I am reading Anthony Williams's "C++ Concurrency in Action, Second Edition", and I encountered this code:
template<typename T>
class lock_free_queue
{
private:
void set_new_tail(counted_node_ptr &old_tail,
counted_node_ptr const &new_tail)
{
node* const current_tail_ptr=old_tail.ptr;
while(!tail.compare_exchange_weak(old_tail,new_tail) &&
old_tail.ptr==current_tail_ptr);
if(old_tail.ptr==current_tail_ptr)
free_external_counter(old_tail);
else
current_tail_ptr->release_ref();
}
public:
void push(T new_value)
{
std::unique_ptr<T> new_data(new T(new_value));
counted_node_ptr new_next;
new_next.ptr=new node;
new_next.external_count=1;
counted_node_ptr old_tail=tail.load();
for(;;)
{
increase_external_count(tail,old_tail);
T* old_data=nullptr;
if(old_tail.ptr->data.compare_exchange_strong(
old_data,new_data.get()))
{
counted_node_ptr old_next={0};
if(!old_tail.ptr->next.compare_exchange_strong(
old_next,new_next))
{
delete new_next.ptr;
new_next=old_next;
}
set_new_tail(old_tail, new_next);
new_data.release();
break;
}
else
{
counted_node_ptr old_next={0};
if(old_tail.ptr->next.compare_exchange_strong(
old_next,new_next))
{
old_next=new_next;
new_next.ptr=new node;
}
set_new_tail(old_tail, old_next);
}
}
}
};
The remaining part is in GitHub, which is from listing_7.19.cpp to listing_7.22.cpp.
Here I encounter some problems, that I can't understand this part:
for(;;)
{
increase_external_count(tail,old_tail);
T* old_data=nullptr;
if(old_tail.ptr->data.compare_exchange_strong(
old_data,new_data.get()))
{
counted_node_ptr old_next={0};
if(!old_tail.ptr->next.compare_exchange_strong(
old_next,new_next))
{
delete new_next.ptr;
new_next=old_next;
}
set_new_tail(old_tail, new_next);
new_data.release();
break;
}
else
{
counted_node_ptr old_next={0};
if(old_tail.ptr->next.compare_exchange_strong(
old_next,new_next))
{
old_next=new_next;
new_next.ptr=new node;
}
set_new_tail(old_tail, old_next);
}
}
First, I think old_tail is something not null, why does it compare this to old_data (which is a nullptr) in this statement?
if(old_tail.ptr->data.compare_exchange_strong( old_data,new_data.get()))
And, what is the above for loop actually doing?
I completely can't understand this.
If the question is why put this:
T* old_data=nullptr;
if(old_tail.ptr->data.compare_exchange_strong(old_data,new_data.get()))
//...
Rather than say this:
if(old_tail.ptr->data.compare_exchange_strong(nullptr,new_data.get()))
//...
That's because compare_exchange_strong() returns the value at the point of compare in the first argument so it requires a modifiable l-value (a reference).
In the specimen code the value returned isn't used. But there isn't a version of the member that accepts a constant.
If the question is "what is the for(;;) loop doing?"
The answer is using compare_exchange_strong() to try and be the thread that either populates data in the tail or adds a new tail to populate.
The logic is lockless (if the relevant atomic operations are lockless on the target platform - not guaranteed) but the downside is that may require several attempts (iterations) for a thread to be 'the winner' and under high load there's a risk of 'live lock' behaviour because it's not guaranteed when a given thread will succeed.
Lock-free implementations sound attractive but sometimes come with downsides that can be worse than locking and completing the work in a single pass.
Of course introducing a lock in the form of a std::mutex doesn't guarantee any fairness and (again) high load scenarios may result in live-lock behaviours. But the point remains that the implementation shown isn't necessarily a silver bullet.
This code demonstrates the first point:
#include <iostream>
#include <atomic>
int main() {
int value1{1};
int value2{2};
std::atomic<int *> x(&value1);
int * e=nullptr;
auto r{x.compare_exchange_strong(e,&value2)};
std::cout<<r<<' '<<e<<' '<<&value1<<'\n';
return 0;
}
Typical Output:
0 0x7ffe378b5340 0x7ffe378b5340
Notice that the address returned for e is the address of value1.