对于一般人,没必要像对待常用公共组件一样,搞清楚每一个点,我们从使用的角度出发,把我们用到的功能读到即可。
https://github.com/quickfix/quickfix
源码就在src\C++下,我们先大致浏览一下。
DataDictionary.cpp:解析诸如FIX42.xml的数据字典
Field.cpp:数据字典中解析预定义的field
Message.cpp:数据字典中解析处理message节点
Http.cpp: 实现http引擎的部分
Socket.cpp:会话层的通信
Session.cpp: 会话层的东西
还有一些其他的文件,略去不说。这里还要注意还有几个子文件夹:fix40/,fix41/,fix42/,fix43/,fix44/,fix50/,fix50sp1。这几个文件夹下是具体实现了该版本的一些头文件。
上篇文章有使用的例子,我们去掉多余部分,拿过来是这样的:
int main( int argc, char** argv )
{
FIX::Initiator * initiator = 0;
try
{
FIX::SessionSettings settings( file );
Application application;
FIX::FileStoreFactory storeFactory( settings );
FIX::ScreenLogFactory logFactory( settings );
initiator = new FIX::SocketInitiator( application, storeFactory,
settings, logFactory );
initiator->start();
application.run();
initiator->stop();
delete initiator;
'''
}
catch ( std::exception & e )
{
'''
}
}
请记住每一行代码,接下来,本文基本是每章讲解本代码中的一行。
就是这一行:FIX::SessionSettings settings( file );
Quickfix中进行数据字典的载入,解析本质是对几个xml文件的解析,是采用pugixml parser,官方网站:pugixml.org - Home。正如官网介绍的那样:
Light-weight, simple and fast XML parser for C++ with XPath support
然后Quickfix中在之上进行了一层自己的封装,形成PUGIXML_DOMAttributes类,PUGIXML_DOMNode类,PUGIXML_DOMDocument类。在头文件”PUGIXML_DOMDocument.h”中进行了定义,如下:
class PUGIXML_DOMAttributes : public DOMAttributes
{
public:
PUGIXML_DOMAttributes( pugi::xml_node pNode )
: m_pNode(pNode) {}
bool get( const std::string&, std::string& );
DOMAttributes::map toMap();
private:
pugi::xml_node m_pNode;
};
/// XML node as represented by pugixml.
class PUGIXML_DOMNode : public DOMNode
{
public:
PUGIXML_DOMNode( pugi::xml_node pNode )
: m_pNode(pNode) {}
~PUGIXML_DOMNode() {}
DOMNodePtr getFirstChildNode();
DOMNodePtr getNextSiblingNode();
DOMAttributesPtr getAttributes();
std::string getName();
std::string getText();
private:
pugi::xml_node m_pNode;
};
/// XML document as represented by pugixml.
class PUGIXML_DOMDocument : public DOMDocument
{
public:
PUGIXML_DOMDocument() throw( ConfigError );
~PUGIXML_DOMDocument();
bool load( std::istream& );
bool load( const std::string& );
bool xml( std::ostream& );
DOMNodePtr getNode( const std::string& );
private:
pugi::xml_document m_pDoc;
};
}
其中大多数函数不需要特别关心,我们只需要重点关心PUGIXML_DOMDocument类中的load()函数,这也是最重要+最复杂的函数。
bool PUGIXML_DOMDocument::load( std::istream& stream )
{
try
{
return m_pDoc.load(stream);
}
catch( ... ) { return false; }
}
bool PUGIXML_DOMDocument::load( const std::string& url )
{
try
{
return m_pDoc.load_file(url.c_str());
}
catch( ... ) { return false; }
}
这个函数就是对给定一个xml路径然后装载后返回一个pugi::xml_document的对象。
上面的类实现了诸如FIX44.xml的载入处理,数据字典中定义了很多结构节点,比如fields,messages,groups等,DataDictionary.cpp是真正对这些xml文件进行解析的源文件。DataDictionary.h中部分源代码如下:
class DataDictionary
{
typedef std::set < int > MsgFields;
typedef std::map < std::string, MsgFields > MsgTypeToField;
typedef std::set < std::string > MsgTypes;
typedef std::set < int > Fields;
typedef std::map < int, bool > NonBodyFields;
typedef std::vector< int > OrderedFields;
typedef message_order OrderedFieldsArray;
typedef std::map < int, TYPE::Type > FieldTypes;
typedef std::set < std::string > Values;
typedef std::map < int, Values > FieldToValue;
typedef std::map < int, std::string > FieldToName;
typedef std::map < std::string, int > NameToField;
typedef std::map < std::pair < int, std::string > , std::string > ValueToName;
// while FieldToGroup structure seems to be overcomplicated
// in reality it yields a lot of performance because:
// 1) avoids memory copying;
// 2) first lookup is done by comparing integers and not string objects
// TODO: use hash_map with good hashing algorithm
typedef std::map < std::string, std::pair < int, DataDictionary* > > FieldPresenceMap;
typedef std::map < int, FieldPresenceMap > FieldToGroup;
public:
DataDictionary();
DataDictionary( const DataDictionary& copy );
DataDictionary( std::istream& stream ) throw( ConfigError );
DataDictionary( const std::string& url ) throw( ConfigError );
virtual ~DataDictionary();
void readFromURL( const std::string& url ) throw( ConfigError );
void readFromDocument( DOMDocumentPtr pDoc ) throw( ConfigError );
void readFromStream( std::istream& stream ) throw( ConfigError );
......
};
....
可以看到DataDictionary类中定义了很多的std::map和std::vector,这些容器都是用来存储从FIX4X.xml文件中解析来的内容,一些映射,但是是否过于繁琐,我没有深究。
比如:
typedef std::map < int, std::string > FieldToName;
表示存储field和实际的字段名的映射,比如8对应BeginString;
typedef std::map < int, Values > FieldToValue;
表示枚举当中的int值跟实际的字段名的映射,比如:
3代表ABSOLUTE;1代表PER_UNIT。
另外需要注意的成员函数readFrom*()系列,底层就是上一章中的类,进行xml的载入。
void DataDictionary::readFromURL( const std::string& url )
throw( ConfigError )
{
DOMDocumentPtr pDoc = DOMDocumentPtr(new PUGIXML_DOMDocument());
if(!pDoc->load(url))
¦ throw ConfigError(url + ": Could not parse data dictionary file");
try
{
¦ readFromDocument( pDoc );
}
catch( ConfigError& e )
{
¦ throw ConfigError( url + ": " + e.what() );
}
}
void DataDictionary::readFromStream( std::istream& stream )
throw( ConfigError )
{
>* DOMDocumentPtr pDoc = DOMDocumentPtr(new PUGIXML_DOMDocument());
if(!pDoc->load(stream))
¦ throw ConfigError("Could not parse data dictionary stream");
readFromDocument( pDoc );
}
>*void DataDictionary::readFromDocument( DOMDocumentPtr pDoc )
throw( ConfigError )
{
// VERSION
DOMNodePtr pFixNode = pDoc->getNode("/fix");
if(!pFixNode.get())
...
}
到这里,数据字典的解析就完成了。简单的理解就是,读入xml文件,然后针对xml文件里的内容,把内容做成映射用map和vector存储。
SessionSettings
/// Container for setting dictionaries mapped to sessions.
class SessionSettings
{
public:
SessionSettings() { m_resolveEnvVars = false; }
SessionSettings( std::istream& stream, bool resolveEnvVars = false ) EXCEPT ( ConfigError );
SessionSettings( const std::string& file, bool resolveEnvVars = false ) EXCEPT ( ConfigError );
''''''
typedef std::map < SessionID, Dictionary > Dictionaries;
std::set < SessionID > getSessions() const;
private:
Dictionaries m_settings;
Dictionary m_defaults;
'''
friend std::istream& operator>>( std::istream&, SessionSettings& ) EXCEPT ( ConfigError );
friend std::ostream& operator<<( std::ostream&, const SessionSettings& );
};
是通过友元函数 operator >>
从任意的流中读取配置,通过一个sessonid的set和一个sessionid->dictionary的map,管理每个段。
若是须要使用QuickFIX开发FIX应用,则须要实现FIX::Application接口,并重载不一样FIX协议版本的MessageCracker::OnMessage接口,如FIX42::MessageCracker。
class Application
{
public:
virtual ~Application() {};
/// Notification of a session begin created
virtual void onCreate( const SessionID& ) = 0;
/// Notification of a session successfully logging on
virtual void onLogon( const SessionID& ) = 0;
/// Notification of a session logging off or disconnecting
virtual void onLogout( const SessionID& ) = 0;
/// Notification of admin message being sent to target
virtual void toAdmin( Message&, const SessionID& ) = 0;
/// Notification of app message being sent to target
virtual void toApp( Message&, const SessionID& )
EXCEPT ( DoNotSend ) = 0;
/// Notification of admin message being received from target
virtual void fromAdmin( const Message&, const SessionID& )
EXCEPT ( FieldNotFound, IncorrectDataFormat, IncorrectTagValue, RejectLogon ) = 0;
/// Notification of app message being received from target
virtual void fromApp( const Message&, const SessionID& )
EXCEPT ( FieldNotFound, IncorrectDataFormat, IncorrectTagValue, UnsupportedMessageType ) = 0;
};
onCreate:当Fix Session创建时调用。
onLogon:当Fix Session登陆成功时调用。
onLogout:当Fix Session退出时调用。
fromAdmin:当收到一个Admin类型消息时调用。
fromApp:当收到一个不属于Admin 类型消息时调用。
toAdmin:当发送一个admin类型消息调用。
toApp:当发送一个非admin(业务类型)消息调用。
admin一般是服务提供方,app是客户端
除了实现FIX::Application接口,还需要重新实现FIX::MessageCracker从具体的FIX协议版本实现继承而来的onMessage方法,crack接口就可以根据message类型匹配到你实现的具体onMessage接口上。
void crack( const Message& message,
const SessionID& sessionID )
{
const FIX::BeginString& beginString =
FIELD_GET_REF( message.getHeader(), BeginString );
crack( message, sessionID, beginString );
}
void crack( const Message& message,
const SessionID& sessionID,
const BeginString& beginString )
{
if ( beginString == BeginString_FIX40 )
((FIX40::MessageCracker&)(*this)).crack((const FIX40::Message&) message, sessionID);
else if ( beginString == BeginString_FIX41 )
((FIX41::MessageCracker&)(*this)).crack((const FIX41::Message&) message, sessionID);
else if ( beginString == BeginString_FIX42 )
((FIX42::MessageCracker&)(*this)).crack((const FIX42::Message&) message, sessionID);
else if ( beginString == BeginString_FIX43 )
((FIX43::MessageCracker&)(*this)).crack((const FIX43::Message&) message, sessionID);
else if ( beginString == BeginString_FIX44 )
((FIX44::MessageCracker&)(*this)).crack((const FIX44::Message&) message, sessionID);
else if ( beginString == BeginString_FIXT11 )
{
if( message.isAdmin() )
{
((FIXT11::MessageCracker&)(*this)).crack((const FIXT11::Message&) message, sessionID);
}
else
{
'''
}
}
}
就是这两行:
FIX::FileStoreFactory storeFactory( settings );
FIX::ScreenLogFactory logFactory( settings );
逻辑比较简单,就是读了上文介绍的settings,然后存下来,存储结构如下:
std::string m_path;
SessionSettings m_settings;
也就是这一行 initiator = new FIX::SocketInitiator( application, storeFactory, settings, logFactory );
这俩大概差不多,先看一个。
主要代码如下:
/**
* Base for classes which act as an acceptor for incoming connections.
*
* Most users will not need to implement one of these. The default
* SocketAcceptor implementation will be used in most cases.
*/
class Acceptor
{
public:
''''''
Acceptor( Application&, MessageStoreFactory&,
const SessionSettings&, LogFactory& ) EXCEPT ( ConfigError );
virtual ~Acceptor();
''''''
/// Poll the acceptor
bool poll( double timeout = 0.0 ) EXCEPT ( ConfigError, RuntimeError );
/// Stop acceptor.
void stop( bool force = false );
/// Check to see if any sessions are currently logged on
bool isLoggedOn();
Session* getSession( const std::string& msg, Responder& );
const std::set& getSessions() const { return m_sessionIDs; }
Session* getSession( const SessionID& sessionID ) const;
const Dictionary* const getSessionSettings( const SessionID& sessionID ) const;
bool has( const SessionID& id )
{ return m_sessions.find( id ) != m_sessions.end(); }
bool isStopped() { return m_stop; }
Application& getApplication() { return m_application; }
MessageStoreFactory& getMessageStoreFactory()
{ return m_messageStoreFactory; }
private:
''''''
static THREAD_PROC startThread( void* p );
typedef std::set < SessionID > SessionIDs;
typedef std::map < SessionID, Session* > Sessions;
thread_id m_threadid;
Sessions m_sessions;
SessionIDs m_sessionIDs;
Application& m_application;
MessageStoreFactory& m_messageStoreFactory;
protected:
SessionSettings m_settings;
private:
LogFactory* m_pLogFactory;
Log* m_pLog;
NullLog m_nullLog;
bool m_firstPoll;
bool m_stop;
};
基本包含了之前介绍的大部分类,如
Session相关的(SessionSettings/set
Application(用于接收并处理消息的)、LogFactory(写日志的对象)
功能就是把配置的每一个session初始化,很简单。
void Acceptor::initialize() EXCEPT ( ConfigError )
{
std::set < SessionID > sessions = m_settings.getSessions();
std::set < SessionID > ::iterator i;
if ( !sessions.size() )
throw ConfigError( "No sessions defined" );
SessionFactory factory( m_application, m_messageStoreFactory,
m_pLogFactory );
for ( i = sessions.begin(); i != sessions.end(); ++i )
{
if ( m_settings.get( *i ).getString( CONNECTION_TYPE ) == "acceptor" )
{
m_sessionIDs.insert( *i );
m_sessions[ *i ] = factory.create( *i, m_settings.get( *i ) );
}
}
if ( !m_sessions.size() )
throw ConfigError( "No sessions defined for acceptor" );
}
这一行:Acceptor/
initiator->start();
SocketAcceptor::onInitialize()
创建 socket 句柄,进行监听端口。SocketAcceptor::onStart()
,检测对端的连接void Acceptor::start() EXCEPT ( ConfigError, RuntimeError )
{
m_stop = false;
onConfigure( m_settings );
onInitialize( m_settings );
HttpServer::startGlobal( m_settings );
if( !thread_spawn( &startThread, this, m_threadid ) )
throw RuntimeError("Unable to spawn thread");
}
其他的操作大同小异,可以自己阅读
SocketAcceptor::onInitialize
主要功能就是对每个session设置监听
void SocketAcceptor::onInitialize(const SessionSettings& s)
EXCEPT ( RuntimeError )
{
short port = 0;
try
{
m_pServer = new SocketServer(1);
std::set sessions = s.getSessions();
std::set::iterator i = sessions.begin();
for( ; i != sessions.end(); ++i )
{
const Dictionary& settings = s.get( *i );
port = (short)settings.getInt( SOCKET_ACCEPT_PORT );
''''''
// 管理监听端口与 SeesionID 的对应关系
m_portToSessions[port].insert(*i);
// 为每个监听的端口创建 Socket 句柄: socket_handle
m_pServer->add( port, reuseAddress, noDelay, sendBufSize, rcvBufSize );
}
}
catch( SocketException& e )
{
''''''
}
}
5.2中的第二步调用
THREAD_PROC Acceptor::startThread( void* p )
{
Acceptor * pAcceptor = static_cast < Acceptor* > ( p );
pAcceptor->onStart();
return 0;
}
回顾所有我们浏览的代码,唯独没有介绍session,最后来看一下。
用factory(初始化心跳、session)
Session* SessionFactory::create( const SessionID& sessionID,
const Dictionary& settings ) EXCEPT ( ConfigError )
{
std::string connectionType = settings.getString( CONNECTION_TYPE );
if ( connectionType != "acceptor" && connectionType != "initiator" )
throw ConfigError( "Invalid ConnectionType" );
if( connectionType == "acceptor" && settings.has(SESSION_QUALIFIER) )
throw ConfigError( "SessionQualifier cannot be used with acceptor." );
// 初始化心跳
HeartBtInt heartBtInt( 0 );
if ( connectionType == "initiator" )
{
heartBtInt = HeartBtInt( settings.getInt( HEARTBTINT ) );
if ( heartBtInt <= 0 ) throw ConfigError( "Heartbeat must be greater than zero" );
}
// 创建 Session 对象
SmartPtr pSession;
pSession.reset( new Session( m_application, m_messageStoreFactory,
sessionID, dataDictionaryProvider, sessionTimeRange,
heartBtInt, m_pLogFactory ) );
return pSession.release();
}
其中session对象内属性太多,挑一些重要的看:
Application(会话)、
SessionID(标识唯一session)、
m_sessionTime/m_logonTime(主要用于之前讲的24小时重新连接,对应配置)、
m_senderDefaultApplVerID/m_targetDefaultApplVerID(发送端/接收端默 Fix 协议版本号)、
m_state(session状态)、
send()(发送消息函数)、
next()(处理收到的消息,比较重要)
精简过的代码如下
void Session::next( const Message& message, const UtcTimeStamp& timeStamp, bool queued )
{
const Header& header = message.getHeader();
try
{
//检查时间
if ( !checkSessionTime(timeStamp) )
{ reset(); return; }
//获取类型,下面根据类型分处理方法
const MsgType& msgType = FIELD_GET_REF( header, MsgType );
//校验时间
const BeginString& beginString = FIELD_GET_REF( header, BeginString );
// make sure these fields are present
FIELD_THROW_IF_NOT_FOUND( header, SenderCompID );
FIELD_THROW_IF_NOT_FOUND( header, TargetCompID );
if ( beginString != m_sessionID.getBeginString() )
throw UnsupportedVersion();
const DataDictionary& sessionDataDictionary =
m_dataDictionaryProvider.getSessionDataDictionary(m_sessionID.getBeginString());
if( m_sessionID.isFIXT() && message.isApp() )
{
ApplVerID applVerID = m_targetDefaultApplVerID;
header.getFieldIfSet(applVerID);
const DataDictionary& applicationDataDictionary =
m_dataDictionaryProvider.getApplicationDataDictionary(applVerID);
DataDictionary::validate( message, &sessionDataDictionary, &applicationDataDictionary );
}
else
{
sessionDataDictionary.validate( message );
}
if ( msgType == MsgType_Logon )
nextLogon( message, timeStamp );
else if ( msgType == MsgType_Heartbeat )
nextHeartbeat( message, timeStamp );
else if ( msgType == MsgType_TestRequest )
nextTestRequest( message, timeStamp );
else if ( msgType == MsgType_SequenceReset )
nextSequenceReset( message, timeStamp );
else if ( msgType == MsgType_Logout )
nextLogout( message, timeStamp );
else if ( msgType == MsgType_ResendRequest )
nextResendRequest( message, timeStamp );
else if ( msgType == MsgType_Reject )
nextReject( message, timeStamp );
else
{
if ( !verify( message ) ) return ;
//内含Session::doTargetTooLow() 来处理序列号过小的消息
// Session::doTargetTooHigh() 来处理序列号过大的消息
m_state.incrNextTargetMsgSeqNum();
}
}
''''''
if( !queued )
nextQueued( timeStamp );
if( isLoggedOn() )
next();
}
经过各种检查后,根据type调用不同的处理方法,然后操作queue进行下次操作。
这里调用的函数太多了,挑一个复杂的看一下。
当收到 type是ResendRequest 消息时,回调用nextResendRequest()
处理:
void Session::nextResendRequest(const Message& resendRequest, const UtcTimeStamp& timeStamp)
{
// ...
// 从缓存拿出需要重传的消息片段(从MessageStore中的消息,如果是FileStore,那么就会从文件中取出)
std::vector < std::string > messages;
m_state.get( beginSeqNo, endSeqNo, messages );
// ...
for ( i = messages.begin(); i != messages.end(); ++i )
{
// 重新计算消息的校验和
// ...
if ( Message::isAdminMsgType( msgType ) )
{
// 跳过管理消息
if ( !begin ) begin = msgSeqNum;
}
else
{
// 在 resend 里会回调 Application::toApp
if ( resend( msg ) )
{
// 有需要跳过的管理消息,则用一条 SeqReset-GapFill 消息替代
if ( begin ) generateSequenceReset( begin, msgSeqNum );
// 发送应用消息
send( msg.toString(messageString) );
m_state.onEvent( "Resending Message: "
+ IntConvertor::convert( msgSeqNum ) );
begin = 0;
appMessageJustSent = true;
}
else
{ if ( !begin ) begin = msgSeqNum; }
}
current = msgSeqNum + 1;
}
// 结尾还有需要跳过的管理消息,需要用一条 SeqReset-GapFill 消息替代
if ( begin )
{
generateSequenceReset( begin, msgSeqNum + 1 );
}
// 序列号同步。为什么在重传借宿后还需要再发送一个 SeqReset-GapFill 消息?
if ( endSeqNo > msgSeqNum )
{
endSeqNo = EndSeqNo(endSeqNo + 1);
int next = m_state.getNextSenderMsgSeqNum();
if( endSeqNo > next )
endSeqNo = EndSeqNo(next);
if ( appMessageJustSent )
beginSeqNo = msgSeqNum + 1;
generateSequenceReset( beginSeqNo, endSeqNo );
}
resendRequest.getHeader().getField( msgSeqNum );
if( !isTargetTooHigh(msgSeqNum) && !isTargetTooLow(msgSeqNum) )
m_state.incrNextTargetMsgSeqNum();
}
作者修行尚浅,这里只是浅读一下源码,由于使用经验不足,肯定对一些知识的认识不足,以后多加改正。