TOPIC: Design Capture as a Paradigm; Design Recovery as a Practical Approach

PRESENTERS: Ira D. Baxter, Ph.D. CEO Semantic Designs, Inc.

DATE: Nov 16, 2016

TIME: 6:30 pm – 8:30 pm

LOCATION: The Advisory Board Company, 12357-C Riata Trace Parkway, Bldg 7, Suite 100, Austin, Texas, United States 78727

COST: FREE

FREE Food & Drinks, Limited Seats

RSVP is required:  https://meetings.vtools.ieee.org/meeting_registration/register/42129

Abstract

Software continues to grow in size. Building software is hard and expensive but only takes finite time. Maintaining it to meet new needs indefinitely once it is built is harder and more expensive; many current systems are 30 to 50 years old. Much of the "maintenance" effort is really rediscovering the program structure and purpose.  Yet we have little theory about how to do this.

This talks suggests that the key problem is loss of the design information and the "obvious" solution is to capture that design while constructing the application.     Program transformation engines seem like a natural foundation, being able to reify specifications into code via formal steps.  One has to augment such a refinement process with rationale as to why specific transformations where chosen. With such a design,  one can define incremental processes for modifying the design, producing new code as a side-effect, and thus providing a full lifecycle model of incremental software maintenance.

Theories are nice, but practice presently requires we handle existing, large software artifacts.  We propose to recover a plausible design from code and then apply incremental techniques, requiring at least "anti"refinements to go from code to abstractions.  The talk such anti-refinements as now-practical data flow pattern matchers in the style of MIT's Programmer Apprentice Project (Rich and Waters, 1987) [http://dspace.mit.edu/handle/1721.1/6054].   We describe the migration of production chemical factories based on this approach.

Presenters/Bios

Dr. Baxter has been building system software since 1969, when he built a timesharing system on Data General Nova serial #3.  In the mid seventies, he built realtime, single user, multi-user systems and locally distributed OSes on 8 bit CPUs.    Realizing that software engineering was largely enhancement of existing code rather than building new code, and that the OS architectures were conceptually similar but shared no code, he went back to graduate school to learn more about reuse of knowledge in software maintenance.     He studied program transformation tools for code generation and modification, obtaining a PhD from UC Irvine in 1990.    At the Schlumberger Computer Science lab, he worked on generation of parallel CM-5 Fortran code for sonic wave models from PDEs.   He spent several years as consulting scientist for Rockwell Automation working on automating factory control. In 1996, he founded Semantic Designs, where he is now CEO and CTO. At SD, he architected DMS, a general purpose program transformation engine, used in commercial software reengineering tasks, and he designed
and implemented PARLANSE, a task-parallel, work-stealing programming language in which DMS is implemented.