One of my primary objections to the current design is the use of unspecified policy parameters in the template parameter list of the main data structures. I think that it is important to limit the number of top-level template parameters to only non-policy parameters.
of course this can be discussed ... i somehow like this explicit way of specifying parameters, but this is probably my personal preference ...
But they're not explicit; the parameters are entirely too vague. If I look at the class definition, I should be able to able to generate some reasonable hypothesis about the kinds of types accepted by the template parameters.
i think it would be great to have a clear policy among all (or all new) boost libraries ... then this can be discussed once and only once ... i would prefer to have clear interface guidelines instead of arguing about personal preferences :)
There's a lot of value in having a consistent design style, but trying to pin one down for all of Boost may be a little bit beyond the scope of this review :) I am curious to hear some other opinions on this.
... i haven't really worked with c++0x, yet ... but isn't this address by variadic templates?
Yes and no. You can certainly support arbitrary policies with variadic templates (basically a tuple of policies). I built a small experimental library that did exactly this in 2009. I didn't like the result because it made the resulting classes so vague.