[boost] Re: Intrusive containers: second iteration

From: David Abrahams (dave_at_[hidden])
Date: 2005-05-03 07:32:34

Olaf Krzikalla <krzikalla_at_[hidden]> writes:

> Hi,
>
> David Abrahams wrote:
>> but if it's true that "nobody seems to care very
>> much," adding the library to the review queue is probably premature at
>> best.
> I'm already aware of this problem too. Unfortunately I don't have a good 
> measurement of 'how much interest is enough interest?'. I've browsed all 
> the (so called) searchable archives of boost.devel, but didn't find 
> anything useful. Either the search function was too limited or the 
> search results couldn't be ordered in threads or there was no 'view 
> complete thread'. 

If you use gmane search, when you click on a message's subject you go
to the complete thread.

> So I have no clue, how many replies indicate enough 
> interest.
>
> And of course you have to keep boost as small as possible. 

I disagree.

> If it becomes too big, nobody will use it anymore, because nobody
> have a complete overview

Too late for that.

> and is able say, what's actually available (and IMHO boost is
> already too big, the fundamentals are coupled too tight with
> high-level code).

We just need better infrastructure for managing Boost's size.  I'm
working on that.

> OTOH intrusive containers are a very fundamental piece of code. You can 
> build all sort of containers on top of it (including STL container, 
> boost::multi_index_container), while you can't do the opposite. But I 
> admit, that the application area is limited - they serve either as a 
> base or are used in perf-critical parts. And as you wrote in another 
> post, the application programmer rarely needs to write really fast code.
> In addition, programmers coming to C++ from 'higher' languages (Java, 
> script lang.) will certainly have problems to understand the differences 
>  between intrusive and non-intrusive containers and hence don't 
> consider intrusive containers even at the right places.
> Nevertheless IMHO intrusive containers belong to boost. They are part of 
> an development process opposite to the popular one (that is, they 
> reintroduce one of the roots of the language instead of trying to 
> emulate yet another JSDK/C# library feature). I wouldn't go as far as to 
> say they belong to std (as someone emailed me), but once in a while a 
> question in a newsgroup pops up, where intrusive containers either fits 
> perfectly or at least have to be considered as a possibility.

It's not yet clear to me how important endogenous (intrusive)
containers are once you have the exogenous kind.  The main question
is: what do we need them for?

-- 
Dave Abrahams
Boost Consulting
www.boost-consulting.com