Re: vector<> can probably never grow to it's maximum size!

[Dhruv Matani <dhruvbird at gmx dot net>]

On Tue, 2004-10-19 at 20:44, Paolo Carlini wrote:
> Dhruv Matani wrote:
> 
> >And tries to explain more concretely what I'm trying to achieve and also
> >shows that it's do able in current C++.
> >
> Why you keep on refusing to deal with one issue at a time?

I know :-( I have to try and consciously stick to one thing at a time,
and learn to control my frivolous nature.

> 
> The below cannot fix, not even in principle, the immediate problem that
> we have with __len overflowing size_type: all the new tricks in __mt_alloc
> are not in effect in this case, ::operator new for sure will never return
> an address == _M_start (half the memory is already used), it will always
> throw, even if you ask only for __old_size + 1. 

Ah! I did not think about this. You are right on this count because
vector first allocates and then deallocates, so no matter what if half
the memory is used, you are doomed.

> Don't try to achieve the
> impossible, concentrate on the case of small memory allocations, managed
> via mt_alloc memory pools: in that case there is rooom for improvement..

Yes, I have shown in the patch that the small size cases can be dealt
with quite well with __mt_alloc<>.


Attached is a patch and test-case which shows that the patch actually
works.

> 
> Paolo.
-- 
        -Dhruv Matani.
http://www.geocities.com/dhruvbird/

The price of freedom is responsibility, but it's a bargain, because
freedom is priceless. ~ Hugh Downs

#include <vector>
int
main()
{
  std::vector<int> iv(2004); // The current year ;-)
  size_t max = 0;
  max = ~max;
  --max;
  max /= sizeof(int);
  int *f, *l;
  f = 0; // This might fail on some machines which have address
	 // registers.
  l = f + max;
  iv.insert(iv.end(), f, l);
}

Attachment: ChangeLog
Description: Text document

Attachment: patch_vector_tcc_19102004_dhruv
Description: Text document