From: Paul Mensonides (pmenso57_at_[hidden])
Date: 2002-10-20 00:01:53
----- Original Message -----
From: "David Abrahams" <dave_at_[hidden]>
> > As some of you know, I recently added a new preprocessor data type to
the
> > preprocessor library. I called it a "set" because it is more or less a
set
[...]
> What are the performance characteristics?
>
> Does it have an O(1) or O(log N) membership test? If not, I don't like
> "SET". Is there any reason to use LIST instead of these beasts? If
> not, maybe we should replace LIST. How many different sequence types
> do we need? I get the sense that the zoo is getting a little
> over-crowded.
What do you mean by "membership test"? It doesn't have any analogy to a
mathematical set if that's what you're getting at. As far as performance
goes, it is superior to lists and virtually as fast as tuples. It is kind
of a cross between the two with some traits of each, but it does not fully
support *all* the traits of both.
As far as a comparison against lists goes, a "set" (or whatever) doesn't
directly support a nil state, so lists are still viable until C++ gets the
C99 macro rules (You can pass nothing as a macro parameter in C99.).
We don't need any more sequence types. I made them initially to be a solid
replacement for tuples--one that doesn't need a size. It just happens that
they are very "algorithmic" in nature, so I just gave them the same support
as lists as well. Altogether, there are only four data types: array, list,
tuple, and "set." Each of them has different properties with various
overlaps.
I've been tooling around with an "includer" that uses both (for example):
#define HEADERS (..., (iostream)(iomanip)(vector)(map)(fstream))
#include ANGLED_INCLUDE()
Which basically includes all the headers in the "set." The ... specifies
"no directory." So you effectively get:
#include <iostream>
#include <iomanip>
#include <vector>
#include <fstream>
So, for instance:
#define HEADERS \
((boost/preprocessor),
(cat.hpp)(stringize.hpp)(tuple/elem.hpp)(control/if.hpp))
#include ANGLED_INCLUDE()
...effectively does this:
#include <boost/preprocessor/cat.hpp>
#include <boost/preprocessor/stringize.hpp>
#include <boost/preprocessor/tuple/elem.hpp>
#include <boost/preprocessor/control/if.hpp>
If I didn't have more than one type of structure, I'd have to put the
directory into the group of files--which I find unacceptable. The structure
difference represents clarity as well as options.
Paul Mensonides