Domain Independent Generative Modeling

1
Domain Independent Generative Modeling

• B. Kusy, A. Ledeczi, M. Maroti,
• P. Volgyesi

2
Modeling of complex software

• Complex software systems that are tightly coupled
  with a physical environment are difficult to
  maintain if they need to change frequently
• Solution generate the code of the final system
  from a centralized set of models enable
  evolution
• Approaches to modeling
• Utilize a single modeling language, designed for
  software modeling (UML)
• Tailor the single language to the needs of a
  specific domain, introduce domain-specific
  modeling language (MIC)

3
A need for extensibility

• Usability of both approaches greatly depends on
  the available tools
• visual model builder, model database manager,
  model analyzer, verifier, interpreter
  transforming abstract models into executables and
  run-time support libraries
• UML approach has an advantage a single
  toolset is sufficient
• MIC approach needs to provide such a toolset for
  each domain-specific language
• GME Generic Modeling Environment
• configurable environment to create
  domain-specific environments
• configurable visual model editor and database
  manager

4
Extensibility through metamodeling

• The cascading design
• Metamodeling tools are used to design a domain
  specific modeling environment.
• This customized environment is then used to
  develop the models of the system.

• Metamodel
• Formal model of DSML.
• Describes the syntax, semantics and presentation
  information of a modeling language using UML, OCL
  and aspects.

5
GME captures declarative models

• Declarative model fully describes the
  relationship between the modeling entities at a
  design time.
• This approach is inflexible in certain domains
• Large models with repetitive structure
• Adaptive systems

Hierarchical signal flow metamodel (SFMeta)
Signal flow example for parallel computing
number of convolutions vs number of available
processors
6
Modeling in a generative manner

• Specify the components of architecture and
  provide an algorithmic description how to
  generate the final architecture.

• What needs to be done?
• The modeling language SFMeta needs to be extended
  with new objects
• Implement an interpreter that transforms
  generative models into declarative models
• We need to avoid the problems that MIC faces
  provide a configurable toolset that would support
  all domain-specific modeling languages

Model transformation
Generative model
Declarative model
7
Outline of our solution for domain independence

• Provide a template metamodel capturing the
  components and relations that are required by
  generative modeling
• Use metamodel composition utilizing template
  metamodel
• Provide code generator/model interpreter that
  executes generative scripts and creates the
  declarative model

8
GenerativeMETA template
New objects
New abstract objects
9
Outline of our solution

• Provide a template metamodel capturing the
  components and relations that are required by
  generative modeling
• Use metamodel composition utilizing template
  metamodel
• Provide code generator/model interpreter that
  executes generative scripts and creates the
  declarative model

10
Metamodel composition example
GenerativeMeta lib
SFMeta lib
11
Metamodel composition example
4
GeneratorScript
GenerativeMeta lib
SFMeta lib
We use inheritance as a technique to achieve
metamodel composition. User needs to decide what
objects participate in generative constructs and
what models contain these constructs
12
Outline of our solution

• Provide a template metamodel capturing the
  components and relations that are required by
  generative modeling
• Use metamodel composition utilizing template
  metamodel
• Provide code generator/model interpreter that
  executes generative scripts and creates the
  declarative model

13
Code generator/model interpreter invocation

• Our generator/interpreter works in 2 stages
• First it traverses a model in the composed
  metamodeling language and based on generator
  scripts it generates code for all generators
• This generator code is then compiled and executed
  by the stage2 model interpreter
• The two stage process creates a new hierarchy
  containing only pure declarative models these
  can be used with the toolkit (interpreters,
  visual model editors,) of the original domain

Generator Execution
14
Conclusions

• We described an approach to generative modeling
  that employs scripts and architectural parameters
  to specify model structure
• Our main concern was reusability that was
  accomplished by metamodel composition and
  two-stage domain-independent model interpretation
• The Generative Modeling clearly supports and
  enhances the strengths of MIC and in particular
  the extensibility of GME
• http//www.isis.vanderbilt.edu/projects/gme/contri
  b/gme_generative.zip