Re: [boost] Nonlocking data structures.

From: Patrick Twohig (p-twohig_at_[hidden])
Date: 2008-04-21 17:49:52

The docs state the pointers are "as close to raw pointers as possible" and
that makes sense, heh.

On Mon, Apr 21, 2008 at 2:20 PM, Tim Blechmann <tim_at_[hidden]> wrote:

> On Mon, 21 Apr 2008 09:37:23 -0700, Patrick Twohig wrote:
>
> > I thought boost::shared_ptr was lock-free and thread safe on the major
> > platforms supported by boost.  I could be wrong though.
>
> there are the following issues with boost::shared_ptr:
>
> thread 1 accesses object, thread 2 frees it object ... one would need a
> thread-local copy of the shared_ptr to dereference it or use hazard
> pointers/pass-the-buck
> --
> shared_ptr<foo_t> foo (new foo_t());
>
> void thread_1(void) {
>    foo->bar();
> }
>
> void thread_2(void) {
>    foo.reset();
> }
> --
>
> copying problem
> if thread 1 updates the pointer entry of foo, then thread 2 updates both
> pointer and reference count entry, then thread 1 updates the reference
> count foo would point to the object managed by bar, using the reference
> count of baz ...
> one would need a way to validate both object and reference count, lance
> suggested a tag, which would add another indirection, or an atomic double-
> word read instruction
> --
> shared_ptr<foo_t> foo (new foo_t());
> shared_ptr<foo_t> bar (new foo_t());
> shared_ptr<foo_t> baz (new foo_t());
>
> void thread_1(void) {
>    foo = bar;
> }
>
> void thread_2(void) {
>    foo = baz;
> }
> --
>
> cheers, tim
>
> --
> tim_at_[hidden]
> http://tim.klingt.org
>
> You can play a shoestring if you're sincere
>  John Coltrane
>
> _______________________________________________
> Unsubscribe & other changes:
> http://lists.boost.org/mailman/listinfo.cgi/boost
>