Subject: Brain dump
From: Andrei Alexandrescu (andrei@cs.washington.edu)
Date: Fri Oct 05 2001 - 18:13:42 PDT
Guys,
I've been racking my brain for ideas on research themes. I like the ones
Craig sent last the most (language interoperability and extensible
languages). Here are a couple more, in case you are interested.
* Self-modifying code
In the first days of computing, von Neumann et al were very excited by the
idea that the program is stored in the same memory as the data is. They were
intuiting a world of fascinating possibilities. With time, however,
self-modifying code proved to be too unwieldy so most languages,
methodologies, and practitioners deemed it a no-no.
It would be worth to revisit the idea and to try give it a theoretical
basis. Self-modifying code has applications in optimization, on-line
compilation, and many things that might be really cool.
Modern processors make self-modifying code of increased interest. Consider
you eliminate an assertion at runtime by overwriting NOPs over it. To a
modern processor, the NOPs are eliminated early by the branch predictor so
basically there is no overhead at all - as if the NOP operation indeed never
existed.
* Parallel language parsing
You have a machine with N processors and a grammar. How do you parse that
grammar so you use all N processors? What kind of grammars lend themselves
to effective parallel parsing? How do you divide parsing among the N
processors? What theory can you put behind that?
* Partial Evaluation
The folks at Indiana Uni are strong on that, but I guess it's an interesting
subject for us, too. In particular I think they missed the link between
partial evaluation and enforcing correctness.
* Compressed object code
This is an idea I had way back when I was an undergrad. VLIW is cool but the
machine code occupies a ton of memory. Besides, the stream of instructions
is bottlenecked by the memory bus. I happen to have an interest in data
compression, and I noticed the two domains playing oh so good together. I
realized that sliding window compression techniques can be applied very
successfully to code, but you have to be careful with jumps. As a result,
you can reduce the size of the assembly code dramatically but, most
important, you greatly increase the throughput with which the processor
processes the code. This is because fetching is done on the compressed code
and decompressing is done inside the processor. To say nothing about passing
the code around on a network etc.
I am sure there is something about this.
* Again on extensible languages
The thing is, how can you ensure that people using your language have access
to the same amenities as the language developer itself has? This is a
constant source of annoyance to me in C++: you can write a class that
behaves almost like a primitive type, but only almost. How can you provide
language tools to implement (and efficiently so) amenities that are
traditionally considered reserved for the language designer - such as new
primitive types, type systems, keywords, non-local flow of control
(exceptions), and so on?
*********
Hope you find this interesting. See you on Sunday!
Cheers,
Andrei
_______________________________________________
Cecil mailing list
Cecil@cs.washington.edu
http://majordomo.cs.washington.edu/mailman/listinfo/cecil
This archive was generated by hypermail 2b25 : Fri Oct 05 2001 - 18:14:03 PDT