使用boost::multi_index高速构建排行榜

使用boost::multi_index高速构建排行榜

前几天在boost的maillist上看到boost::multi_index将要支持ranked_index（邮件内容见附件2），这实乃我等苦逼写排行榜的人的福音。大家再也不用去分析rank_tree里的内容了，故拿出来和大家一起分享。

ranked_index其内部实现和rank_tree是一样的。但其优点是集成在multi_index内部，使用上很方便，而且支持ranked_unique和ranked_non_unique两种索引。

话不多说，首先从http://tinyurl.com/kemwk8q下载下来multi_index的库文件，然后替换boost_1.58中的multi_index，有一个文件夹boost/multi_index和两个文件boost/multi_index_container.hpp,
boost/multi_index_container_fwd.hpp须要替换，替换之后，我们就能够把附件1里的代码拷贝到vs2008里执行啦，输出结果为：

245044518

1

0

了解过rank_tree组件的同学肯定知道是怎么回事了，ranked_index和rank_tree一样比普通的map多了两个函数。其相应关系是：

ranked_index	rank_tree	说明
nth	find_by_rank	名次到迭代器
rank	rank	迭代器到名次

附件1：

// test_rank.cpp : Defines the entry point for the console application.

//

#include "stdafx.h"

#include <boost/multi_index_container.hpp>

#include <boost/multi_index/member.hpp>

#include <boost/multi_index/indexed_by.hpp>

#include <boost/multi_index/ranked_index.hpp>

#include <boost/multi_index/ordered_index.hpp>

#include <iostream>

namespace bmi = boost::multi_index;

typedef unsigned int uint32_t;

typedef int int32_t;

struct Data

{

int32_t key;

uint32_t uin;

};

struct tag_key{};

struct tag_uin{};

typedef boost::multi_index_container<

Data, bmi::indexed_by<

bmi::ranked_non_unique<

bmi::tag<tag_key>,

bmi::member<Data, int32_t, &Data::key>

>,

bmi::ordered_unique<

bmi::tag<tag_uin>,

bmi::member<Data, uint32_t, &Data::uin>

>

>

> container_t;

typedef container_t::index<tag_key>::type key_index_t;

typedef container_t::index<tag_uin>::type uin_index_t;

int _tmain(int argc, _TCHAR* argv[])

{

container_t c;

Data data;

data.key = 200;

data.uin = 245044518;

c.insert(data);

data.key = 100;

data.uin = 503063727;

c.insert(data);

key_index_t& key_index = boost::get<tag_key>(c);

uin_index_t& uin_index = boost::get<tag_uin>(c);

std::cout << key_index.nth(1)->uin << std::endl;

std::cout << key_index.rank(key_index.nth(1))
<< std::endl;

std::cout << key_index.rank(bmi::project<tag_key>(c, uin_index.find(503063727)))
<< std::endl;

return 0;

}

附件2：

Message: 2

Date: Wed, 22 Apr 2015 19:34:04 +0000 (UTC)

From: Joaquin M Lopez Munoz <joaquin@tid.es>

To: boost-users@lists.boost.org

Subject: [Boost-users] [multi_index] Announcing preview of ranked

indices

Message-ID: <loom.20150422T212903-565@post.gmane.org>

Content-Type: text/plain; charset=utf-8

Hi,

It's been 11 years since first proposed, but here it is at last. I'm

releasing a prereview of ranked indices for Boost.MultiIndex:

struct person

{

int         age;

std::string name;

};

typedef multi_index_container<

person,

indexed_by<

ranked_non_unique<member<person,int,&person::age> >,

ordered_non_unique<member<person,std::string,&person::name>>>>

multi_t;

multi_t m={{40,"Joe"},{25,"Jill"},{30,"Kurt"},{32,"Sue"}};

auto it=m.emplace(31,"Maggie").first;

std::cout<<m.rank(it)<<" persons younger than "<<it->name<<"\n";

auto n=m.size()/2;

std::cout<<n<<" persons are less than "<<m.nth(n)->age<<" years old\n";

Ranked indices add a bunch of extra rank-related capabilities to the

interface of ordered indices. The rank of an element is its numerical

position in the index. nth(i) returns an iterator to the element with

rank i, whereas rank(it) is the inverse operation. Both execute in

log(n) time.

Additionally, the member functions

find_rank

lower_bound_rank

upper_bound_rank

equal_range_rank

range_rank

behave as their "_rank"-less counterparts except they return ranks rather

than iterators.

One drawback of ranked indices wrt ordered indices (other than their

slower perfomance and higher memory consumption) is that deletion of

elements is log(n) (in ordered indices it is constant time).

Download lib preview: http://tinyurl.com/kemwk8q

Tutorial: http://tinyurl.com/q36c9f7

Reference: http://tinyurl.com/pvoulgr

For people interested in this, I'd appreciate if you could download the

source, copy it on top of Boost 1.58 (1.57 should work as well), play a bit

with the new feature and report your results. Also, opinions on naming

are welcome. Target version for releasing is Boost 1.59, so we have

plenty of time to discuss the design before it is final.

Thank you,

Joaqu?n M L?

pez Mu?oz

Telef?

nica

------------------------------