home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.lang.scheme
- Path: sparky!uunet!brunix!brunix!mj
- From: mj@cs.brown.edu (Mark Johnson)
- Subject: Re: Are interpreters now as fast as compiled code used to be?
- Message-ID: <1992Dec23.233727.24184@cs.brown.edu>
- Sender: news@cs.brown.edu
- Organization: Brown University Department of Computer Science
- References: <1992Dec17.191058.28471@prodigy.bc.ca> <JAFFER.92Dec23012204@camelot.ai.mit.edu> <JAFFER.92Dec23115358@camelot.ai.mit.edu>
- Date: Wed, 23 Dec 1992 23:37:27 GMT
- Lines: 47
-
-
- What's the point of this thread? Is someone trying to argue that
- Scheme doesn't need a compiler or the very best, most efficient
- implementations available?
-
- As I've said before, Schemers should think long and hard about the
- ``ecological niche'' that Scheme occupies in the programming language
- universe, and design the language and implementations accordingly.
-
- I think that Scheme and Lisp-like languages are located in between the
- arguably higher-level functional and logic programming languages and
- lower-level languages like C. I think Scheme runs a great risk of
- being ``squeezed out'' as the higher-level languages become more
- efficient and the lower-level languages gain higher-level features.
-
- Reading the net, one gets the idea that the only thing that Scheme is
- useful for is teaching high-school students. If that's true, then we
- shouldn't worry about implementation efficiency at all, but instead
- worry about designing environments for absolute beginners. If this is
- the case, the R5RS committee should include more high-school teachers
- and fewer CS people.
-
- But I see two other important applications for Scheme.
-
- One is for embedded interpreters. I don't know a lot about this
- application, but I'd guess that efficiency is not so important here,
- but compact code size and small memory requirements are.
-
- The second is for large semi-numerical computation. There are a
- number of applications that mix symbolic and numerical computation and
- deal with comparatively large amounts of data. For example, in my
- area, natural language processing, it's common to do statistical
- manipulations of corpora that are 100Mb or longer. I don't see any
- reason why a Scheme implementation shouldn't be able to offer *better*
- performance than C on more symbolically-oriented such applications,
- provided that some care is taken in the numerical area and that the
- compilation technology is good.
-
- Any comments?
-
- Mark
-
-
-
- Mark Johnson
- Cognitive Science, Box 1978
- Brown University
-