Olaf van der Spek wrote:

On Fri, Jun 16, 2017 at 2:42 PM, Peter Dimov via Boost wrote:

...

...

Hi,

Does Boost have anything like concurrency TS' atomic_shared_ptr?

No, only http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2632.html

Any plans to add it? IMO it'd be nice to have.

Sure, why not. https://github.com/boostorg/smart_ptr/commit/2964ed2379cf47778b18954b8d3501f...

Gottlob Frege wrote:

...

https://github.com/boostorg/smart_ptr/commit/2964ed2379cf47778b18954b8d3501f...

I have enormous faith in you, however, this is the simplest implementation I've seen. I assume that is because it is not lock-free? Or is it because it doesn't actually work?

I made it a bit more complex: https://github.com/boostorg/smart_ptr/commit/67d897a5330e99821adcebc82f605dd... The reason it was so simple was because it was using the already existing atomic functions. And no, it's not lock-free. Lock-freedom requires significantly more complexity (and a double-width CAS).

Olaf van der Spek wrote:

On Sat, Jun 17, 2017 at 10:49 PM, Peter Dimov via Boost wrote:

...

I made it a bit more complex:

Hi Peter,

Wow, that was fast. Thanks!

You're welcome. :-)

...

The reason it was so simple was because it was using the already existing atomic functions. And no, it's not lock-free. Lock-freedom requires significantly more complexity (and a double-width CAS).

Why not improve the existing functions and keep using those?

Since atomic_shared_ptr is a separate type, there's room for it for a spinlock, which on balance is a bit better than using a separate spinlock pool as the free functions do.