Boost logo

Threads-Devel :

From: David Abrahams (dave_at_[hidden])
Date: 2007-02-03 22:28:19


"Felipe Magno de Almeida" <felipe.m.almeida_at_[hidden]> writes:

> On 2/3/07, David Abrahams <dave_at_[hidden]> wrote:
>> "Felipe Magno de Almeida" <felipe.m.almeida_at_[hidden]> writes:
>>
> [snipped]
>
>>
>> Yes, but if you use the SSO you can (usually) do stack allocation and
>> (on 32-bit platforms) pass the address across the thread boundary in
>> that long. Am I missing something?
>
> I think so, because the stack allocation (which is already done, but
> without any option for heap allocation) would still require
> synchronization in boost::thread constructor, since the constructor
> couldn't return before the copy happened. The heap allocation frees
> the constructor from this synchronization since we don't have any
> copies.

Of course. Silly me.

-- 
Dave Abrahams
Boost Consulting
www.boost-consulting.com

Threads-Devel list run by bdawes at acm.org, david.abrahams at rcn.com, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk