A Generative Approach to Model Interpreter Evolution Jing Zhang, Jeff Gray, and Yuehua Lin

{zhangj, gray, liny} @ cis.uab.eduDept. of Computer & Information Sciences

University of Alabama at Birmingham

http://www.cis.uab.edu/softcom/

Funded by the DARPA Information Exploitation Office (DARPA/IXO), under theProgram Composition for Embedded Systems (PCES) program

Evolution of models and interpreters in terms of meta-model changes

∆MM: The changes made to the meta-models∆M: The changes reflected in the domain models∆I: The changes reflected in the model interpreters

Interpreter1

Model1

Meta-model1

Define

Interpret

Interpretern

Modeln

Meta-modeln

Define

Interpret

Interpreter0

Model0

Meta-model0

Define

Interpret

∆M 1

∆MM 1

∆I 1

∆M 2

∆MM 2

∆I 2

∆M n

∆MM n

∆I n……

……

……

Based

on

Example: Old/New Metamodel and model

Example: Old/New InterpreterCBuilderAtom *GetStartState( CBuilderModel *StateDiagram) { CBuilderAtom *startstate = null; const CBuilderAtomList *states = StateDiagram->GetAtoms("State"); POSITION pos=states->GetHeadPosition(); while(pos) { CBuilderAtom *st = states->GetNext(pos); CBuilderConnectionList *cons = st->GetInConnections("Transition"); if ( cons == null ) if ( startstate == null) startstate = st; else <<ERROR: more than one state has no InConnections>> } ASSERT ( startstate !=null ); return startstate;}

CBuilderAtom *GetStartState( CBuilderModel *StateDiagram)

{ const CBuilderAtomList *startstates = StateDiagram->GetAtoms("StartState");

ASSERT(startstates->GetCount()==1);

CBuilderAtom *startstate = startstates->GetHead();

return startstate;

}

Technical Challenges

Lack of formally-written model interpreter Different developers may program interpreters in various ways Hard to maintain and evolve such subjective realizations of

model interpreters Lack of formal specification for metamodel transformation

Metamodel transformation specifications must include the entire knowledge for the underlying interpreter evolution

∆MM ∆I

Lack of support for parsing and invasively transforming program source code from higher-level models.

Utilize a mature program transformation engine: The Design Maintenance System (DMS)

?

Model Interpreter Evolution Architecture (MIEA)

void CComponent::InvokeEx(CBuilder &builder) { Interpreter aInterpreter; CString fileName; char *specFile=new char[fileName.GetLength()]; strcpy(specFile, fileName); …. }

Interpreters’

void CComponent::InvokeEx(CBuilder &builder) { Interpreter aInterpreter; CString fileName; if(!aInterpreter.selectSpecAspects(fileName)) { return; } … }

Interpreters

Metamodel’Metamodel

modelmodel

Models

Model XformEngine

Model Xform Specification

Models’

modelmodel

Modeling API’Modeling API

DMS rewriting rules for evolving intepreter(1)

rule ChangeName (id:identifier): expression_statement -> expression_statement ="\id -> GetModels(\“State\");" -> "\id -> GetModels(\“StartState\");".

• Differences of names for any model entities, relationships and attributes

rule ChangeModelType (id:identifier): expression_statement -> expression_statement = “\id -> GetAtoms(\“State\”);” -> “\id -> GetModels(\“State\”);”.

• Differences of model types

DMS rewriting rules for evolving intepreter(2)

rule ChangeAttrType (): declaration_statement -> declaration _statement = “CString State_Text;” -> “int State_Text;”.

DMS rewriting rules for evolving intepreter(3)• Differences of attribute types

Conclusion Ideal Goal:

Support the (semi)-automation of model interpreter evolution in terms of metamodel changes and modeling API changes

Proposed Solution: Model Interpreter Evolution Architecture

Obstacles: Informal model interpreter Hard to maintain the fidelity mapping of high-level abstract

models to the low-level source code Possible solution approaches to be investigated:

Attach program rewriting rules to the existing model transformation specification

Develop a new high-level specification for model transformation that can generate the rewriting rules

Investigate an intelligent model comparison technique to generate the rules through human interactions