without problems.
- Elements in a tuple can be accesed by type, but if there are more than one variables with the same type, it wont compile.
std::tuple<string, string, int> a { "Daniel", "Garcia", 1969 };
auto x = std::get<2>(a); // int x = 1969
auto y = std::get<int>(a); // int x = 1969
auot z = std::get<string>(a) // Error.
Time (chrono) library
In c++1 we have now the namespace std::chrono
.
Example of a timepoint:
#include <chrono>
using std;
using chrono;
auto t1 = steady_clock::now();
// ...
auto t2 = steady_clock::now();
auto diff = t2 - t1;
cout << duration_cast<milliseconds>(diff).count() << "\n";
Containers
- Move semantics are supported in all containers
- SUSPECT of any
std::move
operations you do, because probably you can do that without them (unless you are using it on a constructor) - Use the move semantics as much as you can, because it will be so much cheaper than copying the values over and over again.
- SUSPECT of any
- Improved support for const_iterator
vector::data()
acess to the underlying array- immutability in elements in
set
andmultiset
push_back
? nah, better theemplace_back
. It will accelerate the application so much that you'll notice it.- with vectors, sometimes your capacity is a lot bigger than your actual size, so in that cases you can use the
shrink_to_fit
so the extra capacity is freed.
Concurrency
Threads
Since C++11, the language itself has now a std::thread solution independent of any OS. This implies:
- A new memory model
thread_local
variables- Atomic types
- Concurrency portable abstractions
- thread
- mutex
- condition_variable
- lock_guard, unique_lock
- promise, future, packaged_task
void f1();
void g() {
std::thread t {f1}; // starts a thread executing f1
other_task();
t.join(); // wait for the thread to finish
}
Simplified argument parsing:
void f1(int x);
struct f1 {
f2(int px) : x{px} {}
void operator()();
int x;
}
void g() {
std::thread t1{f1, 10}; // Run f1(10)
std::thread t2{f2{20}}; // Run f2{20}.operator()()
std::thread t3{ [] { f1(42); } }; // Run lambda
t1.join();
t2.join();
t3.join();
}
Shared objects access
What to do when you have multiple threads messing around with your variables and making your program "unstable"?
[MUTEX]: has entered the room.
[SPEED GAINED BY THE THREADS]: has exited the room
But if you really have to use them, please use lock_guard
:
std::mutex m;
int x = 0;
void f() {
std::lock_guard<std::mutex> l{m}; // Acquires the lock
++x;
} // releases the lock
void g() {
using namespace std;
thread t1(f);
thread t2(f);
t1.join();
t2.join();
cout << x << endl;
}