Subject: Re: what we're up against
From: Matthai Philipose (matthai@cs.washington.edu)
Date: Fri Mar 09 2001 - 10:59:59 PST
I've heard similar stuff from other grad students in CS (though the vast
majority are fine).
I think the complaint has a tiny bit of truth in it. There is a lot of work in
static analysis that is extremely incremental, and the results of which are
used for optimizing programs, with very marginal gains. Another field with a
similar "reputation" for the same reasons is computer architecture,
microprocessor design in particular. The same is true of many kinds of
analyses that make unsound conclusions via "dynamic analyses"... people have
been doing it for a while, both informally and formally, but the insights
added have been (in the overwhelming majority of cases) interesting heuristics
at best.
One way to tell if there is any truth in the matter is to make a list of
programming systems advancements in the last 15 years that have/have some
chance of making a impact discernible to other computer scientists and the
rest of the world. Some examples off the top of my head:
1. Stuff exemplified by Java:
-- well-defined semantics and static typing, including e.g. parametric
polymorphism, of OO langs
-- implementation (both static whole program analysis & dynamic
compilation and JIT) that support large programs with decent performance
-- reflection (interface to and implementation of Meta-level capabilities
in a practical language)
-- componentization and dynamic update
2. The ability to analyze Word (scalable analyses Henglein, Steensgaard, Das,
etc):
-- Non-standard typing/constraint-based analyses
-- The use of these analyses for finding program errors a la Dan Weise's
toolkits
3. Typed IRs, proof-carrying code, language-based security
-- Precise definition and fast checking of important program properties
-- Increased confidence in compiler correctness
Some of the technical work underlying these things, was of course done long
ago. But even in these cases, reducing the work to mainstream practice was I
think a good achievement. I'm sure others can add more. But I think just this
short list says that there is work that is both technically challenging and
practically important going on.
How to do something to get the word out is more difficult. I think stuff like
Java helps tremendously. One thing I feel is that people don't have an idea of
the breadth of applicabilty of program analysis/understanding. Most people
thing that program analysis = program optimization = C program optimization =
5% speedup with a whole lot of effort, beyond the long-known optimizations.
One way to counter this might be lessen the focus of compiler classes on
program optimization, and use non-standard analyses for many different classes
of clients (perhaps difficult to do in a 1 qtr class); stuff like program
understanding and verification, database query compilation/optimization, OS
security, etc comes to mind. Students should understand that any time they're
dealing with stuff that looks like a programming language (i.e. some syntax
and associated semantics), they can bring about a whole range of techniques
from programming systems to abstract, build, optimize, verify and understand
the stuff.
Matthai
_______________________________________________
Cecil mailing list
Cecil@cs.washington.edu
http://majordomo.cs.washington.edu/mailman/listinfo/cecil
This archive was generated by hypermail 2b25 : Fri Mar 09 2001 - 11:00:04 PST