Re: 2nd: [boost] shared_ptr proposal

From: Greg Colvin (gcolvin_at_[hidden])
Date: 2001-07-16 13:57:53

From: Beman Dawes <bdawes_at_[hidden]>
> At 06:35 AM 7/13/2001, Peter Dimov wrote:
> 
>  >I think that Christian has a valid question that is not addressed by 
> Modern
>  >C++ Design:
>  >
>  >What are the complexity requirements of the boost::shared_ptr<> member
>  >functions?
> 
> They are currently unspecified.  At least for now, that's probably 
> best.  See below.
> 
>  >He asks, if I understand correctly, not how boost::shared_ptr is 
> currently
>  >_implemented_, but how it's _defined_.
>  >
>  >A question that will become more important if/when shared_ptr moves into
>  >std.
> 
> My guess is the committee won't want to specify complexity requirements, 
> because they limit implementors and complicate the specification without 
> apparent benefit to users.  Almost certainly, someone will point out that 
> some error-checking implementations might become non-conforming if they 
> have to meet a stringent complexity requirement.

Ditto.

> In other words, it will be seen as a quality of implementation issue, not a 
> specification issue.
> 
> But remember that anticipating how the LWG will look at a proposal isn't a 
> always possible.  There is a very wide range of viewpoints present, 
> including vendors, implementors, teachers, presenters, authors.  Some view 
> a library through the lens of user problem reports, for example, and so see 
> many issue very differently from typical Boost members.
> 
>  >An interesting related question is "why does shared_ptr have use_count?"
> 
> As an aid to writing test cases and debugging displays.  One of the 
> progenitors had use_count(), and it was useful in tracking down bugs in a 
> complex project that turned out to have cyclic-dependencies.

Right.  I would prefer to make use_count an optional feature, since
it constrains implementations unnecessarily, and I haven't seen any
use for it other than testing and debugging.

I used to maintain STL-style requirements tables for my auto_ptr and
counted_ptr proposals, an idea which might be worth resurrecting for
shared_ptr.  I forget the details, but the counted_ptr requirements
were stated so as to be implementable without actually counting.