Re: [I18nrp] Mappings for IDNA2008 ?

[Martin J. Dürst <[e-mail]>]

Hello Patrik,

On 2019/02/13 20:02, Patrik Fältström wrote:
> 
> 
>> On 13 Feb 2019, at 11:31, Martin J. Dürst <[e-mail]> wrote:
>>
>> Hello Patrik, others,
>>
>>> On 2019/02/13 17:25, Patrik Fältström wrote:

>>> If f(x)=x’ and in the locale used f(x) should be x and not x’, you have dependencies between f(x) and f’(x) if you want f(f’(x)) to be x’ and f(f”(x)) to be x.
>>
>> Well, I have to admit that I didn't explicitly talk about that case in
>> my 'proof'. For functions in general (e.g. a function such as f(x) =
>> (x+7) mod 15), you are right.
>>
>> But f(x) (the general mapping function) is idempotent (i.e. f(x) =
>> f(f(x)), which means that you can map as many times as you want, you
>> always get the same result). This is also true for any locale-dependent
>> mapping function.
> 
> I can not say it is. It might be a requirement that you set on the function. I don’t know enough about various languages in the world, bidirectionality and punctuation. See latest discussions about punctuation in Armenian language and script and it’s impact on registration.

Do you have a pointer?

>> If it were not the case, human users would be very
>> surprised: a maps to A, but then A maps to a: That wouldn't make any sense.
> 
> No, we talk about f(a)=b and f(b)=c as one thing and f(a)=b and f(b)=a as a different one.
> 
>> Also, and even more important, for the general mapping function, if x is
>> allowed in IDNs, then f(x) = x.
> 
> That’s another requirement. Maybe in one locale you DO map from one to another?
> 
>> This may not be true for a
>> locale-dependent mapping function (i.e. it would be possible for a
>> German mapping function to map (Hungarian) ȁ to ä, although that's not
>> needed because German keyboards won't produce ȁ in the first place).
> 
> It might produce it via copy and paste.

Yes, that would be an issue. But in that case, it was probably intended 
to be a Hungarian ȁ, not a mistyped ä, and therefore mapping would be 
doubtful.

>> So if f(x)=x', that would mean that x is not allowed in IDNs. That would
>> mean that it is impossible that in a specific locale, the desired result
>> of f(f'(x)) is x. It may be different from x', i.e. y, but we can easily
>> get this by defining that f'(x) = y.
>>
>> So applying the (in most if not all cases very small) locale-specific
>> mapping before the general mapping always will work.
> 
> I don’t understand what your goal is.

I tried to explain it in general in my mail to Andrew, and will repeat 
it here:

 >>>>
My general direction of arguing is that if things are reasonably
designed (i.e. idempotent,...), then it makes a lot of sense to split
the mapping into a general one and some locale-specific ones, and the
locale-specific ones will go before the general one. Also, each
locale-specific one should be rather small (the Turkish one, often
brought as an example, essentially consists of mappings for 2 letters).
 >>>>

> What I am saying is that what mapping is done when a string enters the system the first time, mapping is to happen and it is context dependent where context can for example be locale.
> 
> What is wrong with that statement?

Nothing. It's perfectly possible that two statements (yours and mine) 
are both true :-). Your statement is about the point in time of where to 
apply a mapping. My statement is about how mappings can be split into a 
locale-specific and a general part.

>     Patrik

Regards,   Martin.