RE: [boost] smart pointer puzzle

From: Dave Harris (brangdon_at_[hidden])
Date: 2002-04-08 16:46:19

In-Reply-To: <KBEEJNHPFAMCGLNKDFEFOEOBDEAA.rdamon_at_[hidden]>
On Mon, 8 Apr 2002 09:44:42 -0400 Richard Damon (rdamon_at_[hidden]) 
wrote:
> > In general, to be useful, you must be able to create a smart
> > pointer from a raw pointer multiple times (or find some way of
> > making "this" a smart pointer).  
> 
> To my experience most *non-invasive* smart pointers have this 
> limitation. An invasive smart pointer put the shared state in the
> object making it easy to adopt a raw pointer. A non-invasive
> pointer gets no such help from the object, to convert the raw
> pointer into a shared pointer normally requires a map of some
> sort, which makes the implementation much more heavy weight. I
> will agree that most smart pointers I have seen where invasive, this are
> much easier to write, but I understand Boosts decision to go 
> non-invasive, there are times when invasive is not an option.

I gather some clever people have recently figured out a compile-time test 
for a class member. We could use this to write a smart pointer which has 
an invasive specialisation for classes which inherit from 
boost::ref_counted, and which defaults to a std::map non-invasive 
implementation otherwise.

Currently I don't use the boost pointer, mainly because of the problems 
with multiple counts. Efficiency is another concern: invasive smart 
pointers tend to be a lot more efficient than non-invasive ones. I think I 
would use a boost pointer like the above, though.

-- Dave Harris