Re: Non-latin email addresses dangerous?

On 1-feb-04, at 19:52, Martin Duerst wrote:

> > This is very dangerous anyone having a non-latin email address can't 
> > expect to receive mail from people who don't master this character 
> > set. So someone who speeks a non-latin language as well as a latin 
> > one, would always have to provide two representations of his/her 
> > email address in order to be reachable in both languages.
> If you have ever looked at business cards of people in such regions, 
> that's
> exactly what they do. Not because they want it that way, but because 
> they
> know that their customers appreciate it. The main piece of 'innovation'
> that you need in order for this to work with email are some clever ways
> of sending MUAs to figure out which sender address to use, so that they
> don't use a non-latin one when they send you email. The rest of the
> problem is mostly pure mechanics, although before we use it, it has
> to be written up.
It's not that easy. What if A and B communicate in Japanese, but they 
also both speak English. Now B wants to give C A's email address. If 
A's software had determined that B only needs to know the Japanese 
character email address then C is going to have trouble, especially 
when there is another step that includes degradation of the encoding.
If if we're going to allow non-latin email addresses, I think the only 
reasonable way to do it is by mandating that the latin equivalent must 
always accompany the non-latin address. Then only in cases where the 
address was typed from paper or some such the latin equivalent is 
unavailable.
I think we should also explore the possibilities of having numeric-only 
email addresses, as this nicely solves the whole mess, be it in a way 
that isn't all that pretty to look at.
A simple way to create numeric addresses would be to concatenate the 
numeric values of all characters in the script used in some way, but 
this has the problem that the addresses get very long very fast. (A 15 
character email address would be something like 50 digits.) A better 
way would be to give each tld, domain and user a sequence number, 
although this has the disadvantage that mapping back and forth is very 
hard. (3 digits for tld, 6 - 9 for domain and ~ 3 - 6 for user would be 
15 digits or less if domains with many users get the short domain 
numbers.) We probably want to include one or two checksum digits.
Another way to handle this would be semantics-based character set 
translation, where a word in a language with character set A is 
replaced by a word with approximately the same meaning in a language 
with characterset B. Note that actually retaining the meaning is only 
an additional benefit when this can be achieved, the real target is to 
create a new email address in the new character set that is relatively 
easy to enter. In order for this to work  everyone would have to use 
large translation tables, but compared to what's needed to implement 
unicode that doesn't seem like a huge problem.