TR46 (was Re: Formal submission of our documents to AD)
Mark Davis ☕
mark at macchiato.com
Thu Oct 8 09:42:03 CEST 2009
There is a working copy at
http://unicode.org/draft/reports/tr46/tr46.htmlwith some fixes made as
a result of your comments. Some responses also
On Wed, Oct 7, 2009 at 16:42, Mark Davis ☕ <mark at macchiato.com> wrote:
> I updated the utility to show the difference between IDNA2003, 2008, and
> Will follow on with responses to your comments, and Vint's request for a
> flow diagram.
> On Mon, Oct 5, 2009 at 04:01, "Martin J. Dürst" <duerst at it.aoyama.ac.jp>wrote:
>> On 2009/10/05 17:03, Mark Davis ☕ wrote:
>> > If you have some particular suggestions regarding items in the document,
>> > can submit them directly via the
>> > *[Feedback<http://www.unicode.org/reports/tr46/#Feedback>
>> > ]* link at the top. If you also want discussion of the topics, then also
>> > one of the Unicode mailing lists. And if you have any other suggestions
>> > how to bridge the compatibility gaps between IDNA2003 implementations
>> > IDNA2008 implementations, those suggestions would be welcome. We had a
>> > number of people from the browser communities at the meetings, and these
>> > were the best we could come up as yet.
>> > *
>> > *Mark
>> Hello Mark,
>> Just some short comments here on this list while rushing through that
>> document. Please forward these wherever appropriate.
>> 1.3.1, Deviations, says "There are a few situations where the strict
>> application of IDNA2008 will *always* result in the resolution of IDNs to
>> different IP addresses than in IDNA2003."
>> The *always* is of course wrong. The document itself later says "Unless
>> the "DE" registry bundles", which is, as far as we know, more or less what
>> they are going to do. The other possibility is of course that the owner of
>> the domain name in question makes sure they get both variants. For the
>> business example that you show, that's not a problem at all, if the business
>> knows about the issue.
I added here ", unless the registry or registrant takes special action", and
dropped "always". Does that work?
>> In 1.3.2, Example 3. "Map http://ÖBB.at <http://xn--bb-eka.at> to http:/
>> phishing.com", is completely weird. If any browser or similar device
>> wants to spoof their users, they have always been able to do this, even
>> without the IETF's or the Unicode Consortium's permission. But such a
>> browser would be very quickly out of business, for obvious reasons.
That is given as an extreme case of what is possible under the spec. While I
agree that that is unlikely, since conformant implementations of IDNA2008
have complete freedom, it is not unlikely that we would see an array of more
subtle interoperability problems resulting.
That being said, I agree that it is probably best to just remove that line.
>> Again in 1.3.2, it says "but adds validity constraints from IDNA2008", but
>> then gives "http://√.com" as an okay example (currently in use, although
>> for domain speculation only), which I'd assume is prohibited in IDNA2008
>> based on the LDH-equivalence rules.
It should be: .. adds bidi validity constraints...
>> (Btw, I'd suggest you remove the links from (most of) your examples,
>> because you shouldn't at the same time claim that there is potential for
>> phishing and make it easy to happen. Another issue is that some of these
>> links don't actually resolve, but they look like the should. (e.g.
Agreed. That is already a TODO mentioned at the top of the document.
>> I don't really like the idea of Compatible Preprocessing (section 1.4) at
>> all. Bypassing IDNA2008 lookup by converting to punycode separately is
>> really going too far. I thought the intent of the document was to use either
>> IDNA2003 or IDNA2008, not to simulate IDNA2003 on top of IDNA2008 at all
>> costs by an additional layer.
By applying 1.4, you get nearly the same effect as "try IDNA2008 then try
IDNA2003". That allows browsers and other clients (including us at Google)
to have a single processing step, without having to maintain two different
The draft had earlier a "hybrid" option, whereby the characters were limited
to those accepted by IDNA2008. To my surprise, at the last meeting the
consensus was to drop that.
>> Section 3, Preprocessing: "(For more about the parts of a URL, including
>> the domain name, see [RFC3987])." I don't know why RFC 3987 is relevant
>> here. It may be misunderstood in that the processing is applied to the whole
>> IRI/URI. Also, RFC 3987 doesn't actually define "domain name", nor does it
>> say which parts (of which it mentions several) of an IRI are domain names.
Thanks, that is not a good reference. What would you recommend?
>> In Section 3, Preprocessing, things such as "URI/IRI %-escapes like %2e
>> for U+002E (.) FULL STOP."
That is only in an illustrative section, and not required. It is describing
what is actually done: the browsers accept %xx escapes. However, %2e is
probably not a good example here: changed to
... %C3%A0 for U+00E0 ( à ) LATIN SMALL LETTER A WITH GRAVE.
> or "U+2488 ( ⒈ ) DIGIT ONE FULL STOP"
This is a tricky case. It has been included for compatibility. Both FF and
IE support this behavior, where the domain name is separated into labels *
after* normalization, while Safari and Chrome separate *before.* And FF,
Safari and Chrome both interpret google%2ecom as "a.com" (haven't checked
I put in a review note, and we can run it by the browser representatives.
(For my part, I think it would be cleaner to break into labels at full stops
(normal & fullwidth, ideographic), and not at all characters that have . in
[Review note: this behavior allows characters whose decompositions contain a
dot and other characters. It is included because it represents the
predominant browser behavior (both FF and IE). Similarly, current browsers
interpret "google%2Ecom" as "google.com". Is there good reason to change
are very similar in my eyes to overlong UTF-8 sequences and such, and
>> therefore have a high potential for security problems.
There are issues with this kind of parsing, although in an analysis of URLs
that actually spoof other sites, these characters did not show up.
People who have to use %2e or U+2488 shoot themselves in their foot, and
>> should feel it sooner rather than later. I cannot understand why this
>> document claims to "avoid... security problems" (in the abstract) and then
>> promotes this kind of stuff. As an aside, RFC 3986 recommends %2E in
>> preference to %2e.
As above, it is reflecting current browser behavior. I can run it by the
>> Clicking on
>> http://unicode.org/cldr/utility/list-unicodeset.jsp?a=[:toNFKC=/\./:], I
>> get some kind of Java exception report at
Sorry, there was a recent change in the libraries that caused that to
malfunction. It should work now, as should the
>> Some of the steps in Section 3 are completely cryptic. As an example, what
>> does "trusted source" in "If any label is in Punycode, and does not come
>> from a trusted source" mean?
Sorry, that was a remnant from an earlier version. Removed. Also added a
[Review note: this could be rewritten for clarity as a step 3a: "Convert any
Punycode labels back to Unicode", with explanation of what a Punycode lable
is, and aborting with an error if such conversion fails.]
> What does "validity criteria" in "Abort with error if the label does not
>> comply with the validity criteria" mean? (a pointer to section 5 would help)
There was one in the main line #4, but added another reference for clarity.
>> Also, in Step 3, you split, which means that in Step 5, you have several
>> strings, but you only return one string ?!
Changed to "the domain_name resulting from Step 2". The splitting into
labels is only to apply validity checks.
>> As for mapping tables, what seems to happen is that e.g. a "ß" is mapped
>> to "ss" for lookup, but not for display. This seems to be really a bad
>> combination: You pretend that the browser (or whatever) distinguishes
>> between "ss" and "ß", but redirect to "ss". From a point of view of a search
>> engine, that may be the right thing to do, but assuming that .de allows
>> separate registrations in those cases where there is a real difference
>> between "ss" and "ß", which I hope they will, the above will be the worst of
>> both worlds. If my name is Straßen, and I own straßen.de<http://strassen.de>,
>> and somebody else owns strassen.de because his/her name is Strassen.de,
>> then we both want to be able to make sure people get to the right place, at
>> least once IDNA2008 is deployed.
This is a real issue. There was a long discussion at the UTC, and this was
felt to be a way out of the even worse problem of indeterminacy of labels
containing ß, final sigma, and especially joiners -- with different
browsers, or different versions of browsers, going to different IP addresses
with the same domain name. The problem is that for years to come, a huge
number of browsers will not support 2008 (look at how many people are still
using IE6). If we could wave a magic wand and change all implementations at
once to use the new scheme to use, it might be feasible; otherwise...
>> Section 5: "[Review Note: Once IDNA2008 is final, the exact specifications
>> can be substituted for the last two bullets, making the above
>> self-contained.]": Doing textual substitution in these cases is a really bad
>> idea. Please keep the pointers.
Thanks. I noticed that #5 really shouldn't be there. As a normative change,
I can only add a review note for that.
For the review note, I modified to be
[Review Note: A previous review note suggested that once IDNA2008 is final,
the exact specification be substituted for the last bullet. However, it
would probably be best to retain the pointer. It does raise another issue,
of whether the BIDI spec should be part of the validity test or not:
IDNA2008 doesn't require it in clients.]
>> 5.1: "Remove block description characters" -> "Remove ideographic
>> description characters" or ""Remove ideographic description block"
>> 5.1: I don't understand how "+ [\u002D]" can add back all valid ASCII.
Typo, added review note.
>> 5.2: This description doesn't help at all. What I think a reader would
>> want to know here is how these two sets differ, not some regexp notation of
>> yet another set.
Good point. I added an editorial note to add a table of example differences.
>> Section 8, Tactics: The title should change, maybe "Background" might
Good point; that title was left over from an earlier version.
> It is completely unbelievable that the Unicode consortium would claim, in
>> one of their TRs (in particular one that seems to be headed for "Technical
>> Standard"), that the difference between "ss" and "ß" is essentially a
>> display issue. Overall, this section just repeats stuff in the other
That is unfortunate wording. The conclusion wasn't really that this
difference was only display, but that the most important feature of the
difference was the display. Modified.
>> Section 9, first question: The entries in the table need an explanation.
Added some text.
>> Section 9, advantages of IDNA2008: Yes, please keep that, it helps a
>> reader getting a balanced overview.
>> Section 9, disadvantages of IDNA2008, you say "More fragile in that future
>> Unicode versions require a manual step to avoid instabilities". I don't
>> understand that.
If Unicode version X changes properties in such a way as to add or remove
characters from PVALID, it requires a manual step to retain the previous
status. That step could have been avoided in the formulation, but wasn't.
Added a note.
>> Section 9, bidi label hopping: quotes on no or both sides.
>> Section 9, "Are the "local" mappings just a UI issue?": This seems to
>> imply that "http://türkıye.com <http://xn--trkye-kva78a.com>" and "
>> http://türkiye.com <http://xn--trkiye-3ya.com>" are different under
>> IDNA2008. In my view, this would be great.
"http://türkıye.com <http://xn--trkye-kva78a.com/>" and
are different under *both* IDNA2008 and IDNA2003. The problem is with <a
href="TÜRKIYE.COM <http://xn--trkiye-3ya.COM>">...</a>. You really don't
want one browser going to http://türkıye.com
<http://xn--trkye-kva78a.com/>and a different browser going to
> Can somebody confirm/deny? (the i/ı issue is in my view almost the only
>> justification for having custom mappings). [If denied, you have to remove
>> the examples.]
>> Also, the answers of the type "Bob clicks on the link, and goes to a bad
>> site." should be changed to "Bob clicks on the link, and doesn't find any
>> site, or goes to a wrong (and potentially malicious) site.
That is reasonable.
>> Also, isn't the idea of IDNA2008 to get people to only use lower case,
>> among else? And shouldn't browsers lowercase domain names in their address
>> field, too (as they already do with ASCII-only ones)?
IDNA2008 references an mapping that lowercases, but it is optional. The
browsers do lowercase (actually show the transformed - NFKC-CaseFolded
version) in the address bar.
>> Also, the relationship between "It is generally understood at the W3C that
>> all attributes that take URLs should take full IRIs, not punycoded-URIs, so
>> for example SVG, MathML, XLink, XML, etc, all take IRIs now, as does HTML5."
>> and its main point isn't clear to me.
That isn't actually clear to me why it is here either. It was due to
someone's previous comment, but it looks out of place. Added a note.
>> The whole document needs careful editing/proofreeding before publication
>> (e.g. map map in Section 9).
Yes. I sent an earlier message about the process; the goal was to get the
content out for review, and follow on later with editing.
>> Section 9, "why does IDNA2003 map map final sigma (ς) to sigma (σ), map
>> eszett (ß) to "ss", and delete ZWJ/ZWNJ?": This is trying to beautify things
>> after the fact. What happened when these were decided upon was that the IETF
>> was looking for a table (they didn't want to create their own, because in
>> that specific WG, that would have opened all the doors for weird
>> script-specific requests), and the Unicode Consortium had a table (what's
>> now in NFKC_CaseFold), and so that was taken.
That was supposed to be conveyed by the phrasing "following the Unicode
Standard". Rereading it, I can see how it would strike you as it did. I
don't think that your interpretation is exactly right either, because we did
have tables that simply normalized, and the working group could have chosen
> There is absolutely no need for domain names to be fully case-insensitive
>> with transitivity and round-trips.
Whether or not transitively is a requirement is a matter of some dispute.
Round-tripping isn't mentioned and wasn't a goal. Moreover, roundtripping
with casing is impossible anyway: "McGowan", once transformed by
casemapping, cannot be restored.
> In some sense, ß and ς are indeed anomalous, but they are full parts of the
>> orthographies of the respective languages, and at least the former is
>> distinguishing, in particular for names.
Nobody is claiming that they are not full parts of the orthographies.
However, for neither IDNA2003 nor IDNA2008 is it claimed that *all* parts of
every language's orthographies, and all the distinctions therein are
representable in domain names. There are trivial examples even in English,
like "can't" vs "cant", which cannot be represented.
>> "The rough consensus among the working group": Which WG?
> IETF IDNA
>> Overall, my impression is that this document isn't yet ready for approval.
>> [Overall, my feeling is that some of the text in this document (not all of
>> it, of course) must feel quite a bit similar (to IETF people) to some of the
>> text in (earlier versions? I haven't had time to read any recent versions)
>> of the Rationale document or some earlier documents, in particular in some
>> draft stages, on the IETF side, that the Unicode side didn't like.]
I think it does reflect some concerns that the Unicode had. It also
represents discussion with browser vendors as to what is feasible. Over
time, once IDNA2008 is fully deployed in registries, then the compatibility
"shim" provided by this specification would hopefully become unnecessary.
>> Regards, Martin.
>> On Mon, Oct 5, 2009 at 00:43, Harald Alvestrand<harald at alvestrand.no
>>> Mark Davis ☕ wrote:
>>>> The IESG may encounter implementation tactics for dealing with the old
>>>>> and new specifications that are controversial.
>>>>> One set of implementation tactics is UTS#46 Unicode IDNA Compatible
>>>>> Preprocessing<http://www.unicode.org/reports/tr46/> (in draft).
>>>>> Yes, that's one of the controversial ones.
>>>> The UTC will be considering that for approval at its upcoming meeting,
>>>>> people with concerns may want to discuss and submit them to the UTC.
>> #-# Martin J. Dürst, Professor, Aoyama Gakuin University
>> #-# http://www.sw.it.aoyama.ac.jp mailto:duerst at it.aoyama.ac.jp
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Idna-update