Re: Best, Simple versus Best

preston@tera.com (Preston Briggs) wrote:
> The entire discussion to this point reminds me of an essay by Richard
> Gabriel (I'm sorry I don't have even a minimal reference) contrasting
> the "MIT/Stanford style of design" with the "New Jersey approach."
[snip]
> Following Lee, he suggests that the MIT/Stanford approach can be
> captured in the phrase "the right thing." The following
> characteristics are important: (I'll try and quote exactly)
[snip]
> The New Jersey approach (also called "worse is better") is slightly
> different. Their characteristics are (deliberately caricatured) as
[snip]


With the advent of Standard ML of New Jersey, Gabriel's 'New Jersey'
reference has become a bit out-dated, since to my knowledge, SML/NJ
tries very hard to do the 'right thing'. Although I have a personal bias
against its syntax, I can admit that SML/NJ has been extraordinarily productive
in the development of relatively simple, but very powerful optimization
techniques.


Re the 'right thing':


The Romans were phenomenally successful at dominating the world for the
better part of a thousand years, even though they used one of the worst
number systems ever devised. If they had survived long enough to develop
computers, I am absolutely confident that Windoze-XCV (??) would utilize
Roman arithmetic throughout, and the computer science community would now
be teaching fast Roman numeral algorithms to their students.


'Elegance' is a concept that appeals to people who, like Archimedes, want
to personally move the world with a long lever. If you're happy to move
the world one micrometer at a time as one of the faceless horde working
for Mr. Gates, then elegance will have no appeal.


Just as Algol-60 and Algol-68 were considerably better than most of their
successors, ML & Haskell have shown us that the lambda-calculus is
considerably better and more useful than the Towers of Babel that computer
scientists have erected over the years. It's pretty amazing what
straightforward beta-expansion can achieve when you unchain it.


--
www/ftp directory:
ftp://ftp.netcom.com/pub/hb/hbaker/home.html
--

Return to the comp.compilers page.
Search the comp.compilers archives again.