[idn] AMC-ACE-Z

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

I have just submitted AMC-ACE-Z as draft-costello-idn-amc-ace-z-00.txt.
You can also find it at:

http://www.cs.berkeley.edu/~amc/charset/amc-ace-z.gz

Recently the ACE design team recommended DUDE, but AMC-ACE-Z did not
exist then.  At this time I personally recommend AMC-ACE-Z.  While it is
not as simple as DUDE, it is still fairly simple, and only a little bit
larger than DUDE when implemented in C.  But it is more efficient than
DUDE for all scripts: DUDE is typically about 30% longer than AMC-ACE-Z.
A comparative evaluation between these and several other ACEs can be
found at:

http://www.cs.berkeley.edu/~amc/charset/ace-eval.gz

(Would anyone like me to submit the evaluation as an internet draft?)

Some have argued that we should select the simplest ACE that is
"efficient enough".  But one could just as well argue that we should
select the most efficient ACE that is "simple enough".  In my opinion,
AMC-ACE-Z is more than simple enough.  And I think most of us here are
much better qualified to speculate on what constitutes "simple enough"
than on what constitutes "efficient enough".

But actually, I don't believe there is any such think as "simple enough"
or "efficient enough".  More efficient is always better, and simpler is
always better, and the two goals must be balanced.

Rather than presume to dictate what is a reasonable length for a name,
I think it would be more prudent to ask dozens of implementors to
spend a few extra hours coding and debugging in order to give millions
registrants more breathing room, in case they need it.

AMC

P.S.  I think Soobok Lee's reordering layer, which is known to improve
DUDE's efficiency, would similarly improve AMC-ACE-Z's efficiency,
but I don't know by how much.  I'm also curious how reordered-DUDE
would compare to non-reordered-AMC-ACE-Z.  (Note:  While DUDE wants
common code points to be in the same aligned block of 16, 256, or
4096, AMC-ACE-Z cares nothing about alignment, but simply wants common
code points to be as close together as possible, so sorting them in
decreasing order of frequency would probably be best.)

One concern I have regarding this reordering layer is who is going to
design the reordering tables for the 32 scripts that Soobok Lee has not
been working on?