Transformation Patterns

1
Transformation Patterns

• Al, Jean, Jeff, Kerry, Krzysztof, William, ??

2
Transformation Patterns

• What is a transformation pattern?
• common patterns I have seen
• structures

3
Examples

• MOF 1.3 ? EMF
• more constructs in MOF than in EMF
• want bi-directional (EMF ? MOF 1.3)
• e.g. MOF constraint into EMF annotation
• annotation had information on how to go back to
  EMF
• Details
• a) Normal EMF ? MOF is pretty easy
• b) But if the annotations have been changed,
  completing the round-trip is a serious problem
• Only allow round-trip if it is consistent
  otherwise (a)

4
More Examples

• UML to MOF
• UML XMI transform to MOF XMI
• example of a tool writing in ATL
• UML Activity to BPEL
• MOF to JMI
• CGI to EJB
• JavaDoc
• Genealogy data

5
More Examples

• Cross-compilation
• OQL to SQL
• etc
• Databases
• Normalizations
• Logical to physical

6
Common pattern
B
A
C
Y
X
Z
7
Concrete ExampleDatabase Logical ? Physical
Class
Table
Attribute
Column
8
Characterizing Transformationsas a whole

• 7) HOT (higher order transformation)
• 6) Ma ? Mb (nothing)
• 5) Ma ? Mb Exists MMa MMp MMb
• 4) Ma ? Mb MMa contains MMb
• 3) Ma ? Mb MMa in MMb
• 2) Ma ? Mb MMa MMb
• 1) Ma ? Ma (within same model)

9
Vertical vs. Horizontal Transformations
10
Kinds of Transformation

• Property-Preserving
• property P we have P(Ma) implies P(Mb) when Ma ?
  Mb
• Evolution
• platform MaMMa ? MaMMa
  semantics-preserving Meaning(Ma) Meaning (Mb)
• functional (change in requirements)
• MaMMa ? MbMMa
• Compilation
• generally semantics-preserving
• Optimization
• one that is P-preserving, Q-reducing
• Refactoring
• changing structure, semantics preserving
• Its like also optimization (Q is -ility or
  niceness metric)

11
Source ? Target

• Cardinality
• all cases
• 11 1M M1 MM
• Effect at type level
• Does it increase or decrease number of types
• For case ?M
• Homogeneous
• Heterogeneous

12
Explosion/Contraction

• State Explosion
• Two states
• adding intermediate states into a transition
• Permutation
• Elaborating all combinations of information
• Version explosion
• multiple models coming in that are variants of
  each other
• Like Query languages
• projection
• selection
• order
• transitive closure

13
Philosophical Interlude

• What is different from Functional programming?
• In functional programming
• main structure is a tree
• In Model-driven approach
• main structure is a graph
• moving from trees to graphs gives more power
• (but text can be better than pictures for some
  uses)
• All the meta-models are OO
• Working with OO structures graphs