Domain-Specific Modeling raises the level of abstraction beyond programming by specifying the solution directly with domain concepts. Within product lines domain-specific approaches are applied to specify variability and then generate final products together with commonality. Such automated product derivation is possible because both the modeling language and generator are made for a particular product line — often inside a single company. We examine which kinds of reuse and product line approaches are applied in industry with domain-specific modeling. Our work is based on empirical analysis of 23 cases and the languages and models created there. The analysis reveals a wide variety and some commonalities in the size of languages and in the ways they apply reuse and product line approaches.

1. © 2019 MetaCase 1
Juha-Pekka Tolvanen, Steven Kelly
jpt@metacase.com
How Domain-Specific Modeling
Languages Address Variability:
Investigation of 23 cases

2. © 2019 MetaCase 2
Agenda
 Introduction
 Research question and method
 Results
 Conclusions and discussion

3. © 2019 MetaCase 3
Languages in product line
development
 Domain-Specific Modeling languages are applied in
industry as one technique to speed up development
 Language definition consist of:
– Metamodel
– Constraints
– Notation
– Generators
– (Tooling)

4. © 2019 MetaCase 4
Research question
How domain-specific
modeling languages
address variability?

5. © 2019 MetaCase 5
Empirical research method
 Analyze created domain-specific languages
– Industrial use
– Cover various kind of product lines
– 23 cases* in total
 Main criteria for selection: access to the language
– Language definitions available, applied in tooling
– Verification from language developers
 All cases done with MetaEdit+
– Tool does not influence to patterns as both full metamodel-
based and “light” customization (profiles) possible
* Not the same 23 as been reported in 2005 – also newer ones

6. © 2019 MetaCase 6
Domain Targets By Size Use Approach
1 Consumer electronics C, HTML, Docs 1 3 1
2 Industrial automation PLC, GUI, DB schema, net config, deploy 1 165 5 1
3 Enterprise applications C#, DB schema 1 6 1
4 Railway signaling Simulation, XML 1 291 6 1
5 Signal Processing Systems Matlab, simulation, XML 1 4 1
6 Oil drilling Cost calculation, documentation 1 3 1
7 Big data applications Java, JSON, CQL, SPARQL, SQL 2 397 4 1
8 Printing process Ruby, XML, Docs 3 55 4 1
9 System performance Gherkin, HTML, Docs 1 145 3 1
10 Consumer electronics JSON 3 72 2 2
11 Telecom service XML 1 61 2 2
12 Medical XML, audit documents, change history 1 63 1 2
13 High-level synthesis System C 1 450 3 2
14 Radio network TTNC-3, simulation/animation 1 5 2
15 An automotive system System specification 3 62 3 2
16 Database applications Java 3 46 1 2
17 Consumer electronics C, localization, docs 1 403 6 3
18 Automotive architecture Simulink, ISO26262 documents, AUTOSAR 3 652 6 3
19 Telecom C, build automation 2 109 3 3
20 Insurance Cobol, DB schema 3 234 6 4
21 Aerospace C#, XSD, JSON, API 2 121 1 5
22 Automotive ECU Python, JSON, Test document, change history 3 64 5 5
23 Software testing Propriety format of state machines 3 317 6 6

7. © 2019 MetaCase 7
Creator and size*
* UML 2.5 = 247

8. © 2019 MetaCase 8
Maturity

9. © 2019 MetaCase 9
Identified Approaches

10. © 2019 MetaCase 10
1. Each model and its elements are
for a single variant
 “Classical DSL”
 No reuse: each variant created and maintained
independently
 Clear process and
organizational
structure

11. © 2019 MetaCase 11
2. Reuse of models or model
elements across multiple variants
 Language has concepts for reuse
– variant 2 can use parts made for variant 1
 Reuse is restricted to the domain concept, e.g. service
 Whitebox or blackbox, have access/governance rules
 Apply different languages on different levels
Subaction
Variant 1 Variant 2

12. © 2019 MetaCase 12
3. Mark/filter/modify reused models
or elements for variants
 Commonly reused models contain variant data
 Responsibility over many variants or whole product line
Bottom up Top-down: e.g.
Sporty does not
have Down button
or stopwatch icon

13. © 2019 MetaCase 13
4. Core models and variant models
 Core team and variant team
– Both core and variants can co-evolve
 Tool and process applied to manage development
Core team Variant team

14. © 2019 MetaCase 14
5. Core models and languages for
restricted variation of core
 Language ensures legal variation
– Variation space is limited to specific aspects
Core team Variant team

15. © 2019 MetaCase 15
6. Multilevel: model elements
become language elements
 A model made by one team became language for others
– Language supports organizational structure and process

16. © 2019 MetaCase 16
Domain Targets By Size Use Approach
1 Consumer electronics C, HTML, Docs 1 3 1
2 Industrial automation PLC, GUI, DB schema, net config, deploy 1 165 5 1
3 Enterprise applications C#, DB schema 1 6 1
4 Railway signaling Simulation, XML 1 291 6 1
5 Signal Processing Systems Matlab, simulation, XML 1 4 1
6 Oil drilling Cost calculation, documentation 1 3 1
7 Big data applications Java, JSON, CQL, SPARQL, SQL 2 397 4 1
8 Printing process Ruby, XML, Docs 3 55 4 1
9 System performance Gherkin, HTML, Docs 1 145 3 1
10 Consumer electronics JSON 3 72 2 2
11 Telecom service XML 1 61 2 2
12 Medical XML, audit documents, change history 1 63 1 2
13 High-level synthesis System C 1 450 3 2
14 Radio network TTNC-3, simulation/animation 1 5 2
15 An automotive system System specification 3 62 3 2
16 Database applications Java 3 46 1 2
17 Consumer electronics C, localization, docs 1 403 6 3
18 Automotive architecture Simulink, ISO26262 documents, AUTOSAR 3 652 6 3
19 Telecom C, build automation 2 109 3 3
20 Insurance Cobol, DB schema 3 234 6 4
21 Aerospace C#, XSD, JSON, API 2 121 1 5
22 Automotive ECU Python, JSON, Test document, change history 3 64 5 5
23 Software testing Propriety format of state machines 3 317 6 6

17. © 2019 MetaCase 17
Further analysis
 Internally developed fell into approaches 1 & 2
– Correlation between increasing explicitness of PL approach
and increasing categories of language creator expertise
R= 0.52, P < 0.01
 Sizes of the languages vary considerably: the largest is
14 times the size of the smallest
 The largest languages were those that have been
applied for a long time (#17, #18, #20, #4)
– R= 0.6, P < 0.005
 No correlation (0.07) was found between Size and
Approach

18. © 2019 MetaCase 18
Conclusions
 Identified 6 approaches applied for addressing
variability with domain-specific modeling languages
– 2 common ones
 Over half of the languages added further PL support
than plain variant specification
 Large difference in size of the languages
– Likely depends on the domain/PL been addressed
 Companies create languages mostly internally
 Further studies needed with other cases and tools

19. © 2019 MetaCase 19
Thank you!
Questions?
Comments?
Experiences?
Counter arguments?
jpt@metacase.com
Juha-Pekka Tolvanen