STL use of not (was: Re: Alpha...)

[Kultarr <kultarr@rocketmail.com>]

---Hary Walsh <hwalsh@tellabs.ie> wrote:
>
> >>>>> "Kasper" == Kasper Peeters <t16@nikhef.nl> writes:
> 
>     Kasper> At this point, I would again like to make a strong
>     Kasper> recommendation in favour of the standard library,
>     Kasper> especially since g++ now has a good implementation of
>     Kasper> 'string' and handles the basic STL containers ('vector',
>     Kasper> 'list' and so on) very well.
> 
> 
> Warning about STL.  In situations where many STL containers are
> needed, it is customary to set a more granular preallocaiton buffer
> size.  This reduces memory wastage considerably.  However in the C++
> draft the methods for manipluating allocator sizes have been removed.
> 
> I personally feel that STL is not necessary.  We should be able to
> short list the containers we need ( vector, map, pair, string ) and
implement
> them ourselves.  If portability is an issue, we can keep the
interface
> of STL.
> 

Liberal use of STL may result in a real code explosion with some of
the dumber C++ compilers.  They just do a piss-poor job with template
generation, and your code size can grow exponentially with little
gain in quality.

Of some of the classes I could offer, I have a reference counting
String class which has been used by others.  It works fairly well,
although, it could probably use a bit of polish.  It has been tested
in several different applications.  I also have a implementations of
lists and a hash table that I could offer.  I have implemented a 
pair of classes for reference counting. Probably some others, but
I think those were what was mentioned on the short list.

OTOH, these classes are not standard and some of them use templates
as well.

Anyway, I'm not pushing an agenda -- just wanted to make the offer.

===
Kultarr

_____________________________________________________________________
Sent by RocketMail. Get your free e-mail at http://www.rocketmail.com