The big STL Algorithms tutorial: the memory header
Sandor Dargo
Posted on February 2, 2022
We are slowly reaching the end of the big STL algorithm tutorial, and in this last but one part we are going to cover a record high 14 operations that are part of the <memory> header. I decided to take all of them because they are quite similar to each other.
uninitialized_copyuninitialized_copy_nuninitialized_filluninitialized_fill_nuninitialized_moveuninitialized_move_nuninitialized_default_constructuninitialized_default_construct_nuninitialized_value_constructuninitialized_value_construct_ndestroydestroy_ndestroy_atconstruct_at
uninitialized_copy / uninitialized_copy_n
std::uninitialized_copy takes an input range and copies the elements to an uninitialized area that is denoted by an iterator pointing at the beginning of the output range.
Potentially, you can also set the execution policy.
The only difference std::uninitialized_copy_n has compared to std::uninitialized_copy is that it doesn't take the input range by two iterators defining the beginning and the end of the input range, but instead it takes the beginning of the range and the size.
#include <algorithm>
#include <iostream>
#include <memory>
#include <string>
#include <tuple>
#include <vector>
int main()
{
std::vector<std::string> v = {"This", "is", "an", "example"};
auto sz = std::size(v);
if(void *pbuf = std::aligned_alloc(alignof(std::string), sizeof(std::string) * sz))
{
try
{
auto first = static_cast<std::string*>(pbuf);
auto last = std::uninitialized_copy(std::begin(v), std::end(v), first);
for (auto it = first; it != last; ++it) {
std::cout << *it << ' ';
}
std::cout << '\n';
std::destroy(first, last);
}
catch(...) {}
std::free(pbuf);
}
std::string* p;
std::tie(p, sz) = std::get_temporary_buffer<std::string>(v.size());
sz = std::min(sz, v.size());
std::uninitialized_copy_n(v.begin(), sz, p);
for (std::string* i = p; i != p+sz; ++i) {
std::cout << *i << ' ';
i->~basic_string<char>();
}
std::return_temporary_buffer(p);
}
uninitialized_move / uninitialized_move_n
std::uninitialized_move and std::uninitialized_move_n - unsurprisingly - work very similarly compared to their copy versions, but instead of copying the items from the input ranges, they move the items.
The range to be moved is either defined by two iterators denoting its beginning and end (uninitialized_move) or by an iterator to its beginning and the number of positions to fill (uninitialized_move_n).
The output range is defined only by its beginning as usual, and as a caller, we have to make sure that it can accommodate all the necessary elements to avoid undefined behaviour.
Before all the other parameters, we can also define an execution policy.
#include <cstdlib>
#include <iomanip>
#include <iostream>
#include <memory>
#include <string>
void print(auto rem, auto first, auto last) {
for (std::cout << rem; first != last; ++first)
std::cout << std::quoted(*first) << ' ';
std::cout << '\n';
}
int main() {
std::string in[] { "Home", "Work!" };
print("initially, in: ", std::begin(in), std::end(in));
if (
constexpr auto sz = std::size(in);
void* out = std::aligned_alloc(alignof(std::string), sizeof(std::string) * sz)
) {
try {
auto first {static_cast<std::string*>(out)};
auto last {first + sz};
std::uninitialized_move(std::begin(in), std::end(in), first);
// comment the previous line and uncomment the next one
// to see uninitialized_move_n in action
// std::uninitialized_move_n(std::begin(in), sz, first);
print("after move, in: ", std::begin(in), std::end(in));
print("after move, out: ", first, last);
std::destroy(first, last);
}
catch (...) {
std::cout << "Exception!\n";
}
std::free(out);
}
}
uninitialized_fill / uninitialized_fill_n
std::uninitialized_fill and std::uninitialized_fill_n fills an uninitialized memory area with a given value.
The range to be filled is either defined by two iterators denoting its beginning and end (uninitialized_fill) or by an iterator to its beginning and the number of positions to fill (uninitialized_fill_n).
In both cases, the value comes after, and execution policy can also be defined.
#include <algorithm>
#include <iostream>
#include <memory>
#include <string>
#include <tuple>
int main()
{
std::string* p;
std::size_t sz;
std::tie(p, sz) = std::get_temporary_buffer<std::string>(4);
std::uninitialized_fill(p, p+sz, "Example");
// comment the previous line and uncomment the next one
// to see uninitialized_fill_n in action
// std::uninitialized_fill_n(p, sz, "Example");
for (std::string* i = p; i != p+sz; ++i) {
std::cout << *i << '\n';
i->~basic_string<char>();
}
std::return_temporary_buffer(p);
}
uninitialized_default_construct / uninitialized_default_construct_n
std::uninitialized_default_construct and std::uninitialized_default_construct_n fills an uninitialized memory area with the default initialized instances of the contained type.
The range to be filled is either defined by two iterators denoting its beginning and end (uninitialized_default_construct) or by an iterator to its beginning and the number of positions to fill (uninitialized_default_construct_n).
In both cases, the value comes after, and execution policy can also be defined.
#include <iostream>
#include <memory>
#include <string>
struct S { std::string m{ "Default value" }; };
int main()
{
constexpr int n {3};
alignas(alignof(S)) unsigned char mem[n * sizeof(S)];
auto first {reinterpret_cast<S*>(mem)};
auto last {first + n};
std::uninitialized_default_construct(first, last);
// comment the previous line and uncomment the next one
// to see uninitialized_default_construct_n in action
// std::uninitialized_default_construct_n(first, n);
for (auto it {first}; it != last; ++it) {
std::cout << it->m << '\n';
}
std::destroy(first, last);
}
We should also note that std::uninitialized_default_construct and std::uninitialized_default_construct_n do not zero-fill the memory area for trivial types!
#include <iostream>
#include <memory>
#include <cstring>
int main()
{
// Notice that for "trivial types" the uninitialized_default_construct
// generally does not zero-fill the given uninitialized memory area.
int v[] { 1, 2, 3, 4 };
const int original[] { 1, 2, 3, 4 };
std::uninitialized_default_construct(std::begin(v), std::end(v));
// comment the previous line and uncomment the next one
// to see uninitialized_default_construct_n in action
// std::uninitialized_default_construct_n(std::begin(v), std::distance(std::begin(v), std::end(v)));
for (const int i : v) { std::cout << i << ' '; }
std::cout << '\n';
// Maybe undefined behavior, pending CWG 1997.
std::cout <<
(std::memcmp(v, original, sizeof(v)) == 0 ? "Unmodified\n" : "Modified\n");
// The result is unspecified.
}
uninitialized_value_construct / uninitialized_value_construct_n
uninitialized_value_construct / uninitialized_value_construct_n has the same signatures as uninitialized_default_construct and uninitialized_default_construct_n.
Besides, they practically work the same way for object types, they both invoke the default constructor of the contained type. However while uninitialized_default_construct and uninitialized_default_construct_n didn't zero-fill trival types (POD types), uninitialized_value_construct / uninitialized_value_construct_n will do it.
Here is a nice little comparison between default and value initialization.
Here is a merged example:
#include <iostream>
#include <memory>
#include <string>
struct S { std::string m{ "Default value" }; };
int main()
{
constexpr int n {3};
alignas(alignof(S)) unsigned char mem[n * sizeof(S)];
auto first {reinterpret_cast<S*>(mem)};
auto last {first + n};
std::uninitialized_value_construct(first, last);
// comment the previous line and uncomment the next one
// to see uninitialized_default_construct_n in action
// std::uninitialized_value_construct_n(first, n);
for (auto it {first}; it != last; ++it) {
std::cout << it->m << '\n';
}
std::destroy(first, last);
// Notice that for "trivial types" the uninitialized_value_construct
// zero-fills the given uninitialized memory area.
int v[] { 1, 2, 3, 4 };
for (const int i : v) { std::cout << i << ' '; }
std::cout << '\n';
std::uninitialized_value_construct(std::begin(v), std::end(v));
// comment the previous line and uncomment the next one
// to see uninitialized_default_construct_n in action
// std::uninitialized_value_construct_n(std::begin(v), std::distance(std::begin(v), std::end(v)));
for (const int i : v) { std::cout << i << ' '; }
std::cout << '\n';
}
destroy / destroy_n / destroy_at
If you have read the code snippets thoroughly in this article, you could already see std::destroy at work and I'm sure you can guess how std::destroy_n works compared to it.
std::destroy and std::destroy_n take a range of objects and invoke the destructor of those. std::destroy takes a pair of iterators, while std::destroy_n takes the beginning of a range and the number of objects to be destroyed. It's also possible to set the execution policy.
Both can be implemented as a loop iterating over the range and in the body they call std::destroy_at which takes only one parameter, a pointer.
#include <memory>
#include <new>
#include <iostream>
struct Tracer {
int value;
~Tracer() { std::cout << value << " destructed\n"; }
};
int main()
{
alignas(Tracer) unsigned char buffer[sizeof(Tracer) * 8];
for (int i = 0; i < 8; ++i) {
new(buffer + sizeof(Tracer) * i) Tracer{i}; //manually construct objects
}
auto ptr = std::launder(reinterpret_cast<Tracer*>(buffer));
std::destroy(ptr, ptr+8);
// you can alternatively try this
// std::destroy_n(ptr, 8);
// or this
// for (int i = 0; i < 8; ++i)
// std::destroy_at(ptr + i);
}
construct_at
std::construct_at takes a memory address of an object of type T and a variable number of parameters and it constructs a T object with all the passed arguments.
#include <iostream>
#include <memory>
struct S {
int x;
float y;
double z;
S(int x, float y, double z) : x{x}, y{y}, z{z} { std::cout << "S::S();\n"; }
~S() { std::cout << "S::~S();\n"; }
friend std::ostream& operator<<(std::ostream& os, const S& o) {
os << "S { x=" << o.x << "; y=" << o.y << "; z=" << o.z << "; };\n";
return os;
}
};
int main()
{
alignas(S) unsigned char storage[sizeof(S)];
S* ptr = std::construct_at(reinterpret_cast<S*>(storage), 42, 2.71828f, 3.1415);
std::cout << *ptr;
std::destroy_at(ptr);
}
Conclusion
This time, we learned about algorithms for dynamic memory management from the <memory> header. To be completely honest with you, in the almost 9 years that I spent with C++, I've never used them and there was no situation when I would have had to use them.
Still, it's good to know about them and even better to avoid dynamic memory management as much as you can and let the compiler do it for us.
In the very last part of this series, we are going to conclude what we learned about STL algorithms, the key points to keep in mind.
Stay tuned!
Connect deeper
If you liked this article, please
- hit on the like button,
- subscribe to my newsletter
- and let's connect on Twitter!
Posted on February 2, 2022
Join Our Newsletter. No Spam, Only the good stuff.
Sign up to receive the latest update from our blog.