CLAM and other environments

In the previous section we have classified CLAM as being a General Purpose Audio Processing Framework. Nevertheless, we have also mentioned that it shares some features with other environments classified in different categories. We may now question whether the overall philosophy of CLAM is adequate or not. Is it really necessary to build an application framework? Wouldn't it be better and simpler to offer a graphical builder à la Max or Kyma? On the other hand, wouldn't it be better to design a completely new programming language more adapted to our domain à la Supercollider?

It is important to remember that CLAM's focus is on research and development of applications. It is not an artistic or creative tool (although such a tool could be developed within the framework). For developing efficient applications and for doing research on low-level details of the algorithms it is necessary to access the programming language level. Visual programming is indeed of great help and ideal for prototyping, and as already mentioned in 1.3 it is a natural evolution of any application framework. On the other hand, as R. Dannenberg points out [Dannenberg, 2004] the ``boxes and arrows'' representation for signal processing systems has stood the test of time. This is no surprise as this representation is directly related to the graphical MoCs explained in section 1.5. That is why CLAM offers the Network Editor (see 3.2.3), its particular graphical MoC visual builder, but this is not designed to substitute the code level.

On the other extreme, one may have the temptation to implement a full-fledged programming language, adapted to this particular domain. This would be an interesting idea to explore if it had the slightest chance of being viable. A programming language has many associated tools, from editors with syntax highlighting to debugging and profiling tools. Offering a completely new programming environment based on a new language is a titanic effort that needs of a very large development team. On the other hand, the language has to offer very unique and outstanding features in order to convince new users that the effort of learning it is worth the while. Supercollider represents the clearest example of such an approach. An now its author, James McCartney questions whether that was a good decision. In [McCartney, 2002] he wanders whether specialized computer music languages are necessary. He concludes that the set of abstractions available in general purpose computer languages are sufficient to build useful dedicated frameworks such as CLAM. This is also our opinion, no specialized computer language is needed, it is much wiser to build tools and frameworks around already existing and proven languages.

As for CLAM's model of computation it is interesting to compare it to three environments that have a sufficiently long tradition and that clearly state their conceptual model3.6: Max (see 2.5), Siren (see 2.4) and Kyma (see 2.5). To our purposes they also represent the extremes in between which CLAM's MoC could stand. On the one hand, Max offers a highly intuitive graphical (and graph-based) MoC while Siren and Kyma offer a very formal object-oriented approach3.7.

It has already been commented that the main feature of CLAM's MoC is that it is object-oriented. Object-oriented techniques have been applied throughout the development process and we have always followed Alan Kay's maxima of ``everything is an object''. As already outlined in chapter 1, we believe that the object-oriented approach not only has many advantages but it also represents the most recommended paradigm when modelling a system. We believe that most opinions against object-orientation come from misunderstanding some basic concepts related to the object-oriented paradigm. It is interesting to note that already in early frameworks such as FORMES (see 2.3.3) the benefits of object-orientation were reported. In [Rodet and Cointe, 1991] the author, Xavier Rodet, states that ``object-orientation matches most of the requirement of Music Composition Synthesis''.

The case of Max is a clear example of this. The authors state that Max is not a dataflow language but rather represents an object-oriented approach [Puckette, 1991a]. Nevertheless they do not use an object-oriented programming language like C++ and decide to base their developments in the C language. Because of this they implement a limited form of polymorphism themselves (see [Puckette, 1991b]) although recognizing that this method has much overhead over using one in a language like C++.

The reasons for not using C++ do not appear clear but two reasons are at least outlined. On one hand in [Puckette, 2002], Miller Puckette, the main author of Max argues against object-orientation in general saying that the transmissibility of practice and ideas does not depend of code reuse, and inheritance is essentially about code reuse. Although the author states that this is a general opinion in the computer science community, we cannot agree with him. The benefits of object-orientation in general and inheritance in particular are always set on two grounds: reuse of code and, most importantly, reuse of concepts. Reuse of code plays no role at all in very important mechanisms related to inheritance like abstract methods and classes or interfaces. Code reuse is a valuable side-effect of inheritance, but not its most important benefit. The author also states that ``(...) elegance matters but it has nothing to do with the creative adaptation of the underlying ideas''. Again, we cannot agree with this opinion. Elegance, understood as a property related to well-structured code and design, is indeed a crucial point for adaptation of any kind. In order to adapt an idea, a concept, or a piece of code, the first thing we have to do is to understand it. And it is much easier to understand an elegant idea, concept or a piece of code.

On the other hand in [Zicarelli, 2002] we find another reason. The author (David Zicarelli in this case) exposes a rather confusing view about dynamic binding. According to him Max uses dynamic binding while C++ uses static binding. We cannot accept this assertion. C++, as most programming languages, accepts both static and dynamic binding (see section 1.1.3 for an explanation on these concepts). As a matter of fact dynamic binding is the base of two of the four different kinds of polymorphism introduced in 1.1.3. It is true that C++, differently to other object-oriented languages like Smalltalk, uses strong typing, but that has nothing to do with the discussion. Symbols can be associated to functions on run-time and the mechanism implemented in Max, based on a pointer to a list of symbols, is simply a primitive version of C++ polymorphism and the pointer to virtual function table that is added to any polymorphic object.

The main drawbacks of the Max model that are reported by other music environments developers are the problems it presents related to extensibility and portability (see [Chaudhary et al., 1999] or [Cook and Scavone, 1999]). In this sense it is also interesting to note how the supposed dynamic features in Max are the reasons why some authors qualify it as a basically static environment. Stephen Travis Pope states in [Pope, 2004] that ``Max (...) provides some object-oriented features, including dynamically typed data, dynamic binding but no inheritance, limited data types, only one data structure for inter-object messaging and few control structures. Max has a static object structure and although jMax and Pd do something about it they are still generally static.''

The authors mention that there is a disagreement in the community whether Max should be considered object-oriented [Puckette, 2002]. Max uses the object-oriented message passing mechanism but not anything else. Because of this and all the reasons previously exposed it is our opinion that Max is not object-oriented.

On the other hand a clear difference between Max or similar environments and CLAM is that their design makes it difficult to process complex data structures. Max's focus is on real-time performance and therefore can handle very well simple control events but its design does not accept naturally other kind of data such as Spectrums and the like.

Finally possibly the biggest diference between Max and CLAM is that CLAM, as it has already been explained, is not a framework but rather a Visual Application.

Siren (see 2.4) is another example of a long standing and well-established paradigm that started with the MODE. Siren is a music oriented framework written in Smalltalk and heavily influenced by the ``Smalltalk-way-of-doing-things''. The motivation for creating the MODE and Siren was ``to build a powerful, flexible, and portable computer-based composer's tool and instrument''. It therefore represents a different approach from that of CLAM as it is completely music-oriented.

Siren presents a complete and mature object-oriented approach and it is indeed one of the main technical goals of the framework [Pope, 2001]. But the model that is used is not modelling a particular domain (music and music composition) but rather the author's point of view and requirements. In [Pope, 2001] he recognizes that although he is not the only user that he has listened to, this assertion is not completely false: the framework has evolved through many years to meet the needs of the author for a particular musical composition.

Now Siren has been extended to other applications rather than simply musical composition and it is also being ported to different languages (see more details in 2.4). The former possibility does not seem a plausible line of future. The author is promoting another more generic framework in parallel (see CSL in section 2.3.3) and we see this as the effort with most guarantees. Nevertheless, the fact that it is being ported to different languages opens up a completely different view as the framework model has always been very much related to that of Smalltalk.

Kyma represents a similar approach, it is not strange as it has also been developed on the Smalltalk environment. Kyma presents a rather extreme object-oriented musical model (see section 2.5) where everything is considered a sound object, from a single timbre to the structure of the whole composition or a transform of several sounds. This underlying model is defended in a rather romantic way as the core and fundamental issue in Kyma. The truth is that the final user has no access to the source code and, on the other hand, the graphical interface has evolved in a completely different direction, now basically focusing on the temporal and graph-based views. The ``fundamental'' model is completely transparent to the final user and is not even mentioned in the program description (see [www-SymbolicSound, ])

It is our opinion that Kyma's object-oriented model is a clear example of overdesign. The conceptual model is not clear and, in some sense, is ``too much object-oriented''3.8, forgetting about the domain that is being modeled and the way that people understand it. There has been a clear tension between the author's model and what the users demanded. This has forced the graphical interface to grow in other directions in order to hide the model. It is clear not the way to go, as these modifications should also have to reach the underlying model.

The fact that CLAM has followed an evolutionary, application and user-driven development process leads to the fact that the framework has not been built around a pre-existing conceptual model but rather the model has emerged as a result of the development process. It has already been stated in 1.3.5 that it is our opinion that frameworks generate metamodels and not the other way around. As already noted, this approach is completely different from that of most musical environments (i.e. Kyma, Siren or Max), which depart from the author's model and build the environment around it.



Subsections
2004-10-18