Generative Programming: Past, present, and future

1
Generative Programming Past, present, and future

• A panel at GPCE 2004
• Vancouver, BC
• October 27, 2004

Jeff Gray, University of Alabama at Birmingham
2
How to teach it? to my mother
Assist in automating tasks that are mundane and
manual, such that error prone nature is removed
and productivity is improved.

• Context Real Estate Legal Forms
• 15 different documents
• 6 different participants with specific contact
  information
• Numerous places within each document where parts
  of contact information is to be inserted

Note Some internal generator living within word
processor
3
How to teach it? to my father-in-law
Assist in mitigating effects of accidental
complexities.
- Reducing all images to common sizes, producing
thumbnails, providing a general index, linking
all pictures in a slide show - Many hours to do
manually about 20 seconds in Web Picture
Creator
4
How to teach it? to my grad students

• The previous were somewhat embarrassingly simple
  examples, but perhaps offers insight into
  essential characteristics of GP that could be
  understood by laypersons
• A key is relating experiences with and
  without the availability of certain tools
• For Grad students
• Provide a 5k SLOC sample application
• Traditional Adaptation Ask them to perform an
  adaptation using traditional development (e.g.,
  plain Java in Eclipse)
• Metaprogramming and reflection Introduce core
  literature and related tools request students to
  perform same changes using OpenJava, Javassist,
  JMangler, etc
• AOSD introduce core literature in AOSD and
  tools request students to perform same changes
  using AspectJ
• Metamodeling introduce core literature and GME
  request students to perform generative tasks on
  domain-specific models
• More info http//www.cis.uab.edu/gray/Pubs/scicop
  -2004.pdf

5
What do we do next?

• Key challenge
• managing the syntactic mismatch between
  abstraction layers
• Examples
• Debugging and testing a DSL program (e.g.,
  debugging YACC output)
• It would be desirable to do this at the
  abstraction of the conceptual artifact
• Version control of models/visualization
• Forced to view differences as flat text files
  (perhaps in XML!), rather than the abstraction
  provided by the modeling tool