skynet vs2013编译版本

Introduction ====== Sproto is an efficient serialization library for C, and focuses on lua binding. It's like Google protocol buffers, but much faster. The design is simple. It only supports a few types that lua supports. It can be easily bound to other dynamic languages, or be used directly in C. In my i5-2500 @3.3GHz CPU, the benchmark is below: The schema in sproto: ``` .Person { name 0 : string id 1 : integer email 2 : string .PhoneNumber { number 0 : string type 1 : integer } phone 3 : *PhoneNumber } .AddressBook { person 0 : *Person } ``` It's equal to: ``` message Person { required string name = 1; required int32 id = 2; optional string email = 3; message PhoneNumber { required string number = 1; optional int32 type = 2 ; } repeated PhoneNumber phone = 4; } message AddressBook { repeated Person person = 1; } ``` Use the data: ```lua local ab = { person = { { name = "Alice", id = 10000, phone = { { number = "123456789" , type = 1 }, { number = "87654321" , type = 2 }, } }, { name = "Bob", id = 20000, phone = { { number = "01234567890" , type = 3 }, } } } } ``` library| encode 1M times | decode 1M times | size -------| --------------- | --------------- | ---- sproto | 2.15s | 7.84s | 83 bytes sproto (nopack) |1.58s | 6.93s | 130 bytes pbc-lua | 6.94s | 16.9s | 69 bytes lua-cjson | 4.92s | 8.30s | 183 bytes * pbc-lua is a google protocol buffers library https://github.com/cloudwu/pbc * lua-cjson is a json library https://github.com/efelix/lua-cjson Lua API ======= ```lua local parser = require "sprotoparser" ``` * `parser.parse` parses a sproto schema to a binary string. The parser is needed for parsing the sproto schema. You can use it to generate binary string offline. The schema text and the parser is not needed when your program is running. ```lua local sproto = require "sproto.core" ``` * `sproto.newproto(sp)` creates a sproto object by a schema string (generates by parser). * `sproto.querytype(sp, typename)` queries a type object from a sproto object by typename. * `sproto.encode(st, luatable)` encodes a lua table by a type object, and generates a string message. * `sproto.decode(st, message)` decodes a message string generated by sproto.encode with type. * `sproto.pack(sprotomessage)` packs a string encoded by sproto.encode to reduce the size. * `sproto.unpack(packedmessage)` unpacks the string packed by sproto.pack. The sproto supports protocol tag for RPC. Use `sproto.protocol(tagorname)` to convert the protocol name to the tag id, or convert back from tag id to the name, and returns the request/response message type objects of this protocol. RPC API ======= There is a lua wrapper for the core API for RPC . Read testrpc.lua for detail. Schema Language ========== Like Protocol Buffers (but unlike json), sproto messages are strongly-typed and are not self-describing. You must define your message structure in a special language. You can use sprotoparser library to parse the schema text to a binary string, so that the sproto library can use it. You can parse them offline and save the string, or you can parse them during your program running. The schema text is like this: ``` # This is a comment. .Person { # . means a user defined type name 0 : string # string is a build-in type. id 1 : integer email 2 : string .PhoneNumber { # user defined type can be nest. number 0 : string type 1 : integer } phone 3 : *PhoneNumber # *PhoneNumber means an array of PhoneNumber. } .AddressBook { person 0 : *Person(id) # (id) is optional, means Person.id is main index. } foobar 1 { # define a new protocol (for RPC used) with tag 1 request Person # Associate the type Person with foobar.request response { # define the foobar.response type ok 0 : boolean } } ``` A schema text can be self-described by the sproto schema language. ``` .type { .field { name 0 : string type 1 : string id 2 : integer array 3 : boolean key 4 : integer # optional tag for map } name 0 : string fields 1 : *field } .protocol { name 0 : string id 1 : integer request 2 : string response 3 : string } .group { type 0 : *type protocol 1 : *protocol } ``` Types ======= * **string** : binary string * **integer** : integer, the max length of an integer is signed 64bit. * **boolean** : true or false You can add * before the typename to declare an array. You can also specify a main index, the array whould be encode as an unordered map. User defined type can be any name in alphanumeric characters except the build-in typenames, and nested types are supported. * Where are double or real types? I have been using Google protocol buffers for many years in many projects, and I found the real types were seldom used. If you really need it, you can use string to serialize the double numbers. * Where is enum? In lua, enum types are not very useful. You can use integer to define an enum table in lua. Wire protocol ======== Each integer number must be serialized in little-endian format. The sproto message must be a user defined type struct, and a struct is encoded in three parts. The header, the field part, and the data part. The tag and small integer or boolean will be encoded in field part, and others are in data part. All the fields must be encoded in ascending order (by tag). The tags of fields can be discontinuous, if a field is nil. (default value in lua), don't encode it in message. The header is a 16bit integer. It is the number of fields. Each field in field part is a 16bit integer (n). If n is zero, that means the field data is encoded in data part ; If n is even (and not zero), the value of this field is n/2-1 ; If n is odd, that means the tags is not continuous, and we should add current tag by (n+1)/2 . Read the examples below to see more details. Notice: If the tag is not declared in schema, the decoder will simply ignore the field for protocol version compatibility. Example 1: ``` person { name = "Alice" , age = 13, marital = false } 03 00 (fn = 3) 00 00 (id = 0, value in data part) 1C 00 (id = 1, value = 13) 02 00 (id = 2, value = false) 05 00 00 00 (sizeof "Alice") 41 6C 69 63 65 ("Alice") ``` Example 2: ``` person { name = "Bob", age = 40, children = { { name = "Alice" , age = 13, marital = false }, } } 04 00 (fn = 4) 00 00 (id = 0, value in data part) 52 00 (id = 1, value = 40) 01 00 (skip id = 2) 00 00 (id = 3, value in data part) 03 00 00 00 (sizeof "Bob") 42 6F 62 ("Bob") 11 00 00 00 (sizeof struct) 03 00 (fn = 3) 00 00 (id = 0, value in data part) 1C 00 (id = 1, value = 13) 02 00 (id = 2, value = false) 05 00 00 00 (sizeof "Alice") 41 6C 69 63 65 ("Alice") ``` 0 Packing ======= The algorithm is very similar to [Cap'n proto](http://kentonv.github.io/capnproto/), but 0x00 is not treated specially. In packed format, the message if padding to 8. Each 8 byte is reduced to a tag byte followed by zero to eight content bytes. The bits of the tag byte correspond to the bytes of the unpacked word, with the least-significant bit corresponding to the first byte. Each zero bit indicates that the corresponding byte is zero. The non-zero bytes are packed following the tag. For example: ``` unpacked (hex): 08 00 00 00 03 00 02 00 19 00 00 00 aa 01 00 00 packed (hex): 51 08 03 02 31 19 aa 01 ``` Tag 0xff is treated specially. A number N is following the 0xff tag. N means (N+1)*8 bytes should be copied directly. The bytes may or may not contain zeros. Because of this rule, the worst-case space overhead of packing is 2 bytes per 2 KiB of inpu