Libidn getaddrinfo-idn.txt -- Proposal for IDN support in POSIX getaddrinfo. Copyright (C) 2003-2016 Simon Josefsson See the end for copying conditions. Background ---------- Libidn is a package for internationalized string handling based on the Stringprep, Punycode and Internationalized Domain Names in Applications (IDNA) specifications. It can be used by applications directly by linking to it, as is done by, e.g., Gnus, KDE, and Mutt. Having each and every application link with and perform its own IDN handling is not a good idea. It bloats the code and makes things unnecessarily complex. Only few applications, such as web browsers and mail clients, will need to do this in the future, to provide good user interfaces for internationalization. See http://josefsson.org/libidn/ for more information. Alternative Approaches ---------------------- There are implementation that modify gethostbyname() to accept UTF-8 strings and perform the IDNA ToASCII operation within gethostbyname(). There are even implementations that assume gethostbyname (on the client host) perform no validation of the string and will send UTF-8 strings out to the DNS server, and perform the IDN-conversion on the DNS server. Some doubts can be raised whether this is an approach that is likely to be standardized. It also lack in functionality: it only provide black-box ToASCII functionality. The application cannot extract the output from the ToASCII operation. More important, there is no way to perform a ToUnicode operation that applications may want to use for display purposes. Furthermore, while the first can support locale specific character sets (e.g., ISO-8859-1), the second approach is bound to either guess the character set, or always use UTF-8. See also the thread rooted in posted to libc-alpha@sources.redhat.com on 08 Jan 2003. What I propose -------------- The getaddrinfo() API should have two new flags, AI_IDN and AI_CANONIDN. Roughly they correspond to IDNA ToASCII and IDNA ToUnicode, but there are several details. Note that strings are still 'char*', i.e. it does not use the "wide" character type, and that the encoding of non-ASCII strings are the current locale's character set (i.e., nl_langinfo(CODESET)). An application that uses AI_IDN signal to the getaddrinfo() implementation that the input host name may be non-ASCII and that the appropriate IDNA ToASCII steps should be carried out on the input, and the output from the ToASCII operation (if any) should be used in the lookup using the current resolver processing. An application that uses AI_CANONIDN signal to the getaddrinfo() implementation that the input host name should be put through the IDNA ToUnicode steps, and the output of that placed in the 'ai_canonname' field of the resulting structure. Normal resolver processing applies to the input string, of course. Consequently, an application that uses AI_IDN|AI_CANONIDN signal to the getaddrinfo() implementation that the input host name may be non-ASCII and should be put through the IDNA ToASCII steps before run through the resolver, and that the input string should also be run through the IDNA ToUnicode steps and the output of that placed in the 'ai_canonname' field. The semantics of AI_CANONNAME|AI_CANONIDN is that instead of running the ToUnicode IDNA steps on the input string, the canonical host name as returned by the resolver for the input string should be used in the ToUnicode IDNA step. Details ------- Four new flags has been proposed; AI_IDN_ALLOW_UNASSIGNED, AI_IDN_USE_STD3_ASCII_RULES for getaddrinfo, and NI_IDN_ALLOW_UNASSIGNED, NI_IDN_USE_STD3_ASCII_RULES for getnameinfo. The implementation is simple, if specified those flag will set the appropriate flag in the call to the IDNA functions. See the RFC for the meaning of those flags. Status ------ The AI_IDN flag has been implemented and shipped as a proof-of-concept patch for GNU Libc with GNU Libidn since January 2003. Binary libc packages with the patch exists for (at least) two GNU/Linux distributions. The AI_CANONIDN flag is not yet implemented. As of March 2004, Libidn has been integrated as an add-on in the GNU Libc CVS repository. The AI_CANONIDN flag has been implemented. The AllowUnassigned and UseSTD3ASCIIRules flags were added. Future ------ Allow non-ASCII in gethostname (and similar functions), if administrator has supplied, e.g., 'option idn' in /etc/resolv.conf? Feedback -------- This document is a work-in-progress and the details may change. Contact me at simon@josefsson.org to discuss changes. ---------------------------------------------------------------------- Permission is granted to anyone to make or distribute verbatim copies of this document, in any medium, provided that the copyright notice and permission notice are preserved, and that the distributor grants the recipient permission for further redistribution as permitted by this notice. Modified versions may not be made.