Juan Carlos Navarro, Jaime Chavarriaga. Using MS Solver Foundation to analyze Feature Models and Configurations EATIS 2016. 2016

1. Juan Carlos Navarro
Jaime Chavarriaga
Using Microsoft
Solver Foundation to
analyze Feature
Models and
Configurations
1

2. Context
• Feature Models are the de facto standard for
documenting, checking and reasoning about the
configurations of a software system
• Well-known Applications:
• Domain Analysis in Software Product Lines
• Engineering of Product Configurators
2

3. Domain Analysis for Product Lines
MS Office
It is a Software
Product Line
Multiple products are
built based on a
predefined set of
assets (components)
3

4. Domain Analysis for Product Lines
MS Office
4

5. Domain Analysis for Product Lines
Feature Model
Represents the
variability of the
products
5

6. • C1 = { Word, Excel, Powerpoint, OneNote}
• C2 = { Word, Excel, Powerpoint, OneNote , Access}
• C3 = { Word, Excel, Powerpoint, OneNote , Access, Publisher}
• C4 = { Word, Excel, Powerpoint, OneNote, Outlook}
• C5 = { Word, Excel, Powerpoint, OneNote , Access, Outlook}
• C6 = { Word, Excel, Powerpoint, OneNote , Access, Publisher, Outlook}
Domain Analysis for Product Lines
Feature
Configuration
Represents a
selection of features
6

7. Domain Analysis for Product Lines
Variability
Analysis
There are features
common to all the
products and
features that are
optional
7
Commonalities Variabilities
• C1 = { Word, Excel, Powerpoint, OneNote}
• C2 = { Word, Excel, Powerpoint, OneNote , Access}
• C3 = { Word, Excel, Powerpoint, OneNote , Access, Publisher}
• C4 = { Word, Excel, Powerpoint, OneNote, Outlook}
• C5 = { Word, Excel, Powerpoint, OneNote , Access, Outlook}
• C6 = { Word, Excel, Powerpoint, OneNote , Access, Publisher, Outlook}

8. Problem (I)
8
Given a Feature
Model for a
Family of
Products…
• Is this model valid?
• Are there constraints that cannot be satisfied at the
same time?
• All the features can be selected by a user ?
• Which features cannot be selected?

9. Related Work
There are some proposals to analyze feature models
Approach Representative works
Propositional Logic Convert FM to
propositional logic
expressions
Batory[2005],
Czarnecki[2005],
Gheyi[2006],
Mendonca[2010]
CSP Convert FMs to Constraint
Satisfaction Problems
Benavides[2005],
Karatas[2010],
Alvarez[2011],
Mazo[2011]
Description Logics Convert FMs to
Ontologies
Zaho[2004],
Fan [2006],
Wang [2007]
Others Use other processing
techniques
Mannion[2002],
Janota[2007],
Gheyi[2008],
Van der Broek [2010] 9

10. Related Work
Existing tools and libraries
Scope Technology
SPLOT Supports feature model and
configuration validation. Also it
supports configuration processes.
Java
FAMA Supports feature model and
configuration validation. Partial
supports configuration processes
Java
FAMILIAR Provides a DSL to process feature
models. Support validation and
configuration processes.
Java
10

11. Problem (II)
• Existing works cannot be used in .Net Platforms
• Existing works do not use CSP + Optimization
Solvers such as the MS Solver Foundation.
11

12. How can we analyse
Feature Models and Configurations
using
Microsoft Solver Foundation?
12

13. Our Approach
❶ Understand semantics for
Feature Models and Configurations
❷ Take the Models and Configurations and
Convert them to CSP/Optimization Problems
❸ Take these CSP/Optimization problems and
Solve it using MS solver Foundation
❹ Provide an
implementation and evaluation
13

14. ❶ Feature Models and
Configurations Semantics
14

15. Feature Model Semantics
15
Root
X
Y
X
Y
Root (Concept).
It must be selected
Mandatory.
If X is selected,
Y must be selected
Optional.
If X is selected,
Y can be (or not)
selected
X
Y1 Yi…
Y1 Yi…
X
X Y
X Y
Or-Group.
If X is selected, one or
more of Y1, .., Yi
must be selected.
Alternative-Group.
If X is selected, only one
of Y1, .., Yi
must be selected.
Requires
If X is selected, Y
must be selected.
Excludes
If X is selected, Y
must be not selected.

16. e.g., a FM for Cell Phones
16
Cellular Phone
Normal
Touch
Screen
External
Memory
KeyPad Small
Size
Large
Size
LCD
Input
Device
Battery
excludes
requires
Stylus

17. Feature Configuration
17
C1 = { CellularPhone, LCD, TouchScreen, InputDevice, Stylus,
Battery, LargeSize}
valid
Cellular Phone
Normal
Touch
Screen
External
Memory
KeyPad Small
Size
Large
Size
LCD
Input
Device
Battery
excludes
requires
Stylus

18. Feature Configuration
18
C1 = { CellularPhone, LCD, TouchScreen, InputDevice, Stylus,
Battery, SmallSize}
invalid
Cellular Phone
Normal
Touch
Screen
External
Memory
KeyPad Small
Size
Large
Size
LCD
Input
Device
Battery
excludes
requires
Stylus

19. ❷ Converting Feature Models
and Configurations to
CSP/Optimization Problems
19

20. Converting a Feature Model
20
Root
X
Y
X
Y
Root (Concept).
𝑟 = 1
Mandatory.
𝑥 = 𝑦
Optional.
𝑥 ≥ 𝑦
X
Y1 Yi…
Y1 Yi…
X
X Y
X Y
Or-Group.
𝑥 = 1 → 𝑦𝑖 ≥ 1
Alternative-Group.
𝑥 = 1 → 𝑦𝑖 = 1
Requires
𝑥 ≤ 𝑦
Excludes
𝑥 + 𝑦 ≤ 1

21. e.g., a FM for Cell Phones
• 𝐶𝑒𝑙𝑙𝑃ℎ𝑜𝑛𝑒 = 1
• 𝐿𝐶𝐷 = 𝐶𝑒𝑙𝑙𝑃ℎ𝑜𝑛𝑒
• 𝐿𝐶𝐷 = 𝑇𝑜𝑢𝑐ℎ𝑆𝑐𝑟𝑒𝑒𝑛 + 𝑁𝑜𝑟𝑚𝑎𝑙
• 𝐼𝑛𝑝𝑢𝑡𝐷𝑒𝑣𝑖𝑐𝑒 = 𝐶𝑒𝑙𝑙𝑃ℎ𝑜𝑛𝑒
• 𝐿𝐶𝐷 = 1 → 𝐾𝑒𝑦𝑃𝑎𝑑 + 𝑆𝑡𝑦𝑙𝑢𝑠 ≥ 1
• 𝐶𝑒𝑙𝑙𝑃ℎ𝑜𝑛𝑒 ≥ 𝐸𝑥𝑡𝑀𝑒𝑚𝑜𝑟𝑦
• 𝐵𝑎𝑡𝑡𝑒𝑟𝑦 = 𝐶𝑒𝑙𝑙𝑃ℎ𝑜𝑛𝑒
• 𝐵𝑎𝑡𝑡𝑒𝑟𝑦 = 𝑆𝑚𝑎𝑙𝑙𝑆𝑖𝑧𝑒 + 𝐿𝑎𝑟𝑔𝑒𝑆𝑖𝑧𝑒
• 𝑁𝑜𝑟𝑚𝑎𝑙 + 𝑆𝑡𝑦𝑙𝑢𝑠 ≤ 1
• 𝑇𝑜𝑢𝑐ℎ𝑆𝑐𝑟𝑒𝑒𝑛 ≤ 𝐿𝑎𝑟𝑔𝑒𝑆𝑖𝑧𝑒
Cellular Phone
Normal
Touch
Screen
External
Memory
KeyPad Small
Size
Large
Size
LCD
Input
Device
Battery
excludes
requires
Stylus

22. ❸Solving CSP/Optimization
Problems using
MS Solver Foundation
22

23. Analysis Operations
Benavides et al. have defined some operations:
• Validating a Feature Model: there is at least one valid
configuration
• Validating a Product Configuration
• Validating a Partial Configuration
• Obtaining all the valid Product Configurations
• Calculating the number of Product Configurations
• Obtaining core features (mandatory)
• Obtaining dead features (non-selectable)
• Obtaining variant features

24. Operations: Validating a FM
ConvertFM( fm ) : ConstraintSystem
ConstraintSystem model;
model = ConstraintSystem.CreateSolver();
:
// introduces variables and constraints
:
return model;

25. Operations: Validating a FM
IsValid( ConstraintSystem model )
ConstraintSolverSolution solution = model.Solve();
return solution.HasFoundSolution;

26. Operations: Obtaining all Products
GetAllProducts( ConstraintSystem model )
List <Configuration> list;
list = new List <Configuration>() ;
ConstraintSolverSolution solution = model.Solve();
while ( solution.HasFoundSolution )
{
list.Add(new Configuration( solution ));
}
return list ;

27. Operations: Core Features
GetDeadFeatures( List<Feature> features, ConstraintSystem
model )
List <Feature> list = new List <Feature>();
foreach ( Feature f in features )
{
CspTerm constraint = model.Equal (0 , f);
model.AddConstraints ( constraint );
ConstraintSolverSolution solution = model.Solve();
if (! solution.HasFoundSolution )
{
list.Add(f);
}
model.RemoveConstraints( constraint );
}
return list ;

28. Operations: Dead Features
GetDeadFeatures( List<Feature> features, ConstraintSystem
model )
List <Feature> list = new List <Feature>();
foreach ( Feature f in features )
{
CspTerm constraint = model.Equal (1 , f);
model.AddConstraints ( constraint );
ConstraintSolverSolution solution = model.Solve();
if (! solution.HasFoundSolution )
{
list.Add(f);
}
model.RemoveConstraints( constraint );
}
return list ;

29. Operations: Validating a Config
IsValidConfiguration(Configuration config , List <Feature>
features , ConstraintSystem model )
foreach ( Feature f in features )
{
if ( config.Contains(f))
model.AddConstraints ( model.Equal (1 , f));
else
model.AddConstraints ( model.Equal (0 , f));
}
ConstraintSolverSolution solution = model.Solve() ;
return solution.HasFoundSolution ;

30. Operations: Partial Config
IsValidPartialConfiguration(Configuration config , List
<Feature> features , ConstraintSystem model )
foreach ( Feature f in features )
{
if ( config.Contains(f))
model.AddConstraints ( model.Equal (1 , f));
}
ConstraintSolverSolution solution = model.Solve() ;
return solution.HasFoundSolution ;

31. ❹ Implementation and
Evaluation
31

32. Implementation & Evaluation
Implementation
https://github.com/FaMoSA/fma.net
• A .Net based library
• Can load SPLOT, FeatureIDE and FAMA feature models.
• Can process feature models and configuration using MS
Solver Foundation
32

33. Implementation & Evaluation
Evaluation
• Test with: Fama Test Suite
http://www.isa.us.es/fama/?FaMa_Test_Suite
• Test with models from SPLOT
http://www.splot-research.org/
• Test with Feature IDE test suite
http://wwwiti.cs.uni-magdeburg.de/iti_db/research/featureide/
Our implementation performs the analysis correctly.
33

34. Validate Model Performance
34
It may take 10 – 26 milliseconds.
Other solvers take between 10 – 30 ms

35. All Products Performance
35
Depends on the # of products.
It invokes the solver for each product.

36. Dead Features Performance
36
Depends on the # of features.
It invokes the solver for each feature.

37. Conclusions
• Our approach
• Implements the operations defined by Benavides et al.
• Works on .Net platforms
• Obtains the same results that the other libraries.
• Performs as fast as the related solutions in Java.
37

38. Conclusions
• However, our implementation
• Does not support optimizations proposed by other
authors (e.g., by combining solvers or by eliminating
non-relevant features from the processing)
• Does not support transforming operations such as slice r
merge of feature models
• Does not support operations for product configuration
such as recommending or optimizing selections.
• Future work is planned to implement optimizations
and additional operations using the MS Solver
Foundation.
38

39. Questions?
Juan Carlos Navarro
juan-navarro@unipiloto.edu.co
Jaime Chavarriaga
ja.chavarriaga908@uniandes.edu.co
39