Skip to content

Why is std::memory_order_relaxed enough for atomic_conditional_increment? #107

@kakashidinho

Description

@kakashidinho

Hi, I'm just trying to understand the rationale behind the using of std::memory_order_relaxed in sp_counted_base_std_atomic.hpp's atomic_conditional_increment's implementation.

atomic_conditional_increment's implementation (click to expand)
inline std::int_least32_t atomic_conditional_increment( std::atomic_int_least32_t * pw ) BOOST_SP_NOEXCEPT
{
    // long r = *pw;
    // if( r != 0 ) ++*pw;
    // return r;

    std::int_least32_t r = pw->load( std::memory_order_relaxed );

    for( ;; )
    {
        if( r == 0 )
        {
            return r;
        }

        if( pw->compare_exchange_weak( r, r + 1, std::memory_order_relaxed, std::memory_order_relaxed ) )
        {
            return r;
        }
    }    
}

AFAIK this function is used by weak_ptr to obtain the shared_ptr if the obj is still alive (ref count is still > 0).
Consider the following case:

shared_ptr A;
if ((A = weak_ptr.lock()) != nullptr) // internally does `if (atomic_conditional_increment() != 0)`
{
   // modify object pointed by A
}

Shouldn't we use std::memory_order_acquire to ensure that the modification of object pointed by A won't be reordered before the nullptr check?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions