Toward The Next [ Next [ Next

1
Toward The Next Next Next Generation
ofMeta-Modeling Tools

• Eric Engstrom (eric.engstrom_at_honeywell.com)
• David Oglesby (david.oglesby_at_honeywell.com)
• Kirk Schloegel (kirk.schloegel_at_honeywell.com)
• Honeywell LaboratoriesHoneywell International,
  Inc

2
The Development Problem

• Current system development is high-risk
• Very complex
• Multiple domains
• Ambiguous requirements
• Subject to rapid change
• Building atop ever-more expansive APIs
• The middleware trend only complicates this
• Meta-programming promises the same

3
Tools to the rescue

• Common Mantra
• Lets use a tool to do ___ for us.
• Build models of systems, fostering
• Higher-level descriptions
• Rapid understanding and evolution
• Separation of concerns
• Generation of code and documentation
• Caveat Fool Tool Fool

4
Model-Based Development
5
Model Based Development Options

• Two options
• Use a generic tool/language applicable for any
  domain
• e.g. UML, SDL, IDEF, etc
• OR
• Use an application or domain-specific tool
• Often limited domain
• Sometimes lacking a large user-base
• Yeilding ? NO COTS!

6
Generic Tool Option

• A Development Solution
• Got Models? Sure!
• Often lack clear semantics
• Without clear semantics, cannot realize full
  benefit of Model-Based Development
• Impose our own semantics
• Effectively ad-hoc
• Often lacking much tool support

7
Semantics Issue 2

• Anything in a model that does not directly
  relate to code can be and will be
  misinterpreted.

8
Domain-Specific Option (aka DSVL)

• Mantra Becomes
• Lets build our own tool to do ____ for us.
• DSVLs are inherently Model-Based
• Well suited to our domain
• Clear semantics
• Address platform/API issues
• Auto-generate code

9
The DSVL Development Problem

• Building DSVLs suffers from the same problems as
  any development effort
• Often complex
• Multiple domains (aka Aspects)
• Ambiguous requirements
• Subject to rapid change
• PLUS
• Reinvented infrastructure
• Lacking clear development methodology

10
The DSVL Solution Meta-Models

• If DSVLs are so great, how about we come up with
  a DSVL for DSVLs?
• Capture the essense of a DSVL
• Provide infrastructure for strong modeling
  semantics
• Encourage a Model-Based view
• Support linkages to external tools
• Codify the generation of code/artifacts

11
DSVL Meta-Modeling Perspective

• Meta-Models can be seen as as DSVL for DSVLs
• Examples of tools for this include
• DOME (Honeywell)
• GME (Vanderbilt University)
• MetaEdit (MetaCASE)
• All allow user to
• Specify logical boxes arrows diagrams
• Provide code/artifact generation support

12
Meta-Modeling for DSVLs Issues

• Multi-Domain/Multi-Aspect Modeling
• COTS Tool Integration
• Complete Code / Artifact Generation
• Model Reuse

13
Multi-Domain/-Aspect Modeling

• Complex systems involve more than one domain,
  needing to
• capture the cross-domain components of systems
• retain domain specific tooling support
• generate artifacts across domains / aspects

14
Multi-Aspect Modeling Example
Archetypes Patterns
Thread Model
Software Model
Hardware Model
Resource Model
Event Model
Data Flow Model
Service / Contract Model
15
Multi-Aspect Modeling Ideas

• Build larger domain meta-models
• Combine all meta-models into one Union Model
• Lose some domain-specificity
• Evolution of domain meta-models difficult
• Build Type-System view of meta-models
• HARD and potentially overly complicated
• Build layered, interacting meta-models
• Represent cross-cutting information
• Allow new aspects of whole System Meta-Model to
  be added dynamically

16
Meta-Modeling for DSVLs Issues

• Multi-Domain/Multi-Aspect Modeling
• COTS Tool Integration
• Complete Code / Artifact Generation
• Model Reuse

17
COTS Tool Integration

• Companies have large investments in commercial
  tools (e.g. Rose)
• (licenses and training)
• Portfolio of work
• DSVLs need to leverage this investment to be
  accepted
• share information in both directions
• cooperative code generation
• allow parts of system to be specified in COTS
  tools

18
COTS Tool Integration Ideas

• Build a generic model repository
• All tools must store into that
• Use common interchange formats
• XML/XMI?
• Use APIs for peer to peer?
• CORBA? COM? SOAP?
• OMGs MDA?

19
Meta-Modeling for DSVLs Issues

• Multi-Domain/Multi-Aspect Modeling
• COTS Tool Integration
• Complete Code / Artifact Generation
• Model Reuse

20
Code / Artifact Generation

• Biggest hurdle to developing DSVLs
• Reuse
• Can generators (or parts of generators) be shared
  or reused?
• Automation
• e.g. code generator generators
• How to cope with COTS tools and multiple domain
  systems?
• Share responsibility with COTS

21
Example Transformation Process
Meta-Model(s)
Artifact Template(s)
Type Inference
Flow Analysis
Template Selection
Template Interpretation
templates and outputs are also models
procedure Foo(Asome_type) is begin A1
B2 Lib.Invert(A,B) for I in 1..B loop
Lib.Prefrobulate(A,B) end loop end Foo
generated artifact
22
Code / Artifact Generation Ideas

• Graph Rewriting
• Capture input and output patterns
• Search and completion issues
• Q Is this easy enough for acceptance?
• XML / XSLT
• Use XML Schemas as Meta Models
• XML ? XML via XSLT is transformation
• Code or documentation is just another output of
  an XSLT transform
• Q Is XSLT sufficient?

23
Meta-Modeling for DSVLs Issues

• Multi-Domain/Multi-Aspect Modeling
• COTS Tool Integration
• Complete Code / Artifact Generation
• Model Reuse

24
Model Reuse

• DSVLs move system design to models
• Just as we reuse code,we need to reuse
  portions of models
• Reuse within a domain
• Model library containing parts of models
• Reuse across domains
• Model library with hooks into different domains?
• Via model transformations?

25
Archetypes Model-Based Patterns
Example Publish/Subscribe Pattern with
Watchdog as an Archetype
Application-Level QoS Parameters
sending object
Max. comm. error prob. ? Timeliness at recv
? Sustained rate ? Safety Impact ?
monitored comm. channel
receiving object
translate application QoS

• Model reuse
• Flexibility
• Extensible code generation
• Explicitly modeled structure

watchdog thread
notify
error-handling object
26
Conclusions

• Meta-Models are like DSVL for DSVLs
• DSVLs then include generic support for generation
  and integration
• Meta-DSVL toolkits still suffer from
• Lack of interaction among DSVLs
• Poor integration with COTS tools
• Difficulty of creating code generators
• Limited of reuse of models and model elements

27
Resources / References

• DOME
• http//www.htc.honeywell.com/dome
• GME
• http//www.isis.vanderbilt.edu/projects/gme
• MetaEdit
• http//www.metacase.com
• OOPSLA Domain Specific Visualization Workshop
  (2002)
• http//www.cis.uab.edu/info/OOPSLA-DSVL2
• Meta-Modeling Resources
• http//www.metamodel.com