Article 74878 of alt.folklore.computers:
From: hoey@aic.nrl.navy.mil (Dan Hoey)
Newsgroups: alt.folklore.computers
Subject: Re: Behavior of sed (was: Re: Interviewing)
Date: 23 Jan 1995 21:50:39 GMT
Organization: Navy Center for Artificial Intelligence
Distribution: usa
In-reply-to: justin@physics20's message of 23 Jan 1995 19:06:34 GMT

Peter Seebach <seebs@solutions.solon.com> sed:

> >s/\([^ ]*\) \([^ ]*\)/\2 \1/g
> >'a b c d e' should produce?

Cute puzzle.

justin@physics20 (Justin R. Bendich) bleated:

> Why do you make such demands on poor sed?  SunOS and VMS (probably
> has same source code) seds produce "b ac d e", which is sed's way of
> crying out for mommy (without the trailing 'g', the behavior is
> correct).  IRIX's sed yields "b a d c e".  I guess that's "right".

No, the former answer is right.  Once you notice that in the second
substitution \1 is "" and \2 is "d" it is easy to see why the correct
result is "b ac d e " (not "b ac d e").  I think I have heard that
some seds refuse to match "[^ ]*" to a null string within s/.../.../g
for fear of an infinite loop.  I would expect that detecting the
infinite loop would be <FM>so trivial that even a writer of U*x system
software would be able to prevent its occurrence without mangling the
semantics in an obscure and undocumented way.</FM>  IRIX's behavior
shows that my expectation is too generous.  Though I don't think this
is SGI's fault--I suspect they are following some brain-dead standard
designed to preserve the historic lossage, like maggots in amber.

> It's hard to see why you'd want to do this...

Haven't you been paying attention?  We are doing this to figure out
who we want to hire.  Would you hire someone who couldn't figure out
that sed was doing the right, if nonobvious, thing?  How would she
ever debug her sed scripts?

Now if you want an _obscure_ sed hack, analyze:

#! /bin/sh
echo "$*" | sed \
's/./&&/g;s/\(.\).*\(.\)/\2&\1/;s/.\(.\)/\1)(&/g;s/..//;s/...$//
s/(\(.\)\1\1)/-/g;s/(\(.\)\1.)/]/g;s/(.\(.\)\1)/[/g;s/(...)/./g;h;G;P;D' \
| more

I'm told that a patent for the above process was applied for, and that
the application reads in part:

 Fantomark strings which, by definition, are imperceptible to man,
 animal, or machine, nonetheless have well-defined streaks, which are
 ordinary mark strings....  If one assigns NOUGHT to quids and BEING
 to quods, and if one calls BIP processing (which amounts to continual
 liquidating and quodizing) by the term BECOMING; and calls
 liquidating DISSOLUTION and quodizing PASSAGE INTO, then the
 intrinsic opposition between quidness and quodness and its continual
 resolution through BIP processing is captured by the language:
 BECOMING continually mediates the DISSOLUTION of BEING into NOUGHT
 and the PASSAGE of NOUGHT INTO BEING.

I offer 57 fantomarks for the cleverest alternative vocabulary
substitions; payment in zorkmids or qatloos is subject to
varying exchange rates.

Dan
Hoey@AIC.NRL.Navy.Mil