The &arts; structure. The idea of &arts; is, that synthesis can be done using small modules, which only do one thing, and then recombine them in complex structures. The small modules normally have inputs, where they can get some signals or parameters, and outputs, where they produce some signals. One module (Synth_ADD) for instance just takes the two signals at it's input and adds them together. The result is available as output signal. The places where modules provide their input/output signals are called ports. A structure is a combination of connected modules, some of which may have parameters coded directly to their input ports, others which may be connected, and others, which are not connected at all. What you can do with &arts-builder; is describing structures. You describe, which modules you want to be connected with which other modules. When you are done, you can save that structure description to a file, or tell &arts; to create such a structure you described (Execute). Then you'll probably hear some sound, if you did everything the right way. Suppose you have an application called mousepling(that should make a pling sound if you click on a button. The latency is the time between your finger clicking the mouse button and you hearing the pling. The latency in this setup composes itself out of certain latencies, that have different causes. In this simple application, latency occurs at these places: The time until the kernel has notified the X11 server that a mouse button was pressed. The time until the X11 server has notified your application that a mouse button was pressed. The time until the mousepling application has decided that this button is worth playing a pling. The time it takes the mousepling application to tell the soundserver that it should play a pling. The time it takes for the pling (which the soundserver starts mixing to the other output at once) to go through the buffered data, until it really reaches the position where the soundcard plays. The time it takes the pling sound from the speakers to reach your ear. The first three items are latencies external to &arts;. They are interesting, but beyond the scope of this document. Nevertheless be aware that they exist, so that even if you have optimized everything else to really low values, you may not necessarily get exactly the result you calculated. Telling the server to play something involves usually one single &MCOP; call. There are benchmarks which confirm that, on the same host with unix domain sockets, telling the server to play something can be done about 9000 times in one second with the current implementation. I expect that most of this is kernel overhead, switching from one application to another. Of course this value changes with the exact type of the parameters. If you transfer a whole image with one call, it will be slower than if you transfer only one long value. For the returncode the same is true. However for ordinary strings (such as the filename of the wav file to play) this shouldn't be a problem. That means, we can approximate this time with 1/9000 sec, that is below 0.15 ms. We'll see that this is not relevant. Next is the time between the server starting playing and the soundcard getting something. The server needs to do buffering, so that when other applications are running, such as your X11 server or mousepling application no dropouts are heard. The way this is done under &Linux; is that there are a number fragments of a size. The server will refill fragments, and the soundcard will play fragments. So suppose there are three fragments. The server refills the first, the soundcard starts playing it. The server refills the second. The server refills the third. The server is done, other applications can do something now. As the soundcard has played the first fragment, it starts playing the second and the server starts refilling the first. And so on. The maximum latency you get with all that is (number of fragments)*(size of each fragment)/(samplingrate * (size of each sample)). Suppose we assume 44kHz stereo, and 7 fragments a 1024 bytes (the current aRts defaults), we get 40 ms. These values can be tuned according to your needs. However, the CPU usage increases with smaller latencies, as the sound server needs to refill the buffers more often, and in smaller parts. It is also mostly impossible to reach better values without giving the soundserver realtime priority, as otherwise you'll often get drop-outs. However, it is realistic to do something like 3 fragments with 256 bytes each, which would make this value 4.4 ms. With 4.4ms delay the idle CPU usage of &arts; would be about 7.5%. With 40ms delay, it would be about 3% (of a PII-350, and this value may depend on your soundcard, kernel version and others). Then there is the time it takes the pling sound to get from the speakers to your ear. Suppose your distance from the speakers is 2 meters. Sound travels at a speed of 330 meters per second. So we can approximate this time with 6 ms. Streaming applications are those that produce their sound themselves. Assume a game, which outputs a constant stream of samples, and should now be adapted to replay things via &arts;. To have an example: when I press a key, the figure which I am playing jumps, and a boing sound is played. First of all, you need to know how &arts; does streaming. Its very similar to the I/O with the soundcard. The game sends some packets with samples to the sound server. Lets say three packets. As soon as the sound server is done with the first packet, it sends a confirmation back to the game that this packet is done. The game creates another packet of sound and sends it to the server. Meanwhile the server starts consuming the second sound packet, and so on. The latency here looks similar like in the simple case: The time until the kernel has notified the X11 server that a key was pressed. The time until the X11 server has notified the game that a key was pressed. The time until the game has decided that this key is worth playing a boing. The time until the packet of sound in which the game has started putting the boing sound is reaching the sound server. The time it takes for the boing (which the soundserver starts mixing to the other output at once) to go through the buffered data, until it really reaches the position where the soundcard plays. The time it takes the boing sound from the speakers to reach your ear. The external latencies, as above, are beyond the scope of this document. Obviously, the streaming latency depends on the time it takes all packets that are used for streaming to be played once. So it is (number of packets)*(size of each packet)/(samplingrate * (size of each sample)) As you see that is the same formula as applies for the fragments. However for games, it makes no sense to do such small delays as above. I'd say a realistic configuration for games would be 2048 bytes per packet, use 3 packets. The resulting latency would be 35ms. This is based on the following: assume that the game renders 25 frames per second (for the display). It is probably safe to assume that you won't notice a difference of sound output of one frame. Thus 1/25 second delay for streaming is acceptable, which in turn means 40ms would be okay. Most people will also not run their games with realtime priority, and the danger of drop-outs in the sound is not to be neglected. Streaming with 3 packets a 256 bytes is possible (I tried that) - but causes a lot of CPU usage for streaming. For server side latencies, you can calculate these exactly as above. There are a lot of factors which influence _CPU usage in a complex scenario, with some streaming applications and some others, some plugins on the server etc. To name a few: Raw CPU usage by the calculations necessary. &arts; internal scheduling overhead - how &arts; decides when which module should calculate what. Integer to float conversion overhead. &MCOP;0 protocol overhead. Kernel: process/context switching. Kernel: communication overhead For raw CPU usage for calculations, if you play two streams, simultaneously you need to do additions. If you apply a filter, some calculations are involved. To have a simplified example, adding two streams involves maybe four CPU cycles per addition, on a 350Mhz processor, this is 44100*2*4/350000000 = 0.1% CPU usage. &arts; internal scheduling: &arts; needs to decide which plugin when calculates what. This takes time. Take a profiler if you are interested in that. Generally what can be said is: the less realtime you do (&ie;. the larger blocks can be calculated at a time) the less scheduling overhead you have. Above calculating blocks of 128 samples at a time (thus using fragment sizes of 512 bytes) the scheduling overhead is probably not worth thinking about it. Integer to float conversion overhead: &arts; uses floats internally as data format. These are easy to handle and on recent processors not slower than integer operations. However, if there are clients which play data which is not float (like a game that should do its sound output via &arts;), it needs to be converted. The same applies if you want to replay the sounds on your soundcard. The soundcard wants integers, so you need to convert. Here are numbers for a Celeron, approx. ticks per sample, with -O2 +egcs 2.91.66 (taken by Eugene Smith hamster@null.ru). This is of course highly processor dependant: convert_mono_8_float: 14 convert_stereo_i8_2float: 28 convert_mono_16le_float: 40 interpolate_mono_16le_float: 200 convert_stereo_i16le_2float: 80 convert_mono_float_16le: 80 So that means 1% CPU usage for conversion and 5% for interpolation on this 350 MHz processor. &MCOP; protocol overheadL &MCOP; does, as a rule of thumb, 9000 invocations per second. Much of this is not &MCOP;s fault, but relates to the two kernel causes named below. However, this gives a base to do calculations what the cost of streaming is. Each data packet transferred through streaming can be considered one &MCOP; invocation. Of course large packets are slower than 9000 packets/s, but its about the idea. Suppose you use packet sizes of 1024 bytes. Thus, to transfer a stream with 44kHz stereo, you need to transfer 44100*4/1024 = 172 packets per second. Suppose you could with 100% cpu usage transfer 9000 packets, then you get (172*100)/9000 = 2% CPU usage due to streaming with 1024 byte packets. That are approximations. However, they show, that you would be much better off (if you can afford it for the latency), to use for instance packets of 4096 bytes. We can make a compact formula here, by calculating the packet size which causes 100% CPU usage as 44100*4/9000 = 19.6 samples, and thus getting the quick formula: streaming CPU usage in percent = 1960/(your packet size) which gives us 0.5% CPU usage when streaming with 4096 byte packets. Kernel process/context switching: this is part of the &MCOP; protocol overhead. Switching between two processes takes time. There is new memory mapping, the caches are invalid, whatever else (if there is a kernel expert reading this - let me know what exactly are the causes). This means: it takes time. I am not sure how many context switches &Linux; can do per second, but that number isn't infinite. Thus, of the &MCOP; protocol overhead I suppose quite a bit is due to context switching. In the beginning of &MCOP;, I did tests to use the same communication inside one process, and it was much faster (four times as fast or so). Kernel: communication overhead: This is part of the &MCOP; protocol overhead. Transferring data between processes is currently done via sockets. This is convenient, as the usual select() methods can be used to determine when a message has arrived. It can also be combined with other I/O sources as audio I/O, X11 server or whatever else easily. However, those read and write calls cost certainly processor cycles. For small invocations (such as transferring one midi event) this is probably not so bad, for large invocations (such as transferring one video frame with several megabytes) this is clearly a problem. Adding the usage of shared memory to &MCOP; where appropriate is probably the best solution. However it should be done transparent to the application programmer. Take a profiler or do other tests to find out how much exactly current audio streaming is impacted by the not using sharedmem. However, its not bad, as audio streaming (replaying mp3) can be done with 6% total CPU usage for &artsd; and artscat (and 5% for the mp3 decoder). However, this includes all things from the necessary calculations up do the socket overhead, thus I'd say in this setup you could perhaps save 1% by using sharedmem. These are done with the current development snapshot. I also wanted to try out the real hard cases, so this is not what everyday applications should use. I wrote an application called streamsound which sends streaming data to &arts;. Here it is running with realtime priority (without problems), and one small serverside (volume-scaling and clipping) plugin: 4974 stefan 20 0 2360 2360 1784 S 0 17.7 1.8 0:21 artsd 5016 stefan 20 0 2208 2208 1684 S 0 7.2 1.7 0:02 streamsound 5002 stefan 20 0 2208 2208 1684 S 0 6.8 1.7 0:07 streamsound 4997 stefan 20 0 2208 2208 1684 S 0 6.6 1.7 0:07 streamsound Each of them is streaming with 3 fragments a 1024 bytes (18 ms). There are three such clients running simultaneously. I know that that does look a bit too much, but as I said: take a profiler and find out what costs time, and if you like, improve it. However, I don't think using streaming like that is realistic or makes sense. To take it even more to the extreme, I tried what would be the lowest latency possible. Result: you can do streaming without interruptions with one client application, if you take 2 fragments of 128 bytes between aRts and the soundcard, and between the client application and aRts. This means that you have a total maximum latency of 128*4/44100*4 = 3 ms, where 1.5 ms is generated due to soundcard I/O and 1.5 ms is generated through communication with &arts;. Both applications need to run realtimed. But: this costs an enormous amount of CPU. This example cost you about 45% of my P-II/350. I also starts to click if you start top, move windows on your X11 display or do disk I/O. All these are kernel issues. The problem is that scheduling two or more applications with realtime priority cost you an enormous amount of effort, too, even more if the communicate, notify each other &etc;. Finally, a more real life example. This is &arts; with artsd and one artscat (one streaming client) running 16 fragments a 4096 bytes: 5548 stefan 12 0 2364 2364 1752 R 0 4.9 1.8 0:03 artsd 5554 stefan 3 0 752 752 572 R 0 0.7 0.5 0:00 top 5550 stefan 2 0 2280 2280 1696 S 0 0.5 1.7 0:00 artscat Busses are dynamically built connections that transfer audio. Basically, there are some uplinks and some downlinks. All signals from the uplinks are added and send to the downlinks. Busses as currently implemented operate in stereo, so you can only transfer stereo data over busses. If you want mono data, well, transfer it only over one channel and set the other to zero or whatever. What you need to to, is to create one or more Synth_BUS_UPLINK objects and tell them a bus name, to which they should talk (⪚ audio or drums). Simply throw the data in there. Then, you'll need to create one or more Synth_BUS_DOWNLINK objects, and tell them the bus name (audio or drums ... if it matches, the data will get through), and the mixed data will come out again. The uplinks and downlinks can reside in different structures, you can even have different &arts-builder;s running and start an uplink in one and receive the data from the other with a downlink. What is nice about busses is, that they are fully dynamic. Clients can plug in and out on the fly. There should be no clicking or noise as this happens. Of course, you should not plug out a client playing a signal, since it will probably not be a zero level when plugged out the bus, and then it will click. &arts;/&MCOP; heavily relies on splitting up things into small components. This makes things very flexible, as you can extend the system easily by adding new components, which implement new effects, fileformats, oscillators, gui elements, ... As almost everything is a component, almost everything can be extended easily, without changing existing sources. New components can be simply loaded dynamically to enhance already existing applications. However, to make this work, two things are required: Components must advertise themselves - they must describe what great things they offer, so that applications will be able to use them. Applications must actively look for components that they could use, instead of using always the same thing for some task. The combination of this: components which say here I am, I am cool, use me, and applications (or if you like, other components) which go out and look which component they could use to get a thing done, is called trading. In &arts;, components describe themselves by specifying values that they support for properties. A typical property for a file-loading component could be the extension of the files that it can process. Typical values could be wav, aiff or mp3. In fact, every component may choose to offer many different values for one property. So one single component could offer reading both, wav and aiff files, by specifying that it supports these values for the property Extension. To do so, a component has to place a .mcopclass file at an appropriate place, containing the properties it supports, for our example, this could look like this (and would be installed in componentdir/Arts/WavPlayObject.mcopclass): Interface=Arts::WavPlayObject,Arts::PlayObject,Arts::SynthModule,Arts::Object Author="Stefan Westerfeld <stefan@space.twc.de>" URL="http://www.arts-project.org" Extension=wav,aiff MimeType=audio/x-wav,audio/x-aiff It is important that the filename of the .mcopclass-file also says what the interface of the component is called like. The trader doesn't look at the contents at all, if the file (like here) is called Arts/WavPlayObject.mcopclass, the component interface is called Arts::WavPlayObject (modules map to folders). To look for components, there are two interfaces (which are defined in core.idl, so you have them in every application), called Arts::TraderQuery and Arts::TraderOffer. You to go on a shopping tour for components like this: Create a query object: Arts::TraderQuery query; Specify what you want. As you saw above, components describe themselves using properties, for which they offer certain values. So specifying what you want is done by selecting components that support a certain value for a property. This is done using the supports method of a TraderQuery: query.supports("Interface","Arts::PlayObject"); query.supports("Extension","wav"); Finally, perform the query using the query method. Then, you'll (hopefully) get some offers: vector<Arts::TraderOffer> *offers = query.query(); Now you can examine what you found. Important is the interfaceName method of TraderOffer, which will tell you the name of the component, that matched the query. You can also find out further properties by getProperty. The following code will simply iterate through all components, print their interface names (which could be used for creation), and delete the results of the query again: vector<Arts::TraderOffer>::iterator i; for(i = offers->begin(); i != offers->end(); i++) cout << i->interfaceName() << endl; delete offers; For this kind of trading service to be useful, it is important to somehow agree on what kinds of properties components should usually define. It is essential that more or less all components in a certain area use the same set of properties to describe themselves (and the same set of values where applicable), so that applications (or other components) will be able to find them. Author (type string, optional): This can be used to ultimately let the world know that you wrote something. You can write anything you like in here, e-mail address is of course helpful. Buildable (type boolean, recommended): This indicates whether the component is usable with RAD tools (such as &arts-builder;) which use components by assigning properties and connecting ports. It is recommended to set this value to true for almost any signal processing component (such as filters, effects, oscillators, ...), and for all other things which can be used in RAD like fashion, but not for internal stuff like for instance Arts::InterfaceRepo. Extension (type string, used where relevant): Everything dealing with files should consider using this. You should put the lowercase version of the file extension without the . here, so something like wav should be fine. Interface (type string, required): This should include the full list of (useful) interfaces your components supports, probably including Arts::Object and if applicable Arts::SynthModule. Language (type string, recommended): If you want your component to be dynamically loaded, you need to specify the language here. Currently, the only allowed value is C++, which means the component was written using the normal C++ API. If you do so, you'll also need to set the Library property below. Library (type string, used where relevant): Components written in C++ can be dynamically loaded. To do so, you have to compile them into a dynamically loadable libtool (.la) module. Here, you can specify the name of the .la-File that contains your component. Remember to use REGISTER_IMPLEMENTATION (as always). MimeType (type string, used where relevant): Everything dealing with files should consider using this. You should put the lowercase version of the standard mimetype here, for instance audio/x-wav. &URL; (type string, optional): If you like to let people know where they can find a new version of the component (or a homepage or anything), you can do it here. This should be standard &HTTP; or &FTP; &URL;. Each namespace declaration corresponds to a module declaration in the &MCOP; &IDL;. // mcop idl module M { interface A { } }; interface B; In this case, the generated C++ code for the &IDL; snippet would look like this: // C++ header namespace M { /* declaration of A_base/A_skel/A_stub and similar */ class A { // Smartwrapped reference class /* [...] */ }; } /* declaration of B_base/B_skel/B_stub and similar */ class B { /* [...] */ }; So when referring the classes from the above example in your C++ code, you would have to write M::A, but only B. However, you can of course use using M somewhere - like with any namespace in C++. There is one global namespace called Arts, which all programs and libraries that belong to &arts; itself use to put their declarations in. This means, that when writing C++ code that depends on &arts;, you normally have to prefix every class you use with Arts::, like this: int main(int argc, char **argv) { Arts::Dispatcher dispatcher; Arts::SimpleSoundServer server(Arts::Reference("global:Arts_SimpleSoundServer")); server.play("/var/foo/somefile.wav"); The other alternative is to write a using once, like this: using namespace Arts; int main(int argc, char **argv) { Dispatcher dispatcher; SimpleSoundServer server(Reference("global:Arts_SimpleSoundServer")); server.play("/var/foo/somefile.wav"); [...] In &IDL; files, you don't exactly have a choice. If you are writing code that belongs to &arts; itself, you'll have to put it into module &arts;. // IDL File for aRts code: #include <artsflow.idl> module Arts { // put it into the Arts namespace interface Synth_TWEAK : SynthModule { in audio stream invalue; out audio stream outvalue; attribute float tweakFactor; }; }; If you write code that doesn't belong to &arts; itself, you should not put it into the Arts namespace. However, you can make an own namespace if you like. In any case, you'll have to prefix classes you use from &arts;. // IDL File for code which doesn't belong to aRts: #include <artsflow.idl> // either write without module declaration, then the generated classes will // not use a namespace: interface Synth_TWEAK2 : Arts::SynthModule { in audio stream invalue; out audio stream outvalue; attribute float tweakFactor; }; // however, you can also choose your own namespace, if you like, so if you // write an application "PowerRadio", you could for instance do it like this: module PowerRadio { struct Station { string name; float frequency; }; interface Tuner : Arts::SynthModule { attribute Station station; // no need to prefix Station, same module out audio stream left, right; }; }; Often, in interfaces, casts, method signatures and similar, &MCOP; needs to refer to names of types or interfaces. These are represented as string in the common &MCOP; datastructures, while the namespace is always fully represented in the C++ style. This means the strings would contain M::A and B, following the example above. Note this even applies if inside the &IDL; text the namespace qualifiers were not given, since the context made clear which namespace the interface A was meant to be used in. Using threads isn't possible on all platforms. This is why &arts; was originally written without using threading at all. For almost all problems, for each threaded solution to the problem, there is a non-threaded solution that does the same. For instance, instead of putting audio output in a separate thread, and make it blocking, &arts; uses non-blocking audio output, and figures out when to write the next chunk of data using select(). However, &arts; (in very recent versions) at least provides support for people who do want to implement their objects using threads. For instance, if you already have code for an mp3 player, and the code expects the mp3 decoder to run in a separate thread, it's usually the easiest thing to do to keep this design. The &arts;/&MCOP; implementation is built along sharing state between separate objects in obvious and non-obvious ways. A small list of shared state includes: The Dispatcher object which does &MCOP; communication. The Reference counting (Smartwrappers). The IOManager which does timer and fd watches. The ObjectManager which creates objects and dynamically loads plugins. The FlowSystem which calls calculateBlock in the appropriate situations. All of the above objects don't expect to be used concurrently (&ie; called from separate threads at the same time). Generally there are two ways of solving this: Require the caller of any functions on this objects to acquire a lock before using them. Making these objects really threadsafe and/or create per-thread instances of them. &arts; follows the first approach: you will need a lock whenever you talk to any of these objects. The second approach is harder to do. A hack which tries to achieve this is available at http://space.twc.de/~stefan/kde/download/arts-mt.tar.gz, but for the current point in time, a minimalistic approach will probably work better, and cause less problems with existing applications. You can get/release the lock with the two functions: Arts::Dispatcher::lock() Arts::Dispatcher::unlock() Generally, you don't need to acquire the lock (and you shouldn't try to do so), if it is already held. A list of conditions when this is the case is: You receive a callback from the IOManager (timer or fd). You get call due to some &MCOP; request. You are called from the NotificationManager. You are called from the FlowSystem (calculateBlock) There are also some exceptions of functions. which you can only call in the main thread, and for that reason you will never need a lock to call them: Constructor/destructor of Dispatcher/IOManager. Dispatcher::run() / IOManager::run() IOManager::processOneEvent() But that is it. For everything else that is somehow related to &arts;, you will need to get the lock, and release it again when done. Always. Here is a simple example: class SuspendTimeThread : Arts::Thread { public: void run() { /* * you need this lock because: * - constructing a reference needs a lock (as global: will go to * the object manager, which might in turn need the GlobalComm * object to look up where to connect to) * - assigning a smartwrapper needs a lock * - constructing an object from reference needs a lock (because it * might need to connect a server) */ Arts::Dispatcher::lock(); Arts::SoundServer server = Arts::Reference("global:Arts_SoundServer"); Arts::Dispatcher::unlock(); for(;;) { /* * you need a lock here, because * - dereferencing a smartwrapper needs a lock (because it might * do lazy creation) * - doing an MCOP invocation needs a lock */ Arts::Dispatcher::lock(); long seconds = server.secondsUntilSuspend(); Arts::Dispatcher::unlock(); printf("seconds until suspend = %d",seconds); sleep(1); } } } The following threading related classes are currently available: Arts::Thread - which encapsulates a thread. Arts::Mutex - which encapsulates a mutex. Arts::ThreadCondition - which provides support to wake up threads which are waiting for a certain condition to become true. Arts::SystemThreads - which encapsulates the operating system threading layer (which offers a few helpful functions to application programmers). See the links for documentation. &MCOP; references are one of the most central concepts in &MCOP; programming. This section will try to describe how exactly references are used, and will especially also try to cover cases of failure (server crashes). An &MCOP; reference is not an object, but a reference to an object: Even though the following declaration Arts::Synth_PLAY p; looks like a definition of an object, it only declares a reference to an object. As C++ programmer, you might also think of it as Synth_PLAY *, a kind of pointer to a Synth_PLAY object. This especially means, that p can be the same thing as a NULL pointer. You can create a NULL reference by assigning it explicitly Arts::Synth_PLAY p = Arts::Synth_PLAY::null(); Invoking things on a NULL reference leads to a core dump Arts::Synth_PLAY p = Arts::Synth_PLAY::null(); string s = p.toString(); will lead to a core dump. Comparing this to a pointer, it is essentially the same as QWindow* w = 0; w->show(); which every C++ programmer would know to avoid. Uninitialized objects try to lazy-create themselves upon first use Arts::Synth_PLAY p; string s = p.toString(); is something different than dereferencing a NULL pointer. You didn't tell the object at all what it is, and now you try to use it. The guess here is that you want to have a new local instance of a Arts::Synth_PLAY object. Of course you might have wanted something else (like creating the object somewhere else, or using an existing remote object). However, it is a convenient short cut to creating objects. Lazy creation will not work once you assigned something else (like a null reference). The equivalent C++ terms would be TQWidget* w; w->show(); which obviously in C++ just plain segfaults. So this is different here. This lazy creation is tricky especially as not necessarily an implementation exists for your interface. For instance, consider an abstract thing like a Arts::PlayObject. There are certainly concrete PlayObjects like those for playing mp3s or wavs, but Arts::PlayObject po; po.play(); will certainly fail. The problem is that although lazy creation kicks in, and tries to create a PlayObject, it fails, because there are only things like Arts::WavPlayObject and similar. Thus, use lazy creation only when you are sure that an implementation exists. References may point to the same object Arts::SimpleSoundServer s = Arts::Reference("global:Arts_SimpleSoundServer"); Arts::SimpleSoundServer s2 = s; creates two references referring to the same object. It doesn't copy any value, and doesn't create two objects. All objects are reference counted So once an object isn't referred any longer by any references, it gets deleted. There is no way to explicitly delete an object, however, you can use something like this Arts::Synth_PLAY p; p.start(); [...] p = Arts::Synth_PLAY::null(); to make the Synth_PLAY object go away in the end. Especially, it should never be necessary to use new and delete in conjunction with references. As references can point to remote objects, the servers containing these objects can crash. What happens then? A crash doesn't change whether a reference is a null reference. This means that if foo.isNull() was true before a server crash then it is also true after a server crash (which is clear). It also means that if foo.isNull() was false before a server crash (foo referred to an object) then it is also false after the server crash. Invoking methods on a valid reference stays safe Suppose the server containing the object calc crashed. Still calling things like int k = calc.subtract(i,j) are safe. Obviously subtract has to return something here, which it can't because the remote object no longer exists. In this case (k == 0) would be true. Generally, operations try to return something neutral as result, such as 0.0, a null reference for objects or empty strings, when the object no longer exists. Checking error() reveals whether something worked. In the above case, int k = calc.subtract(i,j) if(k.error()) { printf("k is not i-j!\n"); } would print out k is not i-j whenever the remote invocation didn't work. Otherwise k is really the result of the subtract operation as performed by the remote object (no server crash). However, for methods doing things like deleting a file, you can't know for sure whether it really happened. Of course it happened if .error() is false. However, if .error() is true, there are two possibilities: The file got deleted, and the server crashed just after deleting it, but before transferring the result. The server crashed before being able to delete the file. Using nested invocations is dangerous in crash resistant programs Using something like window.titlebar().setTitle("foo"); is not a good idea. Suppose you know that window contains a valid Window reference. Suppose you know that window.titlebar() will return a Titlebar reference because the Window object is implemented properly. However, still the above statement isn't safe. What could happen is that the server containing the Window object has crashed. Then, regardless of how good the Window implementation is, you will get a null reference as result of the window.titlebar() operation. And then of course invoking setTitle on that null reference will lead to a crash as well. So a safe variant of this would be Titlebar titlebar = window.titlebar(); if(!window.error()) titlebar.setTitle("foo"); add the appropriate error handling if you like. If you don't trust the Window implementation, you might as well use Titlebar titlebar = window.titlebar(); if(!titlebar.isNull()) titlebar.setTitle("foo"); which are both safe. There are other conditions of failure, such as network disconnection (suppose you remove the cable between your server and client while your application runs). However their effect is the same like a server crash. Overall, it is of course a consideration of policy how strictly you try to trap communication errors throughout your application. You might follow the if the server crashes, we need to debug the server until it never crashes again method, which would mean you need not bother about all these problems. An object, to exist, must be owned by someone. If it isn't, it will cease to exist (more or less) immediately. Internally, ownership is indicated by calling _copy(), which increments an reference count, and given back by calling _release(). As soon as the reference count drops to zero, a delete will be done. As a variation of the theme, remote usage is indicated by _useRemote(), and dissolved by _releaseRemote(). These functions lead a list which server has invoked them (and thus owns the object). This is used in case this server disconnects (&ie; crash, network failure), to remove the references that are still on the objects. This is done in _disconnectRemote(). Now there is one problem. Consider a return value. Usually, the return value object will not be owned by the calling function any longer. It will however also not be owned by the caller, until the message holding the object is received. So there is a time of ownershipless objects. Now, when sending an object, one can be reasonable sure that as soon as it is received, it will be owned by somebody again, unless, again, the receiver dies. However this means that special care needs to be taken about object at least while sending, probably also while receiving, so that it doesn't die at once. The way &MCOP; does this is by tagging objects that are in process of being copied across the wire. Before such a copy is started, _copyRemote is called. This prevents the object from being freed for a while (5 seconds). Once the receiver calls _useRemote(), the tag is removed again. So all objects that are send over wire are tagged before transfer. If the receiver receives an object which is on their server, of course they will not _useRemote() it. For this special case, _cancelCopyRemote() exists to remove the tag manually. Other than that, there is also timer based tag removal, if tagging was done, but the receiver didn't really get the object (due to crash, network failure). This is done by the ReferenceClean class. &GUI; elements are currently in the experimental state. However, this section will describe what is supposed to happen here, so if you are a developer, you will be able to understand how &arts; will deal with &GUI;s in the future. There is some code there already, too. &GUI; elements should be used to allow synthesis structures to interact with the user. In the simplest case, the user should be able to modify some parameters of a structure directly (such as a gain factor which is used before the final play module). In more complex settings, one could imagine the user modifying parameters of groups of structures and/or not yet running structures, such as modifying the ADSR envelope of the currently active &MIDI; instrument. Another thing would be setting the filename of some sample based instrument. On the other hand, the user could like to monitor what the synthesizer is doing. There could be oscilloscopes, spectrum analyzers, volume meters and experiments that figure out the frequency transfer curve of some given filter module. Finally, the &GUI; elements should be able to control the whole structure of what is running inside &arts; and how. The user should be able to assign instruments to midi channels, start new effect processors, configure his main mixer desk (which is built of &arts; structures itself) to have one channel more and use another strategy for its equalizers. You see - the GUI elements should bring all possibilities of the virtual studio &arts; should simulate to the user. Of course, they should also gracefully interact with midi inputs (such as sliders should move if they get &MIDI; inputs which also change just that parameter), and probably even generate events themselves, to allow the user interaction to be recorded via sequencer. Technically, the idea is to have an &IDL; base class for all widgets (Arts::Widget), and derive a number of commonly used widgets from there (like Arts::Poti, Arts::Panel, Arts::Window, ...). Then, one can implement these widgets using a toolkit, for instance &Qt; or Gtk. Finally, effects should build their &GUI;s out of existing widgets. For instance, a freeverb effect could build it's &GUI; out of five Arts::Poti thingies and an Arts::Window. So IF there is a &Qt; implementation for these base widgets, the effect will be able to display itself using &Qt;. If there is Gtk implementation, it will also work for Gtk (and more or less look/work the same). Finally, as we're using &IDL; here, &arts-builder; (or other tools) will be able to plug &GUI;s together visually, or autogenerate &GUI;s given hints for parameters, only based on the interfaces. It should be relatively straight forward to write a create &GUI; from description class, which takes a &GUI; description (containing the various parameters and widgets), and creates a living &GUI; object out of it. Based on &IDL; and the &arts;/&MCOP; component model, it should be easy to extend the possible objects which can be used for the &GUI; just as easy as it is to add a plugin implementing a new filter to &arts;.