Re: 2nd: [boost] shared_ptr proposal

From: williamkempf_at_[hidden]
Date: 2001-07-13 13:00:53

--- In boost_at_y..., Christian Thäter <chth_at_g...> wrote:
> > Excuse me for breaking in where I probably don't belong, but I'm 
not 
> > sure I'm understanding what you're asking.  Do you expect to have 
the 
> > following semantics?
> > 
> > shared_ptr<foo> p1(new foo());
> > shared_ptr<foo> p2(p1);
> > p1.reset(new foo());
> > assert(p1.get() == p2.get());
> 
> Yes
> 
> > That certainly seems to be what you mean when you say "they 
should 
> > not break the shared resource as they do now so either update all 
> > copies or none."
> > 
> > If this is what you mean, I don't find such semantics to be 
> > intuitive, or even useful for anything but the most obscure 
designs.  
> > I wouldn't argue against inclusion of a smart pointer that works 
this 
> > way (provided concrete uses can be illustrated), but I'd 
definately
> > argue against any desire to change shared_ptr<> itself to behave 
in 
> > this manner.
> 
> That leads to the Question "Who owns a shared resource?".

All references share the ownership.  That does not mean that if I 
reset one reference I want or expect all other references to be reset 
as well.

> The actual state
> is some lazy-ownership where anyone who holds a shared_pointer can 
be 
> member in sharing it but can't reset it for the entire 'sharing-
community',
> the only use of reset() is to detach from the existing resource and
> reuse the pointer in some manner.

Exactly.

> I think reusing a object in such a manner is a poor design.

It's the most natural design.  This is the same semantics employed by 
languages that have garbage collection, though the implementation is 
different.  Calling this design "poor" is awfully strong in the face 
of that.

> In
> std::auto_ptr the ownership is strict defined and has no such
> side-effects.

There are no side effects for shared_ptr<> as it is today.  What you 
are proposing would be a side effect.

> Thats why i promoted a all or nothing practice for
> consistency reasons. So I deprecate the the current change-policy in
> boost::shared_ptr, well changeing to a new semantic might be 
unacceptable,
> but i think moving the reset and assign to be protected or removing 
them
> should be possible, making the current boost::shared_ptr immutable 
and
> more precise defined.

It is very precisely defined as is, and the ability to reassign the 
reference is key to the whole concept.

> Also i would like to see some kind of semantic like
> i proposed where 'ownership' is more coherent than actually.

Ownership isn't more coherent with your proposed semantics.  Every 
pointer *still* shares ownership with a single object.  The only 
thing you've changed is the semantics of assigning a new value to the 
pointer, and the proposed semantics cause a side effect to occur.

Bill Kempf