xref: /barrelfish-2018-10-04/usr/eclipseclp/documents/embedding/embremote.tex

% BEGIN LICENSE BLOCK
% Version: CMPL 1.1
%
% The contents of this file are subject to the Cisco-style Mozilla Public
% License Version 1.1 (the "License"); you may not use this file except
% in compliance with the License.  You may obtain a copy of the License
% at www.eclipse-clp.org/license.
% 
% Software distributed under the License is distributed on an "AS IS"
% basis, WITHOUT WARRANTY OF ANY KIND, either express or implied.  See
% the License for the specific language governing rights and limitations
% under the License. 
% 
% The Original Code is  The ECLiPSe Constraint Logic Programming System. 
% The Initial Developer of the Original Code is  Cisco Systems, Inc. 
% Portions created by the Initial Developer are
% Copyright (C) 2006 Cisco Systems, Inc.  All Rights Reserved.
% 
% Contributor(s): Kish Shen, IC-Parc
% 
% END LICENSE BLOCK
%
%% $Id: embremote.tex,v 1.2 2008/09/17 18:13:00 kish_shen Exp $
%
% Author:       Kish Shen, IC-Parc
%

%----------------------------------------------------------------------
\chapter{Remote Tcl Interface}
\label{chapremote}
%HEVEA\cutdef[1]{section}
%----------------------------------------------------------------------

This chapter describes the remote Tcl interface, which allows a separate
external Tcl program to interact with {\eclipse} in much the same fashion as
the embedding Tcl interface (see chapter~\ref{chaptcl}). Like the embedding
interface, Tcl and {\eclipse} code communicates by sending and receiving 
streams of bytes via I/O queues (the {\bf ec_rpc} mechanism is
implemented on top of these queues). 
The interface can thus be
used in similar fashion to the embedding interface, e.g.\ for the
development of graphical user interfaces to an ECLiPSe application, with the
difference that the {\eclipse} program is a separate program and not
embedded into the Tcl program.

The main features of the interface are:
\begin{itemize}
\item The connection between the Tcl and {\eclipse} processes are established
via sockets using TCP network protocol. Thus the Tcl process can be run on
a different machine and platform from the {\eclipse} process. 
\item The Tcl process can be attached to any running {\eclipse} process,
including an {\eclipse} embedded into another host language.
\item More than one Tcl (or other remote) process can be attached to a
single {\eclipse} via the remote interface. 
\item For the programmer, the embedding and remote interfaces are largely
similar, and once the connection is established in the remote interface,
the same code on the {\eclipse} and Tcl sides can be used for both interfaces.
\end{itemize}

The remote interface thus offers more flexibility than an embedding
interface in how the Tcl code can be connected to an {\eclipse}
program. However, as the Tcl and {\eclipse} processes are not as tightly
coupled as in an embedded interface, the speed of communications between
the Tcl and {\eclipse} processes is likely to be slower.

\section{Basic Concepts of the Interface}

The interface is used by starting separate {\eclipse} and Tcl processes, and
then {\bf attaching} the Tcl process to the {\eclipse} process. Once
attached, the Tcl and {\eclipse} processes can communicate much as in the
embedded interface: {\eclipse} goals can be sent from the Tcl side to the
{\eclipse} side via the remote predicate call ({\bf ec_rpc}) mechanism, 
and further I/O queues can be established between the {\eclipse} and Tcl
processes to allow streams of bytes to sent from one side to the other.

The attached Tcl
process can also be detached from the {\eclipse} process. This
disconnection will terminate and clean-up the links between the two
processes. Thus, typically, if the programmer wants to allow a particular
application to be usable through both the Tcl remote and embedding
interfaces, the only code that needs to be specific to one or the other
interface is the code associated with starting and termination of the
application (the attach and detach operations in the case of the remote
interface). 

The interaction between the Tcl and {\eclipse} is mediated by a version of
thread-like control flow of the embedded interface. The
interface distinguishes two `sides': the Tcl side, which is the Tcl
process, and the {\eclipse} side, which is generally the {\eclipse}
process\footnote{The {\eclipse} side may be more complicated than a simple {\eclipse}, as it can be
an embedded {\eclipse}, or the {\eclipse} process and other attached remote
processes.}. At any given time,
either the {\eclipse} side or the Tcl side has `control'. When the Tcl side
has control, execution of the {\eclipse} process is suspended. When the
{\eclipse} side has control, the Tcl side cannot initiate the execution of
ec_rpc goals. The interface can implicitly transfer control from one side
to the other (e.g.\ when processing synchronous I/O), or it can be
explicitly transferred.

An {\eclipse} process can have several attached remote processes. Each
remote process is identified by a {\bf control} name, which is the
{\eclipse} name for a special control connection between the two sides. 

\section{Loading the Interface}
{\sloppy
Before using the interface, 
the Tcl program must first load a Tcl-package called {\bf
remote_eclipse}, which can be loaded as follows:
\index{remote_eclipse}

\begin{quote}\begin{verbatim}
lappend auto_path "/location/of/my/eclipse/lib_tcl"
package require remote_eclipse
\end{verbatim}\end{quote}
% It might also be necessary to provide information about where the
% DLLs or shared library files can be found. The details are
% platform-specific:
% 
% \begin{itemize}
% \item On UNIX systems, the {\tt LD\_LIBRARY\_PATH}
% environment variable must include the path:
% \begin{quote}
% {\tt <eclipse\_dir>/lib/<architecture\_os>/}
% \end{quote}
% where {\tt <architecture\_os>} is for example {\tt i386\_linux} or 
% {\tt sparc\_sunos5}.
% \item On Windows NT, the {\tt PATH} environment variable must contain the path:
% \begin{quote}
% \begin{verbatim}
% <eclipse_dir>\lib\i386_nt\
% \end{verbatim}
% \end{quote}
% \end{itemize}

An {\eclipse} program, onto which the Tcl program will attach, also needs to
be started.
}

\section{Attaching and Initialising the Interface}

To use the interface, the Tcl program needs to be
attached to the {\eclipse} program. The attach request is initiated on the
{\eclipse} side, by calling the predicate
\bipref{remote_connect/3}{../bips/kernel/externals/remote_connect-3.html} \footnote{Instead of {\bf remote_connect/3}, the more
flexible \bipref{remote_connect_setup/3}{../bips/kernel/externals/remote_connect_setup-3.html} and \bipref{remote_connect_accept/6}{../bips/kernel/externals/remote_connect_accept-6.html}
pair of predicates can be used. See the reference manual entries for these predicates for more
details.} from {\eclipse}. The Tcl program is then attached to the
{\eclipse} program by executing the procedure {\bf
ec_remote_init}\index{ec_remote_init (Tcl command)} from Tcl. If no error
occurs, then the connection is established and the interface is set up.


In more detail, the {\eclipse} predicate remote_connect/3  establishes a
socket listening for the connection from the Tcl side. It prints out, on
the stream {\tt log_output}, the
hostname and the port number that the Tcl side should connect to:

\begin{verbatim}
[eclipse 1]: remote_connect(Host/Port, Control, _InitGoal).
Socket created at address chicken.icparc.ic.ac.uk/25909
\end{verbatim}

On the Tcl side, ec_remote_init is called with the hostname and port number
given by remote_connect/3:

\begin{verbatim}
ec_remote_init chicken.icparc.ic.ac.uk 25909
\end{verbatim}

\begin{description}
\item[\index{ec_remote_init (Tcl remote interface)}{\bf ec_remote_init} {\it host port ?init_command? ?pass? ?format?}]\ \\
	Initialise the remote Tcl interface on the Tcl side. A
	corresponding {\bf remote_connect/3} must have been started on the
	{\eclipse} side, which specifies the hostname ({\it host}) and port
	number ({\it port}) to connect to. The optional {\it
	init_command\/} is an is a Tcl command that will be invoked at the
	end of the attachment, before control is handed over to the
	{\eclipse} side (see section~\ref{remote-control} for more
	details). {\it pass\/} and {\it format\/} are optional arguments for a simple
	security check: they specify an {\eclipse} term that will be
	matched against a corresponding term (using \txtbipref{==/2}{(==)/2}{../bips/kernel/termcomp/EE-2.html}) on the
	{\eclipse} side before the connection is allow to proceed ({\it pass\/}
	will be sent to the {\eclipse} side in EXDR format\footnote{See
	section~\ref{secexdrtcl} for more on EXDR format}; the default is
	an empty string, which is what \bipref{remote_connect_setup/3}{../bips/kernel/externals/remote_connect_setup-3.html}
	expects).

\end{description}

If successful, some initial links are established between the two sides,
such as the control connection and the connection to allow rpc goals to be
sent from the Tcl to the {\eclipse} side. 
After the attachment, optional user-defined
initialisations are performed on both sides (via the InitGoal argument on
the {\eclipse} side, and the init_command on the Tcl side), and the two
sides can then interact. Initially, the control is given to the Tcl side,
and {\it remote_connect/3} returns only when control is handed over to the
{\eclipse} side. 

As part of the attachment process, the {\eclipse} name of the control
connection  is passed to the Tcl side. This can be accessed by the user
using the command:

\begin{description}
\item[\index{ec_control_name (Tcl remote interface)}{\bf ec_control_name}]\ \\ returns the {\eclipse} name of the control
connection. An error is raised if this procedure is called before
an attachment to {\eclipse} is made.
\end{description}

Unimplemented functionality error will be raised if the Tcl or
{\eclipse} side are incompatible with each other. This can happen if
one side is outdated, e.g.\ if the remote Tcl interface used and the {\eclipse}
being connected to are not from the same version of {\eclipse}. In this
case, it is best to update both sides to the latest version of {\eclipse}.

\subsection{A Note on Security}

Once a Tcl side is attached to an {\eclipse}, the Tcl side can execute 
{\eclipse} goals on the {\eclipse} side via the {\bf ec_rpc} mechanism. This 
may be a security concern as
this gives the Tcl side as much access to the resources on the {\eclipse}
side as the {\eclipse} process itself, even though the Tcl side can
potentially be anywhere reachable from the {\eclipse} side via TCP.
However, the connection must be initiated from the {\eclipse} side, and the
attachment process must follow a protocol in order for a successful
attachment. Nevertheless,
if a third party somehow knew which Address to connect to, and follows the
protocol, it can `steal' the connection to {\eclipse}. No
authentication is performed by the simple \verb'remote_connect_setup/3',
but \bipref{remote_connect_accept/6}{../bips/kernel/externals/remote_connect_accept-6.html} does allow a simple authentication
where it can require the Tcl side to send an {\eclipse} term that matches
the one specified in calling the predicate. This is done before the Tcl
side is given the ability to run rpc goals on the {\eclipse} side. 

It is also possible to limit the remote connection to the same machine as the
{\eclipse} process by specifying `localhost' as the host name in the
Host/Port address of \verb'remote_connect/3'. The Tcl side must also
use `localhost' for the Host name in its client connection. 

Each peer queue is created by creating a new server socket on the
{\eclipse} side and then accepting a client connection from the Tcl
side. The accept command is told where the client connection is from, and
the client host is checked against the client's host from the attachment,
to ensure that the same host has been connected. If not, the {\eclipse}
side will reject the particular connection. At this point, the security has
probably been compromised, and the two side should disconnect.

Note also that by default, none of the information sent through the queues
between the remote side and the {\eclipse} side is encrypted. If the
programmer requires these communication channels to be secure, then such
encryptions need to be provided by the programmer.

\section{Type Conversion Between Tcl and {\eclipse}}

The EXDR ({\eclipse} External Data Representation, see
chapter~\ref{chapexdr}) representation is fully supported by the
interface. The same type conversions commands as in the embedding Tcl
interface, described in section~\ref{secexdrtcl}
(ec_write_exdr, ec_read_exdr, ec_tcl2exdr, ec_exdr2tcl), are available.

\section{Executing an {\eclipse} Goal From Tcl}

An {\eclipse} predicate can be invoked from the Tcl side using the
remote {\eclipse} predicate call (ec_rpc) facility. This should be the
main method of interacting and communicating with {\eclipse} in the remote
interface. Information can be sent to {\eclipse} via bindings for (input)
arguments when the call is made; and
results returned from {\eclipse} via the bindings made to (output)
arguments:

\begin{description}
\item[\index{ec_rpc (Tcl remote interface)}{\bf ec_rpc} {\it goal ?format?}]\ \\
        Remote {\eclipse} predicate call.
        It calls goal in the default module. The goal should be simple
	in the sense that it can only succeed, fail or throw.
        Any choice-points the goal leaves will be discarded.

	Calls to {\bf ec_rpc} can be nested
	and can be used from within Tcl queue event handlers. However, an
	{\bf ec_rpc} cannot be issued while {\eclipse} side has control.

	If no format argument is given, the goal is assumed to be in
	{\eclipse} syntax. If a {\it format} argument is provided,
	the {\eclipse} goal is constructed from {\it goal} and {\it format},
	according to the conversion rules explained in section \ref{secexdrtcl}.

        On success, {\bf ec_rpc} returns the (possibly more instantiated)
	goal as a Tcl data structure (in EXDR format), otherwise "fail" or 
        "throw" respectively.
\end{description}

\section{Communication via Queues}

Queues should be used to set up long-term I/O links between {\eclipse} and
Tcl. An example would be the main output from an application that is to be
displayed by a Tcl window.  Streams of bytes can be sent along the queue
from one side to the other: on one side, data is written to the queue; and
when the queue is flushed, the data is sent to the other side, which can
now read the data. The data can either be sent as normal strings (where
each byte represents a character) using the normal I/O calls, or they can
be in EXDR format, in which case both sides need to read and write using
EXDR.

On the Tcl side, the queue is seen as a Tcl I/O
channel. On the {\eclipse} side, a queue is
seen as an {\eclipse} I/O stream, which has a unique (numeric) ID, the
stream number, and has a user supplied symbolic
name. These all refer to the same queue. Queues are created using the symbolic names, and the Tcl side
maintains tables of the correspondence between Tcl channel names, symbolic
names and stream numbers. The built-in Tcl I/O commands accepts the Tcl
channel name for operating on the queue, and for 
compatibility with the embedding interface, many of the Tcl remote
interface commands refer to the queue using the stream number. The
interface provides commands to inter-convert the various names so that the
right name can be used for a particular command.


There are two types of queues: 

\begin{description}
\item[synchronous] These queues are unidirectional,
i.e.\ either for sending data from {\eclipse} to Tcl (from-{\eclipse}), or
from Tcl to {\eclipse} (to-{\eclipse}). 
These streams are synchronous because the interface ensures that the
sending and receiving of data across the queue are synchronised. This is
achieved by transferring control between
{\eclipse} and Tcl in a coroutine-like manner to ensure that data that is
sent from one side is processed on the other. 


{\sloppy
These queues are  designed to be compatible with the queues
created via ec_queue_create of the embedded interface 
(see section~\ref{ecqueueconnect}). Their actual implementations are
different, in that the queues in the embedded case are memory queues and
the synchronous queue use socket connections. The interface tries to
minimise the difference by buffering where
possible at either ends of the  socket connection. However, there is an overhead for doing
this, and not all differences can be hidden. This is discussed in more detail in section~\ref{remotediff}.
}


\item[asynchronous] These are bi-directional --
data can be sent both ways. Sending or receiving data on these queues does
not necessarily transfer control between {\eclipse} and
Tcl. In particular, it is not possible to request data from the other side
if the queue is empty: such an operation would simply block. 
This is because such queues map
directly to the socket connections with no buffering, and there is no concept of a socket
being empty. Generally, it is up to the programmer to co-ordinate the
transfer and processing of the data. 

They have no direct
equivalent in the embedding Tcl interface, but some uses of the embedding
Tcl interface queues, such as writing data from one side without a
corresponding reader on the other side, are better approximated by the
asynchronous queues than the synchronous queues.  They can also be more
efficient in that there is no buffering of the data is performed by the
interface. 

\end{description}

\subsection{Queue Data Handlers}
\label{remotehandles}

The processing of data on queues (synchronous and to some extent
asynchronous) can be performed via {\it handlers}. A handler is a piece of
code (a procedure in Tcl, a goal in {\eclipse}) whose execution is
data-driven: it is invoked to deal
with the transfer of data on a queue on their respective sides. 

In {\eclipse}, the handler goal is invoked using the events mechanism. That
 is, an event is raised, and the event handler goal associated with the event
 (see \bipref{set_event_handler/2}{../bips/kernel/event/set_event_handler-2.html}) is then executed when {\eclipse} has control. 

A handler can be called under two situations:

\begin{description}
\item[Data consumer] To consume data that has been sent over from the other
side. Here, the other side has sent data over the queue, invoking the handler.
The handler is expected to read the data off the queue and process it.
An example of a data consumer handler is a Tcl handler which is invoked
when the {\eclipse} side sends data that is intended to be displayed on a Tcl
window. The handler would be
invoked to read the data off the queue and display it on the window. 

\item[Data provider] To provide data that has been requested by the other side. In this
 case, the handler is expected to generate the data and write the data onto
 the queue, and send it to the other side.  For example, on the Tcl side, a
 Tcl handler might be invoked to ask for inputs from the user via the GUI.
 Note that these data providers can only exist for the synchronous
 queues.

\end{description}

For each queue and for a particular direction of data flow, a
handler can be defined on either the Tcl or the {\eclipse} side, but not
both. The handler either consumes or provides data as described above. The
reason that handlers cannot be defined on both sides 
is that this avoids possible infinite loop of alternately
triggering the data provider and the data consumer.

\subsection{Synchronous Queues}


These queues can be created on the Tcl side. This is done with
the {\bf ec_queue_create} command from within Tcl code:

\begin{description}
\item[\index{ec_queue_create (Tcl remote interface)}ec_queue_create {\it
eclipse_stream_name mode ?command? ?event?}]\ \\
        Creates a synchronous queue between Tcl and {\eclipse} sides. On
        the Tcl side, a Tcl channel is created. On the {\eclipse} side, the 
        queue would be given the symbolic name {\it eclipse_stream_name}. 
        The {\it mode} argument indicates the direction of the queue, and
        can either be fromec or toec\footnote{For compatibility with previous
        versions of the embedded Tcl interface, the mode can also be
        specified as r (equivalent to fromec) or w (equivalent to
        toec). These can be somewhat confusing as read/write status depends
        on from which side the queue is viewed (a read queue in {\eclipse} is a
        write queue in Tcl).}.
        The procedure returns a channel identifier for use in commands
        like {\bf puts}, {\bf read}, {\bf ec_read_exdr},
        {\bf ec_write_exdr} or {\bf close}. The optional arguments {\it
        command\/} and {\it event\/} specifies the data handler for the
        queue: {\it command\/} is the name
        of the Tcl
        procedure for handling the data, with its user defined arguments. 
        {\it event} is the name of the event that will be
        raised on the {\eclipse} side (see the section~\ref{remotehandles}
        for more details). As a handler can only be defined for one side,
        either {\it event\/} or {\it command\/} should be undefined
        (\verb'{}'). 
\item[\index{ec_queue_close (Tcl remote interface)}ec_queue_close {\it
eclipse_stream_name}]\ \\
	Closes the (synchronous or asynchronous) queue with the {\eclipse} name of {\it
	ec_stream_name}. The queue is closed on both the Tcl and {\eclipse}
	sides, and bookkeeping information for the queue is removed. 

\end{description}

It is strongly recommended that the queues should be used for long-term I/O
connections between the two sides, and so the queues should not be created
and closed on a short-term basis. For quick interchange of data, it is
recommended that the user use the {\bf ec_rpc} mechanism.

\subsubsection{Handlers for a Synchronous From-{\eclipse} Queue}

\paragraph{Tcl Handler for From-{\eclipse} Queue}

For a from-{\eclipse} queue, the Tcl handler {\it command\/} would be a
data consumer. This handler is initiated when
{\eclipse} side initially has control and flushes
the queue (by calling \bipref{flush/1}{../bips/kernel/iostream/flush-1.html}). With a Tcl handler defined,
control is transferred to the Tcl
side, where {\it command\/} is invoked to consume the data. When the
handler finishes, control is returned to the {\eclipse} side. The general
sequence of actions are:

\vspace{0.5cm}
\begin{center}
\begin{tabular}{l|l}
{\eclipse} side & Tcl side\\
\hline
\parbox{6.5cm}{Writes to the from-{\eclipse} queue} &\\

\parbox{6.5cm}{Flush the from-{\eclipse} queue} &\\

          & \parbox{6.5cm}{Handler invoked to handle data on the from-{\eclipse} queue}\\
\parbox{6.5cm}{{\eclipse} returns from flush, and continue executing the following code}&\\
\end{tabular} 
\end{center}
\vspace{0.5cm}

The Tcl handler is specified by {\it command\/} in {\bf
ec_queue_create}. {\it command\/} includes the name of the Tcl procedure to
invoke, and any user defined arguments. When the handler is invoked, two
additional arguments are appended:
the {\eclipse} stream number for the queue, and the number
of bytes that has been sent on the queue. This command should read the data
off the queue and process it. The following predefined Tcl
data consumer handlers are provided:

\begin{description}
\item[\index{ec_stream_to_window_sync (Tcl remote interface)}ec_stream_to_window_sync {\it tag text_widget stream_nr length}]\ \\
        Read {\it length\/} bytes from the specified queue and insert the data
        at the end of the existing {\it text_widget}, using {\it tag} as the tag
        for the text. If this is invoked as a handler for a from-{\eclipse}
        queue, {\it length\/} and {\it stream_nr\/} would be supplied when
        the  handler is invoked.


\item[\index{ec_stream_output_popup (Tcl remote interface)}ec_stream_output_popup {\it label_text stream_nr length}]\ \\
        Pops up a window displaying the {\it label_text},
        a text field displaying the contents of the specified queue stream,
        and an ok-button for closing. The data is read as normal
        strings. This is the default Tcl fromec handler that will be called
        if {\bf ec_create_queue} did not define one.

\end{description} 

\paragraph{An example from-{\eclipse} queue with Tcl handler}

To create the queue on the Tcl side with a Tcl handler:

\begin{verbatim}
Tcl code :  ec_queue_create myqueue fromec {ec_stream_to_window_sync red textwin} {}
\end{verbatim}

Note that the last \verb'{}' specifies that there is no {\eclipse}
handler. This is the actual default for this argument, so it could be
missed out. After creating the queue, it can be used on the {\eclipse}
side. The programmer can write to the queue, and to send the data to the
Tcl side, the queue should be flushed:

\begin{verbatim}
ECLiPSe code :
   ...
   write(myqueue, hello),
   flush(myqueue),
   ...
\end{verbatim}

When the queue is flushed as shown above, then control is handed over to
Tcl, and the Tcl handler, in this case {\bf ec_stream_to_window_sync},
would be invoked. This reads the data on the queue (hello, and anything
else that has been written since the last flush), and puts it into the text
widget textwin, with the tag red. The procedure is also called with the
{\eclipse} stream number for the queue and the number of bytes sent as extra
arguments. The textwin widget and the tag red must
be defined already in the Tcl program (presumably  `red' means printing the
text in red colour); if no tag is desired, \verb'{}' can be used. 

The procedure {\bf ec_stream_to_window_sync} is predefined in the
interface, but here is a slightly
simplified version of it:

\begin{verbatim}
proc ec_stream_to_window_sync {Tag Window Stream Length} {

    set channel [ec_streamnum_to_channel $Stream]
    set data [read $channel $Length]

    $Window insert end $data $Tag
    $Window see end
}
\end{verbatim}

\paragraph{{\eclipse} Handler for From-{\eclipse} Queue}

Currently, the Tcl remote interface does not support {\eclipse} handlers
(which will be a data provider) for from-{\eclipse} queues. Thus, the {\it
event\/} argument for {\bf ec_queue_create} is currently a dummy argument
that is ignored. The available alternative is to use {\bf ec_rpc} to obtain
the required information: instead of reading from a from-\eclipse queue, an
ec_rpc should be called with argument(s) left to be filled in by the
{\eclipse} side with the required data. 

\subsubsection{Handlers for a Synchronous To-{\eclipse} Queue}

\paragraph{Tcl Handler for a To-{\eclipse} Queue}

\label{toeclipse-tclhandler}

For a to-{\eclipse} queue, the Tcl handler {\it command\/} defined in {\bf
ec_queue_create} would be a data
producer. This handler is initiated when {\eclipse} side has control, and
reads from the to-{\eclipse} queue, which is initially empty. With a Tcl
handler defined, control is transferred to the Tcl side, where {\it
command\/} is invoked to provide the data. The handler should write the
data to the queue, and call the Tcl remote interface command {\bf ec_flush}
to send the data to {\eclipse} side. When the handler finishes,
control is returned to the {\eclipse} side, and the read operation is
performed to read the now available data. The general sequence of actions are:

\vspace{0.5cm}
\begin{center}
\begin{tabular}{l|l}
{\eclipse} side & Tcl side\\
\hline
\parbox{6.5cm}{Reads an empty to-{\eclipse} queue} &\\
         & \parbox{6.5cm}{Handler invoked to supply data to the
         to-{\eclipse} queue. The data is written to the queue and flushed
         with {\bf ec_flush}}\\
\parbox{6.5cm}{{\eclipse} returns from the initial read operation, reading
         the data supplied by the Tcl handler, and continue execution the following code}&\\
\end{tabular}
\end{center}
\vspace{0.5cm}

The Tcl remote interface command {\bf ec_flush}, instead of the standard
Tcl {\bf flush} command,  should be used to flush a
queue so that the data would be transferred and processed on the {\eclipse}
side. {\bf ec_flush} should be used both inside the Tcl data provider
handler, and also to invoke an {\eclipse} data consumer handler (see the
next section).

\begin{description}
\item[\index{ec_flush (Tcl remote interface)}{\bf ec_flush} {\it
eclipse_streamnum ?nbytes?}]\ \\
   If the Tcl side has control, flushes the (synchronous or asynchronous) queue
   with the {\eclipse} stream number {\it eclipse_streamnum} and hands over
   control briefly to {\eclipse} to read the data. Control is
   then returned to Tcl. {\it nbyte\/} is an optional argument that specifies the
   number of bytes being sent. If this argument is missing, the data sent
   must be a single EXDR term in the case of the synchronous queue. There
   is no restriction for the asynchronous queues, but it is the
   programmer's responsibility that the read operation does not block.
\end{description}

Normally, data is written to the queue using standard Tcl output commands,
and the amount of data written is not known. However, the programmer may
have kept track of the number of bytes written inside the handler, and thus
know how many bytes will be sent. In this case, {\bf ec_flush} can be
called with the number of bytes supplied as a parameter. It is the
programmer's responsibility to ensure that this information is accurate.
Without nbytes, the output is restricted to EXDR terms for synchronous
queues. The reason for this is because the data is sent through a socket
connection, and without knowing the amount of data, it is not possible in
general to know when the data ends, unless the data sent has implicit
boundaries, like an EXDR term.

For the use of {\bf ec_flush} inside a Tcl data provider handler, the
sequence of events that appears to the user is that the {\bf ec_flush}
flushes the data, and the Tcl side then continues executing Tcl code
until the handler's execution is finished. Control is then returned to
{\eclipse}, where the original read operation can now read the available
data. The actual sequence of event is slightly more complex for synchronous
queues: when {\bf
ec_flush} is invoked, control is actually transferred to {\eclipse}, and
the data flushed is then read into a buffer by {\eclipse}, which then
returns control to Tcl to continue the execution of the handler. When the
handler finally finishes, control returns to {\eclipse}, and the original
read operation reads the data from the buffer and continues. This extra
complexity should be transparent to the programmer except when the
intermediate {\eclipse} read to buffer does not complete (e.g.\ because
{\it nbytes\/} is greater than the actual amount of data sent). 

The Tcl handler is specified by {\it command\/} in {\bf
ec_queue_create}. {\it command\/} includes the name of the Tcl procedure to
invoke, and any user defined arguments. When the handler is invoked, an
additional argument is appended:
the {\eclipse} stream number for the queue. This command should get the
data required, output it onto the queue, and call {\bf ec_flush} to flush
the data to {\eclipse} side. If the command does not flush data to
{\eclipse}, {\eclipse} will print a warning and return control to Tcl side.

The following predefined Tcl data producer
handler is provided:

\begin{description}
\item[\index{ec_stream_input_popup (Tcl remote interface)}{\bf ec_stream_input_popup} {\it label_text stream_nr}]\ \\
        Pops up a window displaying the label_text, an input field
        and an ok-button. The text typed into the input field will
        be written into the specified queue stream {\it stream_nr}, which
        is the {\eclipse} stream number for the queue. If this command is
        invoked as a handler for a to-{\eclipse} queue, {\it stream_nr}
        will be automatically appended by the interface. There should be no
        unflushed data already on the queue when this command is invoked.
\end{description}

\paragraph{An example to-{\eclipse} queue with Tcl handler}

To create the queue on the Tcl side with a Tcl-handler:

\begin{verbatim}
Tcl code :  
ec_queue_create myqueue toec \
    {ec_stream_input_popup "Input for myqueue:"} {}
\end{verbatim}

This associates the pre-defined Tcl data producer handler {\bf
ec_input_popup} with myqueue. The last \verb'{}' specifies that there is no
{\eclipse} handler and can be omitted as that is the default. This queue
can now be used on the {\eclipse} side in a demand driven way, i.e.\
{\eclipse} side can read from the queue:

\begin{verbatim}
ECLiPSe code :

    ...
    read(myqueue, Data),
    ...
\end{verbatim}

When the {\eclipse} side reads from myqueue, and the queue contains no data
on the {\eclipse} side, then
control will be handed over to Tcl, and {\bf ec_input_popup} 
invoked. This pops up a Tcl window, with the label ``Input for myqueue:''
with a text entry widget, asking the user to supply the requested data. The
data is then sent back to the {\eclipse} side.

Here is a slightly simplified version (there are no buttons)
of {\bf ec_stream_input_popup}:

\begin{verbatim}
set ec_stream_input_string {}

proc ec_stream_input_popup {Msg Stream} {
    global ec_stream_input_string

    toplevel .ec_stream_input_box
    label .ec_stream_input_box.prompt  -width 40 -text $Msg
    entry .ec_stream_input_box.input -bg white -width 40 \
        -textvariable ec_stream_input_string
    bind .ec_stream_input_box.input <Return> {destroy .ec_stream_input_box}

    ;# pack the popup window
    pack .ec_stream_input_box.prompt -side top -fill x
    pack .ec_stream_input_box.input -side top -fill x

    tkwait window .ec_stream_input_box
    puts -nonewline [ec_streamnum_to_channel $Stream] $ec_stream_input_string
    ;# flush the output to ECLiPSe with the length of the input
    ec_flush $Stream [string length $ec_stream_input_string]
}

\end{verbatim}

Data is flushed to the {\eclipse} side using {\bf ec_flush}. The {\bf puts}
needs the Tcl channel name of the queue to write to, and this is provided
via the Tcl remote interface command {\bf ec_streamnum_to_channel} (see
section~\ref{translate-remote-qnames}). {\bf ec_flush} is called with two
arguments in this case, both the queue number ({\it Stream}), and the
length of the data that is sent. Note that this makes the assumption that
no other unflushed data has been written to the queue. 

\paragraph{{\eclipse} Handler for a To-{\eclipse} Queue}

For a to-{\eclipse} queue, the {\eclipse} handler would be a data
consumer. This handler is initiated when Tcl initially has control, and
flushes data on a queue using {\bf ec_flush}. Control is transferred to
{\eclipse}, and if the {\eclipse} handler is defined, this is invoked to
consume the data. When the handler returns, control is returned to Tcl,
which continues executing the code after the flush. The general sequence of
actions are:

\vspace{0.5cm}
\begin{center}
\begin{tabular}{l|l}
{\eclipse} side & Tcl side\\
\hline
	& \parbox{6.5cm}{Outputs data onto the to-{\eclipse} queue}\\
	& \parbox{6.5cm}{Calls {\bf ec_flush} to send data to {\eclipse} side}\\ 
\parbox{6.5cm}{The {\eclipse} handler associated with the queue is called to consume and
process the data} &\\
	& \parbox{6.5cm}{Execution continues after the {\bf ec_flush}}\\
\end{tabular}
\end{center}
\vspace{0.5cm}

The {\eclipse} handler is specified by the {\it event\/} argument of {\bf
ec_queue_create}. This specifies an event that will be raised on the
{\eclipse} side when data is written to a previously empty queue. The
{\eclipse} side does not see this data, and the event not raised, until the
data is flushed by {\bf ec_flush} and copied by {\eclipse} to its
buffer and, if the buffer was initially empty, the event would then be raised.

The programmer should define the event handler associated with {\it
event}.

\paragraph{An example to-{\eclipse} queue with {\eclipse} handler}

To create the queue on the Tcl side with an {\eclipse}-handler:

\begin{verbatim}
Tcl code: 
    ec_queue_create myqueue toec {} remoteflush_myqueue
\end{verbatim}

Note that the \verb'{}' is needed to specify that there is no Tcl handler. 
It defines \verb'remoteflush_myqueue' as the event that will be raised when
the queue is flushed by {\bf ec_flush} on the Tcl side. 

The event handler needs to be defined on the {\eclipse} side:

\begin{verbatim}
ECLiPSe code:

:- set_event_handler(remoteflush_myqueue, read_myqueue/0).

...
read_myqueue :-
      read_exdr(myqueue, Data),
      process(Data).

\end{verbatim}

This read handler assumes that the data is written using EXDR format. So on
the Tcl side, the data should be written using EXDR format:

\begin{verbatim}
Tcl code:

     ...
     ec_write_exdr [ec_streamname_to_channel myqueue] $data
     ec_flush [ec_streamname_to_streamnum myqueue]
     ...
\end{verbatim}


\subsection{Asynchronous Queues}

Asynchronous queues are created on the Tcl side using the Tcl command {\bf
ec_async_queue_create}:

\begin{description}
\item[\index{ec_async_queue_create (Tcl remote interface)}ec_aysnc_queue_create {\it eclipse_stream_name ?mode?
?fromec_command? ?toec_event?}]\ \\
	Creates a socket stream between {\eclipse} and Tcl with the name
	{\it eclipse_stream_name\/} on the {\eclipse} side. The created
	stream is bidirectional, and can be written to or read from at both
	ends. The {\it mode\/} argument is for compatibility with the {\it
	ec_aysnc_queue_create\/} command of the embedded interface only,
	and has no effect on the nature of the queue.  The procedure
	returns a channel identifier for use in commands like {\bf puts},
	{\bf read}, {\bf ec_read_exdr}, {\bf ec_write_exdr} or {\bf
	close}. Unlike the synchronous queues, only data consumer handlers
	can be defined: if a {\it fromec_command} argument is provided,
	this command is set as the Tcl data consumer handler to be called
	when data arrives on the Tcl end of the socket. If {\it toec_event\/}
	is given, it specifies the event that will be raised on the
	{\eclipse} side when data is flushed by {\it ec_flush} on the Tcl
	side. 
\item[\index{ec_queue_close (Tcl remote interface)}ec_queue_close {\it
eclipse_stream_name}]\ \\
	Closes the (synchronous or asynchronous) queue with the {\eclipse}
	name of {\it 	ec_stream_name}. The queue is closed on both the Tcl and {\eclipse}
	sides, and bookkeeping information for the queue is removed. 



\end{description}

Asynchronous queues are bi-directional queues which allows data transfer
between {\eclipse} and Tcl sides without transfer of control. 
In the case where a Tcl data consumer
handler is defined in {\it fromec_command}, which is invoked on the Tcl
side when the queue is flushed on the {\eclipse} side, the {\eclipse} side
will carry on execution while the handler is invoked on the Tcl side. 

These queues are designed to allow for more efficient transfer of data
between {\eclipse} and Tcl than the synchronous queues. 

For data transfer from
{\eclipse} to Tcl, the intended use is that a Tcl data consumer handler
 would be invoked as the data becomes available on
the Tcl side, after being flushed from the {\eclipse} side. 
Note that control is not handed over to Tcl side in this case:
the Tcl handler is
invoked and executed on the Tcl side while {\eclipse} side still has control, with the
restriction that the Tcl handler is unable to issue {\bf ec_rpc}
goals because {\eclipse} side still retains control. Another difference
with the synchronous  from-{\eclipse} queue is that the handler would read
from the queue in non-blocking mode, i.e.\ it will read whatever data is
available on the queue at the Tcl side and never wait for more data. 
If more data become
available, the handler would be invoked again. The following Tcl handler is
pre-defined for the asynchronous queue for handling from-{\eclipse} data:

\begin{description}
\item[\index{ec_stream_to_window (Tcl remote interface)}ec_stream_to_window {\it tag text_widget stream_nr length}]\ \\
        Inserts all the current contents of the specified queue 
        at the end of the existing {\it text_widget}, using {\it tag} as
        the tag for the text. 

\end{description}


For data transfer from Tcl to {\eclipse}, 
the queue can be used either asynchronously or synchronously. 
If the
queue is used asynchronously, then the standard Tcl command {\bf flush}
should be used to flush the queue. There would not be any transfer of
control, and so there would not be an
immediate corresponding read on the {\eclipse} side. In fact, no handler
would be invoked automatically on the {\eclipse} side, even when control is
transferred. 
Output and flush
operations do not block on the Tcl side, as the Tcl side of the queue is
put into non-blocking mode, so that the data is buffered and the operations
carried out when they will not block. It is the programmer's responsibility
to write and call the code to read the data from the queue on the {\eclipse} side when
the {\eclipse} side is given control.

This asynchronous use to send data to {\eclipse} should be useful when
the queue is used as an auxiliary data
channel, where the main data is sent either via {\bf ec_rpc} or another
queue. The desired effect is that data can be sent on the auxiliary channel
without triggering processing on the {\eclipse} side until it is told to do
so on the main data channel, which would be handled synchronously. 

To use the queue synchronously for
to-{\eclipse} data, {\bf ec_flush} should be used to flush the queue on the
Tcl side. 
With {\bf ec_flush}, control will be handed over to the
{\eclipse} side to process the data: the goal associated with the event {\it toec_event} is
executed, and this goal should read the data from the queue. 
Unlike the synchronous to-{\eclipse}
queues, the data is not buffered, and the handler goal is called every time
{\bf ec_flush} is invoked, rather than only when the queue is empty. This
should normally not make any difference, as the handler should empty all
the contents of a queue each time it is invoked.

The goal is called with two optional arguments: the first argument is the
event name, the second argument is the `culprit' of the form
\verb'rem_flushio(Queue,Len)', indicating that this event is caused by a
remote flush, where Queue is the {\eclipse} stream number,
and Len is the number of bytes sent (this is supplied by {\bf ec_flush}, if
{\bf ec_flush} does not supply a length, then Len is the atom
\verb'unknown'.

\subsubsection{Examples for asynchronous queue}

\paragraph{Using the queue asynchronously: to-{\eclipse}}

An example of using an asynchronous queue asynchronously to send data to
{\eclipse} is in the tracer for Tk{\eclipse} development tools. Here the trace line is
printed on a synchronous from-{\eclipse} queue, and handled by a Tcl data
consumer handler which prints the trace line and waits for the user to type
in a debugger command. This debugger command is sent to the {\eclipse}-side
using an asynchronous queue, which is read by the {\eclipse} side when it
returns. Here is a much simplified version of the code:

\begin{verbatim}
Tcl code:
    ...
    ec_queue_create debug_traceline fromec handle_trace_line
    ec_async_queue_create debug_input toec ;# no handlers
    ...
\end{verbatim}

During the initialisation of the development tools, the Tcl code creates
the from-{\eclipse} queue where the trace-line information is sent
(\verb'debug_traceline'), and the asynchronous queue (used only in a
to-{\eclipse} direction) for sending the debugger commands to {\eclipse}
(creep, leap, skip etc.). Note that as this queue is used asynchronously,
there are no handlers associated with it.

On the {\eclipse} side, when a goal with a spy-point is executed, this
raises an event that calls the predicate \verb'trace_line_handler/2' which 
should output the trace-line, and wait for a debug command, process the
command, and carry on:

\begin{verbatim}
trace_line_handler(_, Current) :-
        % Current contains information on the current execution state
        % from this a trace line Traceline (a string) can be created
        make_current_traceline(Current, Traceline),
        % send the traceline to Tcl side
        write_exdr(debug_traceline, Traceline),
        flush(debug_traceline),	
        % flush will return when the Tcl handler has finished
        read_exdr(debug_input, Cmd),
        % read the command from debug_input and process it
        interpret_command(Cmd, Current).
\end{verbatim}

The trace-line handler is called with the second argument set to a
structure that contain information on the current execution state
(\verb'Current'), from this, a trace-line (the debug port name, depth, goal
being traced etc.) can be constructed: \verb'Traceline' is the string that
should be printed, e.g.

\begin{verbatim}
  (1) 1 CALL  append([1, 2, 3], [], L)
\end{verbatim}
This is sent as an EXDR term to the Tcl side using the synchronous
queue \verb'debug_traceline'. When \verb'flush/1' is
called, control is handed over to the Tcl to handle the data, and the
Tcl data consumer handler {\bf handle_trace_line} is invoked:

\begin{verbatim}
proc handle_trace_line {stream length} {
	global tkecl

        $ec_tracer.trace.text insert end \
             [ec_read_exdr [ec_streamnum_to_channel $stream]]
        configure_tracer_buttons active

        ;# wait for a tracer command button to be pressed...
        tkwait variable tkecl(tracercommand)
	configure_tracer_buttons disabled
        ec_write_exdr [ec_streamname_to_channel debug_input] \
             $tkecl(tracercommand)
        flush [ec_streamname_to_channel debug_input]
}
\end{verbatim}

As this is invoked as a handler, the {\eclipse} stream number ({\it
stream}) and number of bytes sent ({\it length}) are appended as
arguments. Note that as the trace-line is written as an EXDR term, the {\it
length\/} information is actually not needed. What the handler does is
simply read the trace-line as an EXDR term, and placing the resulting
string onto the tracer text window \verb'$ec_tracer_trace.text'. Next, {\bf
configure_tracer_buttons active} is called. This code is not shown, but
what it does is to enable the buttons for the debugger commands so that the
user can press them. There are buttons for the debugger commands such as
`leap', `creep' etc. When one of this button is pressed, the global
variable \verb'tkecl(tracercommand)' is set to the corresponding command,
and the handler continues its execution beyond the {\bf tkwait}. The
buttons are disabled, the command sent to {\eclipse} side on the
\verb'debug_input' queue using {\bf flush}. This is the asynchronous
sending of data on the asynchronous queue:
control is {\it not\/} handed
over to {\eclipse} to process this command. Instead, the execution on the
Tcl side carries on (and happens to finish immediately after the {\bf
flush}. Control is then returned to the {\eclipse} side as the Tcl handler
has finished, and the {\eclipse} side continues execution after the
\verb'flush(debug_traceline)' goal. Next, \verb'debug_input' is read for
the tracer command, and this command is acted on.

\paragraph{Using the queue synchronously: to-{\eclipse}}


If the Tcl remote interface command {\bf
ec_stream_input_popup} (see section~\ref{toeclipse-tclhandler}) is used to
send data to the {\eclipse}-side (in section~\ref{toeclipse-tclhandler},
the command was initiated by a read operation on the {\eclipse} side; here
the command is invoked directly when Tcl side has control), then the
following is a possible {\eclipse} handler:

\begin{verbatim}
Tcl code:

;# create the asynchronous queue, with
;#  from-ECLiPSe Tcl consumer handler: data_to_window 
;#  to-ECLiPSe ECLiPSe handler event:  flush_myqueue
ec_async_queue_create myqueue {data_to_window textwin} flush_myqueue

...
;# get input for the queue and send to ECLiPSe side
ec_stream_input_popup "Data:" [ec_channel_to_streamnum myqueue]
...

ECLiPSe code:

:- set_event_handler(flush_myqueue, read_remote_data/2).

% Len is known when ec_stream_input_popup is used to send data
read_remote_data(_Event, rem_flushio(Queue,Len)) :-
	read_string(Queue, "", Len, Data), 
	process(Data).

\end{verbatim}

The {\eclipse} code defines \verb'read_remote_data/2' as the handler for
to-{\eclipse} data sent with {\bf ec_flush} on the Tcl side. This handler
is called when control is handed over to {\eclipse} side to read the
data. Both the two optional arguments are used in this handler. The second
argument supplies the {\eclipse} stream number for the queue and the length
of data written. As the data is
sent by explicitly calling {\bf ec_stream_input_popup}, the length of the
data sent is known, so \verb'read_string/4' can be used to read the exact amount of
data. In the asynchronous queue, it is generally the programmer's
responsibility to ensure that the read will not block.

\paragraph{Using the queue asynchronously: from-{\eclipse}}

The example \verb'ec_async_queue_create' also defines a Tcl data consumer
handler to handle data sent on the from-{\eclipse} direction, with a user
defined argument of the text window that the data will be sent to. Here is
a simple procedure which reads the data on the queue and places it on the
text window specified:

\begin{verbatim}
Tcl code:

proc data_to_window {Window Stream} {
    set channel [ec_streamnum_to_channel $Stream]

    $Window insert end [read $channel]
}

\end{verbatim}

The {\it Stream} argument is appended by the interface when the handler is
invoked, and is the {\eclipse} stream number of the queue. The procedure
simply reads the data from the corresponding Tcl channel and display the
data on {\it Window}, the text window specified by the programmer.


\subsection{Reusable Queue Names}

{\eclipse} stream names are global in scope, so using fixed queues names
like `myqueue' might cause name conflicts with other modules, if the
programmer intend the remote Tcl code to be usable with other {\eclipse} code.
One way to avoid name clashes is to dynamically composing queue names using
the name of the control connection:


\begin{verbatim}
Tcl code:
        append queue_name [ec_control_name] myqueue
        ec_queue_create $queue_name fromec {ec_stream_output_popup red textwin}
\end{verbatim}

The user specified name `myqueue' is appended to the control name of the
remote connection to give a unique queue name. On the {\eclipse} side, 
the code will also need to use the dynamic name:

\begin{verbatim}
        :- local variable(remote_control). 

        ...
        % code fragment to remember the control name
        remote_connect(Addr, Control, _),
        setval(remote_control, Control), 
        ...

        ...
        % code fragment to use the queue
        getval(remote_control, Control),
        concat_atom([Control, myqueue], QName),
        ...
        write(QName, hello), flush(QName),
        ...

\end{verbatim}


\subsection{Translating the Queue Names}
\label{translate-remote-qnames}

The remote queues connecting {\eclipse} and Tcl are given different names
on the two sides. The remote Tcl interface
keeps track of the {\eclipse} names for the queues on the Tcl side. On the
{\eclipse} side, the queue has a stream number, as well as possibly several
symbolic aliases. The interface only keeps track of one symbolic name --
the one that is supplied in {\it ec_queue_connect\/} and {\it
ec_async_queue_create}. If the {\eclipse} stream number was supplied in
these commands, then the stream number is also considered the symbolic name
for the queue as well. The Tcl interface provides several commands to convert the names
from one form to another:

\begin{description}
\item[\index{ec_streamname_to_channel (Tcl remote interface)}ec_streamname_to_channel {\it eclipse_name}]\ \\
	Returns the Tcl channel name for the remote queue with the
	symbolic name {\it eclipse_name}. 
\item[\index{ec_streamnum_to_channel (Tcl remote
	interface)}ec_streamnum_to_channel {\it eclipse_stream_number}]\ \\
	Returns the Tcl channel name for the remote queue with the
	{\eclipse} stream number {\it eclipse_stream_number}.
\item[\index{ec_channel_to_streamnum (Tcl remote interface)}
	ec_channel_to_streamnum {\it channel}]\ \\
	Returns the {\eclipse} stream number for the remote queue with the
	Tcl channel name {\it channel}.
\item[\index{ec_streamname_to_streamnum (Tcl remote interface)}ec_streamname_to_streamnum {\it eclipse_name}]\ \\
	Returns the {\eclipse} stream number for the remote queue with the
	symbolic name {\it eclipse_name}.
\item[\index{ec_stream_nr (Tcl remote interface)}ec_stream_nr {\it eclipse_name}]\ \\
	This is an alias for {\bf ec_streamname_to_streamnum} for
	compatibility with embedded interface.
\end{description}


\section{Additional Control and Support}
\label{remote-control}

The remote interface provides additional support for controlling the
interaction of the Tcl and {\eclipse} sides, such as explicit transfer of
control between {\eclipse} and Tcl, and the disconnection of the Tcl
and {\eclipse} sides. The interface also provides support for special
user-defined commands to be executed during these events.

\subsection{Initialisation During Attachment}


In an application, after the Tcl side has been attached, typically some
application specific initialisation needs to be performed, such as setting
up various data queues between the two sides, and defining the actions to
take when the two sides are disconnected. On both sides, these
initialisations can be performed immediately after attachment.  On the Tcl
side, such
actions can be specified in the optional {\it init_command} argument of
{\it ec_remote_init}. On the {\eclipse} side, such actions can be specified
in the `InitGoal' (last) argument of {\it remote_connect/3}. InitGoal can
be a built-in, or a user-defined goal.

\subsection{Disconnection and Control Transfer Support}

Disconnection should normally be performed when the {\eclipse} application
has finished using the GUI provided by the particular attached remote process.
The disconnection
may be initiated from either side. In addition to cleaning up and closing
all the remote queues connecting the two sides, the disconnection would
trigger the execution of user definable procedures on both sides (through an
event on the {\eclipse} side, and a call-back on the Tcl side), which can
be used to perform extra application specific cleanup and shutdown
routines. 

For the transfer of control from Tcl to {\eclipse} and vice versa, 
user-definable call-backs are made. This is to enable to define application
specific restrictions on what the GUI is allowed to do when the {\eclipse}
side has the control (for example, the GUI may have a button that sends an
rpc goal to {\eclipse} when pressed. Such a button could be disabled by the
call-back when control is transferred to {\eclipse} and reenabled when
control is transferred back to Tcl).

Note that there are two types of transfer of control from {\eclipse} to
Tcl: 1) when the control is implicitly yielded (e.g.\ initiating I/O
from {\eclipse} with Tcl, or returning after an rpc call); 2) when the
control is handed over by yielding explicitly (e.g.\ by calling
\bipref{remote_yield/1}{../bips/kernel/externals/remote_yield-1.html} in {\eclipse}). With implicit yield, the Tcl side is
expected to eventually handed back control implicitly to {\eclipse}, and
not to explicitly hand control over to {\eclipse} before this. Thus two
call-backs are provided when control is yield to Tcl: one is executed
whenever the control is yielded, and the other is only executed when the
control is explicitly yielded. Thus when control is explicitly yielded,
both call-backs are executed. This can be useful for example by defining
the explicit yield call-back to enable a button on the Tcl side that will
explicitly transfer control back to {\eclipse} when pressed, which should
only be enabled when {\eclipse} explicitly yielded to Tcl.

On the {\eclipse} side, an event is raised when the two sides
disconnect. The event's name is the control stream's name. The user can
define a handler for this event to allow user-defined action to take place
on the {\eclipse} side on disconnection. The simplest way to define this
handler is to do it during the connection, via the last argument of \bipref{remote_connect/3}{../bips/kernel/externals/remote_connect-3.html}.

\subsubsection{Tcl side}

\begin{description}
\item[\index{ec_running (Tcl remote interface)}ec_running]\ \\
	checks if the {\eclipse} side has control. Returns 1 if {\eclipse}
	side has control, 0 otherwise. If that is the case,
	then the Tcl side cannot issue an ec_rpc goal. Note that ec_running
	will return 1 before connection and after disconnection.
\item[\index{ec_connected (Tcl remote interface)}ec_connected]\ \\
        checks if the Tcl side is currently attached to {\eclipse}. Returns
        1 if there is a connection to {\eclipse} (i.e.\ it is attached), 0
        otherwise. 
\item[\index{ec_resume (Tcl remote interface)}ec_resume]\ \\
	explicitly hand-over control to {\eclipse}. Tcl side must have
	control when this command is called (i.e. {\it ec_running\/} must
	be false). This command returns when {\eclipse} side yields the
	control back to the Tcl side. Meanwhile, the Tcl process is not
        suspended as the Tcl event loop is entered while waiting for the
	yield. 
\item[\index{ec_running_set_commands (Tcl remote interface)}ec_running_set_commands {\it ?start? ?end? ?yield? ?disconnect?}]\ \\
	set up commands that will be called just before control is handed
	over to {\eclipse} ({\it start}), when control is handed back from
	{\eclipse} ({\it end}), when {\eclipse} explicitly yields control
	({\it yield}), and when the Tcl side is
	disconnected by the {\eclipse} side ({\it
	disconnect}). The {\it start} and {\it end} commands are called
	both when control change hands explicitly (e.g.\ via {\it
	ec_resume}), or implicitly (e.g.\ by making an rpc call or
	performing I/O on a synchronous remote queue). An explicit yield
	from {\eclipse} will in addition call the {\it yield\/} command, {\it
	after\/} the {\it start\/} command is executed.

	The default for each command is that no command will be called.

\item[\index{ec_disconnect (Tcl remote interface)}ec_disconnect {\it ?side?}]\ \\
	disconnect the Tcl process from the {\eclipse} process. This closes
	all the connections between the two sides. The {\eclipse} side will
	abort from what it was doing. After disconnection, the
	two sides can no longer communicate, and {\it ec_running\/} will be
	set. The optional argument {\it side\/} specifies which side, tcl
	or eclipse,
	initiated the disconnection. For user's Tcl program, this will
	normally be the default tcl. If the disconnect is initiated from
	the Tcl side, this command will cause the {\eclipse} side to also
	close its connections to this remote connection, as well as
	raising the disconnect event in {\eclipse} associated with this
	remote connection. If the disconnect was initiated from
	the {\eclipse} side, then {\it ec_disconnect\/} will be called
	automatically with {\it side\/} set to eclipse, and the disconnect
	command set up by {\it ec_running_set_commands\/} will be executed.
\end{description}

\subsubsection{{\eclipse} side}

\begin{description}
\item[\index{remote_yield/1}remote_yield(+Control)]\ \\
	Explicitly yields control from {\eclipse} to the remote side with
	the control stream {\it	Control}. {\eclipse} execution will suspend
	until control is transferred back to {\eclipse}. This predicate
	returns when {\eclipse} side resumes control.
	
\item[\index{remote_disconnect/1}remote_disconnect(+Control)]\ \\
	Initiates disconnection from the remote side specified by {\it
	Control}. This will close all connections between {\eclipse} and
	the remote side, on both sides. It will also cause an event {\it
	Control} to be raised. 
\end{description}

Note that if the {\eclipse} process is halted normally, then {\eclipse}
will try to disconnect from every remote side it may be connected to.

\section{Example}


\begin{figure}[htb]
\begin{verbatim}
% Tcl side will create the from-ECLiPSe queue gui_output, which will also
% automatically flush at the end of every line (and so transfer the data
% to the Tcl side)


disconnect_handler :-  % just terminate ECLiPSe execution
        writeln("Terminating...."),
        halt.


:-      remote_connect(localhost/5000, control, 
        set_event_handler(control, disconnect_handler/0)),
        % for simplicity, the host and port are fixed.
        % Name for control connection is also fixed.
        % when remote_connect returns, the gui_output queue will be
        % connected to Tcl side already.
        % eclipse side initialisation follows is just to set up the 
        % handler for disconnection....
        writeln(gui_output, "Connected...").

% this is for yielding control to Tcl. No need to specify Control explicitly
tclyield :-
        remote_yield(control).
\end{verbatim}
\caption{Example use of interface: {\eclipse} code}
\label{remote-eclipse}
\end{figure}


\begin{figure}[htb]
{\small
\begin{verbatim}
# ECLIPSEDIR has to be set to where your ECLiPSe is at
lappend auto_path [file join $ECLIPSEDIR) lib_tcl]

package require remote_eclipse 

proc terminate {} {
    destroy .
}

# disable the terminate button when control is transferred to ECLiPSe
proc disable_button {} {
    .b configure -state disabled
}

# enable the terminate button when control is transferred back to Tcl
proc enable_button {} {
    .b configure -state normal
}

# the initialisation procedure, called when the remote side is attached. 
# this creates the gui_output from-ECLiPSe queue, and then uses an
# ec_rpc to cause the queue to flush at every newline
proc initialise {} {
	ec_queue_create gui_output fromec
	ec_rpc "set_stream_property(gui_output, flush, end_of_line)"
}

# this button initiates disconnection and terminates the Tcl program
pack [button .b -text "Terminate" -command "ec_disconnect; terminate"] 
# only start and end commands given; the others default to no commands
ec_running_set_commands disable_button enable_button

# disable button initially as ECLiPSe side has initial control
disable_button  

# the attachment to ECLiPSe includes the initialisation ec_init, which
# creates the gui_output queue. The connection is at the fixed port address
ec_remote_init localhost 5000 initialise
# Tcl side initially has control, hand it over to ECLiPSe...
ec_resume resume
\end{verbatim}}
\caption{Example use of interface: Tcl code}
\label{remote-tcl}
\end{figure}

Figures~\ref{remote-eclipse} and \ref{remote-tcl} shows a simple example of
the use of the interface. To try this program, the user should start an
{\eclipse}, compile the {\eclipse} program. This will suspend on the
\verb'remote_connect/3', waiting for the Tcl program to attach. The Tcl
program should then be started on the same machine, and the attachment will
be connected using the fixed host and port address. During the
initialisation when attaching, 
a from-{\eclipse} queue is created, which is set
to flush at every newline. This is done by an {\bf ec_rpc} goal after the
queue is created. As no Tcl data consumer handler is specified, the default
Tcl data consumer handler {\bf ec_stream_output_popup} handles and display the data on the Tcl side on a
pop-up window. The Tcl side
GUI has just one button which allows the application to terminate. This
button is disabled when {\eclipse} side has control, and this is done by
setting up the appropriate call-backs to disable and enable the button in
{\it ec_running_set_commands\/} on the Tcl side.

On disconnection, which can be initiated either by pressing the button on
the Tcl side, or by quitting from {\eclipse}, the handlers for
disconnection ensures that both the {\eclipse} and Tcl program terminates. 

\section{Differences From the Tcl Embedding Interface}
\label{remotediff}

The remote Tcl interface is designed to be largely compatible with the
embedded Tcl interface, so that a user GUI can be written that allows
either interfaces to be used, while sharing most of the code in both
{\eclipse} and Tcl. An example of this is the Tkeclipse development tools.

Some remote specific code would need to be written. This includes code to
handle the connection and disconnection of the Tcl and {\eclipse}
sides: there is no equivalent to ec_cleanup in the embedded Tcl interface,
as the termination of the {\eclipse} side should be handled in {\eclipse}. In
addition, the user may need to provide code to restrict the interaction
within the Tcl/Tk GUI when control is transferred to {\eclipse}. Aside from
this, the rest of the code should be reusable, if the user exercises some
care.

The supported compatible methods of communicating between Tcl and
{\eclipse} in the two interfaces is via the {\bf ec_rpc} calls and the use
of the I/O queues. The {\bf ec_rpc} mechanism should behave the same in
both interfaces. For the queues, there are some differences because the
queues are in-memory queues in the embedded interface, but are socket
channels in the remote interface. This leads to the following differences:

\begin{itemize}

\item Data will not appear on the other side until the queue is
      flushed on the side that is generating the output. To make code
      compatible, output queues should be always flushed.

\item In general, reading data via a blocking socket requires the
      size of the data to be explicitly specified, except when I/O is done
      via the EXDR primitives, where size is implicitly specified. Only
      EXDR format is supported when data is sent from Tcl to {\eclipse}.
      That is, a write handler for a Tcl write channel must write the data
      in EXDR format. For output from the {\eclipse} side, the Tcl read
      handler would be supplied with the length of the output when it is
      invoked, and this information must be used if the Tcl read is not
      done via an EXDR primitive. However, in general it is strongly
      suggested that only EXDR formatted data should be sent via the queues
      in both direction.

\item An I/O operation on the stream may block if there is no 
      handler to consume/produce the data on the other side. If a handler
      is specified via the {\it command} argument, then a corresponding
      handler in the Tcl side will be invoked at the correct place when the
      {\eclipse} side produces output or request input.

\item The  channel identifier is not of the form {\tt
      ec_queueX}. To make code portable, the name of the channel should be
      obtained from the {\eclipse} stream symbolic name or number via the
      commands {\bf ec_streamname_to_channel} or {\bf
      ec_streamnum_to_channel}.

\item Asynchronous queues are bi-directional. In the embedding interface,
      there are no asynchronous queues, and {\bf ec_async_queue_create} is
      aliased to {\bf ec_queue_create}, and a uni-directional queue is
      created. Thus for compatibility, these queues should only be used
      in one direction. 
\end{itemize}

In summary, to write code that will work for both the remote and embedded
interfaces, the data should be sent using EXDR format, flush always
performed, and a handler ({\it command\/} argument) provided. The Tcl
channel identifier should not be constructed explicitly.

However, there may be cases where the two interfaces need to be
distinguished. For example, if the Tcl side is to perform some operations
on the file system for the {\eclipse} side (e.g.\ selecting a file via a
GUI), then with the remote interface, the two sides might not have access
to the same file systems, and being able to distinguish whether the
interface is remote or embedded allows the user to provide code to handle
this.  

To obtain information on which interface is being used, use the command:

\begin{description}
\item[\index{ec_interface_type (Tcl remote interface)}ec_interface_type]\ \\
	returns remote for the remote interface, and embedded for the
        embedding interface.
\end{description}

%HEVEA\cutend