[boost] Re: shared_ptr performance

From: Beman Dawes (beman_at_[hidden])
Date: 1998-12-20 17:27:49

At 09:23 PM 12/19/98 -0700, Greg Colvin wrote:

>I'm doing more testing, and still not seeing any meaningful
>differences between direct and indirect shared_ptr.  So I
>remain inclined to publish just the indirect form for now,
>since it seems quite reasonable to choose less space when
>no large or reliable speed differences are to be seen, and
>I see no point in adding complexity just to allow a choice
>that makes no difference.  What would change my mind would
>be a test that showed a real speed advantage for the direct
>version.

While it is true that the indirect form uses less data space, it may
use more program space if there are a lot of dereferences.  And on
small chips with single execution units the indirect is likely to be
slower.

I waver as to which should be the default, but still think that both
direct and indirect should be available via partial specialization as
Dietmar suggested.

Let's make up our minds about this and get the library posted this
week! 

--Beman 

------------------------------------------------------------------------
E-group home: http://www.eGroups.com/list/boost
Free Web-based e-mail groups by eGroups.com