Subject: Re: [boost] Coverity Static Code Analysis
From: Stefan Seefeld (seefeld_at_[hidden])
Date: 2009-02-05 08:39:45
Mathias Gaunard wrote:
> If low coupling is maintained, I don't see how having dependencies is
> such a problem.
> AFAIK, the interfaces of boost libraries have kept stable once they
> were accepted in release.
I don't think this is true. Sometimes interfaces have changed, which has
caused quite some discussion on this list.
I'm pointing this out because I believe it would be naive to assume that
relying on external APIs comes at no cost. There is always a trade-off
to make. Reuse of design and code is often a good thing, but not always.
Add to that an additional conceptual complexity. If you can express the
design of your software in simple terms, you win. However, if, in order
to understand this design, users have to first grasp very generic
concepts documented and coded elsewhere, it makes it already harder to
understand. The idiosynchratic / proverbial
'AbstractModelImplementationFactoryProvider' from Java also exists (with
different spelling) in modern C++.
> A tuple is just an enhancement to the language to express your
> algorithms with. Some people even consider C++ with that kind of
> library addition to be a new language altogether, and that is why they
> use boost.
-- ...ich hab' noch einen Koffer in Berlin...
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk