<chapter>
<title>Client-Server Mechanisms</title>
<para>
The &cpp; class <classname>EST_Server</classname> provides the core
mechanisms required for simple client-server applications. It is
currently used to implement the
<literal role=program>fringe_client</literal>
program and client-server mechanisms for SIOD. It is planned to
use it to re-implement the festival client-server mechanism.
</para>
<para>
Servers have types and names. When a server is started it
records it's type, name and location in a `services' file. When
a client wishes to connect to a server it looks for it's
location in that file by giving a name and type.
</para>
<para>
Once connected a client must present a magic cookie to the
server as a simple form of authentication. Once authenticated
the client sends requests, consisting of a package name,
operation name and a set of arguments, to the server. The server
responds with an error report or a sequence of values.
</para>
<para>
An instance of <classname>EST_Server</classname> embodies each
side of the client-server relationship. In the server an
instance of <classname>EST_Server</classname> is created and
told how to process requests from clients, a call to the
<function role=method>run()</function> method then starts the
server. In a client an instance of <classname>EST_Server</classname>
represents the server, and calls to the
<function role=method>execute(...)</function> method send
requests to the server.
</para>
<sect1>
<title>The Services Table</title>
<para>
The first problem which needs to be addressed by any
client-server system is how the client finds the
server. Servers based on <classname>EST_Server</classname>
handle this problem by writing a record into a file giving
their name, type and location. Clients can then look servers
up by namd and type.
</para>
<para>
By default the file <filename>.estServices</filename> is used
for this purpose, meaning that each user has their own list of
servers. An alternative file could be specified to record
public services.
</para>
<para>
The services table also provides a simple authorisation
mechanism. Each server records a random string in the table,
and clients must send this string before making any
requests. Thus people who can't read the services table can't
make requests of the server, and the file permissions on the
services table can be used to control access to the server.
<important>
<para>
This `magic cookie' authorisation scheme is not very
secure. The cookie is sent as plain text over the
network and so anyone who can snoop on the network can
break the security.
</para>
<para>
A more secure `challange-responce' authorisation scheme
should be implemented.
</para>
</important>
</para>
<para>
The in-file format of the services table is based on the
&java; properties file format. A typical file might look as
follows:
<programlisting>
#Services
fringe.type=fringe
fringe.host=foo.bar.com
fringe.cookie=511341634
fringe.port=56362
fringe.address=123.456.789.654
siod.type=siod
siod.cookie=492588950
siod.host=foo.bar.com
siod.address=123.456.789.654
siod.port=56382
labeling.type=fringe
labeling.host=foo.bar.com
labeling.cookie=511341634
labeling.port=56362
labeling.address=123.456.789.654
</programlisting>
This file lists three services, a
<literal role=program>fringe</literal> server with the default
name of <literal>fringe</literal>, a scheme interpreter
running as a server, also with the default name, and a second
<literal role=program>fringe</literal> server named
<literal>labeling</literal>.
</para>
<para>
The programing interface to the services table is provided by
the
<link linkend='est-servicetable'><classname>EST_ServiceTable</classname></link>
class.
</para>
&docpp-EST-ServiceTable-2;
</sect1>
<sect1>
<title>Writing Clients and Servers</title>
<para>
If a service type (that is a sub-class of
<classname>EST_Server</classname>) has already been defined
for the job you need to do, creating clients and servers is
quite straight forward. For this section I will use the
<classname>EST_SiodServer</classname> class, which defines a
simple &scheme; execution service service, as an example.
</para>
<sect2>
<title>A Simple Server</title>
<para>
To run a siod server we have to read the server table,
create the server object and update the table, then start the
service running.
</para>
<para>
First we read the default service table.
<programlisting arch='cpp'>
EST_ServiceTable::read();
</programlisting>
Now we create the new scheme service called "mySiod". The
<literal>EST_Server::sm_sequential</literal> parameter to the
server constructor tells the server todeal with one client
at a time. The <literal>NULL</literal> turns off trace
output, replace this with <literal>&cout</literal> to see
what the server is doing.
<programlisting arch='cpp'>
EST_SiodServer *server
= new EST_SiodServer(EST_Server::sm_sequential,
"mySiod",
NULL);
</programlisting>
Write the table back out so clients can find us.
<programlisting arch='cpp'>
EST_ServiceTable::write();
</programlisting>
Create the object which handles the client requests. The
<parameter>handler</parameter> object actually does the work
the client requests. <classname>EST_SiodServer</classname>
provides the obvious default handler (it executes the scheme
code and returns the results), so we use that.
<programlisting arch='cpp'>
EST_SiodServer::RequestHandler handler;
</programlisting>
Finally, start the service. This call never returns.
<programlisting arch='cpp'>
server->run(handler);
</programlisting>
</para>
</sect2>
<sect2 id='simple-client'>
<title>A Simple Client</title>
<para>
A client is created by reading the service table, and then
asking for a server by name. Again the
<literal>NULL</literal> means `no trace output'.
<programlisting arch='cpp'>
EST_ServiceTable::read();
EST_SiodServer *server
= new EST_SiodServer("mySiod", NULL);
</programlisting>
Now we have a representation of the server we must connect
before we can do anything. We can connect and dissconnect a
server object any number of times over it's life. This may
or may not have some meaning to the server. The return value
of the connect operation tells us if we managed to connect.
<programlisting arch='cpp'>
if (server->connect() != connect_ok)
EST_sys_error("Error Connecting");
</programlisting>
Once we are connected we can send requests to the
server. The siod server executes scheme for us, assume that
the function <literal>get_sexp()</literal> returns something
we want evaluated.
<programlisting arch='cpp'>
LISP expression = get_sexp();
</programlisting>
We pass arguments to requests in an <classname>Args</classname>
structure, a special type of
<classname>EST_Features</classname>. The sod server wants
the expression to execute as the value of
<literal>"sexp"</literal>.
<programlisting arch='cpp'>
EST_SiodServer::Args args;
args.set_val("sexp", est_val(expression));
</programlisting>
As in the server, the behaviour of the client is defined by
a `handler' object. The handler
<classname>EST_SiodServer</classname> defines for us does
nothing with the result, leaving it for us to deal with in
the <classname>EST_Features</classname> structure
<literal>handler.res</literal>. Again this is good enough
for us.
<programlisting arch='cpp'>
EST_SiodServer::ResultHandler handler;
</programlisting>
Finally we are ready to send the request to the server. The
siod server provides only one operation, called
<literal>"eval"</literal> in package
<literal>"scheme"</literal>, this is the evaluate-expression
operation we want. The return value of
<function>execute(...)</function> is true of everything goes
OK, false for an error. For an error the message is the
value of <literal>"ERROR"</literal>.
<programlisting arch='cpp'>
if (!server->execute("scheme", "eval", args, handler))
EST_error("error from siod server '%s'",
(const char *)handler.res.String("ERROR"));
</programlisting>
Now we can get the result of the evaluation, it is returned
as the value of <literal>"sexp"</literal>.
<programlisting arch='cpp'>
LISP result = scheme(handler.res.Val("sexp"));
</programlisting>
Although this may seem a lot of work just to evaluate one
expression, once a connection is established, only the three
steps set arguments, execute, extract results need to be
done for each request. So the following would be the code
for a single request:
<programlisting arch='cpp'>
args.set_val("sexp", est_val(expression));
if (!server->execute("scheme", "eval", args, handler))
[handle error]
LISP result = scheme(handler.res.Val("sexp"));
</programlisting>
</para>
</sect2>
<sect2>
<title>A Specialised Server</title>
<para>
If you need to create a server similar to an existing one
but which handles requests slightly differently, all you
need to do is define your own
<classname>RequestHandler</classname> class. This class has
a member function called
<function role=method>process()</function> which deos the
work.
</para>
<para>
Here is a varient on the siod server which handles a new
operation <literal>"print"</literal> which evaluates an
expression and prints the result to standard output as well
as retruning it. (In this example some details of error
catching and so on necessary for dealing with scheme are
omitted so as not to obscure the main points).
</para>
<para>
First we define the handler class. It is a sub-class of the
default handler for siod servers.
<programlisting arch='cpp'>
class MyRequestHandler : public EST_SiodServer::RequestHandler
{
public:
virtual EST_String process(void);
};
</programlisting>
Now, we define the processing method. For any operation
other than <literal>"print"</literal> we call the default
siod handler. (<function>leval</function> and
<function>lprint</function> are functions provided by the
siod interpreter).
<programlisting arch='cpp'>
EST_String MyRequestHandler::process(void)
{
if (operation == "print")
{
// Get the expression.
LISP sexp = scheme(args.Val("sexp"));
// Evaluate it.
LISP result = leval(sexp, current_env);
// Print it.
lprint(result);
// Return it.
res.set_val("sexp", est_val(result));
return "";
}
else
// Let the default handler deal with other operations.
return EST_SiodServer::RequestHandler::process();
}
</programlisting>
And now we can start a server which understands the new
operation.
<programlisting arch='cpp'>
MyRequestHandler handler;
server->run(handler);
</programlisting>
</para>
</sect2>
<sect2>
<title>A Client Which Handles Multiple Results</title>
<para>
Servers have the option to return more than one value for a
single request. This can be used to return the results of a
request a piece at a time as they become available, for
instance &festival; returns a waveform for each sentence in
a piece of text it is given to synthesise.
</para>
<para>
Clearly a simple client of the kind described
<link linkend='simple-client'>above</link> which gets the
result of a request as a result of the call to
<function role=method>execute(...)</function> can't handle
multiple results of this kind. This is what the handler
object is for.
</para>
<para>
I'll asuume we need a client to del with a varient on the
normal siod sever which returns multiple values, say it
evaluates the expression in each of a number of environments
and returns each result separately. I'll also assume that
the work to be done for each result is defined by the fucntion
<function>deal_with_result()</function>.
</para>
<para>
Most of the client will be the same as for
<link linkend='simple-client'>the simple client</link>,
the exception is that we use our own result handler rather
than the default one.
<programlisting arch='cpp'>
class MyResultHandler : public EST_SiodServer::ResultHandler
{
public:
virtual void process(void);
};
</programlisting>
As for the server's request handler, the beahviour of the
result handler is defeined by the
<function role=method>process()</function> method of the
handler.
<programlisting arch='cpp'>
EST_String MyResultHandler::process(void)
{
// Get the result.
LISP result = scheme(handler.res.Val("sexp"));
// And deal with it.
deal_with_result(result);
}
</programlisting>
With this definition in place we can make requests to the
server as follows.
<programlisting arch='cpp'>
MyResultHandler handler;
if (!server->execute("scheme", "multi-eval", args, handler))
[handle errors]
</programlisting>
The <function>deal_with_result()</function> function will be
called on each result which is returned. If anything special
needs to be done with the final value, it can be done after
the call to <function role=method>execute(...)</function>
as in the simple client example.
</para>
</sect2>
</sect1>
<sect1>
<title>Creating a new Service</title>
<para>
</para>
<sect2>
<title>Commands</title>
<para>
</para>
</sect2>
<sect2>
<title>Results</title>
<para>
</para>
</sect2>
</sect1>
<sect1>
<title>The Network Protocol</title>
<para>
</para>
</sect1>
</chapter>
<!--
Local Variables:
mode: sgml
sgml-doctype:"speechtools.sgml"
sgml-parent-document:("speechtools.sgml" "chapter" "book")
sgml-omittag:nil
sgml-shorttag:t
sgml-minimize-attributes:nil
sgml-always-quote-attributes:t
sgml-indent-step:2
sgml-indent-data:t
sgml-exposed-tags:nil
sgml-local-catalogs:nil
sgml-local-ecat-files:nil
End:
-->
