This Library Contains

• Contributors are eagerly required.
• please contact me: rockeet@gmail.com

• serialization framework(vs boost.serialization/google.protocolbuffer)
• Basic

Can be used in protocol parsing, big/small data serialization, even in very small object serialize, performance is good. (such as key/data serialization in BerkeleyDB), febird.DataIO is very fast: (30~1000 times faster than boost.binary_archive), and very low memory usage (ONLY 2 PTR for deserialization).

• Declarational syntax
• Such as: DATA_IO_LOAD_SAVE(Point3D, &x&y&z)
• Support all primitives and stl containers and std::string
• Byteswaping Support by different serializer
• NativeDataInput, NativeDataOutput: No byte swaping
• LittleEndianDataInput, LittleEndianDataOutput: Compile time select if or not swap
• BigEndianDataInput, BigEndianDataOutput: Compile time select if or not swap
• Optional version numbering (where boost.serialization always do)

• Use BerkelyDB as std::map and std::multimap

febird.DataIO is used for auto key/value serialization and deserialization, the interface is very simple and clean. See the code for detail.

• febird.rpc

A C++ remote procedure call without an IDL supporting, it based on the serialization framework. febird.rpc provide convenient usage and fast performance, and an uniform coding style.

• most stl algorithm(sort/heap/find...) in C and for both C/C++

Provide high performance(same as stl) through C-Preprocessor Tricky.

• threaded red-black tree implementation
• Kernel is written in C, and C++ interface(interface is similiar with std::map/set) is supported also

My implementation is fast(no compare function overhead for primitive types) but has no code explosion, because I used a C-preprocessor tricky.

• MultiIndexing (Foreign-Key in relational data base)
• All indexing pointers are in one node, didn't need separated node
• std::map/set can used for implementing Foreign Key indexing, but separated map/set is need, which hold a pointer to primary index(which hold data)
• Compare to boost.multi_index, boost.multi_index didn't need separated node for indexing, but it is compile-time-dynamic multi-index, and its compile speed is very slow
• my multi-index is runtime-dynamic multi-index, multi-index can be added or dropped at runtime, and compile speed is very fast
• so my multi-index can be use as a DataBase storage engine, such as MySQL's Storage engine.
• support 'compute sums in log(n) time'

All of the SQL aggregate functions: SUM, AVG, MAX, MIN, COUNT for a range can be implemented by the algorithm. see trb_sum_test, Implementation: Recompute aggregation in subtree after Rotatation

• hash_strmap
• A very fast hash string keyed hash map
• 40 times faster than std::map<std::string, ...>
• 10 times faster than std::unordered_map<std::string, ...>
• Saved at least 50% memory usage than std::map and unordered_map
• Expose STL container interface
• Support sorting and searching (after sort, insert and erase should be prohibited)
• No extra memory for sorting (just sort and relink)
• Support sorting by key(string) and value(mapped type)
• sorting by key(string) then search by string prefix
• Other minor features
• Linked list radix sorting algorithm
• Perfectly supporting variable-length keys, or arbitrary-length string, my implementation is not by padding 0 bytes/chars:

 let m = array.length, n = sum(array[*].key.length), maxkeylen=max([*].key.length);
 the time comlexity of my algorithm is O(n), but the traditional algorithm is O(m*maxkeylen).

See the code for detail: RadixSortTest, Implementation: radix_sort.h radix_sort.cpp

Chinese Introduction

febird 库包含以下部分

• 急切期望合作贡献者
• 请联系我: rockeet@gmail.com

• 序列化框架

功能类似于 boost.serializaiton 或 google.protocolbuffer, 可以用在协议解析，大/小数据的序列化，有极高的性能（比boost.binary_archive快30~80倍），甚至对于非常小的对象也非常适用，例如只有几个字节的对象，这在序列化BerkeleyDB中key/data这么小的对象（可能只是一个整数/变长整数）时要明显优于其它库。

• BerkeleyDB的序列化封装

构建在该库之上，可以象使用std::map一样使用BerkeleyDB

• 不需要IDL的rpc

基于该序列化框架，使用几个宏，很方便的自动完成函数参数的序列化，比MFC的MessageMap还要方便。

• stl算法的C版本，同时提供C++接口

使用了一种C预处理器技巧，提供和stl相当的性能(例如febird_sort远快于qsort，与stl::sort相当)。

• 线索红黑树（Threaded Red-Black Tree, C核心+C/Cpp接口，仿std::set/map）

• 在提供高性能的同时，没有C++的代码膨胀问题

因为使用了一个C预处理(C-preprocessor)技巧，详见：线索红黑树中的C预处理迭代

• 实现了动态多重索引（好比数据库，可以为不同 Key 分别创建索引，外键概念）
• 不同的索引只消耗一个内存结点，不需要另一个分离的结点
• stl::map/set 也可以实现多重索引，但是需要另一个的指针 map/set，这个额外的指针 map/set 的结点和主结点是分离的内存结点
• 对比 boost.multi_index，它也是单节点实现的多重索引，但 boost.multi_index 是编译时确定的多重索引，编译速度超慢
• 我的多重索引是运行时动态的，可以在运行时添加或删除一个索引，编译速度超快
• 因此，我的多重索引可以作为数据库存储引擎（比如 MySQL 存储引擎）
• 实现了 O(log(n)) 时间复杂度的聚集(aggregation)算法

使用该算法，所有SQL聚集函数：SUM, AVG, MIN, MAX, COUNT 的时间复杂度均是O(log(n)) ，并且，还可以在O(log(n))的时间内找出第n大的元素；更重要的，聚集算法是开放的，可插入的（使用函数指针）。

• hash_strmap: Key 为 string 的 hash map，我希望有人能找到这个世界上性能更好的 string map
• 提供 stl container 兼容的接口，使用上和 hash_map/unordered_map<string, Value, Hash, Equal> 几乎完全相同
• 32 字节的 Key ，3G 的 CPU，单线程，QPS 能达到 35,000,000
• 同样的测试代码，不管是在 Windows 还是 linux 下，都比 hash_map/unordered_map<string, Value> 快 10 倍，并且使用的内存更少（wordcount程序中节省50%以上的内存）
• MSVC 中的 unordered_map 有问题，性能出奇地低（查找性能看上去好像是O(n)的），几乎无法完成测试，只能使用 stdext::hash_map

使用时请尽量checkout最新版

@see

我的CSDN博客
我的C++序列化框架介绍
持久化的map:使用BerkeleyDB