From mcharity Fri Jun 19 14:38:13 -0400 1998
Subject: State of the art

From the preface of John Scott Russell's (engineer and naval architect)
massive "The Modern System of Naval Architecture", 1865 (135 years ago)...

      We the passing generation have had to
    grope our way out of the dark slowly and
    painfully, with trial and error.  But
    what has to be pardoned to us can no
    longer be pardoned to our successors, to
    whom we bequeath the costly knowledge and
    painful experiences that have cost us so
    dear, but which we have gladly earned,
    and now painstakingly contribute for
    their instruction, and the advancement of
    their future...

Heartening to think we may write this ourselves,
about computer and software engineering,
before too many more years have passed.

Mitchell Charity


    (John Scott Russell, from the preface of
     "The Modern System of Naval Architecture", 1865.
     Quoted in Henry Petroski's "John Scott Russell",
     Engineering column of American Scientist,
     v86/1998Jan-Feb, p21)

Russell worked on, among other things, the
first Atlantic cable, and steam-only crossing.


From mcharity Tue Jun 23 16:21:19 -0400 1998
Subject: Re: State of the art

Curious indeed - I'd not heard that.

My metric is unfortunately subjective.  Every once in a while, for
some years now, I re-explore the maze of twisty little passages which
is our collective software accomplishment, with the objective of
perhaps finally being able to stitch together a productive development
environment for myself.  It remains infeasible, but I've noticed
changes in the maze.

Path lengths are increasing, in part because systems are
"cross-pollinating", incorporating each others ideas.  The components
may be variously misdesigned, but atleast there is recognition of
need, and some real progress.  To get some set of components, one now
need only contemplate kludging together n systems with some glue,
rather than 2n with big gobs of it.  It is no longer entirely hopeless
to go looking for an embedable interpreter which permits multiple
instances in one address space, mixed gc policies, and has a good
foreign function interface.  It is still improbable... but no longer
hopeless.

And the character of the dead-ends, the gotchas, has been changing.
The most common reaction used to be "Why would anyone want to go
there?!?".  Now some dead-ends have notes saying "Perhaps this was an
unfortunate place to put a wall."  Or even a small crowd of people
trying to dig through.  "Why would anyone want to unlink symbols in a
running program?; Oh, ok.; Maybe we should do an expressive linker
language under program control..."

It seems a bootstrap problem.  Hugely escalating resources blunted by
a maze which is now mostly historical artifact.  The sheer magnitude
of dissipation is impressive, but the maze seems to be losing ground.
Unless something changes the ablative dynamic (poor intellectual
policy might create new walls; an economic downturn might pull people
out of the maze), I'd expect a non-linear breakthrough at some point.

As for when, and tying back to the objective of a radically more
productive environment, I have lately found it heartening that almost
every component I care about now seems to have a mob headed vaguely
towards it.  Some are quite slow, and some don't see it (the Australian
Magma folks, for instance, almost have a demonstration you can program
using real math... but think it just a research tool for number
theorists).  But having mobs gets me away from worrying whether some
company might fold, language not make it, clueful people rathole, or
architecture be misdesigned.  There is still contingency, but also a
bit of the warm fuzzy feeling that things might be starting to converge.