Path: utzoo!attcan!uunet!mstan!amull
From: amull@Morgan.COM (Andrew P. Mullhaupt)
Newsgroups: comp.unix.i386
Subject: Re: Permuted indices
Summary: Too clever by half
Message-ID: <1235@s8.Morgan.COM>
Date: 15 Jul 90 03:32:31 GMT
References: <KDAHHS@utower.gopas.sub.org> <1990Jul08.224741.1366@virtech.uucp> <1990Jul12.044550.19213@ico.isc.com>
Distribution: comp
Organization: Morgan Stanley & Co. NY, NY
Lines: 36

In article <1990Jul12.044550.19213@ico.isc.com>, rcd@ico.isc.com (Dick Dunn) writes:
> amull@Morgan.COM (Andrew P. Mullhaupt) writes:
> > Has anyone ever made a utility which converts Bermuda indices into
> > human readable form?
> A fun posting, but I don't get much from it.  What's wrong with a permuted
> index?  Seems pretty obvious to me:  Think of useful words that might have
> something to do with the topic, look them up until you find something that
> helps.  It's far more useful than the traditional index in one particular
> sense: context.  For example, if you want to know how to do something to a
> directory, it's easy to go to "directory" in the perverted index and scan
More useful? Not in my experience. A permuted index is a good way to
inflate the number of references you have to read through to get your
hit. I think of the inconvenience as resulting from collisions in a
hash table; you cannot locate the desired reference as quickly, and
much more importantly (in UNIX manuals) you cannot rapidly determine
that you are searching in a manual which doesn't have relevant 
material. (I.e. the misses are a lot more costly in the Bermuda Index.)

My experience with UNIX documentation is that a nontrivial part of
the search is finding the set of manuals which have relevant information.
I usually have a good idea which manuals are not necessary, but I often
check them all anyway; _which_ FM is usually the hard part of RTFM.
It seems to me that in those UNIXes which provide normal indices,
this doesn't take as long. I think the reason is that Bermuda indices
put you in the linear search mode a lot more than the normal indices,
where you stay in binary search, and fewer page faults are involved.

Now if the thing weren't printed, and you got to search it in an
editor, I might not think Bermuda indices were stupid. But the ones
I'm talking about generally are printed, and just seem to get in the
way. I think they've outsmarted themselves. Just like the people who
invented the electric pen. The old fashioned ones were, when properly
made, just fine.

Later,
Andrew Mullhaupt