MVILAIN@G.BBN.COM (Marc Vilain) (12/13/88)
BBN Science Development Program
AI Seminar Series Lecture
PLAN ANALYSIS OF PROGRAMS
Stanley Letovsky
Department of Computer Science
Yale University
(LETOVSKY-STANLEY@YALE.EDU)
BBN Labs
10 Moulton Street
4th floor large conference room
10:30 am, Thursday December 15
***********************************
*** ***
*** Note: not the usual room! ***
*** ***
***********************************
Computer programs are more than just concatenations of
instructions to a machine; they are also compositions of programming
plans. Conventional languages make the instructions to the machine
explicit, but they often obscure the plans, causing difficulties for
program maintainers, who operate mostly at the level of changing the
plans in the code. Plan analysis is the task of determining what plans
are implemented in a given program. Automatic plan analysis may provide
the basis for intelligent documentation tools which can provide
maintainers with high level summaries of programs, and answer questions
about the goals and plans in the code.
This talk presents an approach to automated plan analysis
of programs based on program transformations. Plan recognition is
modelled as program transformation within a wide-spectrum language, in
which the expressions in the code that make up a plan are rewritten into
a new expression describing the corresponding goal. Exhaustive
application of this recognition process yields a new version of the
target program from which optimizations and implementation details have
been removed. This version can be used to provide summary documentation
of programs. The history of transformation applications provides
information about what plans were found in the program. This
information can be used to answer questions about the motivation for
particular pieces of code.
Analysis methods are presented within the transformational
framework for analyzing several problematic types of programming
plans. These include imperative plans with side effects, looping
plans, plans involving abstract datatypes, and plans involving
conditionals. A working prototype transformational analyzer, called
CPU, has been constructed and will be described in the talk.
-------