Let me see if I can clarify the situation.
Once a version of Unicode is issued, the consortium makes no retroactive
changes. Thus if someone claims and correctly implements conformance to
"Unicode Version X", their implementation will remain conformant to that
version forever.
The corrigenda *can be* applied to previous versions, if someone wants
to: but in that case someone would claim and implement conformance to
"Unicode Version X *plus* Corrigendum Y".
On the larger issue of problems in
http://www.ietf.org/internet-drafts/draft-iab-idn-nextsteps-05.txt, the
consortium has communicated a number of problems in that document to the
IAB and the authors, but has never gotten a response on any of those
problems. While the issue appears to be moot, since the document is
being issued without consideration of those flaws, I'll copy the
comments here for those interested. The original comments were on
http://www.iab.org/documents/drafts/draft-iab-idn-nextsteps-02.txt; but
the text has not improved regarding these issues since that point.
=====
The UTC strongly supports many of the goals of the document, including
especially improving the security of IDNs, and updating the version of
Unicode used in NamePrep and StringPrep (since the old version of
Unicode they require excludes or hampers many languages). There are,
however, a number of areas of concern.
As a general issue, we'd urge closer cooperation between the IAB and the
Unicode consortium on the document, so that the character encoding and
software internationalization issues can be reviewed by experts in the
field, and accurately represented in the document.
The chief area of concern is section 4.3.
4.3. Combining Characters and Character Components
One thing that increases IDNA complexity and the need for
normalization is that combining characters are permitted. Without
them, complexity might be reduced enough to permit more easy
transitions to new versions. The community should consider whether
combining characters should be prohibited entirely from IDNs. A
consequence of this, of course, is that each new language or script
would require that all of its characters have Unicode assignments to
specific, precomposed, code points, a model that the Unicode
Consortium has rejected for Roman-based scripts. For non-Roman
scripts, it seems to be the Unicode trend to define such code points.
At some level, telling the users and proponents of scripts that, at
present, require composing characters to work the issues out with the
Unicode Consortium in a way that severely constrains the need for
those characters seems only appropriate. The IAB and the IETF should
examine whether it is appropriate to press the Unicode Consortium to
revise these policies or otherwise to recommend actions that would
reduce the need for normalization and the related complexities.
The descriptions and recommendations in this section are simply not
feasible. They do not recognize the fundamental importance of combining
marks as an integral component of a great many scripts, nor do they
recognize the fundamental need for compatibility that is required of the
Unicode Standard. Asking for combining characters to be removed is akin
to asking English vowels to be removed, and all possible syllables to be
encoded instead. There are, as well, a number of purely factual errors.
For example, "it seems to be the Unicode trend to define such code
points" is simply incorrect. This section serves no purpose but to
betray a basic lack of understanding of scripts; it needs to be removed
entirely.
A second area of major concern is Section 2.2.3.
2.2.3. Normalization and Character Mappings
Unicode contains several different models for representing
characters. The Chinese (Han)-derived characters of the "CJK"
languages are "unified", i.e., characters with common derivation and
similar appearances are assigned to the same code point. European
characters derived from a Greek-Roman base are separated into
separate code blocks for "Latin", Greek and Cyrillic even when
individual characters are identical in both form and semantics.
Separate code points based on font differences alone are generally
prohibited, but a large number of characters for "mathematical" use
have been assigned separate code points even though they differ from
base ASCII characters only by font attributes such as "script",
"bold", or "italic". Some characters that often appear together are
treated as typographical digraphs with specific code points assigned
to the combination, others require that the two-character sequences
be used, and still others are available in both forms. Some Roman-
based letters that were developed as decorated variations on the
basic Latin letter collection (e.g., by addition of diacritical
marks) are assigned code points as individual characters, others must
be built up as two (or more) character sequences using "composing
characters".
This section betrays a lack of understanding of the fundamental
differences between Han characters and the scripts Latin, Greek, and
Cyrillic.
Many of these differences result from the desire to maintain backward
compatibility while the standard evolved historically, and are hence
understandable. However, the DNS requires precise knowledge of which
codes and code sequences represent the same character and which ones
do not. Limiting the potential difficulties with confusable
characters (see Section 2.2.6) requires even more knowledge of which
characters might look alike in some fonts but not in others. These
variations make it difficult or impossible to apply a single set of
rules to all of Unicode. Instead, more or less complex mapping
tables, defined on a character by character basis, are required to
"normalize" different representations of the same character to a
single form so that matching is possible.
The Unicode consortium *does* supply a precise mechanism for determining
when two strings represent the same underlying abstract characters.
These do supply a single set of rules to all of Unicode, based on a set
of data that is in the Unicode Character Database.
This paragraph also conflates the confusable issue with character
equivalence. These are separate issues: there are great many instances
where characters are confusable where they are not at all equivalent
(such as zero and the letter O).
... The fact
that most or all scripts included in Unicode have been initially
incorporated by copying an existing standard more or less intact has
impact on the optimization of these algorithms and on forward
compatibility. Even if the language is known and language-specific
rules can be defined, dependencies on the language do not disappear.
Any canonicalization operations that depend on more than short
sequences of text is not possible to do without context. DNS lookups
and many other operations do not have a way to capture and utilize
the language or other information that would be needed to provide
that context.
First, it is neither "most" nor "all". Very few scripts,
proportionately, have been incorporated by copying an existing standard.
Second, "Any canonicalization operations that depend on more than short
sequences of text is not possible to do without context...." is
difficult to make sense of. One would have to explain the sense of
"canonicalization" that is being discussed. It could be as trivial as
"language-based canonicalization is impossible without language
information", which is true, but above the document argues against using
language-based equivalences on a global basis (and for very good reason!)
===
Other areas of concern:
(more properly "Roman", see below)
The common modern practice in the naming of the script is to use the
term "Latin", not "Roman". Whether or not one thinks that should not
have been the case, insisting on older terms is pointless, and not
germane to the purpose of the document.
When writing or typing the label (or word), a script must be selected
and a charset must be picked for use with that script.
This is confusing charset, keyboard and script. Saying "a script must be
selected" is *neither* true from the user's perspective, nor does it at
all match the implementation pipeline from keypress to storage of a
label. What may have been confusing for the authors is that sometimes
keyboards that are listed for selection are sorted by script; that does
not, however, mean that a "script is selected".
The proper word, if more substantial changes are not made to the
wording, would be "a keyboard must be selected". (Even that is a quite
odd, since it implies that that is done each time a user types a label.)
If that charset, or the local charset being used by the relevant
operating system or application software, is not Unicode, a further
conversion must be performed to produce Unicode. How often this is
an issue depends on estimates of how widely Unicode is deployed as
the native character set for hardware, operating systems, and
applications. Those estimates differ widely, with some Unicode
advocates claiming that it is used in the vast majority of systems
and applications today. Others are more skeptical, pointing out
that:
o ISO 8859 versions [ISO.8859.1992] and even national variations of
ISO 646 [ISO.646.1991] are still widely used in parts of Europe;
o code-table switching methods, typically based on the techniques of
ISO 2022 [ISO.2022.1986] are still in general use in many parts of
the world, especially in Japan with Shift-JIS and its variations;
o that computing, systems, and communications in China tend to use
one or more of the national "GB" standards rather than native
Unicode;
o and so on.
Not all charsets define their characters in the same way and not all
pre-existing coding systems were incorporated into Unicode without
changes. Sometimes local distinctions were made that Unicode does
not make or vice versa. Consequently, conversion from other systems
to Unicode may potentially lose information.
Most of this section is unnecessary and the thrust of it is misleading.
The only issue is "local distinctions" are lost when converting to
Unicode; that doesn't happen when converting from any of the examples
listed. This passage implies that there are significant problems in
mapping to Unicode in doing IDN, and there simply aren't.
... Worse, one needs to be reasonably
familiar with a script and how it is used to understand how much
characters can reasonably vary as the result of artistic fonts and
typography. For example, there are a few fonts for Latin characters
that are sufficiently highly ornamented that an observer might easily
confuse some of the characters with characters in Thai script.
The confusion of Latin with Thai is a red herring. It would take an
exceedingly contrived scenario for it to present a problem. There are
plenty of realistic scenarios involving confusables across, say, Latin
and Cyrillic.
... IDNA
prohibits these mixed-directional (or bidirectional) strings in IDN
labels, but the prohibition causes other problems such as the
rejection of some otherwise linguistically and culturally sensible
strings. As Unicode and conventions for handling so-called
bidirectional ("BIDI") strings evolve, the prohibition in IDNA should
be reviewed and reevaluated.
Deviating from the practices already built into IRI would be a mistake.
As the document recognizes above, it cannot be a goal to represent all
possible "linguistically and culturally sensible strings" in IDNs. The
restrictions on BIDI are ones that have achieved broad consensus as the
minimal ones to help avoid some fairly serious security issues.
4.1.2. Elimination of word-separation punctuation
... We might even
consider banning use of the hyphen itself in non-ASCII strings or,
less restrictively, strings that contained non-Roman characters.
This section is not well motivated. The authors need to justify why such
characters represent a problem (and one of such a serious nature that
hyphens should be disallowed).
-----
* Section 2.2.3: "characters that are essentially identical will not match"
What is meant by "essentially identical"? Does this mean identical in
appearance, identical in internal representation, identical in
semantics, canonically equivalent (same NFC forms), or compatible
equivalent (same NFKC forms)? The intent needs to be clarified,
otherwise the statement is subject to misinterpretation.
* Section 2.2.3: "This Unicode normalization process [does not account
for] equivalences that are language or script dependent"
Which what is meant by "script-dependent equivalences"? Can you provide
an example?
* Section 2.2.3: "U+00F8 [...] and U+00F6 [...] are considered to match
in Swedish"
"Match" needs some clarification. In accordance with Swedish standards,
when collating with Swedish locale, all major implementations match
these characters at the first and second level, but not at a lower
level. Thus they are not exact matches: this might be better phrased in
terms of equivalence.
* Section 2.2.3: "Even if the language is known and language-specific
rules can be defined, dependencies on the language do not disappear"
It is unclear what this means. Could you give an example?
* Section 2.2.1: "Those characters are not treated as equivalent
according to the Unicode consortium while...".
This is somewhat ad hominem. It should rather be "...according to the
Unicode Standard while..."
* Section 2.2.1: "..confusion in Germany, where the U+00F8 character is
never used in the language".
That is not true, there are entries with that character in the Duden
dictionary.
* Section 2.2.4: "This is because [...] some glyphs [...] have been
assigned different code points in Unicode".
This is incorrect: glyphs are not assigned to code points; characters are.
* Section 2.2.6: "Is the answer the same for words two [sic] different
languages that translate into each other?".
This is completely orthogonal to IDNs (cf "Is 'cat' the same as 'gato'
or the same as 'katze'?").
* Section 2.2.7: "the IESG statement [...] that a registry should have a
policy about the scripts, languages, code points and text directions".
This appears to not be an accurate paraphrase of
(http://www.ietf.org/IESG/STATEMENTS/IDNstatement.txt). That document
rather says a registry "MIGHT want to prevent particular characters",
"MIGHT want to automatically generate a list of (...) strings and
suggest that they also be registered" and lastly "it is suggested that a
registry act conservatively". There is no such thing as "SHOULD" wording
and, for instance, text direction is not mentioned.
* Section 2.2.8: "This maybe [...] because many other applications are
internally sensitive only to the appearance of characters and not to
their representation".
This is reversed. The vast majority of application are internally
sensitive only to the representation, not to the appearance. Exceptions
would be OCR, for example.
* Section 2.2.8: "A change in a code point assignment (...) may be
extremely disruptive".
This suggests that the consortium capriciously changes code points.
After the merger with ISO 10646 there was only one point at which the
Unicode consortium changed code points: Unicode 2.0.0 (July, 1996): The
characters in the Korean Hangul block were moved to be part of a new,
larger block with all 11,152 Hangul syllables.
As a result of the disruption that this caused, the Unicode Consortium
and ISO/IEC SC2 resolved never to change code points in the future, and
no changes have ever been done since.
* Section 3.1.1: "...such as code points assigned to font variations...".
Which characters are these referring to? Is it to just characters that
are resolved by an NFKC normalization, or does it refer to others?
* Section 4.5: "the whois protocol itself (...) is ASCII-only".
This appears to be inaccurate. The Whois protocol
(http://www.ietf.org/rfc/rfc3912.txt?number=3912) has no mechanisms to
indicate which character encoding is being used, but the protocol is
8-bit clean and it is indeed used so by many (for instance, DENIC has a
UTF-8 implementation up and running).
-----
(a later comment, on the 04 draft)
In my opinion (not speaking for the UTC here), not only were the issues
from the UTC not addressed, if anything the document regressed. It now
contains the following text:
The IAB and the IETF should examine whether it is appropriate to
press the Unicode Consortium to revise these policies or otherwise to
recommend actions that would reduce the need for normalization and
the related complexities. However, participants in the Unicode
Technical Committee have told us, on behalf of the Committee, that
they would not consider adding the number of precomposed characters
required to support existing languages and scripts, much less new
ones. So this option may not be feasible. Retaining combining
characters without further global restrictions may leave us "stuck"
at Unicode 3.2, leading either to incompatibility differences in
applications that otherwise use a modern version of Unicode (while
IDN remains at Unicode 3.2) or to painful transitions to new
versions.
This wording makes it seem as if it is just some whim of the UTC that
prevents it from adding all possible combinations of sequences that
would involve combining marks and then removing all combining marks.
That is no whim; such a change would be a massive disruption for the
encoding, somewhat akin to changing the encoding for English to be on
the syllable level rather than individual letters. Moreover, none of the
other issues raised by the UTC appear to have been even considered. Is
there any way that we can set up a more effective method of communication?
=====
(I'll add one further comment to the above -- it is a complete non
sequitur to imply that the IETF can't move beyond Unicode 3.2 because of
combining marks.)
Mark
Jeffrey Hutzelman wrote:
On Fri, 21 Apr 2006, Simon Josefsson wrote:
On the other hand, potential security issues caused by instability in
the original (erroneous) definition are at least as serious as the
potential incompatibilities caused by the change.
Can you expand on this?
Can you give an example of how a system that contain components that
all properly implement StringPrep and the NFKC from Unicode 3.2, that
is without the PR-29 fix, have a serious security issues?
I haven't seen any such example. I believe examples can be
constructed when one StringPrep implementation in a system implement
the original Unicode 3.2 NFKC semantics and one component implement
the "fixed" NFKC. If you don't see it, I can try to produce a
complete scenario.
The problem is that the original NFKC isn't just "different"; it's
_unstable_. That is, there exist strings for which you get different
results depending on _how many times_ you apply NFC or NFKC. Systems
which use normalization to ensure that a security-related identifier is
always represented in the same way may break in the face of an unstable
normalization.
Of course, this only happens for the same highly-unlikely sequences...
incompatibilities, despite that the change is not meant to modify
how the version of Unicode that IDNA reference is implemented.
Except I still think you're trying to put words in the UTC's mouth that it
did not intend to say. As I understand it, a corrigendum like this _is_
intended to modify the existing standard, including existing references.
It's roughly the equivalent of RFC errata - they're saying "there's a bug
in the text; it says X but we always meant for the spec to be Y".
Now, you can argue that the IETF should take the position that any of its
specs which refer to Unicode 3.2 and were published prior to corrigendum
#5 should be construed to mandate the use of the original, flawed
algorithms for NFC and NFKC. But you seem to be taking the position that
the _Unicode Technical Committee_ had that intent with respect to existing
users of its standard, and I don't think that's true.
I also don't think the IETF should explicitly mandate the use of the
flawed algorithm by implemntations of preexisting specs. As it turns out,
one of the justifications for making the retroactive change to Unicode was
that a significant number of implementations, including whatever reference
or test implementation they used when writing the spec _and_ the sample
code which appears _in_ the spec, get it right.
-- Jeff
_______________________________________________
Ietf@xxxxxxxx
https://www1.ietf.org/mailman/listinfo/ietf
_______________________________________________
Ietf@xxxxxxxx
https://www1.ietf.org/mailman/listinfo/ietf