# First garbage collector

Today I tried to learn how to make a simple garbage collector for a future project, it didn't need to be fast nor complex nor optimal, it just needed to work. After lots of searching and reading, I was able to make something work.

So, how can I improve it from here? At some point I guess I'll have to ditch the std::array (the small size is just for testing purposes). Do I actually need to use new and delete, or can I just keep using STL containers?

For what I've read, I have a couple options: mark and compact, copying garbage collection, generational garbage collector. Which of them would be the more natural evolution from what I've done?

#include <array>
#include <iostream>
#include <string>

using std::cout;
using std::endl;

template <typename T>
struct Handle {
bool marked {};
T value {};
Handle * next {};
};

template <typename T>
class Heap {
std::array<Handle<T>, 10> heap;

Handle<T> * root {};
Handle<T> * free_list {};

void push(Handle<T> * handle) {
handle->next = free_list;
free_list = handle;
}

void reset() {
for (size_t i = 0; i < heap.size(); ++i) {
heap[i].marked = false;
}
}

void mark(Handle<T> * handle) {
if (handle && !handle->marked) {
handle->marked = true;
mark(handle->next);
}
}

void sweep() {
free_list = nullptr;
for (size_t i = 0; i < heap.size(); ++i) {
if (!heap[i].marked) {
heap[i].value = {};
push(&heap[i]);
}
}
}

public:
Heap() {
for (size_t i = 0; i < heap.size(); ++i) {
push(&heap[i]);
}
}

template <typename U>
Handle<T> * allocate(U && u) {
if (!free_list) {
reset();
mark(root);
sweep();
if (!free_list) {
}
}
Handle<T> * handle = free_list;
free_list = free_list->next;
handle->value = std::forward<T>(u);
handle->next = nullptr;
return handle;
}

Handle<T> * keep_alive(Handle<T> * handle) {
handle->next = root;
root = handle;
return handle;
}

void signal() {
root = nullptr;
}
};

int main() {

Heap<std::string> heap;

auto a = heap.allocate("A");
auto b = heap.keep_alive(a);

a = heap.allocate("B");
auto c = heap.keep_alive(a);

a = heap.allocate("C");
a = heap.allocate("D");
a = heap.allocate("E");

cout << a->value << endl;
cout << b->value << endl;
cout << c->value << endl;

return 0;
}

• Why on earth would you want a GC in C++. The C++ resource management is so much better already. WE have fine grain deterministic garbage collection built into the libraries (They are called smart pointers). If you want to make the language worse the feel free to add a GC but you should prefer to use the facilities already in the language that make GC obsolete. – Martin York Feb 16 '17 at 20:12
• @LokiAstari I'm building a scripting language in C++ similar to Lisp and std::shared_ptrs won't work if there's cyclic references. – João Pires Feb 16 '17 at 20:57
• Then use std::weak_ptrs where you can introduce a cycle. That is what they are there for. – Martin York Feb 17 '17 at 0:16
• Like, I said, I'm building a scripting language, you can't break cycles if you don't know what will happen at runtime. – João Pires Feb 17 '17 at 0:31
• You should take a look at gcpp – Guillaume Racicot Feb 17 '17 at 4:00

From a design and interface point of view, I find the use of:

auto a = heap.allocate("A");
cout << a->value << endl;


unnatural. I would expect a to act like a pointer to std::string since it's the value returned from a function named allocate using an object of type Heap<std::string>.

It would be more natural if:

1. *a evaluated to std::string& (std::string const& if a were const)
2. a-> evaluated to std::string* (std::string const* if a were const)

Then, you could replace a->value by *a.

auto a = heap.allocate("A");
cout << *a << endl;


P.S. I haven't spent enough time delving into the implementation of Heap to suggest how it can be done.

• I believe that implementing the pointer-like semantics for the allocated objects can be done by implementing the Handle::operator*() and Handle::operator->() operators. – Maarten Bamelis Feb 17 '17 at 10:51
• @MaartenBamelis, that was my first thought. However, the return type of Heap::allocate is Handle<T>*, not Handle<T>. The internals of Heap need to be changed significantly to allow for such a change. – R Sahu Feb 17 '17 at 16:11
• an important detail that I missed! In my opinion, the added value of the pointer-like semantics does warrant rethinking the Heap class. As you recognized before, it feels more natural and intuitive that way. – Maarten Bamelis Feb 17 '17 at 17:21

I'm not sure there's such a thing as a simple mark-sweep garbage-collector. Many start off that way, but they all seem to grow more complex, particularly as they get tuned for performance.

I'm not an expert in the domain, but I can review some of the C++ style issues.

Firstly, thanks for providing a well-written and complete program with test cases. That always helps the reviewer! I compiled using GCC with my usual warnings enabled, and the only criticism it made was an Effective C++ recommendation: class Heap’ has pointer data members but does not override its copy constructor or assignment operator. I don't think you need them, so:

Heap(const Heap&) = delete;
Heap operator=(const Heap&) = delete;


There's a numeric literal 10 which seems to be arbitrary - it seems to be maximum number of roots. It's probably better as a named constant, to make this more obvious. You might even want to make it a template parameter of your class.

Loops such as this one:

void reset() {
for (size_t i = 0; i < heap.size(); ++i) {
heap[i].marked = false;
}
}


are (IMO) more readable if written with range-based for:

void reset() {
for (auto& handle: heap)
handle.marked = false;
}


You use std::forward() but don't include <utility>; similarly for std::bad_alloc and <new>. Don't rely on your implementation's transitive includes, but always include the documented headers for what you use. I'd also move the <iostream> stuff down to live with your main(), as that's really only part of your test suite.

My edited version:

#include <array>
#include <new>
#include <utility>

template <typename T>
struct Handle {
bool marked {};
T value {};
Handle * next {};
};

static const size_t heap_capacity = 10;

template <typename T, size_t N = heap_capacity>
class Heap {

Heap(const Heap&) = delete;
Heap operator=(const Heap&) = delete;

std::array<Handle<T>, N> heap{};

Handle<T> * root {};
Handle<T> * free_list {};

void push(Handle<T> * handle) {
handle->next = free_list;
free_list = handle;
}

void reset() {
for (auto& handle: heap)
handle.marked = false;
}

void mark(Handle<T> * handle) {
if (handle && !handle->marked) {
handle->marked = true;
mark(handle->next);
}
}

void sweep() {
free_list = nullptr;
for (auto& handle: heap) {
if (!handle.marked) {
handle.value = {};
push(&handle);
}
}
}

public:
Heap() {
for (auto& handle: heap) {
push(&handle);
}
}

template <typename U>
Handle<T> * allocate(U && u) {
if (!free_list) {
reset();
mark(root);
sweep();
if (!free_list) {
}
}
Handle<T> * handle = free_list;
free_list = free_list->next;
handle->value = std::forward<T>(u);
handle->next = nullptr;
return handle;
}

Handle<T> * keep_alive(Handle<T> * handle) {
handle->next = root;
root = handle;
return handle;
}

void signal() {
root = nullptr;
}
};

#include <string>
#include <iostream>
int main() {
using std::cout;
using std::endl;

Heap<std::string> heap;

auto a = heap.allocate("A");
auto b = heap.keep_alive(a);

a = heap.allocate("B");
auto c = heap.keep_alive(a);

a = heap.allocate("C");
a = heap.allocate("D");
a = heap.allocate("E");

cout << a->value << endl;
cout << b->value << endl;
cout << c->value << endl;

return 0;
}
`