Mark Davis ☕
mark at macchiato.com
Wed Dec 2 18:42:00 CET 2009
2009/12/2 Patrik Fältström <patrik at frobbit.se>
> Protocol elements, in the broader sense, should only consist of labels
> that are stable and "ok" according to IDNA2008, including bidi rules,
> PVALID in tables (or context rules ok) etc.
> No upper case, no unallocated, no dissallowed characters etc.
> I am afraid that is the only thing we 100% can guarantee.
> And this was one of the things that was confusing in IDNA2003. Some
> people thought mapped codepoints where ok, and got confused in cases
> dns returned other codepoints.
The mapping phase in IDNA2003 worked. Every implementation of IDNA2003
followed it; it was always predictable.
It may be the case that it was confusing on the registration side, and it
doesn't hurt to remove it there. So while there isn't an established
benefit, the impact is minimal.
On the lookup side, it is different. There has been no hard evidence that
mapping was a problem (excluding the 4 deviation characters, since we're
discussing them elsewhere); no figures, no data, just a lot of handwaving.
So there is no established benefit to dropping them, and a significant
negative impact for dropping them.
> Then there are boundary cases that might work, depending on use. Very
> hard to describe scenarios as many parameters are included in "works".
> PTR records match? Cert match? Spamassassin rules increasing score if
> not fqdn is in use? User interface and keyboard mappings?
> On 2 dec 2009, at 16.06, Lisa Dusseault <lisa.dusseault at gmail.com>
> > I'd like to try to unpack some of the different use cases we're
> > talking about a little more.
> > ISTM that use cases where the person following the link is the person
> > who is typing it in, are use cases that locale-dependent mapping might
> > be most useful. If I'm in a locale where Ȱ (x230) is considered to
> > be
> > the capitalized version of o (ASCII o), it might very well be most
> > helpful to make that mapping. Use cases where the same user is typing
> > in the domain names that then looks them up include:
> > - typing links in the address bar
> > - typing mail address in the To field of an email
> > - Writing a Web page, blog post or email, wherein I check that the
> > links work before posting/sending my document
> > In contrast, the use case where the person looking up the domain
> > FȰȰ.example is not the person who typed it in, then in most cases we
> > no longer know the intent or locale of the person who typed in the
> > domain. It may be the same locale as the person who is looking up the
> > domain but it may not be. The person who typed in the domain may have
> > intended fȱȱ.example or foo.example, and may have tested that before
> > sending/posting the link, but we no longer have that information. Use
> > cases include:
> > - Following a HTTP link in any Web page, document, blog post, email,
> > etc
> > - Using a mailto link (explicit or implicit), e.g. when one person
> > sends me another person's email address
> > We probably would all agree that people follow links while Web
> > browsing far more often than they type them in, and even when typing
> > in, auto-complete probably drastically reduces the new cases of
> > from-scratch mapping and lookup.
> > However, we probably have quite different assumptions about how much
> > Internet activity takes place among users of a consistent locale. Can
> > we assume that Patrik wants ß interpreted as ß because he communicat
> > es
> > mostly in Swedish with Swedish users and mostly reads Swedish Web
> > pages? Or must we assume that Patrik also gets email from german and
> > swiss senders, and also reads Web pages (perhaps in English!) written
> > by German users who expected different mappings? I am sure this
> > depends heavily on our model of a user, and whether we're using
> > ourselves as hypothetical examples or not.
> > One slightly more solid question for browsers is, would it be entirely
> > crazy to have different mapping algorithms for typed-in domain names
> > than for links followed? There might be a locale-dependent mapping as
> > well as a global mapping. (I assume that having every established
> > locale mapping installed would be complete craziness.)
> > Another question is: when posted links are followed, how often do we
> > know the locale where the link was authored? Not that the browser
> > following the link would necessarily be able to apply the mappings of
> > the locale in which it was authored, but would it be slightly better
> > to apply a global mapping than a mapping from a different locale?
> > Do any authoring software clients fix up links as the user types?
> > When I type a link in a document, the authoring software often makes
> > that link active. Is there any software that automatedly lower-cases?
> > If so, would such software also be likely to map to PVALID characters
> > before the doc is finished?
> > Lisa
> > On Tue, Dec 1, 2009 at 12:45 PM, Shawn Steele
> > <Shawn.Steele at microsoft.com> wrote:
> >> One example I discussed with Patrik yesterday, was whether locale
> >> might affect mapping. I'd like to get better insight into the general
> >> understanding of that.
> >>> 1. Could locale determine whether a PVALID character should be
> >>> mapped
> >>> into another PVALID character prior to following the rules to turn
> >>> into an ALABEL? I believe the consensus answer is probably SHOULD
> >>> NOT
> >>> or MUST NOT because that would make domains with that valid
> >>> character
> >>> unreachable by software using those locale rules.
> >> I agree.
> >>> 2. Could locale determine whether, or how, a DISALLOWED character is
> >>> mapped into a PVALID character prior to getting an ALABEL?
> >> No, for several reasons:
> >> A) If I email you a link that contains a DISALLOWED character, your
> >> machine/environment MUST map it to the same thing my machine did.
> >> Otherwise I say "you have funny charges from travelling, visit Bank.org
> >> to correct it." You are trying to pay for your flight home so you
> >> type "Bank.org" into the computer in the kiosk in the foreign
> >> airport, and if it uses different mapping rules you could end up as
> >> a phishing site. You don't want VISA.com to go to a vısa.com just
> >> because you're using a Turkish airport browser.
> >> B) If I travel myself, I need consistent behavior regardless of the
> >> machine I'm using.
> >> C) If I see an international advertisement, the domains need to go
> >> to the same server, regardless of who and how and where the person
> >> is typing in the link.
> >> D) A server or relay wouldn't necessarily know the context the user
> >> expected when interpreting a forwarded request.
> >> E) It'd be a support nightmare.
> >> F) I'm not sure if it is practical to create APIs that enable this
> >> distinction. (We (software community, not just my company) already
> >> have problems selecting the correct locale specific behavior for
> >> sorting and formatting, etc., so we'd be bound to get it wrong at
> >> least some of the time.)
> >> -Shawn
> > _______________________________________________
> > Idna-update mailing list
> > Idna-update at alvestrand.no
> > http://www.alvestrand.no/mailman/listinfo/idna-update
> Idna-update mailing list
> Idna-update at alvestrand.no
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Idna-update