From: Beman Dawes (beman_at_[hidden])
Date: 2000-05-10 07:28:00
At 08:45 PM 5/7/00 +0300, Peter Dimov wrote:
>What is the Boost policy on alternative implementations?
>... what is the Boost policy on alternative *interfaces*?
Thanks for raising these two questions; we don't have a policy but
If one implementation isn't significantly better than the other, why
supply both? It is very confusing to users. On the other hand, if
one implementation is significantly better, why supply the poorer
one? Of course, two implementations may have different
characteristics, so it isn't always a case of "better" and "poorer",
but the user confusion issue still exists.
Beyond that, its boring to keep reworking implementations. Sure
someone else can do a bit better job that the author, but who cares?
Let's move on to new libraries. Boost should probably have a page of
Challenges for people who are looking for something to do, so they
don't just endlessly reimplement code that is already working well
It seems to me that it just too confusing for users if boost has
alternate interfaces to the same underlying functionality, except in
exceptional circumstances. Normally, boost should just pick one of
the alternates and stick with it.
What may happen, of course, is that we start out with a library with
one interface, and then along comes someone with a much better
interface. If it really is MUCH better, not just a bit better, and
it matters, then we might want to obsolete the old interface and
supply the new one under a different name or namespace. But
stability is an important and highly desirable attribute of
libraries. And one interface can be better than another in a
theoretical sense, but it would have to bring real benefit from the
user's viewpoint to actually matter.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk