[idn] BRACE: Bi-mode RACE

["Adam M. Costello" <amc@cs.berkeley.edu>]

Okay, never mind that FACE stuff.  :)

I've come to understand why FACE and SACE could not hope to match RACE's
efficiency for the kinds of strings RACE is designed for.

On the other hand, RACE does not come close to FACE's efficiency for
strings that use ASCII letters and digits in addition to non-ASCII
characters, which I think will be common (consider non-English phrases
that refer to things with ASCII names).  RACE is also less efficient for
strings in which many characters come from a single row, but not all
(like Japanese text with many kana and some kanji).

BRACE (Bi-mode RACE) is an extension of RACE that is efficient for all
UTF-16 strings.  It uses the same mechanism as RACE (with a few added
variations), and the mode-switching mechanism from FACE.  The cost of
the improved efficiency is increased complexity.

The spec, accompanied by an implementation in ANSI C, is here:

http://www.cs.berkeley.edu/~amc/charset/brace

AMC