Re: [boost] [MultiIndex] "Compatible Sorting Criterion"

Subject: Re: [boost] [MultiIndex] "Compatible Sorting Criterion"
From: Joaquin M Lopez Munoz (joaquin_at_[hidden])
Date: 2008-11-01 15:41:41

David Abrahams <dave <at> boostpro.com> writes:

> 
> 
> The docs mention that the definition is cumbersome.  I think you could
> probably do much better with a formulation that relies on the concept of
> being partitioned with respect to a predicate function.  See
> 
>   www.open-std.org/jtc1/sc22/wg21/docs/papers/2001/n1313.html
> 
> and the wording changes (now in the standard) at:
> 
>   http://anubis.dkuug.dk/jtc1/sc22/wg21/docs/lwg-defects.html#270

I'm glad you bring this up: I already knew your work on N1313
by the time I was writing the compatible sorting thing, and decided
to go for my formulation instead because the partition phrasing
obscures, in my opinion, the intuition behind this: One
can use a sorted range of X's to find embedded Y's if sorting of
X's implies sorting of Y's. This intuition is a strong leit motif
in composite keys, for instance.

The partition formulation is undeniably neater technically but
IMHO it leaves the reader perplexed as to what their actual
applications are. All of this is highly subjective, of course.

Off topic: It is a pity that the partition-based semantic extension
to binary_search and similar was not also applied somehow to
std::[multi]set, as the motivations in N1313 extend naturally to
this case. Boost.MultiIndex ordered indices do provide this
extension and I know this has been the reason for its usage
in certain contexts besides its multi-indexing capabilities.

Joaquín M López Muñoz
Telefónica, Investigación y Desarrollo