Melanee and MetaDepth are two deep modeling tools based on a common set of core concepts such as the orthogonal classi cation architecture and deep instantiation. However, at the present time, they have di erent foci. MetaDepth is a textual modeling tool focusing on the needs of software developers while Melanee is a graphical modeling tool focusing on modeling language creation. The question arises, therefore, as to how di erent/similar the tools actually are and how their features relate to each other. This paper addresses this question by comparing the two tools and investigating the extent to which the modeling features they o er to users overlap.
Melanee [
As far as the data modeling features of the two tools are concerned, they seem
to be very similar despite their focus on di erent target audiences. They are both
based on the orthogonal classi cation architecture [
The paper is structured as follows. The next section clari es the terminology used in this paper since the two tools sometimes use di erent terms for the same concept. This is followed by a comparison of the tools' modeling architectures (Section 3), support for deep instantiation (Section 4), handling of attributes and edges (Section 5) and inheritance rules (Section 6). The paper then presents a small summary of the results of the comparison in a table (Section 7) and a discussion of their signi cance. Finally, Section 8 concludes with some closing remarks. 2
When comparing two modeling tools that sometimes use a di erent vocabulary for similar concepts, it is important to de ne the terminology used. Hence, this chapter highlights the terminological di erences in the two tools' avors of deep modeling and explains which terminology is adopted for the remainder of the paper.
The three core terms used to refer to modeling content in MetaDepth are Model, Node and Edge, while Melanee uses the terms Level, Entity and Connection to refer to the same basic concepts. Of these, the term Model probably has the potential to cause the most confusion. In MetaDepth a Model is the container for all model elements in a single classi cation level, whereas in Melanee the term Model, or more speci cally Deep Model, encompasses all classi cation levels (i.e. models). In Melanee, the term used to refer to a container for content at a given ontological Level is Level. To avoid confusion, in the remainder of the paper we use the term Level to refer to container for model content at a given ontological level, and Model for the union of all ontological levels.
There are also di erences between Melanee and MetaDepth in the terms used to refer to level content. In particular, model elements representing classes/objects in the problem domain are named Nodes in MetaDepth and Entities in Melanee, while model elements representing links/associations are called Edges in MetaDepth and Connections in Melanee. To avoid confusion, in the rest of the paper we use the MetaDepth terms { namely, Node and Edge. Both tools regard Nodes and Edges as Clabject, and employ the term potency to describe over how many subsequent classi cation levels a clabject can be instantiated. Hence, these two terms (Clabject and potency ) are used in the following without ambiguity.
Another di erence of terminology between MetaDepth and Melanee is the use of attribute and eld to refer to the same thing. Melanee uses the term Attribute, while MetaDepth uses Field. In the following, the term Attribute is used as it is widely used in the UML technology space. Finally, the MetaDepth literature uses a technical term, linguistic extension, for introducing new types into a model at levels which are not the most abstract. We also use this term with the same meaning in the rest of this paper.
Figure 1 shows canonical representations from the literature of the MetaDepth
architecture (Figure 1(a)) and Melanee architecture (Figure 1(b)) side-by-side.
The two gures show that both approaches are based on the orthogonal
classi cation architecture (OCA) in which one level-spanning meta-model classi es
the content of the ontological levels from a linguistic (i.e. tool) point of view. In
the MetaDepth version of the OCA (Figure 1(a)), the linguistic meta-model
is displayed on the left of the gure labeled Linguistic meta-model , and the
ontological levels are arranged vertically to the right and use the @-notation to
indicate their potency. The Melanee version of the OCA (Figure 1(b)) displays
the linguistic meta-model at the top of the gure labeled L2, while the ontological
levels are arranged horizontally and are labeled using the O pre x, e.g. O0. The
Melanee example additionally shows the real world which is modeled at level L0.
This real world exists in the MetaDepth architecture too but is not displayed
in Figure 1(a). In both architectures, model elements are not only classi ed by
their linguistic types but can also be classi ed by an ontological type residing
at a higher ontological level. Multiple ontological classi cation is not allowed in
MetaDepth but is allowed Melanee. In [
Linguistic meta-model Type potency:int supers * * lcag fceO iltoo tan on isn* Instance Clabject
* Field (a)
Task name@1: String start: Date iiifttssccaegnnu»O snuEabmvjaeelcu=ta“:etSivoiotannrnils:nuttoaTgalnoaticgsoeikcnOa”fl @1 LL12 n li «
ionnsttoalnocgeicOafl @0 exam: Evaluation subject=“Maths” start=“22/10/12”
L0
Level * Clabject content potency O0
O1
Task2 name1:String start2:Date ???
Element name
Feature * feature durability
name0:String=evaluation start1:Date subject1:String
Evaluation (b) Method Attribute mutability O2
subject0:String=Maths start0:Date=22/10/13
It can be observed that both linguistic meta-models share the concept of Clabject which is a concatenation of the words Class and Object. This term re ects the type instance duality of model elements residing in the middle classi cation levels. They are at the same time instances of their types at the ontological level above and types for their instances one ontological level below. In the MetaDepth version of the OCA, this dual nature of clabjects is explicitly modeled by de ning Clabject as a subclass of both Type and Instance.
For one ontological type to conform to another in these architectures, certain conditions have to be ful lled by the instances. The set of attributes of the instance has to conform to the set of attributes de ned by the ontological type, the connection cardinalities de ned by the ontological type have to be satis ed by ontological instances, and the rules for potency have to be satis ed. These three criteria are the same in MetaDepth and Melanee. 4
Melanee and MetaDepth are both based on deep instantiation. MetaDepth allows a potency for levels, clabjects, and elds to be con gured while Melanee allows potencies to be de ned for clabjects, operations, attributes and their values. The fact that levels do not have potencies in Melanee has the advantage that an unlimited number of classi cation levels can be de ned without thinking of the deepness of a deep model when assigning a potency to the initial classi cation level. However, it has the disadvantage that potencies have to be de ned for each clabject in the level. Futhermore, MetaDepth allows models (i.e. levels) to be imported into other models for reuse. In Melanee only level content such as clabjects and packages can be linked.
The linguistic meta-model element, Operation, shown in the linguistic metamodel of Melanee in Figure 1(b) is not available in MetaDepth. Hence, in MetaDepth only attributes can be added to clabjects.
Potencies facilitate deep instantiation [
The rules for potency reduction during instantiation are identical for Melanee and MetaDepth. Instances always have a potency that is one lower than the potency of their type. Model elements with a negative potency cannot exist, and so, model elements with a potency value 0 cannot be instantiated. They therefore correspond to concepts such as objects and slots, when compared to the UML technical space.
Both tools allow instantiation over an arbitrary number of levels using socalled star potency. When instantiating a model element with star potency, the instances can have a star potency themselves or a potency with a non-negative Integer value. This is an advantage when de ning frameworks in the form of a modeling language for instantiation. In such a scenario, a modeler often does not know in what environment the framework will be used and thus cannot de ne appropriate constraints on the instantiation of the provided types.
Strict meta-modeling [
Linguistic extension, the ability to add new types at any intermediate level without being classi ed by an ontological type, is supported by both tools. 5
There are some signi cant di erences between MetaDepth and Melanee when it comes to the de nition of attributes and edges. In MetaDepth an attribute does not have to be present (repeated) at all intermediate levels. The only level at which an attribute is mandatory is the lowest level, where it plays the role of a slot. However, attributes can also be added to intermediate levels to hold default values. In Melanee, an attribute has to be present throughout the whole instantiation tree of a type. Melanee, in contrast to MetaDepth, allows nergrained control of attributes by assigning a value potency (mutability ) to them in addition to the attribute potency (durability ). The durability de nes over how many subsequent levels an attribute can exist (i.e. endure) and the mutability de nes over how many levels the value of an attribute can change. Together they allow modelers to capture precise details about the ownership, visibility and changeability of attributes across classi cation levels, akin to features such as tags, tag values, statics variables and constraints in the UML space. For example it is possible to de ne that an attribute belongs only to the type facet of a clabject and thus is not derived by its instances (by assigned it durability 0), or that an attribute's default value de ned at the class level cannot be changed by its instances (by assigning it mutability 0).
MetaDepth o ers richer functionality than Melanee when it comes to connecting clabjects with each other. Melanee only allows edges to be represented by connection classes, where each Edge is a fully edged clabject containing attributes, operations and connection ends identifying the connected clabjects. MetaDepth goes beyond this by also allowing edges to be represented by simple references which are essentially attributes whose value is a reference to another clabject.
Both approaches support connection diversi cation [
MetaDepth supports deep references (Figure 2) which are not available in Melanee. Deep references help when nodes need to be connected but the direct types are not known at the time of modeling. In Figure 2, for example, a Page
Page
* C omponent p1: Page
mapTable: TableVisualizer
@1 TableVisualizer: …
JavaScript
DEEP LANGUAGE DEFINITION
DEEP LANGUAGE
USAGE
DEEP LANGUAGE
USAGE instance shall be connected with a Component instance. Component however is instantiated at one more level in contrast to Page resulting in potency 0 Page instances being connected to indirect potency 0 instances of Component. In the example Component is instantiated as TableVisualizer . The Page instances shall be connected with instances of this direct instance of Component. Deep references in MetaDepth support the creation of such references whose direct type is not currently known. By assigning @0 to the reference between Page and Component it is declared that Page instances shall be connected with indirect Component instances of potency 0.
The multiplicity (cardinality) semantics of MetaDepth restricts the number
of instances of a connection at the level below. Each instance then de nes its own
multiplicity for the following level. In Melanee, the multiplicity constraint of the
type level is satis ed if the sum of the multiplicities of all connection instances
is in the range of the multiplicity de ned at the type level. Hence, a
multiplicity constraint not only constrains the immediate level below but all following
ontological classi cation levels. Approaches providing ner grained control of
multiplicity to Melanee are also feasible as described in [
Enumerations can be de ned in both Melanee and MetaDepth. Melanee allows an enumeration to be de ned in a deep model and then used by clabjects across all classi cation levels. MetaDepth de nes enumerations within levels which can then be used at all subsequent levels. 6
Melanee and MetaDepth follow similar rules concerning inheritance. Subtypes must have a potency equal to or higher than their supertype. Attributes can be repeated in subtypes in order to override values de ned in supertypes. Inheritance relationships can exist at all levels including intermediate levels and the most concrete level. Hence, inheritance works for both type and instance facets of clabjects, enabling a sort of prototype-based modelling. Moreover, multiple inheritance is supported by both approaches.
When it comes to typing of participants in an inheritance relationship Melanee and MetaDepth di er. Unlike MetaDepth, Melanee allows supertypes g EmployeeType OnlineMarketing : C i v i l S e r v a n t f name="Online Marketing " ; (a) (b) which do not have an ontological type. MetaDepth always forces a supertype to have an ontological type which is compatible with the ontological type of its subtypes. Compatible means that the ontological type of the supertype has to be equal to, or a supertype of, the ontological type of the subtype. Hence, the Melanee inheritance example shown in Figure 3(a) is not allowed in MetaDepth. To make the example valid in MetaDepth CivilServant would have to be assigned EmployeeType as its ontological type as shown in the MetaDepth version of the example in Figure 3(b).
This restriction prevents the de nition of abstract supertypes with a potency of 0 if the subtypes have ontological types in MetaDepth. In Melanee this technique is commonly used to represent abstract classes, especially in the context of the power type pattern. In MetaDepth, an abstract class always has to be ontologically typed and have a potency one less than the potency of the ontological type. In most cases this new potency is di erent to 0 for instances residing in middle classi cation levels. To overcome this problem the abstract keyword is introduced in MetaDepth's textual syntax as shown in Figure 3(b).
The di erence between MetaDepth's programming language orientation and Melanee's structural modeling orientation can be seen in the example in Figure 4. MetaDepth does not allow inheritance in the middle level of the example where the supertype (A1 ) is an ontological instance of a di erent ontological type, A, to the subtype (B1 ), which is of type B. The problem is that A and B are not related by inheritance at the top ontological level. MetaDepth prevents this because it is possible that a constraint written on A is applied to instances of B in such a model as shown in Figure 4. Such behavior is not easily foreseen by programmers and is thus prohibited in MetaDepth in favor of simpler and more reliable constraint de nition. Melanee, however would alA1: A
B1: B a: A1
b: B1
low such a situation since B could be assumed to be a subtype of A in a case
where the sets of instances are overlapping and hence it is correct to have the
same constraint also applied to instances of B. Melanee's reasoning services [
In addition to pure generalizations as supported in MetaDepth, Melanee adds the ability to de ne generalization sets which can mark the instance set as complete/incomplete and overlapping/disjoint. Furthermore, names (a.k.a discriminants) can be given to these generalization sets. 7
In the previous sections the languages of Melanee and MetaDepth were compared in terms of their implementation of potency, their handling of attributes and edges, classi cation semantics, and their semantics for inheritance. The results are summarized in Table 1 and Table 2. This shows that the number of common features (highlighted cells) roughly matches the number of di erences in all four of these areas.
Leap Potency y cnStar Potency teStrict Meta Modeling o P Potency Declaration
This paper has presented a comparison between the Melanee and MetaDepth approaches to deep modeling and revealed that even though the two languages might look quite similar at rst sight and share a lot of common ideas, they have signi cant di erences in their detailed terminology and semantics. The next step is to extend the comparison to include a broader range of deep modeling tools in order to create a concrete set of criteria for comparing them. Once such an exhaustive set of features has been de ned it can be used as the basis for empirical studies into which language features should be targeted to which user groups and for de ning guidelines for choosing deep modeling tools. We are working on building transformations between the two tools to bridge their semantic di erences. We hope this paper may serve as a starting point for providing criteria to compare deep modeling tools, and encourage other researchers to compare their tools to MetaDepth and Melanee, providing further comparison points. Acknowledgements. Work partially sponsored by the Spanish MINECO (TIN2014-52129-R), and the Madrid Region (S2013/ICE-3006).