non-armlet search speed

The current search routine is designed to work with multi-byte alphabets.
When experimenting with my search armlet, I found that the multi-byte
alphabet support slows down the search by a significant amount.  Switching
from multi-byte to single-byte sped up my benchmark from 2 or 2.5 min (it
was a long time ago, so I can't remember the exact numbers) to 1.5 min on my
NX70v (the armlet search, by comparison, on this benchmark should be under
10 seconds).  It would be easy to implement two searches, one for multi-byte
alphabet machines and one for single-byte alphabet machines.  In fact, we
already have something like this: the armlet only works for single-byte
systems (and of course the code checks if it's appropriate to use the
armlet).  It would take very little work to make this work in 68K mode, too.
(I could just re-use most of the armlet code, which would be going full
circle since I debugged the armlet initially as a 68K routine.)

But I don't know if anybody but other than me cares about the search speed.

Alex

--
Dr. Alexander R. Pruss
Department of Philosophy
Georgetown University
Washington, DC 20057-1133  U.S.A.
e-mail: ap85 at georgetown.edu
online papers and home page: www.georgetown.edu/faculty/ap85
----------------------------------------------------------------------------
-
   "Philosophiam discimus non ut tantum sciamus, sed ut boni efficiamur."
       - Paul of Worczyn (1424)