Re: iterator_adaptor and statefull policies.

From: rogeeff (rogeeff_at_[hidden])
Date: 2001-12-14 01:19:01

--- In boost_at_y..., Jeremy Siek <jsiek_at_c...> wrote:
> Hi Gennadiy,
> 
> On 12/14/01 12:46 AM, "rogeeff" <rogeeff_at_m...> wrote:
> 
> > Hi,
> > 
> > I have a question about iterator_adaptor. The iterator_adaptor 
seems
> > to assume that Policies are stateless. For example, it never copy
> > them in copy constructor and iterator. Why it was designed that 
way?
> 
> Sure it does. The m_iter_p member gets copied, and the policies 
object
> Is inside m_iter_p, so it gets copied.

Oops. Sorry.

> 
> > Another related questions:
> > Why does not iterator_adapter has a copy_constructor/assignment
> > accepting Base iterator?
> 
> There is a constructor that accepts the Base iterator, though of 
course it
> is not technically a copy constructor.
> 
> As for the assignment operator... I guess we didn't see a real need 
for it.
> Do you see a real need for it?

Rationale for my questions is performance considerations. Let say I 
need to process input stream ~ 20000 times per second. I have 
iterator over this stream which is implemented using iterator_adaptor 
over light weight Base iterator and statefull Policies ( meaning that 
copy constructor for the
m could be expensive). I have at least 2 choises: Instantiate my 
iterator every time (in which case I would like to prevent Policies 
from copying) or I can instantiate it once and than reset every time 
with new Base iterator (plus Policies initialize call). Any other 
variants?

> 
> > Why does not iterator_adapter has a constructor accepting some
> > argument to be passed to policy constructor?
> 

> This was considered, but we felt that would complicate things. And 
you can
> just create a make_xxx_iterator function that hides the 
complication of
> creating the policy object explicitly.

make_xxx_iterator function would not prevent Policies copying.

[SNIP]

> 
> Cheers,
> Jeremy
> 
> -- 
> Jeremy Siek                          http://www.osl.iu.edu/~jsiek
> Ph.D. Student, Indiana Univ. B'ton   email: jsiek_at_o...
> C++ Booster (http://www.boost.org)   office phone: [phone]

Gennadiy.