Re: [I18nrp] Mappings for IDNA2008 ?

["Asmus Freytag (c)" <[e-mail]>]

On 2/4/2019 4:25 PM, John Levine wrote:
> In article <[e-mail]> you write:
>>> Are there any published IDNA2008 mappings?  As far as I can tell,
>>> everyone uses one from UTS46 by default, and it's not very good.
>>>
>> Examples of its badness, please.
> As I understand it:
>
> If you speak Turkish, the case folding is wrong.

The mapping says to take CAPITAL I WITH DOT to 0069 + 0307 (which is 
lowercase i + dot).

This is formally correct, because the sequence looks precisely like "i", 
but probably not helpful unless Turkish IDNs use that convention over 
just using lowercase 'i'.

In my personal view no registries should support any soft-dotted letter 
followed by dot above as that leads to an immediate homoglyph issue. So, 
yes, that mapping is indeed questionable.

>
> If you speak Persian, the joiners are wrong.


The Root Zone will be like IDNA2003 here, disallowing the joiners. Yes, 
stuff looks "wrong" without in some languages - worse for Sinhala (used 
in Sri Lanka) than for Persian, by the way.

This one is a no-win situation. Some zones may support these, others 
won't - what is a generic mapping to do?


>
> If you speak Arabic, the mapping or lack thereof between
> ASCII and Arabic digits is often wrong.
>
> If you speak Chinese, the whole thing is wrong because Chinese users
> expect their ASCII pinyin to be turned into Chinese.

That can't be a unique mapping. I imagine it would be more like the T9 
phone input.

Many, many Chinese characters share the same pronunciation (and 
therefore same pinyin).

And pinyin isn't ASCII - it has accents.

>
> IDNA 2008 said very clearly that good mappings depend on the user's
> context, with the language being a large part of that context.


None of these examples, except the first one using U+0307 strike me as 
"bad" -- only as not tailored.

I was hoping you had more examples where the mapping choices are 
inappropriate for most (or even all) languages.

A./