From: fred bertsch (fred.bertsch_at_[hidden])
Date: 2007-10-19 18:12:25
> Thanks for your comments. Both options seems reasonable. I think the
> user of a splay_tree expects splaying as the default option, so I'm
> starting to think that avoiding const versions could be a better choice
> (generic code using non-const containers would work). Users that are
> sure about thread-safety can uncast the container before passing it to
> the algorithm. Anyway, I'm afraid my opinion might change again ;-)
I like this approach. A splay tree is inherently non-const. Attempting
to support const operations is a mistake for a variety of reasons.
Removing all const operations on a collection will have a number of
negative consequences, but I'd like to believe that a lot of that is
required when you have a collection that behaves like a splay tree.
I can think of a bunch of ways of softening the blow to the STL
collection concept, but they all revolve around hiding the fact that
you don't want to do non-const lookups with a splay tree. I'd rather
advertise that behavior rather than trying to hide it.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk