1. Creating DSLs that fit the users’ needs
Jordi Cabot - AtlanMod
Panel @ MiSE 2012
http://modeling-languages.com
@softmodeling
© AtlanMod - atlanmod-contact@mines-nantes.fr 1

2. Before Starting
 Most DSLs should have never been created in
the first place (many are too UML-like)
 It’s a strong investment! (will you get your
ROI?)
 GML are not as bad as people say
 Avoid reinventing the wheel
© AtlanMod - atlanmod-contact@mines-nantes.fr 2

3. Current situation
 Good language workbenches for creating
abstract and concrete syntaxes for your DSL
 Technology is there (well, usability of the tools
could improve…)
 but process and quality aspects are missing
© AtlanMod - atlanmod-contact@mines-nantes.fr 3

4. Process
 DSLs are domain-specific but still it’s a non
domain-expert who creates the DSL
 Collaboro aims to enable a more collaborative
process
 … but we already talked about this yesterday
(check http://modeling-
languages.com/collaboro-slides-available/)
© AtlanMod - atlanmod-contact@mines-nantes.fr 4

5. Quality
 We know what quality properties make sense
for models (e.g. satisfiability) but they do not
translate well to DSLs.
 Quality of DSL = user experience
 Evaluating user experience is a challenging
task
– We cannot develop specific techniques for
each different DSL
– Need the participation of users
© AtlanMod - atlanmod-contact@mines-nantes.fr 5

6. Quality: Dealing with users is not easy
© AtlanMod - atlanmod-contact@mines-nantes.fr 6

7. Quality
 2 research lines
– Evaluating how domain experts are using
the DSL in practice
– Automatic Testing of the user experience
© AtlanMod - atlanmod-contact@mines-nantes.fr 7

8. Approach 1 - Corpus analysis (post
mortem)
 Analysis of repositories of DSL models (i.e.
instances of the DSL under analysis)
 We can analyze:
– (meta) classes that are never used <-
irrelevant?
– Clusters in the DSL <- two subDSLs?
– Complex structures (clones) that appear
often <- is the DSL missing an important
element?
© AtlanMod - atlanmod-contact@mines-nantes.fr 8

9. Approach 2 - Automatic testing of the
user experience (interactive)
 Specially for concrete syntaxes
 Reuse what we know from web interaction and
design
– Small changes can make a huge difference
– Even seasoned designers fail to predict upfront
what would work
– Different user profiles may require different
concrete syntaxes
 What about A/B testing for DSLs?
© AtlanMod - atlanmod-contact@mines-nantes.fr 9

10. Automatic testing of the user
experience (interactive)
 Which syntax for expressing transformation
rules is better?
– It’s up to the users to choose!!!
Rule newMachine Rule newMachine
LHS RHS
c1
c1 c1 m1
m1 m1 equivalent
{del}
c2
c2 c2 m2 m2
{new}
Extended (declarative) notation Compacted (operational) notation
© AtlanMod - atlanmod-contact@mines-nantes.fr 10