Section 1.2.2 of the applications-layer Host Requirements specification [RFC1123]). This principle applies to IDNA. In applying the principle to registries as the source ("sender") of all registered and useful IDNs, registries are responsible for being conservative about what they register and put out in the Internet. For IDNs to work well, zone administrators (registries) must have and require sensible policies about what is registered -- conservative policies -- and implement and enforce them. Conversely, lookup applications are expected to reject labels that clearly violate global (protocol) rules (no one has ever seriously claimed that being liberal in what is accepted requires being stupid). However, once one gets past such global rules and deals with anything sensitive to script or locale, it is necessary to assume that garbage has not been placed into the DNS, i.e., one must be liberal about what one is willing to look up in the DNS rather than guessing about whether it should have been permitted to be registered. If a string cannot be successfully found in the DNS after the lookup processing described here, it makes no difference whether it simply wasn't registered or was prohibited by some rule at the registry. Application implementers should be aware that where DNS wildcards are used, the ability to successfully resolve a name does not guarantee that it was actually registered.
While there are strong arguments for any domain name that is placed "on the wire" -- transmitted between systems -- to be in the zero- ambiguity forms of A-labels, it is inevitable that programs that process domain names will encounter U-labels or variant forms. An application that implements the IDNA protocol [RFC5891] will always take any user input and convert it to a set of Unicode code points. That user input may be acquired by any of several different input methods, all with differing conversion processes to be taken into consideration (e.g., typed on a keyboard, written by hand onto some sort of digitizer, spoken into a microphone and interpreted by a speech-to-text engine, etc.). The process of taking any particular user input and mapping it into a Unicode code point may be a simple one: if a user strikes the "A" key on a US English keyboard, without any modifiers such as the "Shift" key held down, in order to draw a Latin small letter A ("a"), many (perhaps most) modern operating system input methods will produce to the calling application the code point U+0061, encoded in a single octet. Sometimes the process is somewhat more complicated: a user might strike a particular set of keys to represent a combining macron followed by striking the "A" key in order to draw a Latin small letter A with a macron above it. Depending on the operating system, the input method chosen by the user, and even the parameters with which the application communicates with the input method, the result might be the code point U+0101 (encoded as two octets in UTF-8 or UTF-16, four octets in UTF-32, etc.), the code point U+0061 followed by the code point U+0304 (again, encoded in three or more octets, depending upon the encoding used) or even the code point U+FF41 followed by the code point U+0304 (and encoded in some form). These examples leave aside the issue of operating systems and input methods that do not use Unicode code points for their character set. In every case, applications (with the help of the operating systems on which they run and the input methods used) need to perform a mapping from user input into Unicode code points. IDNA2003 used a model whereby input was taken from the user, mapped (via whatever input method mechanisms were used) to a set of Unicode code points, and then further mapped to a set of Unicode code points using the Nameprep profile [RFC3491]. In this procedure, there are two separate mapping steps: first, a mapping done by the input method (which might be controlled by the operating system, the application, or some combination) and then a second mapping performed by the Nameprep portion of the IDNA protocol. The mapping done in Nameprep includes a particular mapping table to re-map some characters to other characters, a particular normalization, and a set of prohibited characters.
Note that the result of the two-step mapping process means that the mapping chosen by the operating system or application in the first step might differ significantly from the mapping supplied by the Nameprep profile in the second step. This has advantages and disadvantages. Of course, the second mapping regularizes what gets looked up in the DNS, making for better interoperability between implementations that use the Nameprep mapping. However, the application or operating system may choose mappings in their input methods, which when passed through the second (Nameprep) mapping result in characters that are "surprising" to the end user. The other important feature of IDNA2003 is that, with very few exceptions, it assumes that any set of Unicode code points provided to the Nameprep mapping can be mapped into a string of Unicode code points that are "sensible", even if that means mapping some code points to nothing (that is, removing the code points from the string). This allowed maximum flexibility in input strings. The present version of IDNA (IDNA2008) differs significantly in approach from the original version. First and foremost, it does not provide explicit mapping instructions. Instead, it assumes that the application (perhaps via an operating system input method) will do whatever mapping it requires to convert input into Unicode code points. This has the advantage of giving flexibility to the application to choose a mapping that is suitable for its user given specific user requirements, and avoids the two-step mapping of the original protocol. Instead of a mapping, IDNA2008 provides a set of categories that can be used to specify the valid code points allowed in a domain name. In principle, an application ought to take user input of a domain name and convert it to the set of Unicode code points that represent the domain name the user intends. As a practical matter, of course, determining user intent is a tricky business, so an application needs to choose a reasonable mapping from user input. That may differ based on the particular circumstances of a user, depending on locale, language, type of input method, etc. It is up to the application to make a reasonable choice.
RFC4690], two key goals of the IDNA2008 design are: o to enable applications to be agnostic about whether they are being run in environments supporting any Unicode version from 3.2 onward. o to permit incrementally adding new characters, character groups, scripts, and other character collections as they are incorporated into Unicode, doing so without disruption and, in the long term, without "heavy" processes (an IETF consensus process is required by the IDNA2008 specifications and is expected to be required and used until significant experience accumulates with IDNA operations and new versions of Unicode). RFC5891] and Tables [RFC5892] documents): o The characters are "letters", marks needed to form letters, numerals, or other code points used to write words in some language. Symbols, drawing characters, and various notational characters are intended to be permanently excluded. There is no evidence that they are important enough to Internet operations or internationalization to justify expansion of domain names beyond the general principle of "letters, digits, and hyphen". (Additional discussion and rationale for the symbol decision appears in Section 7.6.) o Other than in very exceptional cases, e.g., where they are needed to write substantially any word of a given language, punctuation characters are excluded. The fact that a word exists is not proof that it should be usable in a DNS label, and DNS labels are not expected to be usable for multiple-word phrases (although they are certainly not prohibited if the conventions and orthography of a particular language cause that to be possible). o Characters that are unassigned (have no character assignment at all) in the version of Unicode being used by the registry or application are not permitted, even on lookup. The issues involved in this decision are discussed in Section 7.7.
o Any character that is mapped to another character by a current version of NFKC is prohibited as input to IDNA (for either registration or lookup). With a few exceptions, this principle excludes any character mapped to another by Nameprep [RFC3491]. The principles above drive the design of rules that are specified exactly in the Tables document. Those rules identify the characters that are valid under IDNA. The rules themselves are normative, and the tables are derived from them, rather than vice versa. RFC0952] where those requirements are imposed by the expected applications. Other applications contexts, such as the later addition of special service location formats [RFC2782] imposed new requirements on zone administrators. For zones that will contain IDNs, support for Unicode version-independence requires restrictions on all strings placed in the zone. In particular, for such zones (the exact rules appear in Section 4 of the Protocol document [RFC5891]): o Any label that appears to be an A-label, i.e., any label that starts in "xn--", must be valid under IDNA, i.e., they must be valid A-labels, as discussed in Section 2 above. o The Unicode tables (i.e., tables of code points, character classes, and properties) and IDNA tables (i.e., tables of contextual rules such as those that appear in the Tables document), must be consistent on the systems performing or validating labels to be registered. Note that this does not require that tables reflect the latest version of Unicode, only that all tables used on a given system are consistent with each other. Under this model, registry tables will need to be updated (both the Unicode-associated tables and the tables of permitted IDN characters) to enable a new script or other set of new characters. The registry will not be affected by newer versions of Unicode, or newly authorized characters, until and unless it wishes to support them. The zone administrator is responsible for verifying validity for IDNA as well as its local policies -- a more extensive set of checks than are required for looking up the labels. Systems looking up or
resolving DNS labels, especially IDN DNS labels, must be able to assume that applicable registration rules were followed for names entered into the DNS. Section 5 of the Protocol document [RFC5891]): o Maintain IDNA and Unicode tables that are consistent with regard to versions, i.e., unless the application actually executes the classification rules in the Tables document [RFC5892], its IDNA tables must be derived from the version of Unicode that is supported more generally on the system. As with registration, the tables need not reflect the latest version of Unicode, but they must be consistent. o Validate the characters in labels to be looked up only to the extent of determining that the U-label does not contain "DISALLOWED" code points or code points that are unassigned in its version of Unicode. o Validate the label itself for conformance with a small number of whole-label rules. In particular, it must verify that: * there are no leading combining marks, * the Bidi conditions are met if right-to-left characters appear, * any required contextual rules are available, and * any contextual rules that are associated with joiner characters (and CONTEXTJ characters more generally) are tested. o Do not reject labels based on other contextual rules about characters, including mixed-script label prohibitions. Such rules may be used to influence presentation decisions in the user interface, but not to avoid looking up domain names. To further clarify the rules about handling characters that require contextual rules, note that one can have a context-required character (i.e., one that requires a rule), but no rule. In that case, the character is treated the same way DISALLOWED characters are treated, until and unless a rule is supplied. That state is more or less equivalent to "the idea of permitting this character is accepted in principle, but it won't be permitted in practice until consensus is reached on a safe way to use it".
The ability to add a rule more or less exempts these characters from the prohibition against reclassifying characters from DISALLOWED to PVALID. And, obviously, "no rule" is different from "have a rule, but the test either succeeds or fails". Lookup applications that follow these rules, rather than having their own criteria for rejecting lookup attempts, are not sensitive to version incompatibilities with the particular zone registry associated with the domain name except for labels containing characters recently added to Unicode. An application or client that processes names according to this protocol and then resolves them in the DNS will be able to locate any name that is registered, as long as those registrations are valid under IDNA and its version of the IDNA tables is sufficiently up to date to interpret all of the characters in the label. Messages to users should distinguish between "label contains an unallocated code point" and other types of lookup failures. A failure on the basis of an old version of Unicode may lead the user to a desire to upgrade to a newer version, but will have no other ill effects (this is consistent with behavior in the transition to the DNS when some hosts could not yet handle some forms of names or record types). Unicode52]), generate different characters or sets of characters. Those operations are not reversible and lose even more information than traditional uppercase or lowercase transformations, but are more useful than those transformations for comparison purposes. Two notable characters of this type are the German character Eszett (Sharp S, U+00DF) and the Greek Final Form Sigma (U+03C2). The former is case folded to the ASCII string "ss", the latter to a medial (lowercase) Sigma (U+03C3).
Section 3.1.2 above, those characters are essential for writing many reasonable mnemonics for certain scripts. However, treating them as valid in IDNA2008, even with contextual restrictions, raises approximately the same problem as exists with Eszett and Final Sigma: strings that were valid under IDNA2003 have different interpretations as labels, and different A-labels, than the same strings under this newer version.
registries that already had a considerable number of labels did when IDNs were first introduced. Section 6, IDNA2003, via Nameprep (see Section 7.5), mapped many characters into related ones. Those mappings no longer exist as requirements in IDNA2008. These specifications strongly prefer that only A-labels or U-labels be used in protocol contexts and as much as practical more generally. IDNA2008 does anticipate situations in which some mapping at the time of user input into lookup applications is appropriate and desirable. The issues are discussed in Section 6 and specific recommendations are made in the Mapping document [IDNA2008-Mapping]. Section 7.2). Note that if the input string was valid under one version and not valid under the other, this condition would not apply. See the first item in Section 7.4.2, below.
3. A fundamental change was made to the semantics of the string that would be inserted in the DNS, e.g., if a decision were made to try to include language or script information in the encoding in addition to the string itself. 4. A sufficiently large number of characters were added to Unicode so that the Punycode mechanism for block offsets would no longer reference the higher-numbered planes and blocks. This condition is unlikely even in the long term and certain not to arise in the next several years.
Consequently, a prefix change should have been, and was, avoided if at all possible, even if it means accepting some IDNA2003 decisions about character distinctions as irreversible and/or giving special treatment to edge cases. RFC3491], a key part of IDNA2003, is a profile of Stringprep [RFC3454]. While Nameprep is a Stringprep profile specific to IDNA, Stringprep is used by a number of other protocols. Were Stringprep to have been modified by IDNA2008, those changes to improve the handling of IDNs could cause problems for non-DNS uses, most notably if they affected identification and authentication protocols. Several elements of IDNA2008 give interpretations to strings prohibited under IDNA2003 or prohibit strings that IDNA2003 permitted. Those elements include the new inclusion information in the Tables document [RFC5892], the reduction in the number of characters permitted as input for registration or lookup (Section 3), and even the changes in handling of right-to-left strings as described in the Bidi document [RFC5893]. IDNA2008 does not use Nameprep or Stringprep at all, so there are no side-effect changes to other protocols. It is particularly important to keep IDNA processing separate from processing for various security protocols because some of the constraints that are necessary for smooth and comprehensible use of IDNs may be unwanted or undesirable in other contexts. For example, the criteria for good passwords or passphrases are very different from those for desirable IDNs: passwords should be hard to guess, while domain names should normally be easily memorable. Similarly, internationalized Small Computer System Interface (SCSI) identifiers and other protocol components are likely to have different requirements than IDNs.
specification operates on an inclusion model, extrapolating from the original "hostname" rules (LDH, see the Definitions document [RFC5890]) -- which have served the Internet very well -- to a Unicode base rather than an ASCII base. 2. Symbol names are more problematic than letters because there may be no general agreement on whether a particular glyph matches a symbol; there are no uniform conventions for naming; variations such as outline, solid, and shaded forms may or may not exist; and so on. As just one example, consider a "heart" symbol as it might appear in a logo that might be read as "I love...". While the user might read such a logo as "I love..." or "I heart...", considerable knowledge of the coding distinctions made in Unicode is needed to know that there is more than one "heart" character (e.g., U+2665, U+2661, and U+2765) and how to describe it. These issues are of particular importance if strings are expected to be understood or transcribed by the listener after being read out loud. 3. Design of a screen reader used by blind Internet users who must listen to renderings of IDN domain names and possibly reproduce them on the keyboard becomes considerably more complicated when the names of characters are not obvious and intuitive to anyone familiar with the language in question. 4. As a simplified example of this, assume one wanted to use a "heart" or "star" symbol in a label. This is problematic because those names are ambiguous in the Unicode system of naming (the actual Unicode names require far more qualification). A user or would-be registrant has no way to know -- absent careful study of the code tables -- whether it is ambiguous (e.g., where there are multiple "heart" characters) or not. Conversely, the user seeing the hypothetical label doesn't know whether to read it -- try to transmit it to a colleague by voice -- as "heart", as "love", as "black heart", or as any of the other examples below. 5. The actual situation is even worse than this. There is no possible way for a normal, casual, user to tell the difference between the hearts of U+2665 and U+2765 and the stars of U+2606 and U+2729 without somehow knowing to look for a distinction. We have a white heart (U+2661) and few black hearts. Consequently, describing a label as containing a heart is hopelessly ambiguous: we can only know that it contains one of several characters that look like hearts or have "heart" in their names. In cities where "Square" is a popular part of a location name, one might well want to use a square symbol in a label as well and there are far more squares of various flavors in Unicode than there are hearts or stars.
The consequence of these ambiguities is that symbols are a very poor basis for reliable communication. Consistent with this conclusion, the Unicode standard recommends that strings used in identifiers not contain symbols or punctuation [Unicode-UAX31]. Of course, these difficulties with symbols do not arise with actual pictographic languages and scripts which would be treated like any other language characters; the two should not be confused. RFC5892] (such as a compatibility character). In IDNA2003, since there is no direct dependency on NFKC (many of the entries in Stringprep's tables are based on NFKC, but IDNA2003 depends only on Stringprep), allocation of a compatibility character might produce some odd situations, but it would not be a problem. In IDNA2008, where compatibility characters are DISALLOWED unless character-specific exceptions are made, permitting strings containing unassigned characters to be looked up would violate the principle that characters in DISALLOWED are not looked up. o The Unicode Standard specifies that an unassigned code point normalizes (and, where relevant, case folds) to itself. If the code point is later assigned to a character, and particularly if the newly assigned code point has a combining class that
determines its placement relative to other combining characters, it could normalize to some other code point or sequence. It is possible to argue that the issues above are not important and that, as a consequence, it is better to retain the principle of looking up labels even if they contain unassigned characters because all of the important scripts and characters have been coded as of Unicode 5.2 (or even earlier), and hence unassigned code points will be assigned only to obscure characters or archaic scripts. Unfortunately, that does not appear to be a safe assumption for at least two reasons. First, much the same claim of completeness has been made for earlier versions of Unicode. The reality is that a script that is obscure to much of the world may still be very important to those who use it. Cultural and linguistic preservation principles make it inappropriate to declare the script of no importance in IDNs. Second, we already have counterexamples, e.g., in the relationships associated with new Han characters being added (whether in the BMP or in Unicode Plane 2). Independent of the technical transition issues identified above, it can be observed that any addition of characters to an existing script to make it easier to use or to better accommodate particular languages may lead to transition issues. Such additions may change the preferred form for writing a particular string, changes that may be reflected, e.g., in keyboard transition modules that would necessarily be different from those for earlier versions of Unicode where the newer characters may not exist. This creates an inherent transition problem because attempts to access labels may use either the old or the new conventions, requiring registry action whether or not the older conventions were used in labels. The need to consider transition mechanisms is inherent to evolution of Unicode to better accommodate writing systems and is independent of how IDNs are represented in the DNS or how transitions among versions of those mechanisms occur. The requirement for transitions of this type is illustrated by the addition of Malayalam Chillu in Unicode 5.1.0.
get an xn-- prefix, but the string that can be displayed to a user appears to be an IDN. IDNA2008 eliminates this artifact. A character is either permitted as itself or it is prohibited; special cases that make sense only in a particular linguistic or cultural context can be dealt with as localization matters where appropriate. RFC 1034) and DNS update messages [RFC2136]) could not be IDNA-aware because they predate IDNA. Other sections of this document provide the needed shielding by ensuring that internationalized domain names entering DNS server databases through such channels have already been converted to their equivalent ASCII A-label forms. Because of the distinction made between the algorithms for Registration and Lookup in Sections 4 and 5 (respectively) of the Protocol document [RFC5891] (a domain name containing only ASCII code points cannot be converted to an A-label), there cannot be more than one A-label form for any given U-label. As specified in clarifications to the DNS specification [RFC2181], the DNS protocol explicitly allows domain labels to contain octets beyond the ASCII range (0000..007F), and this document does not change that. However, although the interpretation of octets 0080..00FF is well-defined in the DNS, many application protocols support only ASCII labels and there is no defined interpretation of these non-ASCII octets as characters and, in particular, no interpretation of case-independent matching for them (e.g., see the clarification on DNS case insensitivity [RFC4343]). If labels containing these octets are returned to applications, unpredictable behavior could result. The A-label form, which cannot contain those characters, is the only standard representation for internationalized labels in the DNS protocol.
so Extension Mechanisms for DNS (EDNS0) [RFC2671] support may be more important (otherwise, queries and responses may be forced to go to TCP instead of UDP). RFC2277]) are applicable to situations in which language identification is used to provide language-specific contexts. The DNS is, by contrast, global and international and ultimately has nothing to do with languages. Adding languages (or similar context) to IDNs generally, or to DNS matching in particular, would imply context-dependent matching in DNS, which would be a very significant change to the DNS protocol itself. It would also imply that users would need to identify the language associated with a particular label in order to look that label up. That knowledge is generally not available because many labels are not words in any language and some may be words in more than one. RFC5892]. This document describes the registries, but it does not specify any IANA actions.
Section 3.1.2). The details for those rules appear in the Tables document. RFC5890].
Campbell, Martin Duerst, Subramanian Moonesamy, Peter Saint-Andre, and Dan Winship, for catching specific errors and recommending corrections. A meeting was held on 30 January 2008 to attempt to reconcile differences in perspective and terminology about this set of specifications between the design team and members of the Unicode Technical Consortium. The discussions at and subsequent to that meeting were very helpful in focusing the issues and in refining the specifications. The active participants at that meeting were (in alphabetic order, as usual) Harald Alvestrand, Vint Cerf, Tina Dam, Mark Davis, Lisa Dusseault, Patrik Faltstrom (by telephone), Cary Karp, John Klensin, Warren Kumari, Lisa Moore, Erik van der Poel, Michel Suignard, and Ken Whistler. We express our thanks to Google for support of that meeting and to the participants for their contributions. Useful comments and text on the working group versions of the working draft were received from many participants in the IETF "IDNABIS" working group and a number of document changes resulted from mailing list discussions made by that group. Marcos Sanz provided specific analysis and suggestions that were exceptionally helpful in refining the text, as did Vint Cerf, Martin Duerst, Andrew Sullivan, and Ken Whistler. Lisa Dusseault provided extensive editorial suggestions during the spring of 2009, most of which were incorporated. IDNA2008-Mapping] by Pete Resnick and Paul Hoffman. In addition, there were many specific contributions and helpful comments from those listed in the Acknowledgments section and others who have contributed to the development and use of the IDNA protocols. [RFC3490] Faltstrom, P., Hoffman, P., and A. Costello, "Internationalizing Domain Names in Applications (IDNA)", RFC 3490, March 2003.
[RFC3492] Costello, A., "Punycode: A Bootstring encoding of Unicode for Internationalized Domain Names in Applications (IDNA)", RFC 3492, March 2003. [RFC5890] Klensin, J., "Internationalized Domain Names for Applications (IDNA): Definitions and Document Framework", RFC 5890, August 2010. [RFC5891] Klensin, J., "Internationalized Domain Names in Applications (IDNA): Protocol", RFC 5891, August 2010. [RFC5892] Faltstrom, P., "The Unicode Code Points and Internationalized Domain Names for Applications (IDNA)", RFC 5892, August 2010. [RFC5893] Alvestrand, H. and C. Karp, "Right-to-Left Scripts for Internationalized Domain Names for Applications (IDNA)", RFC 5893, August 2010. [Unicode52] The Unicode Consortium. The Unicode Standard, Version 5.2.0, defined by: "The Unicode Standard, Version 5.2.0", (Mountain View, CA: The Unicode Consortium, 2009. ISBN 978-1-936213-00-9). <http://www.unicode.org/versions/Unicode5.2.0/>. [IDNA2008-Mapping] Resnick, P. and P. Hoffman, "Mapping Characters in Internationalized Domain Names for Applications (IDNA)", Work in Progress, April 2010. [RFC0952] Harrenstien, K., Stahl, M., and E. Feinler, "DoD Internet host table specification", RFC 952, October 1985. [RFC1034] Mockapetris, P., "Domain names - concepts and facilities", STD 13, RFC 1034, November 1987. [RFC1035] Mockapetris, P., "Domain names - implementation and specification", STD 13, RFC 1035, November 1987. [RFC1123] Braden, R., "Requirements for Internet Hosts - Application and Support", STD 3, RFC 1123, October 1989. [RFC2136] Vixie, P., Thomson, S., Rekhter, Y., and J. Bound, "Dynamic Updates in the Domain Name System (DNS UPDATE)", RFC 2136, April 1997.
[RFC2181] Elz, R. and R. Bush, "Clarifications to the DNS Specification", RFC 2181, July 1997. [RFC2277] Alvestrand, H., "IETF Policy on Character Sets and Languages", BCP 18, RFC 2277, January 1998. [RFC2671] Vixie, P., "Extension Mechanisms for DNS (EDNS0)", RFC 2671, August 1999. [RFC2782] Gulbrandsen, A., Vixie, P., and L. Esibov, "A DNS RR for specifying the location of services (DNS SRV)", RFC 2782, February 2000. [RFC3454] Hoffman, P. and M. Blanchet, "Preparation of Internationalized Strings ("stringprep")", RFC 3454, December 2002. [RFC3491] Hoffman, P. and M. Blanchet, "Nameprep: A Stringprep Profile for Internationalized Domain Names (IDN)", RFC 3491, March 2003. [RFC3743] Konishi, K., Huang, K., Qian, H., and Y. Ko, "Joint Engineering Team (JET) Guidelines for Internationalized Domain Names (IDN) Registration and Administration for Chinese, Japanese, and Korean", RFC 3743, April 2004. [RFC3987] Duerst, M. and M. Suignard, "Internationalized Resource Identifiers (IRIs)", RFC 3987, January 2005. [RFC4290] Klensin, J., "Suggested Practices for Registration of Internationalized Domain Names (IDN)", RFC 4290, December 2005. [RFC4343] Eastlake, D., "Domain Name System (DNS) Case Insensitivity Clarification", RFC 4343, January 2006. [RFC4690] Klensin, J., Faltstrom, P., Karp, C., and IAB, "Review and Recommendations for Internationalized Domain Names (IDNs)", RFC 4690, September 2006. [RFC4713] Lee, X., Mao, W., Chen, E., Hsu, N., and J. Klensin, "Registration and Administration Recommendations for Chinese Domain Names", RFC 4713, October 2006.
[Unicode-UAX31] The Unicode Consortium, "Unicode Standard Annex #31: Unicode Identifier and Pattern Syntax, Revision 11", September 2009, <http://www.unicode.org/reports/tr31/tr31-11.html>. [Unicode-UTS39] The Unicode Consortium, "Unicode Technical Standard #39: Unicode Security Mechanisms, Revision 2", August 2006, <http://www.unicode.org/reports/tr39/tr39-2.html>.