path: root/comm/MsgStream.cc

/*
 *	The Unix Channel
 *
 *	by Michel Beaudouin-Lafon
 *
 *	Copyright 1990-1993
 *	Laboratoire de Recherche en Informatique (LRI)
 *
 *	UchMessage streams
 *
 *	$Id$
 *	$CurLog$
 */

#include "MsgStream.h"
#include "Message.h"
#include "error.h"

//	#define DEBUG

#ifdef DEBUG
#include <stdio.h>
#define DBG(inst)	inst
#else
#define DBG(inst)
#endif

/*?class UchMsgStream
An object of class \typ{UchMsgStream} is a stream that sends and receives messages:
we call it a message stream.

Messages are sent with the function \fun{Send}.
Output messages are normally buffered to increase performance, but synchronous mode is available.
The output buffer can be flushed by the application, or by the stream itself if needed
(for instance when the buffer is full, or when a synchronous communication is needed).

Incoming messages are handled by the virtual function \fun{HandleRead} of class \typ{UchChannel}:
it is redefined so that incoming bytes are packed into messages.
When a full message is available, \fun{HandleRead} calls the virtual
function \fun{NewMessage}. In the class \typ{UchMsgStream}, this virtual function does
nothing but discarding the message.

Messages that need an answer are handled in the following way:
the sender calls the function\fun{Ask}, and is blocked until the answer is received;
any incoming messages are stored for later processing.
When the answer arrives, \fun{Ask} calls the virtual function \fun{ConvertAnswer}.
The returned value of \fun{ConvertAnswer} is then returned by \fun{Ask},
thus returning to the application the reply to its question.

On the receiver's side, the following happens:
when a message sent with \fun{Ask} is received, \fun{NewMessage} is called as usual,
but an argument indicates that it is a question that needs an answer.
The receiver must then use \fun{Reply} to send its answer.
Messages can be sent before replying, but it is not possible to send a question
on a message stream that is waiting for an answer:
this would result in a deadlock since the other party is already waiting for an answer
and is buffering other incoming messages.

The default output mode is buffered (i.e. asynchronous).
The buffered output is automatically flushed when a question is sent,
or when the output buffer is full, or explicitly with \fun{Flush}, or with the second argument of \fun{Send}.
The output mode can be switched to a synchronous mode where each message is sent immediately.
% It can also be switched to a locked synchronous mode: this is a synchronous mode where
% the sender waits for each message to be processed before proceeding its execution.
% This is mainly useful for debugging purposes.

\medskip
To use a message stream, a program needs to derive the class \typ{UchMsgStream}
in order to redefine the virtual functions \fun{NewMessage} and \fun{ConvertAnswer}.
\fun{NewMessage} treats the incoming message, and will probably call \fun{Send} and \fun{Ask}.
The top level of the program needs just call \fun{HandleRead} in a forever loop.

Most of the time, a program will use several message streams
(for instance to manage several clients).
In this case a channel set is the best way to implement the application:
the definitions in the class \typ{UchMsgStream} of the functions
\fun{HandleRead}, \fun{HandleWrite} and \fun{HandleSelect} make it easy
to use this class in combination with the class \typ{UchMultiplexer}.
The virtual function \fun{HandeWrite} of channels is redefined to flush the output buffer.
The virtual function \fun{HandleSelect} of channels is redefined to handle the incoming messages
that were buffered while waiting for an answer.
As for the single stream situation, you need just derive the class \fun{UchMsgStream} to
redefine the virtual functions \fun{NewMessage} and \fun{ConvertAnswer}.
?*/

// ---- no byte swapping ...
// ---- no locked sync mode ...

/*?nextdoc?*/
UchMsgStream :: UchMsgStream ()
: UchStream (), InBuffer (), OutBuffer (), Buffered ()
{
	OutSize = 128;
	State = WAITING;
	BufferedMessages = FALSE;
	WaitingForAnswer = FALSE;
	WaitingReply = FALSE;
	Sync = FALSE;
}

/*?
These constructors are similar to those of class \typ{UchSocket}.
?*/
UchMsgStream :: UchMsgStream (UchAddress* bindTo, UchAddress* connectTo)
: UchStream (bindTo, connectTo), InBuffer (), OutBuffer (), Buffered ()
{
	OutSize = 128;
	State = WAITING;
	BufferedMessages = FALSE;
	WaitingForAnswer = FALSE;
	WaitingReply = FALSE;
	Sync = FALSE;
}

// *** this copy constructor might be automatically generated
/*?nodoc?*/
UchMsgStream :: UchMsgStream (const UchMsgStream& ms)
: UchStream (*(UchMsgStream*)&ms), InBuffer (), OutBuffer (), Buffered (*(UchMsgBuffer*)&ms.Buffered)
{
      OutSize = ms.OutSize;
      State = ms.State;
      BufferedMessages = ms.BufferedMessages;
      WaitingForAnswer = ms.WaitingForAnswer;
      WaitingReply = ms.WaitingReply;
      Sync = ms.Sync;
}

/*?nextdoc?*/
void
UchMsgStream :: InputBuffer (int min, int grow, int max)
{
	InBuffer.SetSizes (min, grow, max);
}

/*?
Set the input and output buffer sizes.
Default sizes are used if these functions are not called.
?*/
void
UchMsgStream :: OutputBuffer (int min, int grow, int max)
{
	InBuffer.SetSizes (min, grow, max);
	OutSize = max;
}

/*?nodoc?*/
UchMsgStream :: ~UchMsgStream ()
{
	Flush ();
	InBuffer.Delete ();
	OutBuffer.Delete ();
	Buffered.Delete ();
}

/*?nodoc?*/
UchChannel*
UchMsgStream :: Copy () const
{
	return new UchMsgStream (*this);
}


// read stream into input buffer
// delete the stream when an oef is received
// return the number of bytes read
//
/*?hidden?*/
int
UchMsgStream :: ReadInput ()
{
	if (! InBuffer.BufLength ())
		InBuffer.NeedSize (128);
	int n = Read (InBuffer);
	if (n <= 0) {
		if (n == -2) {
//printf ("ReadInput: growing\n", n);
			InBuffer.Grow ();
			n = Read (InBuffer);
		}
//printf ("ReadInput: %d bytes\n", n);
		if (n < 0)
			SysError (ErrWarn, "UchMsgStream::HandleRead");
		if (n <= 0)
			Delete ();
	}
	return n;
}

/*?
This virtual function is called when an end of file is read, 
meaning that the communication is terminated.
It is also called if an error occurs while reading.
You can check the value of the global variable \var{errno}:
if it is non zero, \fun{Delete} was called because of an error.
By default this function does nothing.
?*/
void
UchMsgStream :: Delete ()
{
	// nothing
}

// process the input buffer
// waitAnswer indicates whether we are waiting for an answer
// this functions uses a very simple automaton:
//	WAITING:	nothing in the buffer
//	GOT_TYPE:	read the 1 byte header mark of a message
//			MSG / ASK / ANS for messages, questions, answers
//			SYNC / ASYNC / OK for sync management
//	GOT_LENGTH:	read the 4 byte header fo a message
//	DONE:		a full message is in the buffer
//
// WaitingReply is true if a question has been received and Reply has not been called yet
// WaitingForAnswer is true when Ask has been called and the answer is not yet there
//
/*?hidden?*/
void
UchMsgStream :: ProcessInput (UchMsgBuffer& buf, bool waitAnswer)
{
DBG (printf ("ProcessInput:	");)
	for (;;) {
		switch (State) {
			case WAITING:
DBG (printf ("WAITING\n");)
				buf.Get ((byte*) &InType);
DBG (if (buf.Error ()) printf ("	buf empty\n");)
				if (buf.Error ())
					return;
				WaitingReply = FALSE;
				switch (InType) {
					case ASK :
DBG (printf ("	waiting reply\n");)
						WaitingReply = TRUE;
						State = GOT_TYPE;
						break;
					case ANS :
DBG (printf ("	answer\n");)
					case MSG :
DBG (if (InType != ANS) printf ("	msg\n");)
DBG (printf ("	got type\n");)
						State = GOT_TYPE;
						break;
					case SYNC :
					case ASYNC :
					case OK :
					default :
DBG (printf ("	unknown !!\n");)
						State = WAITING;
				}
				if (State != GOT_TYPE)
					break;
				// fallthrough
				
			case GOT_TYPE:
DBG (printf ("GOT_TYPE\n");)
				if (! buf.Peek ((lword*) &InLength))
					return;
				buf.NeedSize ((int) InLength - buf.BufLength ());
				State = GOT_LENGTH;
				// fallthrough
				
			case GOT_LENGTH:
DBG (printf ("GOT_LENGTH\n");)
				if (buf.BufLength () < InLength)
					return;
				State = DONE;
				// fallthrough
			
			case DONE:
DBG (printf ("DONE\n");)
				if (waitAnswer) {
					if (InType == ANS) {
DBG (printf ("got answer\n");)
						WaitingForAnswer = FALSE;
						// answer still in the buffer
						// the answer is converted and
						// the buffer is flushed in Ask
						return;
					} else {
						// store incoming message in a separate buffer
						BufferedMessages = TRUE;
						Buffered.Append (InType);
						Buffered.Append (buf.Buffer (), InLength);
DBG (printf ("buffering\n");)
					}
				} else {
					if (InType == MSG || InType == ASK) {
DBG (printf ("new message\n");)
						// pass a fake buffer to the handler
						UchMsgBuffer fake (buf, InLength);
						if (! NewMessage (fake, WaitingReply))
							return;
						// *** this return breaks the assumption that
						// *** ProcessInput empties the buffer.
						// *** this is assumed in HandleRead/HandleSelect
						// *** because BufferedMessages is reset to FALSE;
					}
DBG (else printf ("discarding\n");)
				}
				buf.Flush (InLength);
				State = WAITING;
				// fallthrough
		}
	}
}

/*?nodoc?*/
void
UchMsgStream :: HandleRead ()
{
	if (BufferedMessages) {
		ProcessInput (Buffered, FALSE);
		BufferedMessages = FALSE;
	}
	int n = ReadInput ();
	if (n <= 0)
		return;
	ProcessInput (InBuffer, FALSE);
}

/*?nodoc?*/
void
UchMsgStream :: HandleWrite ()
{
	Flush ();
}

/*?nodoc?*/
bool
UchMsgStream :: HandleSelect ()
{
	if (BufferedMessages) {
		ProcessInput (Buffered, FALSE);
		BufferedMessages = FALSE;
	}
	return FALSE;
}


/*?
This virtual function is called whenever a complete message is in the buffer.
The buffer contains exactly one message, so that you can use \com{buf.Get(&msg)}
to extract the message from the buffer.
\var{ask} is TRUE if the message was sent with \fun{Ask}.
In this case an answer must be sent back with \fun{Reply}.
Messages can be sent before replying; they will be buffered by the receiver for later processing;
thus, you cannot use \fun{Ask} before replying.
\fun{NewMessage} function must return TRUE if it handled the message, else FALSE.
If it returns FALSE, it will be called again with the same arguments next time data arrives on this channel.
In the class \typ{UchMsgStream} this function does nothing and returns TRUE;
if \var{ask} is TRUE, it replies immediately with an empty message.
?*/
bool
UchMsgStream :: NewMessage (UchMsgBuffer&, bool ask)
{
	if (ask) {
		UchMessage	dummy;
		
		Reply (dummy);
	}
	return TRUE;
}

/*?
Send a message.
If \var{flush} is TRUE, the output buffer will be flushed.
This also happens when the message stream is in synchronous mode,
or if the output UchMsgBuffer.has exceeded its flush size (see \fun{FlushSize}).
?*/
void
UchMsgStream :: Send (UchMessage& msg, bool flush)
{
	OutBuffer.Append ((byte) MSG);
	OutBuffer.Append (msg);
	if (flush || Sync || OutBuffer.BufLength () >= OutSize)
		Flush ();
}

/*?
Send a message and wait for an answer.
Incoming messages that are received while waiting for the answer are kept for later processing.
The answer message is returned by calling the virtual function \fun{ConvertAnswer}.
?*/
void*
UchMsgStream :: Ask (UchMessage& msg)
{
	if (WaitingReply) {
		Error (ErrWarn, "UchMsgStream::Ask", "cannot ask before replying");
		return 0;
	}
	OutBuffer.Append ((byte) ASK);
	OutBuffer.Append (msg);
	Flush ();
	WaitingForAnswer = TRUE;

	do {
#ifdef DEBUG
		printf ("waiting for answer\n");
#endif
		int n = ReadInput ();
#ifdef DEBUG
		printf ("	got %d bytes\n", n);
#endif
		if (n <= 0)
			return 0;
		ProcessInput (InBuffer, TRUE);
	} while (WaitingForAnswer);
	
	// ProcessInput did not flush so that we can convert the answer
	// *** If ProcessInput would return a void*, we could do all this stuff in it ...
	UchMsgBuffer fake (InBuffer, InLength);
	void* res = ConvertAnswer (fake);
	InBuffer.Flush (InLength);
	State = WAITING;
	return res;
}

/*?
This function must be used instead of \fun{Send} to send a reply to a message sent by \fun{Ask}.
?*/
void
UchMsgStream :: Reply (UchMessage& msg)
{
	if (! WaitingReply) {
		Error (ErrWarn, "UchMsgStream::Reply", "out of phase reply discarded");
		return;
	}
	OutBuffer.Append ((byte) ANS);
	OutBuffer.Append (msg);
	Flush ();
	WaitingReply = FALSE;
}


/*?
This function is called by \fun{Ask} when the answer is in the buffer,
in order to convert it into an object usable by the application.
?*/
void*
UchMsgStream :: ConvertAnswer (UchMsgBuffer&)
{
	return 0;
}

/*?
Flush the output buffer.
This function is called automatically when the buffer exceeds its flush size,
or after each message when this stream is in synchronous mode.
It is also called from \fun{Ask} to wait for the answer.
?*/
void
UchMsgStream :: Flush ()
{
	if (! OutBuffer.BufLength ())
		return;
	int n = Write (OutBuffer);
	OutBuffer.Flush (n);
}

/*?
Send a buffer containing a message.
If \var{flush} is TRUE, the output buffer will be flushed.
This also happens when the message stream is in synchronous mode,
or if the output UchMsgBuffer.has exceeded its flush size (see \fun{FlushSize}).
The buffer {\em must} contain a converted message.
This can be used for instance from inside \fun{NewMessage} to resend
the incoming message to another client, without having to convert
the buffer to a message.
?*/
void
UchMsgStream :: Send (UchMsgBuffer& buf, bool flush)
{
	OutBuffer.Append ((byte) MSG);
	OutBuffer.Append (buf.Buffer (), buf.BufLength ());
	if (flush || Sync || OutBuffer.BufLength () >= OutSize)
		Flush ();
}

#ifdef DOC
/*?
This function defines the size of the output buffer that triggers automatic flushing
in asynchronous mode. By default the flush size is the maximum size of the
output buffer. As a consequence, it is changed by \fun{OutBuffer}.
?*/
void
UchMsgStream :: FlushSize (int n)
{ }

/*?nextdoc?*/
void
UchMsgStream :: SetSyncMode (bool s)
{ }

/*?
A message stream can be in synchronous or asynchronous mode.
In asynchronous mode output is buffered while in synchronous mode it is not.
Synchronous mode is usually less efficient than asynchronous mode
because it makes more system calls to transfer data;
however synchronous mode can be useful for debugging applications.
?*/
bool
UchMsgStream :: GetSyncMode ()
{ }

#endif /* DOC */