22 Dec
2016
22 Dec
'16
10:36 p.m.
On 19 Dec 2016, at 20:10, Noel Chiappa
<jnc(a)mercury.lcs.mit.edu> wrote:
On a related note, great as my respect is for Ken and Doug for their work on
early Unix (surely the system with the greatest bang/buck ratio ever), I have
to disagree with them about Multics. In particular, if one is going to have a
system as complex as modern Unices have become, one might as well get the
power of Multics for it. Alas, we have the worst of both worlds - the size,
_without_ the power.
This is slightly tangential, but I think it's fairly hard to take a simple system and
turn it into a complicated system without ending up with a mess (and I claim that modern
Unix is a mess).
I spent a bunch of my life programming in Common Lisp: CL was famously thought of in the
about 1990 as an *extremely* large language with many baroque complexities: the standards
document I think was over a thousand pages, even with lots of things you actually need
missing, and there were features which people regarded as hard to implement efficiently
without special hardware support (they were wrong, but it was a common thing for people to
say at the time).
And whatever CL was it was not particularly pretty or elegant as a language, at least on
the surface: the standards people valued agreement with each other and easy compatibility
with older Lisps over elegance, so there were just lots of things which were there for no
really good reason other than that they had been there in older Lisps.
(Of course 'extremely large & baroque language' means 'a tiny fraction
the size of modern C++', but this was in the early 1990s before the true horror of
C++ had become apparent.)
A lot of people were just derisive about CL: in particular the Scheme people. Scheme was
this tiny and *extremely* elegant language. Programming in Scheme was just nice, because
it was so small: R4RS seems to be 55 pages, including formal semantics and macros, I
can't find R3RS in page-countable form, but it must have been 40 pages or something
(no macros). And tail-call elimination with first-class continuations: all the guilty
pleasure of GO TO without the guilt.
Scheme was just great, except you couldn't do anything useful because it was so
minimal. But apart from that, great.
So now I use Racket which is a sort of Scheme which has eaten too much and read too many
computer science text books. And although it's just an enormously nice language, the
moment you try to use some of its more industrial parts (structures or the object system)
you realise *just how careful the design of those things in CL was*. Nothing about the CL
design was pretty, but it was designed by people who had both used the system themselves
*and talked to many other users*, and addressed the things that made their lives hard.
So, of course, I still program in CL because, though it's ugly as shit, it's
very very sorted out.
So, finally getting to the point: I think it's significantly hard, to take small
elegant systems and turn them into large industrial systems: if you want large industrial,
you need to design for large industrial and not worry too much about elegance. Unix is
kind of the poster child for what happens when you don't do that.
--tim