You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
tdemultimedia/doc/artsbuilder/mcop.docbook

2275 lines
61 KiB

<!-- <?xml version="1.0" ?>
<!DOCTYPE chapter PUBLIC "-//KDE//DTD DocBook XML V4.2-Based Variant V1.1//EN" "dtd/kdex.dtd">
To validate or process this file as a standalone document, uncomment
this prolog. Be sure to comment it out again when you are done -->
<chapter id="mcop">
<title>&MCOP;: Object Model and Streaming</title>
<sect1 id="mcop-overview">
<title>Overview</title>
<para>
&MCOP; is the standard &arts; uses for:
</para>
<itemizedlist>
<listitem>
<para>
Communication between objects.
</para>
</listitem>
<listitem>
<para>
Network transparency.
</para>
</listitem>
<listitem>
<para>
Describing object interfaces.
</para>
</listitem>
<listitem>
<para>
Language independancy.
</para>
</listitem>
</itemizedlist>
<para>
One major aspect of &MCOP; is the <emphasis>interface description
language</emphasis>, &IDL;, in which many of the &arts; interfaces and
<acronym>API</acronym>s are defined in a language independent way.
</para>
<para>
To use &IDL; interfaces from C++, is compiled by the &IDL;
compiler into C++ code. When you implement an interface, you derive from
the skeleton class the &IDL; compiler has generated. When you use an
interface, you do so using a wrapper. This way, &MCOP; can use a
protocol if the object you are talking to is not local - you get network
transparency.
</para>
<para>
This chapter is supposed to describe the basic features of the object
model that results from the use of &MCOP;, the protocol, how do use
&MCOP; in C++ (language binding), and so on.
</para>
</sect1>
<sect1 id="interfaces">
<title>Interfaces and &IDL;</title>
<para>
Many of the services provided by &arts;, such as modules and the sound
server, are defined in terms of <acronym>interfaces</acronym>.
Interfaces are specified in a programming language independent format:
&IDL;.
</para>
<para>
This allows many of the implementation details such as the format of
multimedia data streams, network transparency, and programming language
dependencies, to be hidden from the specification for the interface. A
tool, &mcopidl;, translates the interface
definition into a specific programming language (currently only C++ is
supported).
</para>
<para>
The tool generates a skeleton class with all of the boilerplate code and
base functionality. You derive from that class to implement the features
you want.
</para>
<para>
The &IDL; used by &arts; is similar to that used by
<acronym>CORBA</acronym> and <acronym>DCOM</acronym>.
</para>
<para>
&IDL; files can contain:
</para>
<itemizedlist>
<listitem>
<para>
C-style #include directives for other &IDL; files.
</para>
</listitem>
<listitem>
<para>
Definitions of enumerated and struct types, as in C/C++.
</para>
</listitem>
<listitem>
<para>
Definitions of interfaces.
</para>
</listitem>
</itemizedlist>
<para>
In &IDL;, interfaces are defined much like a C++ class or C struct,
albeit with some restrictions. Like C++, interfaces can subclass other
interfaces using inheritance. Interface definitions can include three
things: streams, attributes, and methods.
</para>
<sect2 id="streams">
<title>Streams</title>
<para>
Streams define multimedia data, one of the most important components of
a module. Streams are defined in the following format:
</para>
<para>
[ async ] in|out [ multi ] <replaceable>type</replaceable> stream <replaceable>name</replaceable> [ , <replaceable>name</replaceable> ] ;
</para>
<para>
Streams have a defined direction in reference to the module, as
indicated by the required qualifiers in or out. The type argument
defines the type of data, which can be any of the types described later
for attributes (not all are currently supported). Many modules use the
stream type audio, which is an alias for float since that is the
internal data format used for audio stream. Multiple streams of the same
type can defined in the same definition uisng comma separated names.
</para>
<para>
Streams are by default synchronous, which means they are continuous
flows of data at a constant rate, such as <acronym>PCM</acronym>
audio. The async qualifier specifies an asynchronous stream, which is
used for non-continuous data flows. The most common example of an async
stream is &MIDI; messages.
</para>
<para>
The multi keyword, only valid for input streams, indicates that the
interface supports a variable number of inputs. This is useful for
implementing devices such as mixers that can accept any number of input
streams.
</para>
</sect2>
<sect2 id="attributes">
<title>Attributes</title>
<para>
Attributes are data associated with an instance of an interface. They
are declared like member variables in C++, and can can use any of the
primitive types boolean, byte, long, string, or float. You can also use
user-defined struct or enum types as well as variable sized sequences
using the syntax sequence&lt;type&gt;. Attributes can optionally be
marked readonly.
</para>
</sect2>
<sect2 id="methods">
<title>Methods</title>
<para>
As in C++, methods can be defined in interfaces. The method parameters
are restricted to the same types as attributes. The keyword oneway
indicates a method which returns immediately and is executed
asynchronously.
</para>
</sect2>
<sect2 id="standardinterfaces">
<title>Standard Interfaces</title>
<para>
Several standard module interfaces are already defined for you in
&arts;, such as <interfacename>StereoEffect</interfacename>, and
<interfacename>SimpleSoundServer</interfacename>.
</para>
</sect2>
<sect2 id="example">
<title>Example</title>
<para>
A simple example of a module taken from &arts; is the constant delay
module, found in the file
<filename>tdemultimedia/arts/modules/artsmodules.idl</filename>. The
interface definition is listed below.
</para>
<programlisting>
interface Synth_CDELAY : SynthModule {
attribute float time;
in audio stream invalue;
out audio stream outvalue;
};
</programlisting>
<para>
This modules inherits from
<interfacename>SynthModule</interfacename>. That interface, defined in
<filename>artsflow.idl</filename>, defines the standard methods
implemented in all music synthesizer modules.
</para>
<para>
The CDELAY effect delays a stereo audio stream by the time value
specified as a floating point parameter. The interface definition has an
attribute of type float to store the delay value. It defines two input
audio streams and two output audio streams (typical of stereo
effects). No methods are required other than those it inherits.
</para>
</sect2>
</sect1>
<sect1 id="more-about-streams">
<title>More About Streams</title>
<para>
This section covers some additional topics related to streams.
</para>
<sect2 id="stream-types">
<title>Stream Types</title>
<para>
There are various requirements for how a module can do streaming. To
illustrate this, consider these examples:
</para>
<itemizedlist>
<listitem>
<para>
Scaling a signal by a factor of two.
</para>
</listitem>
<listitem>
<para>
Performing sample frequency conversion.
</para>
</listitem>
<listitem>
<para>
Decompressing a run-length encoded signal.
</para>
</listitem>
<listitem>
<para>
Reading &MIDI; events from <filename
class="devicefile">/dev/midi00</filename> and inserting them into a
stream.
</para>
</listitem>
</itemizedlist>
<para>
The first case is the simplest: upon receiving 200 samples of input the
module produces 200 samples of output. It only produces output when it
gets input.
</para>
<para>
The second case produces different numbers of output samples when given
200 input samples. It depends what conversion is performed, but the
number is known in advance.
</para>
<para>
The third case is even worse. From the outset you cannot even guess how
much data 200 input bytes will generate (probably a lot more than 200
bytes, but...).
</para>
<para>
The last case is a module which becomes active by itself, and sometimes
produces data.
</para>
<para>
In &arts;s-0.3.4, only streams of the first type were handled, and most
things worked nicely. This is probably what you need most when writing
modules that process audio. The problem with the other, more complex
types of streaming, is that they are hard to program, and that you don't
need the features most of the time. That is why we do this with two
different stream types: synchronous and asynchronous.
</para>
<para>
Synchronous streams have these characteristics:
</para>
<itemizedlist>
<listitem>
<para>
Modules must be able to calculate data of any length, given enough
input.
</para>
</listitem>
<listitem>
<para>
All streams have the same sampling rate.
</para>
</listitem>
<listitem>
<para>
The <function>calculateBlock()</function> function will be called when
enough data is available, and the module can rely on the pointers
pointing to data.
</para>
</listitem>
<listitem>
<para>
There is no allocation and deallocation to be done.
</para>
</listitem>
</itemizedlist>
<para>
Asynchronous streams, on the other hand, have this behavior:
</para>
<itemizedlist>
<listitem>
<para>
Modules may produce data sometimes, or with varying sampling rate, or
only if they have input from some filed descriptor. They are not bound by
the rule <quote>must be able to satisfy requests of any size</quote>.
</para>
</listitem>
<listitem>
<para>
Asynchronous streams of a module may have entirely different sampling
rates.
</para>
</listitem>
<listitem>
<para>
Outgoing streams: there are explicit functions to allocate packets, to
send packets - and an optional polling mechanism that will tell you when
you should create some more data.
</para>
</listitem>
<listitem>
<para>
Incoming streams: you get a call when you receive a new packet - you
have to say when you are through with processing all data of that
packet, which must not happen at once (you can say that anytime later,
and if everybody has processed a packet, it will be freed/reused)
</para>
</listitem>
</itemizedlist>
<para>
When you declare streams, you use the keyword <quote>async</quote> to
indicate you want to make an asynchronous stream. So, for instance,
assume you want to convert an asynchronous stream of bytes into a
synchronous stream of samples. Your interface could look like this:
</para>
<programlisting>
interface ByteStreamToAudio : SynthModule {
async in byte stream indata; // the asynchronous input sample stream
out audio stream left,right; // the synchronous output sample streams
};
</programlisting>
</sect2>
<sect2 id="async-streams">
<title>Using Asynchronous Streams</title>
<para>
Suppose you decided to write a module to produce sound
asynchronously. Its interface could look like this:
</para>
<programlisting>
interface SomeModule : SynthModule
{
async out byte stream outdata;
};
</programlisting>
<para>
How do you send the data? The first method is called <quote>push
delivery</quote>. With asynchronous streams you send the data as
packets. That means you send individual packets with bytes as in the
above example. The actual process is: allocate a packet, fill it, send
it.
</para>
<para>
Here it is in terms of code. First we allocate a packet:
</para>
<programlisting>
DataPacket&lt;mcopbyte&gt; *packet = outdata.allocPacket(100);
</programlisting>
<para>
The we fill it:
</para>
<programlisting>
// cast so that fgets is happy that it has a (char *) pointer
char *data = (char *)packet-&gt;contents;
// as you can see, you can shrink the packet size after allocation
// if you like
if(fgets(data,100,stdin))
packet-&gt;size = strlen(data);
else
packet-&gt;size = 0;
</programlisting>
<para>
Now we send it:
</para>
<programlisting>
packet-&gt;send();
</programlisting>
<para>
This is quite simple, but if we want to send packets exactly as fast as
the receiver can process them, we need another approach, the <quote>pull
delivery</quote> method. You ask to send packets as fast as the receiver
is ready to process them. You start with a certain amount of packets you
send. As the receiver processes one packet after another, you start
refilling them with fresh data, and send them again.
</para>
<para>
You start that by calling setPull. For example:
</para>
<programlisting>
outdata.setPull(8, 1024);
</programlisting>
<para>
This means that you want to send packets over outdata. You want to start
sending 8 packets at once, and as the receiver processes some of them,
you want to refill them.
</para>
<para>
Then, you need to implement a method which fills the packets, which could
look like this:
</para>
<programlisting>
void request_outdata(DataPacket&lt;mcopbyte&gt; *packet)
{
packet-&gt;size = 1024; // shouldn't be more than 1024
for(int i = 0;i &lt; 1024; i++)
packet-&gt;contents[i] = (mcopbyte)'A';
packet-&gt;send();
}
</programlisting>
<para>
Thats it. When you don't have any data any more, you can start sending
packets with zero size, which will stop the pulling.
</para>
<para>
Note that it is essential to give the method the exact name
<methodname>request_<replaceable>streamname</replaceable></methodname>.
</para>
<para>
We just discussed sending data. Receiving data is much much
simpler. Suppose you have a simple ToLower filter, which simply converts
all letters in lowercase:
</para>
<programlisting>
interface ToLower {
async in byte stream indata;
async out byte stream outdata;
};
</programlisting>
<para>
This is really simple to implement; here is the whole implementation:
</para>
<programlisting>
class ToLower_impl : public ToLower_skel {
public:
void process_indata(DataPacket&lt;mcopbyte&gt; *inpacket)
{
DataPacket&lt;mcopbyte&gt; *outpacket = outdata.allocPacket(inpacket-&gt;size);
// convert to lowercase letters
char *instring = (char *)inpacket-&gt;contents;
char *outstring = (char *)outpacket-&gt;contents;
for(int i=0;i&lt;inpacket-&gt;size;i++)
outstring[i] = tolower(instring[i]);
inpacket-&gt;processed();
outpacket-&gt;send();
}
};
REGISTER_IMPLEMENTATION(ToLower_impl);
</programlisting>
<para>
Again, it is essential to name the method
<methodname>process_<replaceable>streamname</replaceable></methodname>.
</para>
<para>
As you see, for each arriving packet you get a call for a function (the
<function>process_indata</function> call in our case). You need to call
the <methodname>processed()</methodname> method of a packet to indicate
you have processed it.
</para>
<para>
Here is an implementation tip: if processing takes longer (&ie; if you
need to wait for soundcard output or something like that), don't call
processed immediately, but store the whole data packet and call
processed only as soon as you really processed that packet. That way,
senders have a chance to know how long it really takes to do your work.
</para>
<para>
As synchronization isn't so nice with asynchronous streams, you should
use synchronous streams wherever possible, and asynchronous streams only
when necessary.
</para>
</sect2>
<sect2 id="default-streams">
<title>Default Streams</title>
<para>
Suppose you have 2 objects, for example an AudioProducer and an
AudioConsumer. The AudioProducer has an output stream and AudioConsumer
has an input one. Each time you want to connect them, you will use those
2 streams. The first use of defaulting is to enable you to make the
connection without specifying the ports in that case.
</para>
<para>
Now suppose the teo objects above can handle stereo, and each have a
<quote>left</quote> and <quote>right</quote> port. You'd still like to
connect them as easily as before. But how can the connecting system
know which output port to connect to which input port? It has no way to
correctly map the streams. Defaulting is then used to specify several
streams, with an order. Thus, when you connect an object with 2 default
output streams to another one with 2 default input streams, you don't
have to specify the ports, and the mapping will be done correctly.
</para>
<para>
Of course, this is not limited to stereo. Any number of streams can be
made default if needed, and the connect function will check that the
number of defaults for 2 object match (in the required direction) if you
don't specify the ports to use.
</para>
<para>
The syntax is as follows: in the &IDL;, you can use the default keyword
in the stream declaration, or on a single line. For example:
</para>
<programlisting>
interface TwoToOneMixer {
default in audio stream input1, input2;
out audio stream output;
};
</programlisting>
<para>
In this example, the object will expect its two input ports to be
connected by default. The order is the one specified on the default
line, so an object like this one:
</para>
<programlisting>
interface DualNoiseGenerator {
out audio stream bzzt, couic;
default couic, bzzt;
};
</programlisting>
<para>
Will make connections from <quote>couic</quote> to
<quote>input1</quote>, and <quote>bzzt</quote> to <quote>input2</quote>
automatically. Note that since there is only one output for the mixer,
it will be made default in this case (see below). The syntax used in the
noise generator is useful to declare a different order than the
declaration, or selecting only a few ports as default. The directions of
the ports on this line will be looked up by &mcopidl;, so don't specify
them. You can even mix input and output ports in such a line, only the
order is important.
</para>
<para>
There are some rules that are followed when using inheritance:
</para>
<itemizedlist>
<listitem>
<para>
If a default list is specified in the &IDL;, then use
it. Parent ports can be put in this list as well, whether they were
default in the parent or not.
</para>
</listitem>
<listitem>
<para>
Otherwise, inherit parent's defaults. Ordering is parent1 default1,
parent1 default2..., parent2 default1... If there is a common ancestor
using 2 parent branches, a <quote>virtual public</quote>-like merging is
done at that default's first occurrence in the list.
</para>
</listitem>
<listitem>
<para>
If there is still no default and a single stream in a
direction, use it as default for that direction.
</para>
</listitem>
</itemizedlist>
</sect2>
</sect1>
<sect1 id="attribute-change-notify">
<title>Attribute change notifications</title>
<!-- TODO: This should be embedded better into the context - I mean: the
context should be written ;-). -->
<para>
Attribute change notifications are a way to know when an attribute
changed. They are a bit comparable with &Qt;'s or Gtk's signals and
slots. For instance, if you have a &GUI; element, a slider, which
configures a number between 0 and 100, you will usually have an object
that does something with that number (for instance, it might be
controlling the volume of some audio signal). So you would like that
whenever the slider is moved, the object which scales the volume gets
notified. A connection between a sender and a receiver.
</para>
<para>
&MCOP; deals with that by being able to providing notifications when
attributes change. Whatever is declared as <quote>attribute</quote> in
the &IDL;, can emit such change notifications, and should do so,
whenever it is modified. Whatever is declared as
<quote>attribute</quote> can also receive such change notifications. So
for instance if you had two &IDL; interfaces, like these:
</para>
<programlisting>
interface Slider {
attribute long min,max;
attribute long position;
};
interface VolumeControl : Arts::StereoEffect {
attribute long volume; // 0..100
};
</programlisting>
<para>
You can connect them using change notifications. It works using the
normal flowsystem connect operation. In this case, the C++ code to
connect two objects would look like this:
</para>
<programlisting>
#include &lt;connect.h&gt;
using namespace Arts;
[...]
connect(slider,"position_changed",volumeControl,"volume");
</programlisting>
<para>
As you see, each attribute offers two different streams, one for sending
the change notifications, called
<function><replaceable>attributename</replaceable>_changed</function>,
<!-- TODO - how do I markup code that is an example - you wouldn't write
attributename in the source, but the name of some attribute
LW: I'm guessing
here, because I know how to markup QT code, but your stuff is different.
Hopefully this will give you inspiration, and we can work out later the fine
tuning if I have it wrong. The line above in the code sample, if it were qt
stuff, I would mark up this way (linebreaks for clarity of markup only, yes I
know it's incorrect!):
<function>connect(<classname>slider</classname>,
<function><replaceable>position</replaceable>_changed</function>,
<classname>volumeControl</classname>,
<function>volume</function>);</function>
You can use <function><replaceable>attributename</function> and even
<function><replaceable>attributename</replaceable>_changed</function>.
If I have the above totally wrong (which is entirely possible!) Some other
elements you might find handy:
<varname>, <type>, <returnvalue>, <constant>, <methodname>
There's also a markup guide at http://madmax.atconnex.net/kde/ that might
help, although unfortunately the programming section is still incomplete. -->
and one for receiving change notifications, called
<function>attributename</function>.
</para>
<para>
It is important to know that change notifications and asynchronous
streams are compatible. They are also network transparent. So you can
connect a change notification of a float attribute of a &GUI; widget has
to an asynchronous stream of a synthesis module running on another
computer. This of course also implies that change notifications are
<emphasis>not synchronous</emphasis>, this means, that after you have
sent the change notification, it may take some time until it really gets
received.
</para>
<sect2 id="sending-change-notifications">
<title>Sending change notifications</title>
<para>
When implementing objects that have attributes, you need to send change
notifications whereever an attribute changes. The code for doing this
looks like this:
</para>
<programlisting>
void KPoti_impl::value(float newValue)
{
if(newValue != _value)
{
_value = newValue;
value_changed(newValue); // &lt;- send change notification
}
}
</programlisting>
<para>
It is strongly recommended to use code like this for all objects you
implement, so that change notifications can be used by other people. You
should however void sending notifications too often, so if you are doing
signal processing, it is probably the best if you keep track when you
sent your last notification, so that you don't send one with every
sample you process.
</para>
</sect2>
<sect2 id="change-notifications-apps">
<title>Applications for change notifications</title>
<para>
It will be especially useful to use change notifications in conjunction
with scopes (things that visualize audio data for instance), gui
elements, control widgets, and monitoring. Code using this is in
<filename class="directory">tdelibs/arts/tests</filename>, and in the
experimental artsgui implementation, which you can find under <filename
class="directory">tdemultimedia/arts/gui</filename>.
</para>
<!-- TODO: can I markup links into the source code - if yes, how? -->
<!-- LW: Linking into the source is problematic - we can't assume people are
reading this on a machine with the sources available, or that they aren't
reading it from a website. We're working on it! -->
</sect2>
</sect1>
<sect1 id="the-mcoprc-file">
<title>The <literal role="extension">.mcoprc</literal> file</title>
<para>
The <literal role="extension">.mcoprc</literal> file (in each user's
home folder) can be used to configure &MCOP; in some ways. Currently,
the following is possible:
</para>
<variablelist>
<varlistentry>
<term>GlobalComm</term>
<listitem>
<para>
The name of an interface to be used for global communication. Global
communication is used to find other objects and obtain the secret
cookie. Multiple &MCOP; clients/servers that should be able to talk to
each other need to have a GlobalComm object which is able to share
information between them. Currently, the possible values are
<quote>Arts::TmpGlobalComm</quote> to communicate via <filename
class="directory">/tmp/mcop-<replaceable>username</replaceable></filename>
folder (which will only work on the local computer) and
<quote>Arts::X11GlobalComm</quote> to communicate via the root window
properties on the X11 server.
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>TraderPath</term>
<listitem>
<para>
Specifies where to look for trader information. You can list more than
one folder here, and separate them with commas, like
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>ExtensionPath</term>
<listitem>
<para>
Specifies from which folders extensions (in the form of shared
libraries) are loaded. Multiple values can be specified comma separated.
</para>
</listitem>
</varlistentry>
</variablelist>
<para>
An example which uses all of the above is:
</para>
<programlisting>
# $HOME/.mcoprc file
GlobalComm=Arts::X11GlobalComm
# if you are a developer, it might be handy to add a folder in your home
# to the trader/extension path to be able to add components without
# installing them
TraderPath="/opt/kde2/lib/mcop","/home/joe/mcopdevel/mcop"
ExtensionPath="/opt/kde2/lib","/home/joe/mcopdevel/lib"
</programlisting>
</sect1>
<sect1 id="mcop-for-corba-users">
<title>&MCOP; for <acronym>CORBA</acronym> Users</title>
<para>
If you have used <acronym>CORBA</acronym> before, you will see that
&MCOP; is much the same thing. In fact, &arts; prior to version 0.4 used
<acronym>CORBA</acronym>.
</para>
<para>
The basic idea of <acronym>CORBA</acronym> is the same: you implement
objects (components). By using the &MCOP; features, your objects are not
only available as normal classes from the same process (via standard C++
techniques) - they also are available to remote servers
transparently. For this to work, the first thing you need to do is to
specify the interface of your objects in an &IDL; file - just like
<acronym>CORBA</acronym> &IDL;. There are only a few differences.
</para>
<sect2 id="corba-missing">
<title><acronym>CORBA</acronym> Features That Are Missing In
&MCOP;</title>
<para>
In &MCOP; there are no <quote>in</quote> and <quote>out</quote>
parameters on method invocations. Parameters are always incoming, the
return code is always outgoing, which means that the interface:
</para>
<programlisting>
// CORBA idl
interface Account {
void deposit( in long amount );
void withdraw( in long amount );
long balance();
};
</programlisting>
<para>
is written as
</para>
<programlisting>
// MCOP idl
interface Account {
void deposit( long amount );
void withdraw( long amount );
long balance();
};
</programlisting>
<para>
in &MCOP;.
</para>
<para>
There is no exception support. &MCOP; doesn't have exceptions - it uses
something else for error handling.
</para>
<para>
There are no union types and no typedefs. I don't know if that is a real
weakness, something one would desperately need to survive.
</para>
<para>
There is no support for passing interfaces or object references
</para>
</sect2>
<sect2 id="corba-different">
<title><acronym>CORBA</acronym> Features That Are Different In
&MCOP;</title>
<para>
You declare sequences as
<quote>sequence<replaceable>type</replaceable></quote> in &MCOP;. There
is no need for a typedef. For example, instead of:
</para>
<programlisting>
// CORBA idl
struct Line {
long x1,y1,x2,y2;
};
typedef sequence&lt;Line&gt; LineSeq;
interface Plotter {
void draw(in LineSeq lines);
};
</programlisting>
<para>
you would write
</para>
<programlisting>
// MCOP idl
struct Line {
long x1,y1,x2,y2;
};
interface Plotter {
void draw(sequence&lt;Line&gt; lines);
};
</programlisting>
</sect2>
<sect2 id="no-in-corba">
<title>&MCOP; Features That Are Not In <acronym>CORBA</acronym></title>
<para>
You can declare streams, which will then be evaluated by the &arts;
framework. Streams are declared in a similar manner to attributes. For
example:
</para>
<programlisting>
// MCOP idl
interface Synth_ADD : SynthModule {
in audio stream signal1,signal2;
out audio stream outvalue;
};
</programlisting>
<para>
This says that your object will accept two incoming synchronous audio
streams called signal1 and signal2. Synchronous means that these are
streams that deliver x samples per second (or other time), so that the
scheduler will guarantee to always provide you a balanced amount of
input data (&eg; 200 samples of signal1 are there and 200 samples
signal2 are there). You guarantee that if your object is called with
those 200 samples signal1 + signal2, it is able to produce exactly 200
samples to outvalue.
</para>
</sect2>
<sect2 id="mcop-binding">
<title>The &MCOP; C++ Language Binding</title>
<para>
This differs from <acronym>CORBA</acronym> mostly:
</para>
<itemizedlist>
<listitem>
<para>
Strings use the C++ <acronym>STL</acronym> <classname>string</classname>
class. When stored in sequences, they are stored <quote>plain</quote>,
that means they are considered to be a primitive type. Thus, they need
copying.
</para>
</listitem>
<listitem>
<para>
longs are plain long's (expected to be 32 bit).
</para>
</listitem>
<listitem>
<para>
Sequences use the C++ <acronym>STL</acronym>
<classname>vector</classname> class.
</para>
</listitem>
<listitem>
<para>
Structures are all derived from the &MCOP; class
<classname>Type</classname>, and generated by the &MCOP; &IDL;
compiler. When stored in sequences, they are not stored
<quote>plain</quote> , but as pointers, as otherwise, too much copying
would occur.
</para>
</listitem>
</itemizedlist>
</sect2>
<sect2 id="implementing-objects">
<title>Implementing &MCOP; Objects</title>
<para>
After having them passed through the &IDL; compiler, you need to derive
from the <classname>_skel</classname> class. For instance, consider you
have defined your interface like this:
</para>
<programlisting>
// MCOP idl: hello.idl
interface Hello {
void hello(string s);
string concat(string s1, string s2);
long sum2(long a, long b);
};
</programlisting>
<para>
You pass that through the &IDL; compiler by calling
<userinput><command>mcopidl</command>
<parameter>hello.idl</parameter></userinput>, which will in turn generate
<filename>hello.cpp</filename> and <filename>hello.h</filename>. To
implement it, you need to define a C++-class that inherits the skeleton:
</para>
<programlisting>
// C++ header file - include hello.h somewhere
class Hello_impl : virtual public Hello_skel {
public:
void hello(const string&amp; s);
string concat(const string&amp; s1, const string&amp; s2);
long sum2(long a, long b);
};
</programlisting>
<para>
Finally, you need to implement the methods as normal C++
</para>
<programlisting>
// C++ implementation file
// as you see string's are passed as const string references
void Hello_impl::hello(const string&amp; s)
{
printf("Hello '%s'!\n",s.c_str());
}
// when they are a returncode they are passed as "normal" strings
string Hello_impl::concat(const string&amp; s1, const string&amp; s2)
{
return s1+s2;
}
long Hello_impl::sum2(long a, long b)
{
return a+b;
}
</programlisting>
<para>
Once you do that, you have an object which can communicate using &MCOP;.
Just create one (using the normal C++ facilities to create an object):
</para>
<programlisting>
Hello_impl server;
</programlisting>
<para>
And as soon as you give somebody the reference
</para>
<programlisting>
string reference = server._toString();
printf("%s\n",reference.c_str());
</programlisting>
<para>
and go to the &MCOP; idle loop
</para>
<programlisting>
Dispatcher::the()-&gt;run();
</programlisting>
<para>
People can access the thing using
</para>
<programlisting>
// this code can run anywhere - not necessarily in the same process
// (it may also run on a different computer/architecture)
Hello *h = Hello::_fromString([the object reference printed above]);
</programlisting>
<para>
and invoke methods:
</para>
<programlisting>
if(h)
h-&gt;hello("test");
else
printf("Access failed?\n");
</programlisting>
</sect2>
</sect1>
<sect1 id="mcop-security">
<title>&MCOP; Security Considerations</title>
<para>
Since &MCOP; servers will listen on a <acronym>TCP</acronym> port,
potentially everybody (if you are on the Internet) may try to connect
&MCOP; services. Thus, it is important to authenticate clients. &MCOP;
uses the md5-auth protocol.
</para>
<para>
The md5-auth protocol does the following to ensure that only selected
(trusted) clients may connect to a server:
</para>
<itemizedlist>
<listitem>
<para>
It assumes you can give every client a secret cookie.
</para>
</listitem>
<listitem>
<para>
Every time a client connects, it verifies that this client knows that
secret cookie, without actually transferring it (not even in a form that
somebody listening to the network traffic could find it out).
</para>
</listitem>
</itemizedlist>
<para>
To give each client the secret cookie, &MCOP; will (normally) put it in
the <filename class="directory">mcop</filename> folder (under
<filename
class="directory">/tmp/mcop-<envar>USER</envar>/secret-cookie</filename>). Of
course, you can copy it to other computers. However, if you do so, use a
secure transfer mechanism, such as <command>scp</command> (from
<application>ssh</application>).
</para>
<para>
The authentication of clients uses the following steps:
</para>
<procedure>
<step>
<para>
[SERVER] generate a new (random) cookie R
</para>
</step>
<step>
<para>
[SERVER] send it to the client
</para>
</step>
<step>
<para>
[CLIENT] read the "secret cookie" S from a file
</para>
</step>
<step>
<para>
[CLIENT] mangle the cookies R and S to a mangled cookie M using the MD5
algorithm
</para>
</step>
<step>
<para>
[CLIENT] send M to the server
</para>
</step>
<step>
<para>
[SERVER] verify that mangling R and S gives just the
same thing as the cookie M received from the client. If yes,
authentication is successful.
</para>
</step>
</procedure>
<para>
This algorithm should be secure, given that
</para>
<orderedlist>
<listitem>
<para>
The secret cookies and random cookies are <quote>random enough</quote>
and
</para>
</listitem>
<listitem>
<para>
The MD5 hashing algorithm doesn't allow to find out the
<quote>original text</quote>, that is the secret cookie S and the random
cookie R (which is known, anyway), from the mangled cookie M.
</para>
</listitem>
</orderedlist>
<para>
The &MCOP; protocol will start every new connection with an
authentication process. Basically, it looks like this:
</para>
<procedure>
<step>
<para>
Server sends a ServerHello message, which describes
the known authentication protocols.
</para>
</step>
<step>
<para>
Client sends a ClientHello message, which includes authentication info.
</para>
</step>
<step>
<para>
Server sends an AuthAccept message.
</para>
</step>
</procedure>
<para>
To see that the security actually works, we should look at how messages
are processed on unauthenticated connections:
</para>
<itemizedlist>
<listitem>
<para>
Before the authentication succeeds, the server will not receive other
messages from the connection. Instead, if the server for instance
expects a <quote>ClientHello</quote> message, and gets an mcopInvocation
message, it will drop the connection.
</para>
</listitem>
<listitem>
<para>
If the client doesn't send a valid &MCOP; message at all (no &MCOP;
magic in the message header) in the authentication phase, but something
else, the connection is dropped.
</para>
</listitem>
<listitem>
<para>
If the client tries to send a very very large message (&gt; 4096 bytes
in the authentication phase, the message size is truncated to 0 bytes,
which will cause that it isn't accepted for authentication) This is to
prevent unauthenticated clients from sending &eg; 100 megabytes of
message, which would be received and could cause the server to run out
of memory.
</para>
</listitem>
<listitem>
<para>
If the client sends a corrupt ClientHello message (one, for which
demarshalling fails), the connection is dropped.
</para>
</listitem>
<listitem>
<para>
If the client send nothing at all, then a timeout should occur (to be
implemented).
</para>
</listitem>
</itemizedlist>
</sect1>
<sect1 id="mcop-protocol">
<title>&MCOP; Protocol Specification</title>
<sect2 id="mcop-protocol-intro">
<title>Introduction</title>
<para>
It has conceptual similarities to <acronym>CORBA</acronym>, but it is
intended to extend it in all ways that are required for real time
multimedia operations.
</para>
<para>
It provides a multimedia object model, which can be used for both:
communication between components in one address space (one process), and
between components that are in different threads, processes or on
different hosts.
</para>
<para>
All in all, it will be designed for extremely high performance (so
everything shall be optimized to be blazingly fast), suitable for very
communicative multimedia applications. For instance streaming videos
around is one of the applications of &MCOP;, where most
<acronym>CORBA</acronym> implementations would go down to their knees.
</para>
<para>
The interface definitions can handle the following natively:
</para>
<itemizedlist>
<listitem>
<para>
Continuous streams of data (such as audio data).
</para>
</listitem>
<listitem>
<para>
Event streams of data (such as &MIDI; events).
</para>
</listitem>
<listitem>
<para>
Real reference counting.
</para>
</listitem>
</itemizedlist>
<para>
and the most important <acronym>CORBA</acronym> gimmicks, like
</para>
<itemizedlist>
<listitem>
<para>
Synchronous method invocations.
</para>
</listitem>
<listitem>
<para>
Asynchronous method invocations.
</para>
</listitem>
<listitem>
<para>
Constructing user defined data types.
</para>
</listitem>
<listitem>
<para>
Multiple inheritance.
</para>
</listitem>
<listitem>
<para>
Passing object references.
</para>
</listitem>
</itemizedlist>
</sect2>
<sect2 id="mcop-protocol-marshalling">
<title>The &MCOP; Message Marshalling</title>
<para>
Design goals/ideas:
</para>
<itemizedlist>
<listitem>
<para>
Marshalling should be easy to implement.
</para>
</listitem>
<listitem>
<para>
Demarshalling requires the receiver to know what type they want to
demarshall.
</para>
</listitem>
<listitem>
<para>
The receiver is expected to use every information - so skipping is only
in the protocol to a degree that:
</para>
<itemizedlist>
<listitem>
<para>
If you know you are going to receive a block of bytes, you don't need to
look at each byte for an end marker.
</para>
</listitem>
<listitem>
<para>
If you know you are going to receive a string, you don't need to read it
until the zero byte to find out it's length while demarshalling, however,
</para>
</listitem>
<listitem>
<para>
If you know you are going to receive a sequence of strings, you need to
look at the length of each of them to find the end of the sequence, as
strings have variable length. But if you use the strings for something
useful, you'll need to do that anyway, so this is no loss.
</para>
</listitem>
</itemizedlist>
</listitem>
<listitem>
<para>
As little overhead as possible.
</para>
</listitem>
</itemizedlist>
<!-- TODO: Make this a table -->
<para>
Marshalling of the different types is show in the table below:
</para>
<informaltable>
<tgroup cols="3">
<thead>
<row>
<entry>Type</entry>
<entry>Marshalling Process</entry>
<entry>Result</entry>
</row>
</thead>
<tbody>
<row>
<entry><type>void</type></entry>
<entry><type>void</type> types are marshalled by omitting them, so
nothing is written to the stream for them.</entry>
<entry></entry>
</row>
<row>
<entry><type>long</type></entry>
<entry>is marshalled as four bytes, the most significant byte first,
so the number 10001025 (which is 0x989a81) would be marshalled
as:</entry>
<entry><literal>0x00 0x98 0x9a 0x81</literal></entry>
</row>
<row>
<entry><type>enums</type></entry>
<entry><para>are marshalled like <type>long</type>s</para></entry>
<entry></entry>
</row>
<row>
<entry><type>byte</type></entry>
<entry><para>is marshalled as a single byte, so the byte 0x42 would be
marshalled as:</para></entry>
<entry><literal>0x42</literal></entry>
</row>
<row>
<entry><type>string</type></entry>
<entry><para>is marshalled as a <type>long</type>, containing the length
of the following string, and then the sequence of characters strings
must end with one zero byte (which is included in the length
counting).</para>
<important>
<para>include the trailing 0 byte in length counting!</para>
</important>
<para><quote>hello</quote> would be marshalled as:</para></entry>
<entry><literal>0x00 0x00 0x00 0x06 0x68 0x65 0x6c 0x6c 0x6f 0x00</literal></entry>
</row>
<row>
<entry><type>boolean</type></entry>
<entry><para>is marshalled as a byte, containing 0 if
<returnvalue>false</returnvalue> or 1 if
<returnvalue>true</returnvalue>, so the boolean value
<returnvalue>true</returnvalue> is marshalled as:</para></entry>
<entry><literal>0x01</literal></entry>
</row>
<row>
<entry><type>float</type></entry>
<entry><para>is marshalled after the four byte IEEE754 representation -
detailed docs how IEEE works are here: <ulink
url="http://twister.ou.edu/workshop.docs/common-tools/numerical_comp_guide/ncg_math.doc.html">http://twister.ou.edu/workshop.docs/common-tools/numerical_comp_guide/ncg_math.doc.html</ulink>
and here: <ulink
url="http://java.sun.com/docs/books/vmspec/2nd-edition/html/Overview.doc.html">http://java.sun.com/docs/books/vmspec/2nd-edition/html/Overview.doc.html</ulink>.
So, the value 2.15 would be marshalled as:</para></entry>
<entry><literal>0x9a 0x99 0x09 0x40</literal></entry>
</row>
<row>
<entry><type>struct</type></entry>
<entry><para>A structure is marshalled by marshalling it's
contents. There are no additional prefixes or suffixes required, so the
structure
</para>
<programlisting>
struct test {
string name; // which is "hello"
long value; // which is 10001025 (0x989a81)
};
</programlisting>
<para>would be marshalled as</para></entry>
<entry>
<literallayout>
0x00 0x00 0x00 0x06 0x68 0x65 0x6c 0x6c
0x6f 0x00 0x00 0x98 0x9a 0x81
</literallayout></entry>
</row>
<row>
<entry><type>sequence</type></entry>
<entry><para>a sequence is marshalled by listing the number of elements
that follow, and then marshalling the elements one by one.</para>
<para>So a sequence of 3 longs a, with a[0] = 0x12345678, a[1] = 0x01
and a[2] = 0x42 would be marshalled as:</para></entry>
<entry>
<literallayout>
0x00 0x00 0x00 0x03 0x12 0x34 0x56 0x78
0x00 0x00 0x00 0x01 0x00 0x00 0x00 0x42
</literallayout>
</entry>
</row>
</tbody>
</tgroup>
</informaltable>
<para>
If you need to refer to a type, all primitive types are referred by the
names given above. Structures and enums get own names (like
Header). Sequences are referred as *<replaceable>normal
type</replaceable>, so that a sequence of longs is <quote>*long</quote>
and a sequence of Header struct's is <quote>*Header</quote>.
</para>
</sect2>
<sect2 id="mcop-protocol-messages">
<title>Messages</title>
<para>
The &MCOP; message header format is defined as defined by this
structure:
</para>
<programlisting>
struct Header {
long magic; // the value 0x4d434f50, which is marshalled as MCOP
long messageLength;
long messageType;
};
</programlisting>
<para>
The possible messageTypes are currently
</para>
<programlisting>
mcopServerHello = 1
mcopClientHello = 2
mcopAuthAccept = 3
mcopInvocation = 4
mcopReturn = 5
mcopOnewayInvocation = 6
</programlisting>
<para>
A few notes about the &MCOP; messaging:
</para>
<itemizedlist>
<listitem>
<para>
Every message starts with a Header.
</para>
</listitem>
<listitem>
<para>
Some messages types should be dropped by the server, as long as the
authentication is not complete.
</para>
</listitem>
<listitem>
<para>
After receiving the header, the protocol (connection) handling can
receive the message completely, without looking at the contents.
</para>
</listitem>
</itemizedlist>
<para>
The messageLength in the header is of course in some cases redundant,
which means that this approach is not minimal regarding the number of
bytes.
</para>
<para>
However, it leads to an easy (and fast) implementation of non-blocking
messaging processing. With the help of the header, the messages can be
received by protocol handling classes in the background (non-blocking),
if there are many connections to the server, all of them can be served
parallel. You don't need to look at the message content, to receive the
message (and to determine when you are done), just at the header, so the
code for that is pretty easy.
</para>
<para>
Once a message is there, it can be demarshalled and processed in one
single pass, without caring about cases where not all data may have been
received (because the messageLength guarantees that everything is
there).
</para>
</sect2>
<sect2 id="mcop-protocol-invocations">
<title>Invocations</title>
<para>
To call a remote method, you need to send the following structure in the
body of an &MCOP; message with the messageType = 1 (mcopInvocation):
</para>
<programlisting>
struct Invocation {
long objectID;
long methodID;
long requestID;
};
</programlisting>
<para>
after that, you send the parameters as structure, &eg; if you invoke the
method string concat(string s1, string s2), you send a structure like
</para>
<programlisting>
struct InvocationBody {
string s1;
string s2;
};
</programlisting>
<para>
if the method was declared to be oneway - that means asynchronous
without return code - then that was it. Otherwise, you'll receive as
answer the message with messageType = 2 (mcopReturn)
</para>
<programlisting>
struct ReturnCode {
long requestID;
&lt;resulttype&gt; result;
};
</programlisting>
<para>
where &lt;resulttype&gt; is the type of the result. As void types are
omitted in marshalling, you can also only write the requestID if you
return from a void method.
</para>
<para>
So our string concat(string s1, string s2) would lead to a returncode
like
</para>
<programlisting>
struct ReturnCode {
long requestID;
string result;
};
</programlisting>
</sect2>
<sect2 id="mcop-protocol-inspecting">
<title>Inspecting Interfaces</title>
<para>
To do invocations, you need to know the methods an object supports. To
do so, the methodID 0, 1, 2 and 3 are hardwired to certain
functionalities. That is
</para>
<programlisting>
long _lookupMethod(MethodDef methodDef); // methodID always 0
string _interfaceName(); // methodID always 1
InterfaceDef _queryInterface(string name); // methodID always 2
TypeDef _queryType(string name); // methodID always 3
</programlisting>
<para>
to read that, you of course need also
</para>
<programlisting>
struct MethodDef {
string methodName;
string type;
long flags; // set to 0 for now (will be required for streaming)
sequence&lt;ParamDef&gt; signature;
};
struct ParamDef {
string name;
long typeCode;
};
</programlisting>
<para>
the parameters field contains type components which specify the types of
the parameters. The type of the returncode is specified in the
MethodDef's type field.
</para>
<para>
Strictly speaking, only the methods
<methodname>_lookupMethod()</methodname> and
<methodname>_interfaceName()</methodname> differ from object to object,
while the <methodname>_queryInterface()</methodname> and
<methodname>_queryType()</methodname> are always the same.
</para>
<para>
What are those methodIDs? If you do an &MCOP; invocation, you are
expected to pass a number for the method you are calling. The reason for
that is, that numbers can be processed much faster than strings when
executing an &MCOP; request.
</para>
<para>
So how do you get those numbers? If you know the signature of the
method, that is a MethodDef that describes the method, (which contains
name, type, parameter names, parameter types and such), you can pass
that to _lookupMethod of the object where you wish to call a method. As
_lookupMethod is hardwired to methodID 0, you should encounter no
problems doing so.
</para>
<para>
On the other hand, if you don't know the method signature, you can find
which methods are supported by using _interfaceName, _queryInterface and
_queryType.
</para>
</sect2>
<sect2 id="mcop-protocol-typedefs">
<title>Type Definitions</title>
<para>
User defined datatypes are described using the
<structname>TypeDef</structname> structure:
</para>
<programlisting>
struct TypeComponent {
string type;
string name;
};
struct TypeDef {
string name;
sequence&lt;TypeComponent&gt; contents;
};
</programlisting>
</sect2>
</sect1>
<sect1 id="why-not-dcop">
<title>Why &arts; Doesn't Use &DCOP;</title>
<para>
Since &kde; dropped <acronym>CORBA</acronym> completely, and is using
&DCOP; everywhere instead, naturally the question arises why &arts;
isn't doing so. After all, &DCOP; support is in
<classname>TDEApplication</classname>, is well-maintained, supposed to
integrate greatly with libICE, and whatever else.
</para>
<para>
Since there will be (potentially) a lot of people asking whether having
&MCOP; besides &DCOP; is really necessary, here is the answer. Please
don't get me wrong, I am not trying to say <quote>&DCOP; is
bad</quote>. I am just trying to say <quote>&DCOP; isn't the right
solution for &arts;</quote> (while it is a nice solution for other
things).
</para>
<para>
First, you need to understand what exactly &DCOP; was written
for. Created in two days during the &kde;-TWO meeting, it was intended
to be as simple as possible, a really <quote>lightweight</quote>
communication protocol. Especially the implementation left away
everything that could involve complexity, for instance a full blown
concept how data types shall be marshalled.
</para>
<para>
Even although &DCOP; doesn't care about certain things (like: how do I
send a string in a network-transparent manner?) - this needs to be
done. So, everything that &DCOP; doesn't do, is left to &Qt; in the
&kde; apps that use &DCOP; today. This is mostly type management (using
the &Qt; serialization operator).
</para>
<para>
So &DCOP; is a minimal protocol which perfectly enables &kde;
applications to send simple messages like <quote>open a window pointing
to http://www.kde.org</quote> or <quote>your configuration data has
changed</quote>. However, inside &arts; the focus lies on other things.
</para>
<para>
The idea is, that little plugins in &arts; will talk involving such data
structures as <quote>midi events</quote> and <quote>songposition
pointers</quote> and <quote>flow graphs</quote>.
</para>
<para>
These are complex data types, which must be sent between different
objects, and be passed as streams, or parameters. &MCOP; supplies a type
concept, to define complex data types out of simpler ones (similar to
structs or arrays in C++). &DCOP; doesn't care about types at all, so
this problem would be left to the programmer - like: writing C++ classes
for the types, and make sure they can serialize properly (for instance:
support the &Qt; streaming operator).
</para>
<para>
But that way, they would be inaccessible to everything but direct C++
coding. Specifically, you could not design a scripting language, that
would know all types plugins may ever expose, as they are not self
describing.
</para>
<para>
Much the same argument is valid for interfaces as well. &DCOP; objects
don't expose their relationships, inheritance hierarchies, etc. - if you
were to write an object browser which shows you <quote>what attributes
has this object got</quote>, you'd fail.
</para>
<para>
While Matthias told me that you have a special function
<quote>functions</quote> on each object that tells you about the methods
that an object supports, this leaves out things like attributes
(properties), streams and inheritance relations.
</para>
<para>
This seriously breaks applications like &arts-builder;. But remember:
&DCOP; was not so much intended to be an object model (as &Qt; already
has one with <application>moc</application> and similar), nor to be
something like <acronym>CORBA</acronym>, but to supply inter-application
communication.
</para>
<para>
Why &MCOP; even exists is: it should work fine with streams between
objects. &arts; makes heavily use of small plugins, which interconnect
themselves with streams. The <acronym>CORBA</acronym> version of &arts;
had to introduce a very annoying split between <quote>the SynthModule
objects</quote>, which were the internal work modules that did do the
streaming, and <quote>the <acronym>CORBA</acronym> interface</quote>,
which was something external.
</para>
<para>
Much code cared about making interaction between <quote>the SynthModule
objects</quote> and <quote>the <acronym>CORBA</acronym>
interface</quote> look natural, but it didn't, because
<acronym>CORBA</acronym> knew nothing at all about streams. &MCOP;
does. Look at the code (something like
<filename>simplesoundserver_impl.cpp</filename>). Way better! Streams
can be declared in the interface of modules, and implemented in a
natural looking way.
</para>
<para>
One can't deny it. One of the reasons why I wrote &MCOP; was speed. Here
are some arguments why &MCOP; will definitely be faster than &DCOP;
(even without giving figures).
</para>
<para>
An invocation in &MCOP; will have a six-<quote>long</quote>-header. That
is:
</para>
<itemizedlist>
<listitem><para>magic <quote>MCOP</quote></para></listitem>
<listitem><para>message type (invocation)</para></listitem>
<listitem><para>size of the request in bytes</para></listitem>
<listitem><para>request ID</para></listitem>
<listitem><para>target object ID</para></listitem>
<listitem><para>target method ID</para></listitem>
</itemizedlist>
<para>
After that, the parameters follow. Note that the demarshalling of this
is extremely fast. You can use table lookups to find the object and the
method demarshalling function, which means that complexity is O(1) [ it
will take the same amount of time, no matter how many objects are alive,
or how many functions are there ].
</para>
<para>
Comparing this to &DCOP;, you'll see, that there are at least
</para>
<itemizedlist>
<listitem><para>a string for the target object - something like
<quote>myCalculator</quote></para></listitem>
<listitem><para>a string like <quote>addNumber(int,int)</quote> to
specify the method</para></listitem>
<listitem><para>several more protocol info added by libICE, and other
DCOP specifics I don't know</para></listitem>
</itemizedlist>
<para>
These are much more painful to demarshall, as you'll need to parse the
string, search for the function, &etc;.
</para>
<para>
In &DCOP;, all requests are running through a server
(<application>DCOPServer</application>). That means, the process of a
synchronous invocation looks like this:
</para>
<itemizedlist>
<listitem>
<para>
Client process sends invocation.
</para>
</listitem>
<listitem>
<para>
<application>DCOPserver</application> (man-in-the-middle) receives
invocation and looks where it needs to go, and sends it to the
<quote>real</quote> server.
</para>
</listitem>
<listitem>
<para>
Server process receives invocation, performs request and sends result.
</para>
</listitem>
<listitem>
<para>
<application>DCOPserver</application> (man-in-the-middle) receives
result and ... sends it to the client.
</para>
</listitem>
<listitem>
<para>
Client decodes reply.
</para>
</listitem>
</itemizedlist>
<para>
In &MCOP;, the same invocation looks like this:
</para>
<itemizedlist>
<listitem>
<para>
Client process sends invocation.
</para>
</listitem>
<listitem>
<para>
Server process receives invocation, performs request and sends result.
</para>
</listitem>
<listitem>
<para>
Client decodes reply.
</para>
</listitem>
</itemizedlist>
<para>
Say both were implemented correctly, &MCOP;s peer-to-peer strategy
should be faster by a factor of two, than &DCOP;s man-in-the-middle
strategy. Note however that there were of course reasons to choose the
&DCOP; strategy, which is namely: if you have 20 applications running,
and each app is talking to each app, you need 20 connections in &DCOP;,
and 200 with &MCOP;. However in the multimedia case, this is not
supposed to be the usual setting.
</para>
<para>
I tried to compare &MCOP; and &DCOP;, doing an invocation like adding
two numbers. I modified testdcop to achieve this. However, the test may
not have been precise on the &DCOP; side. I invoked the method in the
same process that did the call for &DCOP;, and I didn't know how to get
rid of one debugging message, so I used output redirection.
</para>
<para>
The test only used one object and one function, expect &DCOP;s results
to decrease with more objects and functions, while &MCOP;s results
should stay the same. Also, the <application>dcopserver</application>
process wasn't connected to other applications, it might be that if many
applications are connected, the routing performance decreases.
</para>
<para>
The result I got was that while &DCOP; got slightly more than 2000
invocations per second, &MCOP; got slightly more than 8000 invocations
per second. That makes a factor of 4. I know that &MCOP; isn't tuned to
the maximum possible, yet. (Comparison: <acronym>CORBA</acronym>, as
implemented with mico, does something between 1000 and 1500 invocations
per second).
</para>
<para>
If you want <quote>harder</quote> data, consider writing some small
benchmark app for &DCOP; and send it to me.
</para>
<para>
<acronym>CORBA</acronym> had the nice feature that you could use objects
you implemented once, as <quote>separate server process</quote>, or as
<quote>library</quote>. You could use the same code to do so, and
<acronym>CORBA</acronym> would transparently decide what to do. With
&DCOP;, that is not really intended, and as far as I know not really
possible.
</para>
<para>
&MCOP; on the other hand should support that from the beginning. So you
can run an effect inside &artsd;. But if you are a wave editor, you can
choose to run the same effect inside your process space as well.
</para>
<para>
While &DCOP; is mostly a way to communicate between apps, &MCOP; is also
a way to communicate inside apps. Especially for multimedia streaming,
this is important (as you can run multiple &MCOP; objects parallely, to
solve a multimedia task in your application).
</para>
<para>
Although &MCOP; does not currently do so, the possibilities are open to
implement quality of service features. Something like <quote>that &MIDI; event
is really really important, compared to this invocation</quote>. Or something
like <quote>needs to be there in time</quote>.
</para>
<para>
On the other hand, stream transfer can be integrated in the &MCOP;
protocol nicely, and combined with <acronym>QoS</acronym> stuff. Given
that the protocol may be changed, &MCOP; stream transfer should not
really get slower than conventional <acronym>TCP</acronym> streaming,
but: it will be easier and more consistent to use.
</para>
<para>
There is no need to base a middleware for multimedia on &Qt;. Deciding
so, and using all that nice &Qt;-streaming and stuff, will easily lead
to the middleware becoming a &Qt;-only (or rather &kde;-only) thing. I
mean: as soon as I'll see the GNOMEs using &DCOP;, too, or something like
that, I am certainly proven wrong.
</para>
<para>
While I do know that &DCOP; basically doesn't know about the data types
it sends, so that you could use &DCOP; without using &Qt;, look at how
it is used in daily &kde; usage: people send types like
<classname>TQString</classname>, <classname>QRect</classname>,
<classname>QPixmap</classname>, <classname>QCString</classname>, ...,
around. These use &Qt;-serialization. So if somebody choose to support
&DCOP; in a GNOME program, they would either have to claim to use
<classname>TQString</classname>,... types (although they don't do so),
and emulate the way &Qt; does the streaming, or they would send other
string, pixmap and rect types around, and thus not be interoperable.
</para>
<para>
Well, whatever. &arts; was always intended to work with or without
&kde;, with or without &Qt;, with or without X11, and maybe even with or
without &Linux; (and I have even no problems with people who port it to
a popular non-free operating systems).
</para>
<para>
It is my position that non-&GUI;-components should be written
non-&GUI;-dependant, to make sharing those among wider amounts of
developers (and users) possible.
</para>
<para>
I see that using two <acronym>IPC</acronym> protocols may cause
inconveniences. Even more, if they are both non-standard. However, for
the reasons given above, switching to &DCOP; is no option. If there is
significant interest to find a way to unite the two, okay, we can
try. We could even try to make &MCOP; speak <acronym>IIOP</acronym>,
then we'd have a <acronym>CORBA</acronym> <acronym>ORB</acronym> ;).
</para>
<para>
I talked with Matthias Ettrich a bit about the future of the two
protocols, and we found lots of ways how things could go on. For
instance, &MCOP; could handle the message communication in &DCOP;, thus
bringing the protocols a bit closer together.
</para>
<para>
So some possible solutions would be:
</para>
<itemizedlist>
<listitem>
<para>
Write an &MCOP; - &DCOP; gateway (which should be possible, and would
make interoperation possible) - note: there is an experimental
prototype, if you like to work on that.
</para>
</listitem>
<listitem>
<para>
Integrate everything &DCOP; users expect into &MCOP;, and try to only do
&MCOP; - one could add an <quote>man-in-the-middle-option</quote> to
&MCOP;, too ;)
</para>
</listitem>
<listitem>
<para>
Base &DCOP; on &MCOP; instead of libICE, and slowly start integrating
things closer together.
</para>
</listitem>
</itemizedlist>
<para>
However, it may not be the worst possibility to use each protocol for
everything it was intended for (there are some big differences in the
design goals), and don't try to merge them into one.
</para>
</sect1>
</chapter>