Download presentation
Presentation is loading. Please wait.
1
Mining Metamodels From Instance Models: The MARS System Faizan Javed Department of Computer & Information Sciences, University of Alabama at Birmingham The Fourth Annual Southeastern Software Engineering Conference (SE) 2 30 th March 2005
2
Motivation Software artifacts like models and source code conform to a schema, and are stored in a repository. Evolution of schema might be required to address new feature requests. Repository artifacts might become obsolete if not transformed to conform to new schema. Example: Java SDK – new versions restructure or rename API’s.
3
Domain-Specific Modeling (DSM) Raises level of abstraction, while narrowing design space to single domain of discourse. Allows construction of models which follow domain abstractions and semantics allowing developers to work with domain concepts. Metamodel: Defines key elements of domain. Models: Defines specific configurations of the domain.
4
DSM modeling with GME GME: Generic Modeling Environment Metamodel: Networking
5
DSM Modeling with GME Model: CompanyA
6
Challenges of Mining Domain Instance Models Metamodel drift: inability to load models due to changes to metamodel. Solution: Infer metamodel from instance models ! Grammar Inference community has done extensive work on similar problem, albeit for programming language domain.
7
Challenges of Mining Domain Instance Models Idea: Apply grammar inference techniques to the metamodel inference problem. Problem: Modeling tools export XML files; mismatch in representation expected by grammar inference techniques. Solution: Translate XML to textual DSL (Domain-Specific Language) !
8
Tools Used in the Project GME: Metamodel described with UML class diagrams, and constraints with OCL. LISA: An interactive environment where users can specify, generate, compile, and execute programs in a newly specified language. DMS: A program transformation system and re-engineering toolkit.
9
Overview of the MARS system
10
From GME models to MRL (Model Representation Language) ……. model NetDiagram { WSGroup ; Perimeter ; Host ; Network ; WSGroup ; Host ; Router ; fields; connections Connection : Port -> Network ; Connection : Host -> Network ; Connection : Port -> Perimeter ; Connection : WSGroup -> Network ; Connection : Host -> Network ; } ……….. XSLT
11
From MRL to Inferred Metamodel ……. model NetDiagram { WSGroup ; Perimeter ; Host ; Network ; WSGroup ; Host ; Router ; fields; connections Connection : Port -> Network ; Connection : Host -> Network ; Connection : Port -> Perimeter ; Connection : WSGroup -> Network ; Connection : Host -> Network ; } ……….. LISA+DMS
12
Inferred vs. Original Metamodel
13
Limitations and Observations Generalization hierarchy can’t be inferred resulting in more elements in inferred metamodel. Domain-Specific Visualization: graphic assigned to metamodel entities cant be inferred. OCL Constraints: capture domain semantics that cant be captured with static diagrams. They are not explicitly indicated in domain models.
14
Related Work The XTRACT System: 1) Infers DTD from XML documents 2) Derives a regular expression for each element in the XML document. 3) Uses the Minimum Description Length (MDL) principle to choose the best DTD from a set of candidate DTDs o ECFG Based System (Chidlovskii): 1) Represents XML documents as structured examples of an unknown ECFG. 2) Uses existing grammar inference techniques to infer the ECFG.
15
For more information: Project Website: http://www.cis.uab.edu/softcom/GenParse/mars.htm
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.