[idn] ACE37

["Edmon" <edmon@neteka.com>]

From: "Adam M. Costello" <amc@cs.berkeley.edu>
> I was basing it on 59 octets.  In the worst case, DUDE would require
> 60 octets for 15 Han characters, but the worst case virtually never
> happens.  In the thousands of names that JPNIC tried, it never happened.
> 59 octets happened a few times, but I think the typical length was 58 or
> 57.  So DUDE can almost always support 15 characters, but usually cannot
> support 16.  I think the boundary between "usually fits" and "usually
> fails" is more useful than the "always fits" number.

I dont think this is very true for Chinese.  Most of the time, characters
would come from different rows causing DUDE to have to use 4 characters for
each Chinese ideograph.  In anycase, if you compare "usually fits" then
ACE37 should still out perform DUDE by a magnitude of at least 5 codepoints,
cause 19 codepoints is the "worst case" scenario (with the 59-octet limit),
but it is likely that it would "usually fit" 20-21 codepoints.  As in my
example with JPNIC in Japanese, it has 25 (>21) codepoints and ACE37 can
still handle it, whereas DUDE fails.

I think the point I am trying to make is that ACE37 greatly increases the
capacity for CJK ideographs, which has been identified as an undeniable
concern through the discussion on this mailing list, without requiring a
complex algorithm and utilizes the simplicity of DUDE.

Edmon